diff --git a/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-0e6e3fa0-317b-4a46-bdb4-752f8e86cd181758538014008-2025_09_22-12.47.12.63/source.csv b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-0e6e3fa0-317b-4a46-bdb4-752f8e86cd181758538014008-2025_09_22-12.47.12.63/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..a675e2aab207418b8ba15e64f8803f3aee9d1d3d --- /dev/null +++ b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-0e6e3fa0-317b-4a46-bdb4-752f8e86cd181758538014008-2025_09_22-12.47.12.63/source.csv @@ -0,0 +1,932 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,3,"genie.py",0,0,"from typing import Dict\n\nimport einops\nimport jax\nimport jax.numpy as jnp\nimport flax.nnx as nnx\nimport orbax.checkpoint as ocp\n\nfrom models.dynamics import DynamicsMaskGIT, DynamicsCausal\nfrom models.lam import LatentActionModel\nfrom models.tokenizer import TokenizerVQVAE\n\n\nclass Genie(nnx.Module):\n """"""Genie model""""""\n\n def __init__(\n self,\n in_dim: int,\n tokenizer_dim: int,\n tokenizer_ffn_dim: int,\n latent_patch_dim: int,\n num_patch_latents: int,\n patch_size: int,\n tokenizer_num_blocks: int,\n tokenizer_num_heads: int,\n lam_dim: int,\n lam_ffn_dim: int,\n latent_action_dim: int,\n num_latent_actions: int,\n lam_patch_size: int,\n lam_num_blocks: int,\n lam_num_heads: int,\n lam_co_train: bool,\n use_gt_actions: bool,\n dyna_type: str,\n dyna_dim: int,\n dyna_ffn_dim: int,\n dyna_num_blocks: int,\n dyna_num_heads: int,\n param_dtype: jnp.dtype,\n dtype: jnp.dtype,\n use_flash_attention: bool,\n decode: bool,\n rngs: nnx.Rngs,\n dropout: float = 0.0,\n mask_limit: float = 0.0,\n ):\n # --- Tokenizer ---\n self.in_dim = in_dim\n self.tokenizer_dim = tokenizer_dim\n self.tokenizer_ffn_dim = tokenizer_ffn_dim\n self.latent_patch_dim = latent_patch_dim\n self.num_patch_latents = num_patch_latents\n self.patch_size = patch_size\n self.tokenizer_num_blocks = tokenizer_num_blocks\n self.tokenizer_num_heads = tokenizer_num_heads\n # --- LAM ---\n self.lam_dim = lam_dim\n self.lam_ffn_dim = lam_ffn_dim\n self.latent_action_dim = latent_action_dim\n self.num_latent_actions = num_latent_actions\n self.lam_patch_size = lam_patch_size\n self.lam_num_blocks = lam_num_blocks\n self.lam_num_heads = lam_num_heads\n self.lam_co_train = lam_co_train\n self.use_gt_actions = use_gt_actions\n # --- Dynamics ---\n self.dyna_type = dyna_type\n self.dyna_dim = dyna_dim\n self.dyna_ffn_dim = dyna_ffn_dim\n self.dyna_num_blocks = dyna_num_blocks\n self.dyna_num_heads = dyna_num_heads\n self.param_dtype = param_dtype\n self.dtype = dtype\n self.use_flash_attention = use_flash_attention\n self.dropout = dropout\n self.mask_limit = mask_limit\n self.decode = decode\n\n self.tokenizer = TokenizerVQVAE(\n in_dim=self.in_dim,\n model_dim=self.tokenizer_dim,\n ffn_dim=self.tokenizer_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_patch_latents,\n patch_size=self.patch_size,\n num_blocks=self.tokenizer_num_blocks,\n num_heads=self.tokenizer_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n if self.use_gt_actions:\n self.action_embed = nnx.Embed(\n self.num_latent_actions, self.latent_action_dim, rngs=rngs\n )\n self.lam = None\n else:\n self.lam = LatentActionModel(\n in_dim=self.in_dim,\n model_dim=self.lam_dim,\n ffn_dim=self.lam_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_latent_actions,\n patch_size=self.lam_patch_size,\n num_blocks=self.lam_num_blocks,\n num_heads=self.lam_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n self.action_embed = None\n if self.dyna_type == ""maskgit"":\n self.dynamics = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n elif self.dyna_type == ""causal"":\n self.dynamics = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=decode,\n rngs=rngs,\n )\n else:\n raise ValueError(f""Invalid dynamics type: {self.dyna_type}"")\n\n def __call__(\n self,\n batch: Dict[str, jax.Array],\n training: bool = True,\n ) -> Dict[str, jax.Array]:\n videos_BTHWC = batch[""videos""]\n tokenizer_outputs = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_indices_BTN = tokenizer_outputs[""indices""]\n latent_actions_BTm11L = None\n action_embeddings_BTm11L = None\n if self.use_gt_actions:\n assert self.action_embed is not None\n action_indices_E = None\n # batch[""actions""]_t is the action taken after obs_t\n action_embeddings_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n action_embeddings_BTm11L = action_embeddings_BT1L[:, :-1]\n else:\n assert self.lam is not None\n lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n outputs = dict(\n video_tokens=jax.lax.stop_gradient(token_indices_BTN),\n latent_actions=(\n action_embeddings_BTm11L\n if self.use_gt_actions\n else latent_actions_BTm11L\n ),\n )\n outputs[""mask_rng""] = batch[""rng""]\n dyna_logits_BTNV, dyna_mask = self.dynamics(outputs, training)\n outputs[""token_logits""] = dyna_logits_BTNV\n outputs[""mask""] = dyna_mask\n mle_indices_BTN = jnp.argmax(outputs[""token_logits""], axis=-1)\n H, W = batch[""videos""].shape[2:4]\n outputs[""recon""] = self.tokenizer.decode(mle_indices_BTN, (H, W))\n if action_indices_E is not None:\n outputs[""lam_indices""] = action_indices_E\n return outputs\n\n def sample(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n maskgit_steps: int = 25,\n ) -> tuple[jax.Array, jax.Array]:\n if self.dyna_type == ""maskgit"":\n return self.sample_maskgit(\n batch, seq_len, maskgit_steps, temperature, sample_argmax\n )\n elif self.dyna_type == ""causal"":\n return self.sample_causal(batch, seq_len, temperature, sample_argmax)\n else:\n raise ValueError(f""Dynamics model type unknown: {self.dyna_type}"")\n\n def sample_maskgit(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n steps: int = 25,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n P: S * N\n """"""\n assert isinstance(self.dynamics, DynamicsMaskGIT)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n init_logits_BSNV = jnp.zeros(\n shape=(*token_idxs_BSN.shape, self.num_patch_latents)\n )\n if self.use_gt_actions:\n assert self.action_embed is not None\n latent_actions_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n latent_actions_BTm11L = latent_actions_BT1L[:, 1:]\n action_tokens_EL = latent_actions_BTm11L.reshape(-1, self.latent_action_dim)\n else:\n assert self.lam is not None\n latent_actions_E = batch[""latent_actions""]\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n # --- Extract submodule state ---\n dynamics_state = nnx.state(self.dynamics)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def maskgit_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, mask_BSN, action_tokens_EL = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_maskgit = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_maskgit, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_maskgit.patch_embed(token_idxs_BSN)\n mask_token_111M = dynamics_maskgit.mask_token.value\n mask_expanded_BSN1 = mask_BSN[..., None]\n vid_embed_BSNM = jnp.where(\n mask_expanded_BSN1, mask_token_111M, vid_embed_BSNM\n )\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_maskgit.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNM += act_embed_BS1M\n unmasked_ratio = jnp.cos(jnp.pi * (step + 1) / (steps * 2))\n step_temp = temperature * (1.0 - unmasked_ratio)\n final_logits_BSNV = dynamics_maskgit.transformer(vid_embed_BSNM) / step_temp\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_BSN = jnp.argmax(final_logits_BSNV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_BSN = jax.random.categorical(_rng, final_logits_BSNV)\n gather_fn = jax.vmap(jax.vmap(jax.vmap(lambda x, y: x[y])))\n final_token_probs_BSN = gather_fn(\n jax.nn.softmax(final_logits_BSNV), sampled_token_idxs_BSN\n )\n final_token_probs_BSN += ~mask_BSN\n # Update masked tokens and logits only\n token_idxs_BSN = jnp.where(mask_BSN, sampled_token_idxs_BSN, token_idxs_BSN)\n logits_BSNV = jnp.where(\n jnp.expand_dims(mask_BSN, -1), final_logits_BSNV, logits_BSNV\n )\n\n # --- Update mask ---\n num_unmasked_tokens = jnp.round(N * (1.0 - unmasked_ratio)).astype(int)\n final_token_probs_flat_BP = einops.rearrange(\n final_token_probs_BSN, ""b s n -> b (s n)""\n )\n idx_mask_P = (\n jnp.arange(final_token_probs_flat_BP.shape[-1])\n <= N - num_unmasked_tokens\n )\n sorted_idxs_BP = jnp.argsort(final_token_probs_flat_BP, axis=-1)\n mask_update_fn = jax.vmap(lambda msk, ids: msk.at[ids].set(idx_mask_P))\n mask_flat_BP = einops.rearrange(mask_BSN, ""b s n -> b (s n)"")\n new_mask_flat_BP = mask_update_fn(mask_flat_BP, sorted_idxs_BP)\n new_mask_BSN = einops.rearrange(new_mask_flat_BP, ""b (s n) -> b s n"", n=N)\n\n new_carry = (\n rng,\n token_idxs_BSN,\n logits_BSNV,\n new_mask_BSN,\n action_tokens_EL,\n )\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # Mask current frame (i.e., t == step_t)\n mask_S = jnp.arange(seq_len) == step_t\n mask_BSN = jnp.broadcast_to(mask_S[None, :, None], (B, seq_len, N)).astype(\n bool\n )\n masked_token_idxs_BSN = current_token_idxs_BSN * ~mask_BSN\n masked_logits_BSNV = current_logits_BSNV * jnp.expand_dims(~mask_BSN, -1)\n\n # --- Initialize and run MaskGIT loop ---\n init_carry_maskgit = (\n step_rng,\n masked_token_idxs_BSN,\n masked_logits_BSNV,\n mask_BSN,\n action_tokens_EL,\n )\n final_carry_maskgit = maskgit_step_fn(init_carry_maskgit, jnp.arange(steps))\n updated_token_idxs_BSN = final_carry_maskgit[1]\n updated_logits_BSNV = final_carry_maskgit[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, init_logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def sample_causal(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n """"""\n assert isinstance(self.dynamics, DynamicsCausal)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n logits_BSNV = jnp.zeros((*token_idxs_BSN.shape, self.num_patch_latents))\n dynamics_state = nnx.state(self.dynamics)\n\n if self.use_gt_actions:\n assert self.action_embed is not None\n latent_actions_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n latent_actions_BTm11L = latent_actions_BT1L[:, 1:]\n action_tokens_EL = latent_actions_BTm11L.reshape(-1, self.latent_action_dim)\n else:\n assert self.lam is not None\n latent_actions_E = batch[""latent_actions""]\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def causal_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step_n: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_causal = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=self.decode,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_causal, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_causal.patch_embed(token_idxs_BSN)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_causal.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNp1M = jnp.concatenate([act_embed_BS1M, vid_embed_BSNM], axis=2)\n final_logits_BTNp1V = (\n dynamics_causal.transformer(vid_embed_BSNp1M, (step_t, step_n))\n / temperature\n )\n final_logits_BV = final_logits_BTNp1V[:, step_t, step_n, :]\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_B = jnp.argmax(final_logits_BV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_B = jax.random.categorical(_rng, final_logits_BV)\n # Update next tokens only\n token_idxs_BSN = token_idxs_BSN.at[:, step_t, step_n].set(\n sampled_token_idxs_B\n )\n logits_BSNV = logits_BSNV.at[:, step_t, step_n].set(final_logits_BV)\n\n new_carry = (rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t)\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # --- Initialize and run causal loop ---\n init_carry_causal = (\n step_rng,\n current_token_idxs_BSN,\n current_logits_BSNV,\n action_tokens_EL,\n step_t,\n )\n final_carry_causal = causal_step_fn(init_carry_causal, jnp.arange(N))\n updated_token_idxs_BSN = final_carry_causal[1]\n updated_logits_BSNV = final_carry_causal[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def vq_encode(self, batch: Dict[str, jax.Array], training: bool) -> jax.Array:\n # --- Preprocess videos ---\n assert self.lam is not None\n video_BTHWC = batch[""videos""]\n lam_output: Dict[str, jax.Array] = self.lam.vq_encode(\n video_BTHWC, training=training\n )\n lam_indices_E = lam_output[""indices""]\n return lam_indices_E\n\n\n# FIXME (f.srambical): add conversion script for old checkpoints\ndef restore_genie_components(\n optimizer: nnx.Optimizer,\n sharding: jax.sharding.NamedSharding,\n rng: jax.Array,\n args,\n) -> nnx.Optimizer:\n """"""Restore pre-trained Genie components""""""\n rng_tokenizer, rng_lam = jax.random.split(rng)\n rngs_tokenizer = nnx.Rngs(rng_tokenizer)\n rngs_lam = nnx.Rngs(rng_lam)\n\n tx = optimizer.tx\n model = optimizer.model\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n step_format_fixed_length=6,\n )\n tokenizer_checkpoint_manager = ocp.CheckpointManager(\n directory=args.tokenizer_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_tokenizer = TokenizerVQVAE(\n in_dim=args.image_channels,\n model_dim=args.tokenizer_dim,\n ffn_dim=args.tokenizer_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n num_blocks=args.tokenizer_num_blocks,\n num_heads=args.tokenizer_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs_tokenizer,\n )\n dummy_tokenizer_optimizer = nnx.Optimizer(dummy_tokenizer, tx)\n dummy_tokenizer_optimizer_state = nnx.state(dummy_tokenizer_optimizer)\n abstract_sharded_tokenizer_optimizer_state = _create_abstract_sharded_pytree(\n dummy_tokenizer_optimizer_state, sharding\n )\n restored_tokenizer = tokenizer_checkpoint_manager.restore(\n step=tokenizer_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_tokenizer_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_tokenizer_optimizer.model, restored_tokenizer.model)\n model.tokenizer = dummy_tokenizer_optimizer.model\n tokenizer_checkpoint_manager.close()\n\n if args.lam_checkpoint:\n lam_checkpoint_manager = ocp.CheckpointManager(\n directory=args.lam_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_lam = LatentActionModel(\n in_dim=args.image_channels,\n model_dim=args.lam_dim,\n ffn_dim=args.lam_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_latent_actions,\n patch_size=args.lam_patch_size,\n num_blocks=args.lam_num_blocks,\n num_heads=args.lam_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs_lam,\n )\n dummy_lam_optimizer = nnx.Optimizer(dummy_lam, tx)\n dummy_lam_optimizer_state = nnx.state(dummy_lam_optimizer)\n abstract_sharded_lam_optimizer_state = _create_abstract_sharded_pytree(\n dummy_lam_optimizer_state, sharding\n )\n restored_lam_optimizer = lam_checkpoint_manager.restore(\n step=lam_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_lam_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_lam_optimizer.model, restored_lam_optimizer.model)\n model.lam = dummy_lam_optimizer.model\n # Remove the LAM decoder to save memory and avoid unnecessary computation.\n del model.lam.decoder\n lam_checkpoint_manager.close()\n\n # Reinitialize the optimizer states\n optimizer = nnx.Optimizer(model, tx)\n return optimizer\n\n\ndef _create_abstract_sharded_pytree(\n pytree_template: nnx.GraphState, sharding_spec: jax.sharding.NamedSharding\n) -> jax.Array:\n """"""Replaces arrays in a pytree with ShapeDtypeStructs having the given sharding.""""""\n\n def map_fn(leaf_template):\n if hasattr(leaf_template, ""shape"") and hasattr(leaf_template, ""dtype""):\n return jax.ShapeDtypeStruct(\n leaf_template.shape, leaf_template.dtype, sharding=sharding_spec\n )\n return leaf_template\n\n return jax.tree_util.tree_map(map_fn, pytree_template)\n",python,tab +2,2180,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"12:47:12 PM [info] Activating crowd-code\n12:47:12 PM [info] Recording started\n12:47:12 PM [info] Initializing git provider using file system watchers...\n",Log,tab +3,2284,"genie.py",0,0,"",python,tab +4,2950,"genie.py",6091,0,"",python,selection_command +5,3088,"genie.py",6055,0,"",python,selection_command +6,3336,"genie.py",6091,0,"",python,selection_command +7,6059,"genie.py",6055,0,"",python,selection_command +8,6191,"genie.py",6006,0,"",python,selection_command +9,6625,"genie.py",5924,0,"",python,selection_command +10,6625,"genie.py",5859,0,"",python,selection_command +11,7229,"genie.py",5924,0,"",python,selection_command +12,7396,"genie.py",5859,0,"",python,selection_command +13,14355,"TERMINAL",0,0,"",,terminal_command +14,18248,"genie.py",5924,0,"",python,selection_command +15,18521,"genie.py",6006,0,"",python,selection_command +16,18523,"genie.py",6055,0,"",python,selection_command +17,18561,"genie.py",6091,0,"",python,selection_command +18,18590,"genie.py",6139,0,"",python,selection_command +19,18622,"genie.py",6175,0,"",python,selection_command +20,18669,"genie.py",6215,0,"",python,selection_command +21,18692,"genie.py",6290,0,"",python,selection_command +22,18857,"genie.py",6334,0,"",python,selection_command +23,19012,"genie.py",6388,0,"",python,selection_command +24,19192,"genie.py",6437,0,"",python,selection_command +25,19394,"genie.py",6473,0,"",python,selection_command +26,20006,"genie.py",6509,0,"",python,selection_command +27,20545,"genie.py",6546,0,"",python,selection_command +28,20546,"genie.py",6570,0,"",python,selection_command +29,20547,"genie.py",6606,0,"",python,selection_command +30,20548,"genie.py",6666,0,"",python,selection_command +31,20704,"genie.py",6702,0,"",python,selection_command +32,20705,"genie.py",6743,0,"",python,selection_command +33,20706,"genie.py",6782,0,"",python,selection_command +34,20708,"genie.py",6804,0,"",python,selection_command +35,20709,"genie.py",6814,0,"",python,selection_command +36,20817,"genie.py",6804,0,"",python,selection_command +37,20887,"genie.py",6814,0,"",python,selection_command +38,21127,"genie.py",6850,0,"",python,selection_command +39,21163,"genie.py",6893,0,"",python,selection_command +40,21192,"genie.py",6964,0,"",python,selection_command +41,21222,"genie.py",7015,0,"",python,selection_command +42,21255,"genie.py",7051,0,"",python,selection_command +43,21397,"genie.py",7015,0,"",python,selection_command +44,21658,"genie.py",6964,0,"",python,selection_command +45,21688,"genie.py",6893,0,"",python,selection_command +46,21811,"genie.py",6850,0,"",python,selection_command +47,22072,"genie.py",6814,0,"",python,selection_command +48,22091,"genie.py",6804,0,"",python,selection_command +49,22126,"genie.py",6782,0,"",python,selection_command +50,22174,"genie.py",6743,0,"",python,selection_command +51,22187,"genie.py",6702,0,"",python,selection_command +52,22218,"genie.py",6666,0,"",python,selection_command +53,22262,"genie.py",6606,0,"",python,selection_command +54,22303,"genie.py",6570,0,"",python,selection_command +55,22419,"genie.py",6546,0,"",python,selection_command +56,22631,"genie.py",6509,0,"",python,selection_command +57,22827,"genie.py",6473,0,"",python,selection_command +58,23000,"genie.py",6437,0,"",python,selection_command +59,23181,"genie.py",6388,0,"",python,selection_command +60,23370,"genie.py",6334,0,"",python,selection_command +61,23533,"genie.py",6290,0,"",python,selection_command +62,23729,"genie.py",6334,0,"",python,selection_command +63,23878,"genie.py",6388,0,"",python,selection_command +64,24023,"genie.py",6437,0,"",python,selection_command +65,24293,"genie.py",6473,0,"",python,selection_command +66,24393,"genie.py",6509,0,"",python,selection_command +67,24574,"genie.py",6546,0,"",python,selection_command +68,24775,"genie.py",6570,0,"",python,selection_command +69,24968,"genie.py",6546,0,"",python,selection_command +70,25257,"genie.py",6509,0,"",python,selection_command +71,25258,"genie.py",6473,0,"",python,selection_command +72,25258,"genie.py",6437,0,"",python,selection_command +73,25478,"genie.py",6388,0,"",python,selection_command +74,26362,"genie.py",6334,0,"",python,selection_command +75,26603,"genie.py",6290,0,"",python,selection_command +76,26640,"genie.py",6215,0,"",python,selection_command +77,26656,"genie.py",6175,0,"",python,selection_command +78,26701,"genie.py",6139,0,"",python,selection_command +79,26745,"genie.py",6091,0,"",python,selection_command +80,26750,"genie.py",6055,0,"",python,selection_command +81,26793,"genie.py",6006,0,"",python,selection_command +82,26872,"genie.py",5924,0,"",python,selection_command +83,26973,"genie.py",5859,0,"",python,selection_command +84,27783,"genie.py",5825,65,"",python,content +85,27786,"genie.py",5837,0,"",python,selection_command +86,30039,"genie.py",5919,0,"",python,selection_command +87,30173,"genie.py",5990,0,"",python,selection_command +88,30341,"genie.py",6004,0,"",python,selection_command +89,34853,"genie.py",6074,0,"",python,selection_command +90,35011,"genie.py",6088,0,"",python,selection_command +91,35051,"genie.py",6128,0,"",python,selection_command +92,35082,"genie.py",6203,0,"",python,selection_command +93,35351,"genie.py",6247,0,"",python,selection_command +94,35506,"genie.py",6301,0,"",python,selection_command +95,36804,"genie.py",6247,0,"",python,selection_command +96,37066,"genie.py",6203,0,"",python,selection_command +97,37092,"genie.py",6128,0,"",python,selection_command +98,37125,"genie.py",6088,0,"",python,selection_command +99,37160,"genie.py",6074,0,"",python,selection_command +100,37282,"genie.py",6004,0,"",python,selection_command +101,54738,"input_pipeline/generate_atari_dataset.py",0,0,"# adapted from https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/rainbow_atari.py\nimport collections\nimport math\nimport os\nimport random\nimport time\nfrom collections import deque\nfrom dataclasses import dataclass\n\nimport gymnasium as gym\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport tyro\nfrom typing import Optional, Any\nfrom torch.utils.tensorboard.writer import SummaryWriter\n\nfrom cleanrl_utils.atari_wrappers import (\n ClipRewardEnv,\n EpisodicLifeEnv,\n FireResetEnv,\n MaxAndSkipEnv,\n NoopResetEnv,\n)\n\ntry:\n from utils import save_chunks # type: ignore\nexcept Exception: # pragma: no cover\n from input_pipeline.utils import save_chunks # type: ignore\nimport json\n\n\n@dataclass\nclass Args:\n exp_name: str = os.path.basename(__file__)[: -len("".py"")]\n """"""the name of this experiment""""""\n seed: int = 1\n """"""seed of the experiment""""""\n torch_deterministic: bool = True\n """"""if toggled, `torch.backends.cudnn.deterministic=False`""""""\n cuda: bool = True\n """"""if toggled, cuda will be enabled by default""""""\n track: bool = False\n """"""if toggled, this experiment will be tracked with Weights and Biases""""""\n wandb_project_name: str = ""cleanRL""\n """"""the wandb's project name""""""\n wandb_entity: Optional[str] = None\n """"""the entity (team) of wandb's project""""""\n capture_video: bool = False\n """"""whether to capture videos of the agent performances (check out `videos` folder)""""""\n save_model: bool = False\n """"""whether to save model into the `runs/{run_name}` folder""""""\n upload_model: bool = False\n """"""whether to upload the saved model to huggingface""""""\n hf_entity: str = """"\n """"""the user or org name of the model repository from the Hugging Face Hub""""""\n\n env_id: str = ""BreakoutNoFrameskip-v4""\n """"""the id of the environment""""""\n total_timesteps: int = 10000000\n """"""total timesteps of the experiments""""""\n learning_rate: float = 0.0000625\n """"""the learning rate of the optimizer""""""\n num_envs: int = 1\n """"""the number of parallel game environments""""""\n buffer_size: int = 1000000\n """"""the replay memory buffer size""""""\n gamma: float = 0.99\n """"""the discount factor gamma""""""\n tau: float = 1.0\n """"""the target network update rate""""""\n target_network_frequency: int = 8000\n """"""the timesteps it takes to update the target network""""""\n batch_size: int = 32\n """"""the batch size of sample from the reply memory""""""\n start_e: float = 1\n """"""the starting epsilon for exploration""""""\n end_e: float = 0.01\n """"""the ending epsilon for exploration""""""\n exploration_fraction: float = 0.10\n """"""the fraction of `total-timesteps` it takes from start-e to go end-e""""""\n learning_starts: int = 80000\n """"""timestep to start learning""""""\n train_frequency: int = 4\n """"""the frequency of training""""""\n n_step: int = 3\n """"""the number of steps to look ahead for n-step Q learning""""""\n prioritized_replay_alpha: float = 0.5\n """"""alpha parameter for prioritized replay buffer""""""\n prioritized_replay_beta: float = 0.4\n """"""beta parameter for prioritized replay buffer""""""\n prioritized_replay_eps: float = 1e-6\n """"""epsilon parameter for prioritized replay buffer""""""\n n_atoms: int = 51\n """"""the number of atoms""""""\n v_min: float = -10\n """"""the return lower bound""""""\n v_max: float = 10\n """"""the return upper bound""""""\n\n # Dataset capture\n capture_dataset: bool = True\n num_episodes_train: int = 10000\n num_episodes_val: int = 500\n num_episodes_test: int = 500\n output_dir: str = ""data/atari_episodes""\n min_episode_length: int = 1\n chunk_size: int = 160\n chunks_per_file: int = 100\n stop_on_complete: bool = True\n\n\ndef make_env(env_id, seed, idx, capture_video, run_name):\n def thunk():\n if capture_video and idx == 0:\n env = gym.make(env_id, render_mode=""rgb_array"")\n env = gym.wrappers.RecordVideo(env, f""videos/{run_name}"")\n else:\n env = gym.make(env_id)\n env = gym.wrappers.RecordEpisodeStatistics(env)\n\n env = NoopResetEnv(env, noop_max=30)\n env = MaxAndSkipEnv(env, skip=4)\n env = EpisodicLifeEnv(env)\n if ""FIRE"" in env.unwrapped.get_action_meanings():\n env = FireResetEnv(env)\n env = ClipRewardEnv(env)\n env = gym.wrappers.ResizeObservation(env, (84, 84))\n env = gym.wrappers.GrayScaleObservation(env)\n env = gym.wrappers.FrameStack(env, 4)\n\n env.action_space.seed(seed)\n return env\n\n return thunk\n\n\nclass NoisyLinear(nn.Module):\n def __init__(self, in_features, out_features, std_init=0.5):\n super().__init__()\n self.in_features = in_features\n self.out_features = out_features\n self.std_init = std_init\n\n self.weight_mu = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.weight_sigma = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.register_buffer(\n ""weight_epsilon"", torch.FloatTensor(out_features, in_features)\n )\n self.bias_mu = nn.Parameter(torch.FloatTensor(out_features))\n self.bias_sigma = nn.Parameter(torch.FloatTensor(out_features))\n self.register_buffer(""bias_epsilon"", torch.FloatTensor(out_features))\n # factorized gaussian noise\n self.reset_parameters()\n self.reset_noise()\n\n def reset_parameters(self):\n mu_range = 1 / math.sqrt(self.in_features)\n self.weight_mu.data.uniform_(-mu_range, mu_range)\n self.weight_sigma.data.fill_(self.std_init / math.sqrt(self.in_features))\n self.bias_mu.data.uniform_(-mu_range, mu_range)\n self.bias_sigma.data.fill_(self.std_init / math.sqrt(self.out_features))\n\n def reset_noise(self):\n self.weight_epsilon.normal_()\n self.bias_epsilon.normal_()\n\n def forward(self, input):\n if self.training:\n weight = self.weight_mu + self.weight_sigma * self.weight_epsilon\n bias = self.bias_mu + self.bias_sigma * self.bias_epsilon\n else:\n weight = self.weight_mu\n bias = self.bias_mu\n return F.linear(input, weight, bias)\n\n\n# ALGO LOGIC: initialize agent here:\nclass NoisyDuelingDistributionalNetwork(nn.Module):\n def __init__(self, env, n_atoms, v_min, v_max):\n super().__init__()\n self.n_atoms = n_atoms\n self.v_min = v_min\n self.v_max = v_max\n self.delta_z = (v_max - v_min) / (n_atoms - 1)\n self.n_actions = env.single_action_space.n\n self.register_buffer(""support"", torch.linspace(v_min, v_max, n_atoms))\n\n self.network = nn.Sequential(\n nn.Conv2d(4, 32, 8, stride=4),\n nn.ReLU(),\n nn.Conv2d(32, 64, 4, stride=2),\n nn.ReLU(),\n nn.Conv2d(64, 64, 3, stride=1),\n nn.ReLU(),\n nn.Flatten(),\n )\n conv_output_size = 3136\n\n self.value_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512), nn.ReLU(), NoisyLinear(512, n_atoms)\n )\n\n self.advantage_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512),\n nn.ReLU(),\n NoisyLinear(512, n_atoms * self.n_actions),\n )\n\n def forward(self, x):\n h = self.network(x / 255.0)\n value = self.value_head(h).view(-1, 1, self.n_atoms)\n advantage = self.advantage_head(h).view(-1, self.n_actions, self.n_atoms)\n q_atoms = value + advantage - advantage.mean(dim=1, keepdim=True)\n q_dist = F.softmax(q_atoms, dim=2)\n return q_dist\n\n def reset_noise(self):\n for layer in self.value_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n for layer in self.advantage_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n\n\nPrioritizedBatch = collections.namedtuple(\n ""PrioritizedBatch"",\n [\n ""observations"",\n ""actions"",\n ""rewards"",\n ""next_observations"",\n ""dones"",\n ""indices"",\n ""weights"",\n ],\n)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass SumSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.zeros(self.tree_size, dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = self.tree[parent * 2 + 1] + self.tree[parent * 2 + 2]\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def total(self):\n return self.tree[0]\n\n def retrieve(self, value):\n idx = 0\n while idx * 2 + 1 < self.tree_size:\n left = idx * 2 + 1\n right = left + 1\n if value <= self.tree[left]:\n idx = left\n else:\n value -= self.tree[left]\n idx = right\n return idx - (self.capacity - 1)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass MinSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.full(self.tree_size, float(""inf""), dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = np.minimum(\n self.tree[parent * 2 + 1], self.tree[parent * 2 + 2]\n )\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def min(self):\n return self.tree[0]\n\n\nclass PrioritizedReplayBuffer:\n def __init__(\n self, capacity, obs_shape, device, n_step, gamma, alpha=0.6, beta=0.4, eps=1e-6\n ):\n self.capacity = capacity\n self.device = device\n self.n_step = n_step\n self.gamma = gamma\n self.alpha = alpha\n self.beta = beta\n self.eps = eps\n\n self.buffer_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_next_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_actions = np.zeros(capacity, dtype=np.int64)\n self.buffer_rewards = np.zeros(capacity, dtype=np.float32)\n self.buffer_dones = np.zeros(capacity, dtype=np.bool_)\n\n self.pos = 0\n self.size = 0\n self.max_priority = 1.0\n\n self.sum_tree = SumSegmentTree(capacity)\n self.min_tree = MinSegmentTree(capacity)\n\n # For n-step returns\n self.n_step_buffer = deque(maxlen=n_step)\n\n def _get_n_step_info(self):\n reward = 0.0\n next_obs = self.n_step_buffer[-1][3]\n done = self.n_step_buffer[-1][4]\n\n for i in range(len(self.n_step_buffer)):\n reward += self.gamma**i * self.n_step_buffer[i][2]\n if self.n_step_buffer[i][4]:\n next_obs = self.n_step_buffer[i][3]\n done = True\n break\n return reward, next_obs, done\n\n def add(self, obs, action, reward, next_obs, done):\n self.n_step_buffer.append((obs, action, reward, next_obs, done))\n\n if len(self.n_step_buffer) < self.n_step:\n return\n\n reward, next_obs, done = self._get_n_step_info()\n obs = self.n_step_buffer[0][0]\n action = self.n_step_buffer[0][1]\n\n idx = self.pos\n self.buffer_obs[idx] = obs\n self.buffer_next_obs[idx] = next_obs\n self.buffer_actions[idx] = action\n self.buffer_rewards[idx] = reward\n self.buffer_dones[idx] = done\n\n priority = self.max_priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n self.pos = (self.pos + 1) % self.capacity\n self.size = min(self.size + 1, self.capacity)\n\n if done:\n self.n_step_buffer.clear()\n\n def sample(self, batch_size):\n indices = []\n p_total = self.sum_tree.total()\n segment = p_total / batch_size\n\n for i in range(batch_size):\n a = segment * i\n b = segment * (i + 1)\n upperbound = np.random.uniform(a, b)\n idx = self.sum_tree.retrieve(upperbound)\n indices.append(idx)\n\n samples = {\n ""observations"": torch.from_numpy(self.buffer_obs[indices]).to(self.device),\n ""actions"": torch.from_numpy(self.buffer_actions[indices])\n .to(self.device)\n .unsqueeze(1),\n ""rewards"": torch.from_numpy(self.buffer_rewards[indices])\n .to(self.device)\n .unsqueeze(1),\n ""next_observations"": torch.from_numpy(self.buffer_next_obs[indices]).to(\n self.device\n ),\n ""dones"": torch.from_numpy(self.buffer_dones[indices])\n .to(self.device)\n .unsqueeze(1),\n }\n\n probs = np.array(\n [self.sum_tree.tree[idx + self.capacity - 1] for idx in indices]\n )\n weights = (self.size * probs / p_total) ** -self.beta\n weights = weights / weights.max()\n samples[""weights""] = torch.from_numpy(weights).to(self.device).unsqueeze(1)\n samples[""indices""] = indices\n\n return PrioritizedBatch(**samples)\n\n def update_priorities(self, indices, priorities):\n priorities = np.abs(priorities) + self.eps\n self.max_priority = max(self.max_priority, priorities.max())\n\n for idx, priority in zip(indices, priorities):\n priority = priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n\nif __name__ == ""__main__"":\n args = tyro.cli(Args)\n assert args.num_envs == 1, ""vectorized envs are not supported at the moment""\n run_name = f""{args.env_id}__{args.exp_name}__{args.seed}__{int(time.time())}""\n if args.track:\n import wandb\n\n wandb.init(\n project=args.wandb_project_name,\n entity=args.wandb_entity,\n sync_tensorboard=True,\n config=vars(args),\n name=run_name,\n monitor_gym=True,\n save_code=True,\n )\n writer = SummaryWriter(f""runs/{run_name}"")\n writer.add_text(\n ""hyperparameters"",\n ""|param|value|\n|-|-|\n%s""\n % (""\n"".join([f""|{key}|{value}|"" for key, value in vars(args).items()])),\n )\n\n # TRY NOT TO MODIFY: seeding\n random.seed(args.seed)\n np.random.seed(args.seed)\n torch.manual_seed(args.seed)\n torch.backends.cudnn.deterministic = args.torch_deterministic\n\n device = torch.device(""cuda"" if torch.cuda.is_available() and args.cuda else ""cpu"")\n\n # env setup\n envs = gym.vector.SyncVectorEnv(\n [\n make_env(args.env_id, args.seed + i, i, args.capture_video, run_name)\n for i in range(args.num_envs)\n ]\n )\n assert isinstance(\n envs.single_action_space, gym.spaces.Discrete\n ), ""only discrete action space is supported""\n\n q_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n optimizer = optim.Adam(q_network.parameters(), lr=args.learning_rate, eps=1.5e-4)\n target_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n target_network.load_state_dict(q_network.state_dict())\n\n rb = PrioritizedReplayBuffer(\n args.buffer_size,\n envs.single_observation_space.shape,\n device,\n args.n_step,\n args.gamma,\n args.prioritized_replay_alpha,\n args.prioritized_replay_beta,\n args.prioritized_replay_eps,\n )\n\n # dataset capture state\n split_targets = {\n ""train"": args.num_episodes_train,\n ""val"": args.num_episodes_val,\n ""test"": args.num_episodes_test,\n }\n # Determine splits to run (order: train -> val -> test)\n splits_in_order = [s for s in [""train"", ""val"", ""test""] if split_targets[s] > 0]\n\n episodes_captured_per_split: dict[str, int] = {\n s: 0 for s in [""train"", ""val"", ""test""]\n }\n file_idx_by_split: dict[str, int] = {s: 0 for s in [""train"", ""val"", ""test""]}\n episode_metadata_by_split: dict[str, list[dict]] = {\n s: [] for s in [""train"", ""val"", ""test""]\n }\n\n obs_chunks: list[np.ndarray] = []\n act_chunks: list[np.ndarray] = []\n\n current_split_idx = 0\n current_split = splits_in_order[0]\n split_dir = os.path.join(args.output_dir, current_split)\n if args.capture_dataset:\n os.makedirs(split_dir, exist_ok=True)\n\n start_time = time.time()\n\n # TRY NOT TO MODIFY: start the game\n obs, _ = envs.reset(seed=args.seed)\n observations_seq: list[np.ndarray] = []\n actions_seq: list[np.ndarray] = []\n for global_step in range(args.total_timesteps):\n # anneal PER beta to 1\n rb.beta = min(\n 1.0,\n args.prioritized_replay_beta\n + global_step * (1.0 - args.prioritized_replay_beta) / args.total_timesteps,\n )\n\n # ALGO LOGIC: put action logic here\n with torch.no_grad():\n q_dist = q_network(torch.Tensor(obs).to(device))\n q_values = torch.sum(q_dist * q_network.support, dim=2)\n actions = torch.argmax(q_values, dim=1).cpu().numpy()\n\n # TRY NOT TO MODIFY: execute the game and log data.\n next_obs, rewards, terminations, truncations, infos = envs.step(actions)\n\n if args.capture_dataset:\n observations_seq.append(next_obs.astype(np.uint8))\n actions_seq.append(actions.astype(np.int64))\n\n if ""final_info"" in infos:\n for info in infos[""final_info""]:\n if info and ""episode"" in info:\n print(\n f""global_step={global_step}, episodic_return={info['episode']['r']}""\n )\n writer.add_scalar(\n ""charts/episodic_return"", info[""episode""][""r""], global_step\n )\n writer.add_scalar(\n ""charts/episodic_length"", info[""episode""][""l""], global_step\n )\n\n continue_capturing_multi = any(\n episodes_captured_per_split[s] < split_targets[s]\n for s in splits_in_order\n )\n if args.capture_dataset and continue_capturing_multi:\n current_len = len(observations_seq)\n if current_len >= args.min_episode_length:\n frames = np.concatenate(observations_seq, axis=0).astype(\n np.uint8\n )\n acts = np.concatenate(actions_seq, axis=0).astype(np.int64)\n\n episode_obs_chunks = []\n episode_act_chunks = []\n start_idx = 0\n while start_idx < current_len:\n end_idx = min(start_idx + args.chunk_size, current_len)\n if end_idx - start_idx < args.chunk_size:\n print(\n f""Warning: Inconsistent chunk_sizes. Episode has {current_len} frames, ""\n f""which is smaller than the requested chunk_size: {args.chunk_size}. ""\n ""This might lead to performance degradation during training.""\n )\n episode_obs_chunks.append(frames[start_idx:end_idx])\n episode_act_chunks.append(acts[start_idx:end_idx])\n start_idx = end_idx\n\n obs_chunks_data = [\n seq.astype(np.uint8) for seq in episode_obs_chunks\n ]\n act_chunks_data = [act for act in episode_act_chunks]\n obs_chunks.extend(obs_chunks_data)\n act_chunks.extend(act_chunks_data)\n\n # Save to the active split\n ep_metadata, obs_chunks, next_file_idx, act_chunks = (\n save_chunks(\n obs_chunks,\n file_idx_by_split[current_split],\n args.chunks_per_file,\n split_dir,\n act_chunks,\n )\n )\n file_idx_by_split[current_split] = next_file_idx\n episode_metadata_by_split[current_split].extend(ep_metadata)\n\n episodes_captured_per_split[current_split] += 1\n\n if (\n episodes_captured_per_split[current_split]\n >= split_targets[current_split]\n ):\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks before switching split '"",\n {current_split},\n ""' for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n obs_chunks = []\n act_chunks = []\n if current_split_idx + 1 < len(splits_in_order):\n current_split_idx += 1\n current_split = splits_in_order[current_split_idx]\n split_dir = os.path.join(\n args.output_dir, current_split\n )\n os.makedirs(split_dir, exist_ok=True)\n else:\n print(\n f""Episode too short ({current_len}), skipping capture...""\n )\n\n observations_seq = []\n actions_seq = []\n\n # TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`\n real_next_obs = next_obs.copy()\n for idx, trunc in enumerate(truncations):\n if trunc:\n real_next_obs[idx] = infos[""final_observation""][idx]\n rb.add(obs, actions, rewards, real_next_obs, terminations)\n\n # TRY NOT TO MODIFY: CRUCIAL step easy to overlook\n obs = next_obs\n\n # ALGO LOGIC: training.\n if global_step > args.learning_starts:\n if global_step % args.train_frequency == 0:\n # reset the noise for both networks\n q_network.reset_noise()\n target_network.reset_noise()\n data = rb.sample(args.batch_size)\n\n with torch.no_grad():\n next_dist = target_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n support = target_network.support # [n_atoms]\n next_q_values = torch.sum(\n next_dist * support, dim=2\n ) # [B, num_actions]\n\n # double q-learning\n next_dist_online = q_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n next_q_online = torch.sum(\n next_dist_online * support, dim=2\n ) # [B, num_actions]\n best_actions = torch.argmax(next_q_online, dim=1) # [B]\n next_pmfs = next_dist[\n torch.arange(args.batch_size), best_actions\n ] # [B, n_atoms]\n\n # compute the n-step Bellman update.\n gamma_n = args.gamma**args.n_step\n next_atoms = data.rewards + gamma_n * support * (\n 1 - data.dones.float()\n )\n tz = next_atoms.clamp(q_network.v_min, q_network.v_max)\n\n # projection\n delta_z = q_network.delta_z\n b = (tz - q_network.v_min) / delta_z # shape: [B, n_atoms]\n l = b.floor().clamp(0, args.n_atoms - 1)\n u = b.ceil().clamp(0, args.n_atoms - 1)\n\n # (l == u).float() handles the case where bj is exactly an integer\n # example bj = 1, then the upper ceiling should be uj= 2, and lj= 1\n d_m_l = (\n u.float() + (l == b).float() - b\n ) * next_pmfs # [B, n_atoms]\n d_m_u = (b - l) * next_pmfs # [B, n_atoms]\n\n target_pmfs = torch.zeros_like(next_pmfs)\n for i in range(target_pmfs.size(0)):\n target_pmfs[i].index_add_(0, l[i].long(), d_m_l[i])\n target_pmfs[i].index_add_(0, u[i].long(), d_m_u[i])\n\n dist = q_network(data.observations) # [B, num_actions, n_atoms]\n pred_dist = dist.gather(\n 1, data.actions.unsqueeze(-1).expand(-1, -1, args.n_atoms)\n ).squeeze(1)\n log_pred = torch.log(pred_dist.clamp(min=1e-5, max=1 - 1e-5))\n\n loss_per_sample = -(target_pmfs * log_pred).sum(dim=1)\n loss = (loss_per_sample * data.weights.squeeze()).mean()\n\n # update priorities\n new_priorities = loss_per_sample.detach().cpu().numpy()\n rb.update_priorities(data.indices, new_priorities)\n\n if global_step % 100 == 0:\n writer.add_scalar(""losses/td_loss"", loss.item(), global_step)\n q_values = (pred_dist * q_network.support).sum(dim=1) # [B]\n writer.add_scalar(\n ""losses/q_values"", q_values.mean().item(), global_step\n )\n sps = int(global_step / (time.time() - start_time))\n print(""SPS:"", sps)\n writer.add_scalar(""charts/SPS"", sps, global_step)\n writer.add_scalar(""charts/beta"", rb.beta, global_step)\n\n # optimize the model\n optimizer.zero_grad()\n loss.backward()\n optimizer.step()\n\n # update target network\n if global_step % args.target_network_frequency == 0:\n for target_param, param in zip(\n target_network.parameters(), q_network.parameters()\n ):\n target_param.data.copy_(\n args.tau * param.data + (1.0 - args.tau) * target_param.data\n )\n\n # optional early stop on dataset completion\n if args.capture_dataset and args.stop_on_complete:\n all_done = (\n all(\n episodes_captured_per_split[s] >= split_targets[s]\n for s in splits_in_order\n )\n and len(splits_in_order) > 0\n )\n if all_done:\n break\n\n envs.close()\n writer.close()\n\n # write metadata for dataset\n if args.capture_dataset:\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n\n os.makedirs(args.output_dir, exist_ok=True)\n metadata_path = os.path.join(args.output_dir, ""metadata.json"")\n if os.path.exists(metadata_path):\n try:\n with open(metadata_path, ""r"") as f:\n metadata = json.load(f)\n except Exception:\n metadata = {}\n else:\n metadata = {}\n\n metadata.setdefault(""env"", args.env_id)\n metadata.setdefault(""num_actions"", int(envs.single_action_space.n))\n for split in [""train"", ""val"", ""test""]:\n metadata.setdefault(f""num_episodes_{split}"", 0)\n metadata.setdefault(f""avg_episode_len_{split}"", 0.0)\n metadata.setdefault(f""episode_metadata_{split}"", [])\n\n for split_key in splits_in_order:\n ep_meta_list = episode_metadata_by_split[split_key]\n if ep_meta_list:\n metadata[f""episode_metadata_{split_key}""].extend(ep_meta_list)\n metadata[f""num_episodes_{split_key}""] = len(\n metadata[f""episode_metadata_{split_key}""]\n )\n metadata[f""avg_episode_len_{split_key}""] = float(\n np.mean(\n [\n ep[""avg_seq_len""]\n for ep in metadata[f""episode_metadata_{split_key}""]\n ]\n )\n )\n\n with open(metadata_path, ""w"") as f:\n json.dump(metadata, f)\n",python,tab +102,57516,"input_pipeline/generate_atari_dataset.py",947,0,"",python,selection_keyboard +103,57718,"input_pipeline/generate_atari_dataset.py",2588,0,"",python,selection_keyboard +104,58240,"input_pipeline/generate_atari_dataset.py",4011,0,"",python,selection_keyboard +105,58674,"input_pipeline/generate_atari_dataset.py",5506,0,"",python,selection_keyboard +106,59886,"input_pipeline/generate_atari_dataset.py",6964,0,"",python,selection_keyboard +107,60071,"input_pipeline/generate_atari_dataset.py",8133,0,"",python,selection_keyboard +108,61944,"input_pipeline/generate_atari_dataset.py",15078,0,"",python,selection_command +109,62561,"input_pipeline/generate_atari_dataset.py",15974,0,"",python,selection_command +110,63542,"input_pipeline/generate_atari_dataset.py",16294,0,"",python,selection_command +111,64117,"input_pipeline/generate_atari_dataset.py",16794,0,"",python,selection_command +112,64450,"input_pipeline/generate_atari_dataset.py",17743,0,"",python,selection_command +113,65853,"input_pipeline/generate_atari_dataset.py",16794,0,"",python,selection_command +114,67087,"input_pipeline/generate_atari_dataset.py",17743,0,"",python,selection_command +115,69704,"input_pipeline/generate_atari_dataset.py",17776,0,"",python,selection_command +116,69861,"input_pipeline/generate_atari_dataset.py",17743,0,"",python,selection_command +117,70122,"input_pipeline/generate_atari_dataset.py",17725,0,"",python,selection_command +118,70183,"input_pipeline/generate_atari_dataset.py",17661,0,"",python,selection_command +119,70184,"input_pipeline/generate_atari_dataset.py",17601,0,"",python,selection_command +120,70219,"input_pipeline/generate_atari_dataset.py",17583,0,"",python,selection_command +121,70246,"input_pipeline/generate_atari_dataset.py",17534,0,"",python,selection_command +122,70281,"input_pipeline/generate_atari_dataset.py",17466,0,"",python,selection_command +123,70317,"input_pipeline/generate_atari_dataset.py",17405,0,"",python,selection_command +124,70388,"input_pipeline/generate_atari_dataset.py",17375,0,"",python,selection_command +125,70389,"input_pipeline/generate_atari_dataset.py",17331,0,"",python,selection_command +126,70424,"input_pipeline/generate_atari_dataset.py",17313,0,"",python,selection_command +127,70453,"input_pipeline/generate_atari_dataset.py",17311,0,"",python,selection_command +128,70492,"input_pipeline/generate_atari_dataset.py",17231,0,"",python,selection_command +129,70520,"input_pipeline/generate_atari_dataset.py",17190,0,"",python,selection_command +130,70550,"input_pipeline/generate_atari_dataset.py",17171,0,"",python,selection_command +131,70576,"input_pipeline/generate_atari_dataset.py",17150,0,"",python,selection_command +132,70621,"input_pipeline/generate_atari_dataset.py",17119,0,"",python,selection_command +133,70652,"input_pipeline/generate_atari_dataset.py",17067,0,"",python,selection_command +134,70694,"input_pipeline/generate_atari_dataset.py",17028,0,"",python,selection_command +135,70721,"input_pipeline/generate_atari_dataset.py",16984,0,"",python,selection_command +136,70770,"input_pipeline/generate_atari_dataset.py",16944,0,"",python,selection_command +137,70860,"input_pipeline/generate_atari_dataset.py",16984,0,"",python,selection_command +138,71121,"input_pipeline/generate_atari_dataset.py",17028,0,"",python,selection_command +139,71132,"input_pipeline/generate_atari_dataset.py",17067,0,"",python,selection_command +140,71170,"input_pipeline/generate_atari_dataset.py",17119,0,"",python,selection_command +141,71193,"input_pipeline/generate_atari_dataset.py",17150,0,"",python,selection_command +142,71235,"input_pipeline/generate_atari_dataset.py",17171,0,"",python,selection_command +143,71263,"input_pipeline/generate_atari_dataset.py",17190,0,"",python,selection_command +144,71296,"input_pipeline/generate_atari_dataset.py",17231,0,"",python,selection_command +145,71325,"input_pipeline/generate_atari_dataset.py",17311,0,"",python,selection_command +146,71362,"input_pipeline/generate_atari_dataset.py",17313,0,"",python,selection_command +147,71387,"input_pipeline/generate_atari_dataset.py",17331,0,"",python,selection_command +148,71441,"input_pipeline/generate_atari_dataset.py",17375,0,"",python,selection_command +149,71456,"input_pipeline/generate_atari_dataset.py",17405,0,"",python,selection_command +150,71502,"input_pipeline/generate_atari_dataset.py",17466,0,"",python,selection_command +151,71547,"input_pipeline/generate_atari_dataset.py",17534,0,"",python,selection_command +152,71583,"input_pipeline/generate_atari_dataset.py",17583,0,"",python,selection_command +153,71635,"input_pipeline/generate_atari_dataset.py",17601,0,"",python,selection_command +154,71635,"input_pipeline/generate_atari_dataset.py",17661,0,"",python,selection_command +155,71747,"input_pipeline/generate_atari_dataset.py",17725,0,"",python,selection_command +156,72195,"input_pipeline/generate_atari_dataset.py",17743,0,"",python,selection_command +157,72196,"input_pipeline/generate_atari_dataset.py",17776,0,"",python,selection_command +158,78697,"input_pipeline/generate_atari_dataset.py",17743,0,"",python,selection_command +159,78877,"input_pipeline/generate_atari_dataset.py",17725,0,"",python,selection_command +160,79002,"input_pipeline/generate_atari_dataset.py",17661,0,"",python,selection_command +161,79162,"input_pipeline/generate_atari_dataset.py",17601,0,"",python,selection_command +162,79469,"input_pipeline/generate_atari_dataset.py",17661,0,"",python,selection_command +163,80003,"input_pipeline/generate_atari_dataset.py",17660,0,"",python,selection_command +164,80246,"input_pipeline/generate_atari_dataset.py",17652,0,"",python,selection_command +165,81554,"input_pipeline/generate_atari_dataset.py",17660,0,"",python,selection_command +166,81785,"input_pipeline/generate_atari_dataset.py",17662,0,"",python,selection_command +167,81800,"input_pipeline/generate_atari_dataset.py",17669,0,"",python,selection_command +168,81843,"input_pipeline/generate_atari_dataset.py",17671,0,"",python,selection_command +169,81880,"input_pipeline/generate_atari_dataset.py",17683,0,"",python,selection_command +170,81900,"input_pipeline/generate_atari_dataset.py",17685,0,"",python,selection_command +171,81954,"input_pipeline/generate_atari_dataset.py",17696,0,"",python,selection_command +172,81968,"input_pipeline/generate_atari_dataset.py",17698,0,"",python,selection_command +173,82029,"input_pipeline/generate_atari_dataset.py",17704,0,"",python,selection_command +174,82037,"input_pipeline/generate_atari_dataset.py",17706,0,"",python,selection_command +175,82070,"input_pipeline/generate_atari_dataset.py",17710,0,"",python,selection_command +176,82086,"input_pipeline/generate_atari_dataset.py",17711,0,"",python,selection_command +177,82128,"input_pipeline/generate_atari_dataset.py",17715,0,"",python,selection_command +178,82171,"input_pipeline/generate_atari_dataset.py",17716,0,"",python,selection_command +179,82445,"input_pipeline/generate_atari_dataset.py",17715,0,"",python,selection_command +180,82649,"input_pipeline/generate_atari_dataset.py",17711,0,"",python,selection_command +181,85761,"input_pipeline/generate_atari_dataset.py",17710,0,"",python,selection_command +182,85982,"input_pipeline/generate_atari_dataset.py",17706,0,"",python,selection_command +183,93334,"input_pipeline/generate_atari_dataset.py",17725,0,"",python,selection_command +184,93517,"input_pipeline/generate_atari_dataset.py",17757,0,"",python,selection_command +185,93654,"input_pipeline/generate_atari_dataset.py",17820,0,"",python,selection_command +186,94416,"input_pipeline/generate_atari_dataset.py",17759,62," observations_seq.append(next_obs.astype(np.uint8))",python,selection_command +187,110098,"TERMINAL",0,0,"",,terminal_command +188,130686,"utils/train_utils.py",0,0,"import jax\nimport optax\nimport operator\n\n\ndef get_lr_schedule(\n lr_schedule: str,\n init_lr: float,\n max_lr: float,\n decay_end: float,\n total_steps: int,\n warmup_steps: int,\n wsd_decay_steps: int,\n) -> optax.Schedule:\n supported_schedules = [""wsd"", ""cos""]\n if lr_schedule == ""cos"":\n assert (\n warmup_steps <= total_steps\n ), ""Warmup steps can't be greater than total steps.""\n return optax.warmup_cosine_decay_schedule(\n init_value=init_lr,\n peak_value=max_lr,\n warmup_steps=warmup_steps,\n decay_steps=total_steps, # Note: decay_steps includes the warmup steps, so we need to pass total value\n end_value=decay_end,\n )\n elif lr_schedule == ""wsd"":\n assert (\n warmup_steps + wsd_decay_steps <= total_steps\n ), ""Warmup and decay period is longer than total steps.""\n schedules = [\n optax.linear_schedule(\n init_value=init_lr, end_value=max_lr, transition_steps=warmup_steps\n ),\n optax.constant_schedule(value=max_lr),\n optax.linear_schedule(\n init_value=max_lr, end_value=decay_end, transition_steps=wsd_decay_steps\n ),\n ]\n boundaries = [warmup_steps, total_steps - wsd_decay_steps]\n return optax.join_schedules(schedules, boundaries)\n else:\n raise ValueError(\n f""Learning rate schedule not supported. Please use one of {supported_schedules}""\n )\n\n\ndef _count_component(component_params):\n """"""Count total parameters in a component.""""""\n params_sizes = jax.tree.map(jax.numpy.size, component_params)\n total_parameters = jax.tree.reduce(operator.add, params_sizes)\n return total_parameters\n\n\ndef count_parameters_by_component(params):\n """"""Count parameters for each component of the model.\n\n Args:\n params: Model parameters from nnx.split(model, nnx.Param, ...)\n\n Returns:\n Dictionary with parameter counts for each component\n """"""\n component_names = list(params.keys())\n print(f""Counting all components: {component_names}"")\n\n counts = {}\n total_params = 0\n\n for name in component_names:\n component_params = params[name]\n count = _count_component(component_params)\n counts[name] = count\n total_params += count\n\n counts[""total""] = total_params\n return counts\n\n\ndef bytes_to_gb(num_bytes):\n return num_bytes / (1024**3)\n\n\ndef print_compiled_memory_stats(compiled_stats):\n """"""from: https://github.com/AI-Hypercomputer/maxtext/blob/b18829fbaa48aec7ac350a03e62248e24c6a76b2/MaxText/max_utils.py#L739""""""\n output_gb = bytes_to_gb(compiled_stats.output_size_in_bytes)\n temp_gb = bytes_to_gb(compiled_stats.temp_size_in_bytes)\n argument_gb = bytes_to_gb(compiled_stats.argument_size_in_bytes)\n alias_gb = bytes_to_gb(compiled_stats.alias_size_in_bytes)\n host_temp_gb = bytes_to_gb(compiled_stats.host_temp_size_in_bytes)\n total_gb = output_gb + temp_gb + argument_gb - alias_gb\n print(\n f""Total memory size: {total_gb:.1f} GB, Output size: {output_gb:.1f} GB, Temp size: {temp_gb:.1f} GB, ""\n f""Argument size: {argument_gb:.1f} GB, Host temp size: {host_temp_gb:.1f} GB.""\n )\n\n\ndef print_compiled_cost_analysis(cost_stats):\n flops = float(cost_stats.get(""flops"", 0.0))\n bytes_accessed = float(cost_stats.get(""bytes accessed"", 0.0))\n gb = bytes_to_gb(bytes_accessed) if bytes_accessed else 0.0\n intensity = (flops / bytes_accessed) if bytes_accessed else float(""nan"")\n print(\n f""FLOPs: {flops:.3e}, Bytes: {bytes_accessed:.3e} ({gb:.1f} GB), ""\n f""Intensity: {intensity:.1f} FLOPs/byte""\n )\n\n\ndef print_mem_stats(label: str):\n """"""from: https://github.com/AI-Hypercomputer/maxtext/blob/7898576359bacde81be25cb3038e348aac1f943b/MaxText/max_utils.py#L713""""""\n print(f""\nMemstats: {label}:"")\n try:\n for d in jax.local_devices():\n stats = d.memory_stats()\n used = round(stats[""bytes_in_use""] / 2**30, 2)\n limit = round(stats[""bytes_limit""] / 2**30, 2)\n print(f""\tUsing (GB) {used} / {limit} ({used/limit:%}) on {d}"")\n except (RuntimeError, KeyError, TypeError) as ex:\n print(f""\tMemstats unavailable, error: {ex}"")\n",python,tab +189,131743,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +190,132034,"input_pipeline/generate_atari_dataset.py",17759,12," ",python,content +191,132825,"input_pipeline/generate_atari_dataset.py",17820,0,"",python,selection_command +192,161836,"input_pipeline/generate_atari_dataset.py",17771,4,"",python,content +193,161865,"input_pipeline/generate_atari_dataset.py",17820,0,"",python,selection_command +194,163416,"input_pipeline/generate_atari_dataset.py",17759,62," observations_seq.append(next_obs.astype(np.uint8))",python,selection_command +195,913224,"input_pipeline/generate_atari_dataset.py",20503,539," (\n ep_metadata,\n file_idx_by_split[current_split],\n obs_chunks,\n act_chunks,\n ) = save_chunks(\n file_idx_by_split[current_split],\n args.chunks_per_file,\n split_dir,\n obs_chunks,\n act_chunks,\n )",python,content +196,913224,"input_pipeline/generate_atari_dataset.py",17759,62," observations_seq.append(obs.astype(np.uint8))",python,content +197,935149,"input_pipeline/generate_atari_dataset.py",20497,0,"",python,selection_mouse +198,935152,"input_pipeline/generate_atari_dataset.py",20496,0,"",python,selection_command +199,942678,"input_pipeline/generate_atari_dataset.py",20567,0,"",python,selection_mouse +200,943461,"input_pipeline/generate_atari_dataset.py",20612,0,"",python,selection_mouse +201,946031,"input_pipeline/generate_atari_dataset.py",20767,0,"",python,selection_mouse +202,947175,"input_pipeline/generate_atari_dataset.py",20975,0,"",python,selection_mouse +203,947679,"input_pipeline/generate_atari_dataset.py",21021,0,"",python,selection_mouse +204,948383,"input_pipeline/generate_atari_dataset.py",21052,0,"",python,selection_mouse +205,948384,"input_pipeline/generate_atari_dataset.py",21051,0,"",python,selection_command +206,969059,"input_pipeline/generate_atari_dataset.py",20814,0,"",python,selection_mouse +207,971058,"input_pipeline/generate_atari_dataset.py",20881,0,"",python,selection_mouse +208,971680,"input_pipeline/generate_atari_dataset.py",20929,0,"",python,selection_mouse +209,973166,"input_pipeline/generate_atari_dataset.py",21022,0,"",python,selection_mouse +210,973204,"input_pipeline/generate_atari_dataset.py",21021,0,"",python,selection_command +211,973996,"input_pipeline/generate_atari_dataset.py",20766,0,"",python,selection_mouse +212,975918,"input_pipeline/utils.py",0,0,"import os\nimport pickle\nimport numpy as np\nfrom array_record.python.array_record_module import ArrayRecordWriter\n\n\ndef save_chunks(file_idx, chunks_per_file, output_dir, obs_chunks, act_chunks=None):\n os.makedirs(output_dir, exist_ok=True)\n\n metadata = []\n while len(obs_chunks) >= chunks_per_file:\n chunk_batch = obs_chunks[:chunks_per_file]\n obs_chunks = obs_chunks[chunks_per_file:]\n act_chunk_batch = None\n if act_chunks:\n act_chunk_batch = act_chunks[:chunks_per_file]\n act_chunks = act_chunks[chunks_per_file:]\n episode_path = os.path.join(output_dir, f""data_{file_idx:04d}.array_record"")\n writer = ArrayRecordWriter(str(episode_path), ""group_size:1"")\n seq_lens = []\n for idx, chunk in enumerate(chunk_batch):\n seq_len = chunk.shape[0]\n seq_lens.append(seq_len)\n chunk_record = {\n ""raw_video"": chunk.tobytes(),\n ""sequence_length"": seq_len,\n }\n if act_chunk_batch:\n assert len(chunk) == len(\n act_chunk_batch[idx]\n ), f""Observation data length and action sequence length do not match: {len(chunk)} != {len(act_chunk_batch[idx])}""\n chunk_record[""actions""] = act_chunk_batch[idx]\n writer.write(pickle.dumps(chunk_record))\n writer.close()\n file_idx += 1\n metadata.append(\n {\n ""path"": episode_path,\n ""num_chunks"": len(chunk_batch),\n ""avg_seq_len"": np.mean(seq_lens),\n }\n )\n print(f""Created {episode_path} with {len(chunk_batch)} video chunks"")\n\n return metadata, file_idx, obs_chunks, act_chunks\n",python,tab +213,975918,"input_pipeline/utils.py",119,0,"",python,selection_command +214,983018,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +215,983019,"input_pipeline/generate_atari_dataset.py",20855,0,"",python,selection_command +216,988307,"input_pipeline/generate_atari_dataset.py",21315,0,"",python,selection_mouse +217,988307,"input_pipeline/generate_atari_dataset.py",21314,0,"",python,selection_command +218,1337253,"input_pipeline/generate_atari_dataset.py",18843,0,"",python,selection_mouse +219,1344637,"input_pipeline/generate_atari_dataset.py",19348,0,"",python,selection_mouse +220,1347187,"input_pipeline/generate_atari_dataset.py",19447,0,"",python,selection_mouse +221,1350194,"input_pipeline/generate_atari_dataset.py",19344,0,"",python,selection_mouse +222,1437790,"input_pipeline/generate_atari_dataset.py",19033,0,"",python,selection_mouse +223,1452025,"input_pipeline/generate_atari_dataset.py",18902,0,"",python,selection_mouse +224,1452973,"input_pipeline/generate_atari_dataset.py",18923,0,"",python,selection_mouse +225,1461757,"input_pipeline/generate_atari_dataset.py",18667,0,"",python,selection_mouse +226,1462837,"input_pipeline/generate_atari_dataset.py",18683,0,"",python,selection_mouse +227,1463992,"input_pipeline/generate_atari_dataset.py",18842,0,"",python,selection_mouse +228,1481938,"input_pipeline/generate_atari_dataset.py",19447,0,"",python,selection_mouse +229,1490604,"input_pipeline/generate_atari_dataset.py",18667,0,"",python,selection_mouse +230,1492380,"input_pipeline/generate_atari_dataset.py",18847,0,"",python,selection_mouse +231,1493068,"input_pipeline/generate_atari_dataset.py",18866,0,"",python,selection_mouse +232,1493068,"input_pipeline/generate_atari_dataset.py",18865,0,"",python,selection_command +233,1494518,"input_pipeline/generate_atari_dataset.py",18968,0,"",python,selection_mouse +234,1496382,"input_pipeline/generate_atari_dataset.py",19016,0,"",python,selection_mouse +235,1496926,"input_pipeline/generate_atari_dataset.py",19131,0,"",python,selection_mouse +236,1496926,"input_pipeline/generate_atari_dataset.py",19130,0,"",python,selection_command +237,1498509,"input_pipeline/generate_atari_dataset.py",18932,0,"",python,selection_mouse +238,1498509,"input_pipeline/generate_atari_dataset.py",18931,0,"",python,selection_command +239,1499115,"input_pipeline/generate_atari_dataset.py",18630,0,"",python,selection_mouse +240,1499115,"input_pipeline/generate_atari_dataset.py",18629,0,"",python,selection_command +241,1534885,"input_pipeline/generate_atari_dataset.py",18704,0,"",python,selection_mouse +242,1534885,"input_pipeline/generate_atari_dataset.py",18703,0,"",python,selection_command +243,1536200,"input_pipeline/generate_atari_dataset.py",18630,0,"",python,selection_mouse +244,1536201,"input_pipeline/generate_atari_dataset.py",18629,0,"",python,selection_command +245,1538266,"input_pipeline/generate_atari_dataset.py",18704,0,"",python,selection_mouse +246,1538267,"input_pipeline/generate_atari_dataset.py",18703,0,"",python,selection_command +247,1547628,"input_pipeline/generate_atari_dataset.py",18787,0,"",python,selection_mouse +248,1550007,"input_pipeline/generate_atari_dataset.py",18840,0,"",python,selection_mouse +249,1551958,"input_pipeline/generate_atari_dataset.py",18853,0,"",python,selection_mouse +250,1725703,"input_pipeline/generate_atari_dataset.py",18753,0,"",python,selection_command +251,1725845,"input_pipeline/generate_atari_dataset.py",18679,0,"",python,selection_command +252,1725919,"input_pipeline/generate_atari_dataset.py",18629,0,"",python,selection_command +253,1726161,"input_pipeline/generate_atari_dataset.py",18679,0,"",python,selection_command +254,1726258,"input_pipeline/generate_atari_dataset.py",18753,0,"",python,selection_command +255,1726559,"input_pipeline/generate_atari_dataset.py",18853,0,"",python,selection_command +256,1726749,"input_pipeline/generate_atari_dataset.py",18753,0,"",python,selection_command +257,1726909,"input_pipeline/generate_atari_dataset.py",18679,0,"",python,selection_command +258,1727079,"input_pipeline/generate_atari_dataset.py",18753,0,"",python,selection_command +259,1727260,"input_pipeline/generate_atari_dataset.py",18853,0,"",python,selection_command +260,1727406,"input_pipeline/generate_atari_dataset.py",18915,0,"",python,selection_command +261,1727525,"input_pipeline/generate_atari_dataset.py",18969,0,"",python,selection_command +262,1727786,"input_pipeline/generate_atari_dataset.py",19019,0,"",python,selection_command +263,1727818,"input_pipeline/generate_atari_dataset.py",19096,0,"",python,selection_command +264,1727860,"input_pipeline/generate_atari_dataset.py",19130,0,"",python,selection_command +265,1727882,"input_pipeline/generate_atari_dataset.py",19180,0,"",python,selection_command +266,1727918,"input_pipeline/generate_atari_dataset.py",19218,0,"",python,selection_command +267,1727936,"input_pipeline/generate_atari_dataset.py",19267,0,"",python,selection_command +268,1728092,"input_pipeline/generate_atari_dataset.py",19357,0,"",python,selection_command +269,1728214,"input_pipeline/generate_atari_dataset.py",19431,0,"",python,selection_command +270,1728367,"input_pipeline/generate_atari_dataset.py",19447,0,"",python,selection_command +271,1728574,"input_pipeline/generate_atari_dataset.py",19431,0,"",python,selection_command +272,1728808,"input_pipeline/generate_atari_dataset.py",19357,0,"",python,selection_command +273,1728820,"input_pipeline/generate_atari_dataset.py",19267,0,"",python,selection_command +274,1728866,"input_pipeline/generate_atari_dataset.py",19218,0,"",python,selection_command +275,1728876,"input_pipeline/generate_atari_dataset.py",19180,0,"",python,selection_command +276,1728909,"input_pipeline/generate_atari_dataset.py",19130,0,"",python,selection_command +277,1728972,"input_pipeline/generate_atari_dataset.py",19096,0,"",python,selection_command +278,1728982,"input_pipeline/generate_atari_dataset.py",19019,0,"",python,selection_command +279,1729023,"input_pipeline/generate_atari_dataset.py",18969,0,"",python,selection_command +280,1729063,"input_pipeline/generate_atari_dataset.py",18915,0,"",python,selection_command +281,1729086,"input_pipeline/generate_atari_dataset.py",18853,0,"",python,selection_command +282,1729218,"input_pipeline/generate_atari_dataset.py",18753,0,"",python,selection_command +283,1729522,"input_pipeline/generate_atari_dataset.py",18679,0,"",python,selection_command +284,1818181,"input_pipeline/generate_atari_dataset.py",18753,0,"",python,selection_command +285,1818378,"input_pipeline/generate_atari_dataset.py",18853,0,"",python,selection_command +286,1819799,"input_pipeline/generate_atari_dataset.py",18851,0,"",python,selection_command +287,1820212,"input_pipeline/generate_atari_dataset.py",18849,0,"",python,selection_command +288,1820283,"input_pipeline/generate_atari_dataset.py",18829,0,"",python,selection_command +289,1824236,"input_pipeline/generate_atari_dataset.py",19003,0,"",python,selection_command +290,1825293,"input_pipeline/generate_atari_dataset.py",19032,0,"",python,selection_command +291,1826024,"input_pipeline/generate_atari_dataset.py",19037,0,"",python,selection_command +292,1827792,"input_pipeline/generate_atari_dataset.py",18969,0,"",python,selection_command +293,1827792,"input_pipeline/generate_atari_dataset.py",18931,0,"",python,selection_command +294,1827793,"input_pipeline/generate_atari_dataset.py",18865,0,"",python,selection_command +295,1827793,"input_pipeline/generate_atari_dataset.py",18864,0,"",python,selection_command +296,1827794,"input_pipeline/generate_atari_dataset.py",18860,0,"",python,selection_command +297,1827794,"input_pipeline/generate_atari_dataset.py",18859,0,"",python,selection_command +298,1827795,"input_pipeline/generate_atari_dataset.py",18851,0,"",python,selection_command +299,1827796,"input_pipeline/generate_atari_dataset.py",18849,0,"",python,selection_command +300,1827891,"input_pipeline/generate_atari_dataset.py",18829,0,"",python,selection_command +301,1834862,"input_pipeline/generate_atari_dataset.py",18919,0,"",python,selection_command +302,1835350,"input_pipeline/generate_atari_dataset.py",17685,0,"",python,selection_command +303,1840353,"input_pipeline/generate_atari_dataset.py",18919,0,"",python,selection_command +304,1842706,"input_pipeline/generate_atari_dataset.py",18918,0,"",python,selection_command +305,1842972,"input_pipeline/generate_atari_dataset.py",18909,0,"",python,selection_command +306,1842979,"input_pipeline/generate_atari_dataset.py",18906,0,"",python,selection_command +307,1843354,"input_pipeline/generate_atari_dataset.py",18969,0,"",python,selection_command +308,1843423,"input_pipeline/generate_atari_dataset.py",19010,0,"",python,selection_command +309,1864866,"input_pipeline/generate_atari_dataset.py",19096,0,"",python,selection_command +310,1864948,"input_pipeline/generate_atari_dataset.py",19130,0,"",python,selection_command +311,1865142,"input_pipeline/generate_atari_dataset.py",19171,0,"",python,selection_command +312,1866070,"input_pipeline/generate_atari_dataset.py",19130,0,"",python,selection_command +313,1866071,"input_pipeline/generate_atari_dataset.py",19130,1,"]",python,selection_command +314,1866308,"input_pipeline/generate_atari_dataset.py",19130,0,"",python,selection_command +315,1866710,"input_pipeline/generate_atari_dataset.py",19098,33," ]",python,selection_command +316,1867084,"input_pipeline/generate_atari_dataset.py",19058,73," idx\n ]",python,selection_command +317,1867159,"input_pipeline/generate_atari_dataset.py",18971,160," final_obs_to_append[idx] = infos[""final_observation""][\n idx\n ]",python,selection_command +318,1867280,"input_pipeline/generate_atari_dataset.py",18933,198," if trunc:\n final_obs_to_append[idx] = infos[""final_observation""][\n idx\n ]",python,selection_command +319,1867899,"input_pipeline/generate_atari_dataset.py",18867,264," for idx, trunc in enumerate(truncations):\n if trunc:\n final_obs_to_append[idx] = infos[""final_observation""][\n idx\n ]",python,selection_command +320,2020825,"input_pipeline/generate_atari_dataset.py",18899,0,"",python,selection_command +321,2021519,"input_pipeline/generate_atari_dataset.py",18965,0,"",python,selection_command +322,2021671,"input_pipeline/generate_atari_dataset.py",19003,0,"",python,selection_command +323,2021813,"input_pipeline/generate_atari_dataset.py",19090,0,"",python,selection_command +324,2021944,"input_pipeline/generate_atari_dataset.py",19130,0,"",python,selection_command +325,2022116,"input_pipeline/generate_atari_dataset.py",19164,0,"",python,selection_command +326,2045818,"input_pipeline/generate_atari_dataset.py",18705,743,"",python,content +327,2045818,"input_pipeline/generate_atari_dataset.py",17759,57," observations_seq.append(next_obs.astype(np.uint8))",python,content +328,2046017,"input_pipeline/generate_atari_dataset.py",18710,0," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]\n observations_seq.append(final_obs_to_append.astype(np.uint8))\n\n # Ensure actions and observations have the same length for saving\n if len(actions_seq) + 1 == len(observations_seq):\n actions_seq.append(actions_seq[-1])\n\n",python,content +329,2046018,"input_pipeline/generate_atari_dataset.py",17759,62," observations_seq.append(obs.astype(np.uint8))",python,content +330,2127255,"input_pipeline/generate_atari_dataset.py",19784,0,"",python,selection_mouse +331,2127256,"input_pipeline/generate_atari_dataset.py",19783,0,"",python,selection_command +332,2290485,"input_pipeline/generate_atari_dataset.py",19716,0,"",python,selection_command +333,2290716,"input_pipeline/generate_atari_dataset.py",19657,0,"",python,selection_command +334,2290757,"input_pipeline/generate_atari_dataset.py",19655,0,"",python,selection_command +335,2290800,"input_pipeline/generate_atari_dataset.py",19584,0,"",python,selection_command +336,2290847,"input_pipeline/generate_atari_dataset.py",19494,0,"",python,selection_command +337,2290934,"input_pipeline/generate_atari_dataset.py",19428,0,"",python,selection_command +338,2290960,"input_pipeline/generate_atari_dataset.py",19407,0,"",python,selection_command +339,2290960,"input_pipeline/generate_atari_dataset.py",19316,0,"",python,selection_command +340,2290961,"input_pipeline/generate_atari_dataset.py",19249,0,"",python,selection_command +341,2290987,"input_pipeline/generate_atari_dataset.py",19212,0,"",python,selection_command +342,2291020,"input_pipeline/generate_atari_dataset.py",19139,0,"",python,selection_command +343,2291060,"input_pipeline/generate_atari_dataset.py",19032,0,"",python,selection_command +344,2291061,"input_pipeline/generate_atari_dataset.py",18932,0,"",python,selection_command +345,2291248,"input_pipeline/generate_atari_dataset.py",18865,0,"",python,selection_command +346,2291498,"input_pipeline/generate_atari_dataset.py",18864,0,"",python,selection_command +347,2291658,"input_pipeline/generate_atari_dataset.py",18860,0,"",python,selection_command +348,2291800,"input_pipeline/generate_atari_dataset.py",18859,0,"",python,selection_command +349,2292249,"input_pipeline/generate_atari_dataset.py",18851,0,"",python,selection_command +350,2292911,"input_pipeline/generate_atari_dataset.py",18849,0,"",python,selection_command +351,2293051,"input_pipeline/generate_atari_dataset.py",18829,0,"",python,selection_command +352,2293268,"input_pipeline/generate_atari_dataset.py",18792,0,"",python,selection_command +353,2293678,"input_pipeline/generate_atari_dataset.py",18829,0,"",python,selection_command +354,2505441,"input_pipeline/generate_atari_dataset.py",18729,0,"",python,selection_command +355,2505467,"input_pipeline/generate_atari_dataset.py",18655,0,"",python,selection_command +356,2506780,"input_pipeline/generate_atari_dataset.py",18729,0,"",python,selection_command +357,2507020,"input_pipeline/generate_atari_dataset.py",18829,0,"",python,selection_command +358,2507058,"input_pipeline/generate_atari_dataset.py",18891,0,"",python,selection_command +359,2507069,"input_pipeline/generate_atari_dataset.py",18991,0,"",python,selection_command +360,2507104,"input_pipeline/generate_atari_dataset.py",19098,0,"",python,selection_command +361,2507127,"input_pipeline/generate_atari_dataset.py",19175,0,"",python,selection_command +362,2507183,"input_pipeline/generate_atari_dataset.py",19238,0,"",python,selection_command +363,2507189,"input_pipeline/generate_atari_dataset.py",19275,0,"",python,selection_command +364,2507228,"input_pipeline/generate_atari_dataset.py",19366,0,"",python,selection_command +365,2507266,"input_pipeline/generate_atari_dataset.py",19428,0,"",python,selection_command +366,2507297,"input_pipeline/generate_atari_dataset.py",19453,0,"",python,selection_command +367,2507326,"input_pipeline/generate_atari_dataset.py",19543,0,"",python,selection_command +368,2507362,"input_pipeline/generate_atari_dataset.py",19617,0,"",python,selection_command +369,2507418,"input_pipeline/generate_atari_dataset.py",19657,0,"",python,selection_command +370,2507499,"input_pipeline/generate_atari_dataset.py",19682,0,"",python,selection_command +371,2507500,"input_pipeline/generate_atari_dataset.py",19742,0,"",python,selection_command +372,2507501,"input_pipeline/generate_atari_dataset.py",19809,0,"",python,selection_command +373,2509630,"input_pipeline/generate_atari_dataset.py",19895,0,"",python,selection_command +374,2509669,"input_pipeline/generate_atari_dataset.py",19936,0,"",python,selection_command +375,2509688,"input_pipeline/generate_atari_dataset.py",19966,0,"",python,selection_command +376,2509708,"input_pipeline/generate_atari_dataset.py",20030,0,"",python,selection_command +377,2509748,"input_pipeline/generate_atari_dataset.py",20055,0,"",python,selection_command +378,2509783,"input_pipeline/generate_atari_dataset.py",20107,0,"",python,selection_command +379,2509836,"input_pipeline/generate_atari_dataset.py",20159,0,"",python,selection_command +380,2509856,"input_pipeline/generate_atari_dataset.py",20201,0,"",python,selection_command +381,2509888,"input_pipeline/generate_atari_dataset.py",20260,0,"",python,selection_command +382,2509925,"input_pipeline/generate_atari_dataset.py",20348,0,"",python,selection_command +383,2509958,"input_pipeline/generate_atari_dataset.py",20422,0,"",python,selection_command +384,2509989,"input_pipeline/generate_atari_dataset.py",20465,0,"",python,selection_command +385,2510043,"input_pipeline/generate_atari_dataset.py",20578,0,"",python,selection_command +386,2510125,"input_pipeline/generate_atari_dataset.py",20689,0,"",python,selection_command +387,2510125,"input_pipeline/generate_atari_dataset.py",20791,0,"",python,selection_command +388,2510126,"input_pipeline/generate_atari_dataset.py",20829,0,"",python,selection_command +389,2510157,"input_pipeline/generate_atari_dataset.py",20914,0,"",python,selection_command +390,2510203,"input_pipeline/generate_atari_dataset.py",20997,0,"",python,selection_command +391,2510241,"input_pipeline/generate_atari_dataset.py",21025,0,"",python,selection_command +392,2510248,"input_pipeline/generate_atari_dataset.py",21050,0,"",python,selection_command +393,2510282,"input_pipeline/generate_atari_dataset.py",21098,0,"",python,selection_command +394,2510323,"input_pipeline/generate_atari_dataset.py",21181,0,"",python,selection_command +395,2510350,"input_pipeline/generate_atari_dataset.py",21211,0,"",python,selection_command +396,2510375,"input_pipeline/generate_atari_dataset.py",21293,0,"",python,selection_command +397,2510411,"input_pipeline/generate_atari_dataset.py",21356,0,"",python,selection_command +398,2510445,"input_pipeline/generate_atari_dataset.py",21395,0,"",python,selection_command +399,2510497,"input_pipeline/generate_atari_dataset.py",21420,0,"",python,selection_command +400,2510518,"input_pipeline/generate_atari_dataset.py",21475,0,"",python,selection_command +401,2510548,"input_pipeline/generate_atari_dataset.py",21505,0,"",python,selection_command +402,2510652,"input_pipeline/generate_atari_dataset.py",21550,0,"",python,selection_command +403,2510652,"input_pipeline/generate_atari_dataset.py",21616,0,"",python,selection_command +404,2510653,"input_pipeline/generate_atari_dataset.py",21660,0,"",python,selection_command +405,2510680,"input_pipeline/generate_atari_dataset.py",21704,0,"",python,selection_command +406,2510723,"input_pipeline/generate_atari_dataset.py",21749,0,"",python,selection_command +407,2510760,"input_pipeline/generate_atari_dataset.py",21815,0,"",python,selection_command +408,2510795,"input_pipeline/generate_atari_dataset.py",21869,0,"",python,selection_command +409,2510859,"input_pipeline/generate_atari_dataset.py",21912,0,"",python,selection_command +410,2510862,"input_pipeline/generate_atari_dataset.py",21956,0,"",python,selection_command +411,2510891,"input_pipeline/generate_atari_dataset.py",22000,0,"",python,selection_command +412,2510929,"input_pipeline/generate_atari_dataset.py",22030,0,"",python,selection_command +413,2510945,"input_pipeline/generate_atari_dataset.py",22095,0,"",python,selection_command +414,2512329,"input_pipeline/generate_atari_dataset.py",22030,0,"",python,selection_command +415,2512806,"input_pipeline/generate_atari_dataset.py",22095,0,"",python,selection_command +416,2513004,"input_pipeline/generate_atari_dataset.py",22120,0,"",python,selection_command +417,2513267,"input_pipeline/generate_atari_dataset.py",22172,0,"",python,selection_command +418,2513317,"input_pipeline/generate_atari_dataset.py",22197,0,"",python,selection_command +419,2513333,"input_pipeline/generate_atari_dataset.py",22230,0,"",python,selection_command +420,2513467,"input_pipeline/generate_atari_dataset.py",22305,0,"",python,selection_command +421,2513974,"input_pipeline/generate_atari_dataset.py",22369,0,"",python,selection_command +422,2513974,"input_pipeline/generate_atari_dataset.py",22400,0,"",python,selection_command +423,2513974,"input_pipeline/generate_atari_dataset.py",22456,0,"",python,selection_command +424,2513975,"input_pipeline/generate_atari_dataset.py",22499,0,"",python,selection_command +425,2513975,"input_pipeline/generate_atari_dataset.py",22611,0,"",python,selection_command +426,2513975,"input_pipeline/generate_atari_dataset.py",22668,0,"",python,selection_command +427,2513976,"input_pipeline/generate_atari_dataset.py",22755,0,"",python,selection_command +428,2513976,"input_pipeline/generate_atari_dataset.py",22884,0,"",python,selection_command +429,2513976,"input_pipeline/generate_atari_dataset.py",22922,0,"",python,selection_command +430,2513977,"input_pipeline/generate_atari_dataset.py",22974,0,"",python,selection_command +431,2513977,"input_pipeline/generate_atari_dataset.py",23026,0,"",python,selection_command +432,2513977,"input_pipeline/generate_atari_dataset.py",23107,0,"",python,selection_command +433,2513978,"input_pipeline/generate_atari_dataset.py",23166,0,"",python,selection_command +434,2513978,"input_pipeline/generate_atari_dataset.py",23253,0,"",python,selection_command +435,2513978,"input_pipeline/generate_atari_dataset.py",23315,0,"",python,selection_command +436,2513980,"input_pipeline/generate_atari_dataset.py",23386,0,"",python,selection_command +437,2513981,"input_pipeline/generate_atari_dataset.py",23424,0,"",python,selection_command +438,2513987,"input_pipeline/generate_atari_dataset.py",23498,0,"",python,selection_command +439,2513988,"input_pipeline/generate_atari_dataset.py",23528,0,"",python,selection_command +440,2514045,"input_pipeline/generate_atari_dataset.py",23563,0,"",python,selection_command +441,2514060,"input_pipeline/generate_atari_dataset.py",23653,0,"",python,selection_command +442,2514090,"input_pipeline/generate_atari_dataset.py",23659,0,"",python,selection_command +443,2514385,"input_pipeline/generate_atari_dataset.py",23684,0,"",python,selection_command +444,2516426,"input_pipeline/generate_atari_dataset.py",23659,0,"",python,selection_command +445,2516507,"input_pipeline/generate_atari_dataset.py",23653,0,"",python,selection_command +446,2516540,"input_pipeline/generate_atari_dataset.py",23563,0,"",python,selection_command +447,2516580,"input_pipeline/generate_atari_dataset.py",23528,0,"",python,selection_command +448,2516622,"input_pipeline/generate_atari_dataset.py",23498,0,"",python,selection_command +449,2516629,"input_pipeline/generate_atari_dataset.py",23424,0,"",python,selection_command +450,2516660,"input_pipeline/generate_atari_dataset.py",23386,0,"",python,selection_command +451,2516702,"input_pipeline/generate_atari_dataset.py",23315,0,"",python,selection_command +452,2516740,"input_pipeline/generate_atari_dataset.py",23253,0,"",python,selection_command +453,2516785,"input_pipeline/generate_atari_dataset.py",23166,0,"",python,selection_command +454,2517013,"input_pipeline/generate_atari_dataset.py",23107,0,"",python,selection_command +455,2517014,"input_pipeline/generate_atari_dataset.py",23026,0,"",python,selection_command +456,2517014,"input_pipeline/generate_atari_dataset.py",22974,0,"",python,selection_command +457,2517015,"input_pipeline/generate_atari_dataset.py",22922,0,"",python,selection_command +458,2517015,"input_pipeline/generate_atari_dataset.py",22884,0,"",python,selection_command +459,2517020,"input_pipeline/generate_atari_dataset.py",22755,0,"",python,selection_command +460,2517052,"input_pipeline/generate_atari_dataset.py",22668,0,"",python,selection_command +461,2517061,"input_pipeline/generate_atari_dataset.py",22611,0,"",python,selection_command +462,2517113,"input_pipeline/generate_atari_dataset.py",22499,0,"",python,selection_command +463,2517122,"input_pipeline/generate_atari_dataset.py",22456,0,"",python,selection_command +464,2517184,"input_pipeline/generate_atari_dataset.py",22400,0,"",python,selection_command +465,2517185,"input_pipeline/generate_atari_dataset.py",22369,0,"",python,selection_command +466,2517194,"input_pipeline/generate_atari_dataset.py",22305,0,"",python,selection_command +467,2517286,"input_pipeline/generate_atari_dataset.py",22230,0,"",python,selection_command +468,2517287,"input_pipeline/generate_atari_dataset.py",22197,0,"",python,selection_command +469,2517289,"input_pipeline/generate_atari_dataset.py",22172,0,"",python,selection_command +470,2517334,"input_pipeline/generate_atari_dataset.py",22120,0,"",python,selection_command +471,2517340,"input_pipeline/generate_atari_dataset.py",22095,0,"",python,selection_command +472,2517455,"input_pipeline/generate_atari_dataset.py",22120,0,"",python,selection_command +473,2517738,"input_pipeline/generate_atari_dataset.py",22172,0,"",python,selection_command +474,2517756,"input_pipeline/generate_atari_dataset.py",22197,0,"",python,selection_command +475,2517774,"input_pipeline/generate_atari_dataset.py",22230,0,"",python,selection_command +476,2517825,"input_pipeline/generate_atari_dataset.py",22305,0,"",python,selection_command +477,2517852,"input_pipeline/generate_atari_dataset.py",22369,0,"",python,selection_command +478,2517878,"input_pipeline/generate_atari_dataset.py",22400,0,"",python,selection_command +479,2518021,"input_pipeline/generate_atari_dataset.py",22369,0,"",python,selection_command +480,2518168,"input_pipeline/generate_atari_dataset.py",22305,0,"",python,selection_command +481,2518217,"input_pipeline/generate_atari_dataset.py",22230,0,"",python,selection_command +482,2518402,"input_pipeline/generate_atari_dataset.py",20055,0,"",python,selection_keyboard +483,2518830,"input_pipeline/generate_atari_dataset.py",17874,0,"",python,selection_keyboard +484,2519957,"input_pipeline/generate_atari_dataset.py",20055,0,"",python,selection_keyboard +485,3153177,"input_pipeline/generate_atari_dataset.py",18866,0,"",python,selection_mouse +486,3153178,"input_pipeline/generate_atari_dataset.py",18865,0,"",python,selection_command +487,3156050,"input_pipeline/generate_atari_dataset.py",18765,0,"",python,selection_command +488,3156475,"input_pipeline/generate_atari_dataset.py",18705,99," # Append the final post-action observation once to capture all observations",python,selection_command +489,3158465,"input_pipeline/generate_atari_dataset.py",18705,161," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()",python,selection_command +490,3158466,"input_pipeline/generate_atari_dataset.py",18705,261," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.",python,selection_command +491,3158466,"input_pipeline/generate_atari_dataset.py",18705,368," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.",python,selection_command +492,3158467,"input_pipeline/generate_atari_dataset.py",18705,445," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)",python,selection_command +493,3158467,"input_pipeline/generate_atari_dataset.py",18705,508," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):",python,selection_command +494,3158467,"input_pipeline/generate_atari_dataset.py",18705,545," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:",python,selection_command +495,3158467,"input_pipeline/generate_atari_dataset.py",18705,636," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]",python,selection_command +496,3158468,"input_pipeline/generate_atari_dataset.py",18705,722," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]\n observations_seq.append(final_obs_to_append.astype(np.uint8))",python,selection_command +497,3158468,"input_pipeline/generate_atari_dataset.py",18705,723," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]\n observations_seq.append(final_obs_to_append.astype(np.uint8))\n",python,selection_command +498,3158468,"input_pipeline/generate_atari_dataset.py",18705,813," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]\n observations_seq.append(final_obs_to_append.astype(np.uint8))\n\n # Ensure actions and observations have the same length for saving",python,selection_command +499,3158469,"input_pipeline/generate_atari_dataset.py",18705,887," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]\n observations_seq.append(final_obs_to_append.astype(np.uint8))\n\n # Ensure actions and observations have the same length for saving\n if len(actions_seq) + 1 == len(observations_seq):",python,selection_command +500,3158469,"input_pipeline/generate_atari_dataset.py",18705,951," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]\n observations_seq.append(final_obs_to_append.astype(np.uint8))\n\n # Ensure actions and observations have the same length for saving\n if len(actions_seq) + 1 == len(observations_seq):\n actions_seq.append(actions_seq[-1])",python,selection_command +501,3158469,"input_pipeline/generate_atari_dataset.py",18705,952," # Append the final post-action observation once to capture all observations\n final_obs_to_append = next_obs.copy()\n # If an episode finished, next_obs for that index is already the reset obs.\n # Swap in the true terminal obs from infos[""final_observation""] for any done envs.\n done_mask = np.logical_or(terminations, truncations)\n for idx, done in enumerate(done_mask):\n if done:\n final_obs_to_append[idx] = infos[""final_observation""][idx]\n observations_seq.append(final_obs_to_append.astype(np.uint8))\n\n # Ensure actions and observations have the same length for saving\n if len(actions_seq) + 1 == len(observations_seq):\n actions_seq.append(actions_seq[-1])\n",python,selection_command +502,3158902,"input_pipeline/generate_atari_dataset.py",18705,953,"",python,content +503,3158905,"input_pipeline/generate_atari_dataset.py",18729,0,"",python,selection_command +504,3161687,"input_pipeline/generate_atari_dataset.py",21003,0,"",python,selection_keyboard +505,3161688,"input_pipeline/generate_atari_dataset.py",23126,0,"",python,selection_keyboard +506,3161689,"input_pipeline/generate_atari_dataset.py",24837,0,"",python,selection_keyboard +507,3161689,"input_pipeline/generate_atari_dataset.py",26926,0,"",python,selection_keyboard +508,3161690,"input_pipeline/generate_atari_dataset.py",28361,0,"",python,selection_keyboard +509,3161692,"input_pipeline/generate_atari_dataset.py",26926,0,"",python,selection_keyboard +510,3161909,"input_pipeline/generate_atari_dataset.py",24837,0,"",python,selection_keyboard +511,3161909,"input_pipeline/generate_atari_dataset.py",23126,0,"",python,selection_keyboard +512,3161930,"input_pipeline/generate_atari_dataset.py",21003,0,"",python,selection_keyboard +513,3161979,"input_pipeline/generate_atari_dataset.py",18729,0,"",python,selection_keyboard +514,3162051,"input_pipeline/generate_atari_dataset.py",17035,0,"",python,selection_keyboard +515,3162052,"input_pipeline/generate_atari_dataset.py",15760,0,"",python,selection_keyboard +516,3162058,"input_pipeline/generate_atari_dataset.py",14415,0,"",python,selection_keyboard +517,3162129,"input_pipeline/generate_atari_dataset.py",13031,0,"",python,selection_keyboard +518,3162135,"input_pipeline/generate_atari_dataset.py",11649,0,"",python,selection_keyboard +519,3162356,"input_pipeline/generate_atari_dataset.py",13031,0,"",python,selection_keyboard +520,3162508,"input_pipeline/generate_atari_dataset.py",14415,0,"",python,selection_keyboard +521,3162677,"input_pipeline/generate_atari_dataset.py",15760,0,"",python,selection_keyboard +522,3162785,"input_pipeline/generate_atari_dataset.py",17035,0,"",python,selection_keyboard +523,3162900,"input_pipeline/generate_atari_dataset.py",18729,0,"",python,selection_keyboard +524,3193679,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +525,3195647,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +526,3200448,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +527,3202133,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +528,3202135,"TERMINAL",0,0,"",,terminal_focus +529,3204182,"TERMINAL",0,0,"source /home/franz.srambical/cleanrl/.venv/bin/activate",,terminal_command +530,3204182,"TERMINAL",0,0,"]633;C]0;franz.srambical@hai-login2:~/jafar",,terminal_output +531,3209822,"TERMINAL",0,0,"ls /fast/project/HFMI_SynergyUnit/jafar_ws/data/breakout/train/",,terminal_command +532,3209842,"TERMINAL",0,0,"]633;Cdata_0000.array_record data_0005.array_record data_0010.array_record data_0015.array_record data_0020.array_record data_0025.array_record data_0030.array_record\r\ndata_0001.array_record data_0006.array_record data_0011.array_record data_0016.array_record data_0021.array_record data_0026.array_record data_0031.array_record\r\ndata_0002.array_record data_0007.array_record data_0012.array_record data_0017.array_record data_0022.array_record data_0027.array_record\r\ndata_0003.array_record data_0008.array_record data_0013.array_record data_0018.array_record data_0023.array_record data_0028.array_record\r\ndata_0004.array_record data_0009.array_record data_0014.array_record data_0019.array_record data_0024.array_record data_0029.array_record\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +533,3224053,"input_pipeline/generate_atari_dataset.py",18655,0,"",python,selection_command +534,3224296,"input_pipeline/generate_atari_dataset.py",18629,0,"",python,selection_command +535,3224374,"input_pipeline/generate_atari_dataset.py",18584,0,"",python,selection_command +536,3224375,"input_pipeline/generate_atari_dataset.py",18510,0,"",python,selection_command +537,3224409,"input_pipeline/generate_atari_dataset.py",18458,0,"",python,selection_command +538,3224428,"input_pipeline/generate_atari_dataset.py",18433,0,"",python,selection_command +539,3224471,"input_pipeline/generate_atari_dataset.py",18431,0,"",python,selection_command +540,3224503,"input_pipeline/generate_atari_dataset.py",18351,0,"",python,selection_command +541,3224544,"input_pipeline/generate_atari_dataset.py",18312,0,"",python,selection_command +542,3224559,"input_pipeline/generate_atari_dataset.py",18286,0,"",python,selection_command +543,3224592,"input_pipeline/generate_atari_dataset.py",18206,0,"",python,selection_command +544,3224637,"input_pipeline/generate_atari_dataset.py",18167,0,"",python,selection_command +545,3224670,"input_pipeline/generate_atari_dataset.py",18141,0,"",python,selection_command +546,3224699,"input_pipeline/generate_atari_dataset.py",18052,0,"",python,selection_command +547,3224736,"input_pipeline/generate_atari_dataset.py",18025,0,"",python,selection_command +548,3224771,"input_pipeline/generate_atari_dataset.py",17978,0,"",python,selection_command +549,3224812,"input_pipeline/generate_atari_dataset.py",17933,0,"",python,selection_command +550,3225149,"input_pipeline/generate_atari_dataset.py",17899,0,"",python,selection_command +551,3225207,"input_pipeline/generate_atari_dataset.py",17874,0,"",python,selection_command +552,3225370,"input_pipeline/generate_atari_dataset.py",17841,0,"",python,selection_command +553,3228520,"input_pipeline/generate_atari_dataset.py",17783,0,"",python,selection_command +554,3229001,"input_pipeline/generate_atari_dataset.py",17841,0,"",python,selection_command +555,3229001,"input_pipeline/generate_atari_dataset.py",17783,0,"",python,selection_command +556,3229110,"input_pipeline/generate_atari_dataset.py",17841,0,"",python,selection_command +557,3229430,"input_pipeline/generate_atari_dataset.py",17783,0,"",python,selection_command +558,3229602,"input_pipeline/generate_atari_dataset.py",17750,0,"",python,selection_command +559,3229768,"input_pipeline/generate_atari_dataset.py",17725,0,"",python,selection_command +560,3229999,"input_pipeline/generate_atari_dataset.py",17750,0,"",python,selection_command +561,3230134,"input_pipeline/generate_atari_dataset.py",17783,0,"",python,selection_command +562,3230279,"input_pipeline/generate_atari_dataset.py",17841,0,"",python,selection_command +563,3236697,"input_pipeline/generate_atari_dataset.py",20044,0,"",python,selection_keyboard +564,3236858,"input_pipeline/generate_atari_dataset.py",22154,0,"",python,selection_keyboard +565,3236985,"input_pipeline/generate_atari_dataset.py",23923,0,"",python,selection_keyboard +566,3237113,"input_pipeline/generate_atari_dataset.py",25891,0,"",python,selection_keyboard +567,3237269,"input_pipeline/generate_atari_dataset.py",27693,0,"",python,selection_keyboard +568,3237361,"input_pipeline/generate_atari_dataset.py",29185,0,"",python,selection_keyboard +569,3237666,"input_pipeline/generate_atari_dataset.py",27693,0,"",python,selection_keyboard +570,3237919,"input_pipeline/generate_atari_dataset.py",25891,0,"",python,selection_keyboard +571,3238357,"input_pipeline/generate_atari_dataset.py",23923,0,"",python,selection_keyboard +572,3238360,"input_pipeline/generate_atari_dataset.py",22154,0,"",python,selection_keyboard +573,3238361,"input_pipeline/generate_atari_dataset.py",20044,0,"",python,selection_keyboard +574,3238363,"input_pipeline/generate_atari_dataset.py",17841,0,"",python,selection_keyboard +575,3238365,"input_pipeline/generate_atari_dataset.py",16384,0,"",python,selection_keyboard +576,3238367,"input_pipeline/generate_atari_dataset.py",15019,0,"",python,selection_keyboard +577,3238368,"input_pipeline/generate_atari_dataset.py",13740,0,"",python,selection_keyboard +578,3238442,"input_pipeline/generate_atari_dataset.py",12248,0,"",python,selection_keyboard +579,3238541,"input_pipeline/generate_atari_dataset.py",10947,0,"",python,selection_keyboard +580,3238803,"input_pipeline/generate_atari_dataset.py",9736,0,"",python,selection_keyboard +581,3239068,"input_pipeline/generate_atari_dataset.py",8451,0,"",python,selection_keyboard +582,3239068,"input_pipeline/generate_atari_dataset.py",7283,0,"",python,selection_keyboard +583,3239104,"input_pipeline/generate_atari_dataset.py",5910,0,"",python,selection_keyboard +584,3239120,"input_pipeline/generate_atari_dataset.py",4375,0,"",python,selection_keyboard +585,3239216,"input_pipeline/generate_atari_dataset.py",2995,0,"",python,selection_keyboard +586,3239216,"input_pipeline/generate_atari_dataset.py",1365,0,"",python,selection_keyboard +587,3239236,"input_pipeline/generate_atari_dataset.py",242,0,"",python,selection_keyboard +588,3239385,"input_pipeline/generate_atari_dataset.py",1365,0,"",python,selection_keyboard +589,3239604,"input_pipeline/generate_atari_dataset.py",2995,0,"",python,selection_keyboard +590,3239751,"input_pipeline/generate_atari_dataset.py",4375,0,"",python,selection_keyboard +591,3239910,"input_pipeline/generate_atari_dataset.py",5910,0,"",python,selection_keyboard +592,3240428,"input_pipeline/generate_atari_dataset.py",7283,0,"",python,selection_keyboard +593,3240428,"input_pipeline/generate_atari_dataset.py",8451,0,"",python,selection_keyboard +594,3240428,"input_pipeline/generate_atari_dataset.py",9736,0,"",python,selection_keyboard +595,3240428,"input_pipeline/generate_atari_dataset.py",10947,0,"",python,selection_keyboard +596,3240539,"input_pipeline/generate_atari_dataset.py",12248,0,"",python,selection_keyboard +597,3240713,"input_pipeline/generate_atari_dataset.py",13740,0,"",python,selection_keyboard +598,3240714,"input_pipeline/generate_atari_dataset.py",15019,0,"",python,selection_keyboard +599,3240714,"input_pipeline/generate_atari_dataset.py",16384,0,"",python,selection_keyboard +600,3240714,"input_pipeline/generate_atari_dataset.py",17841,0,"",python,selection_keyboard +601,3240715,"input_pipeline/generate_atari_dataset.py",20044,0,"",python,selection_keyboard +602,3240832,"input_pipeline/generate_atari_dataset.py",22154,0,"",python,selection_keyboard +603,3240833,"input_pipeline/generate_atari_dataset.py",23923,0,"",python,selection_keyboard +604,3254481,"genie.py",0,0,"",python,tab +605,3254482,"genie.py",5992,0,"",python,selection_command +606,3264784,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +607,3264784,"input_pipeline/generate_atari_dataset.py",17759,0,"",python,selection_command +608,3267765,"input_pipeline/generate_atari_dataset.py",0,0,"# adapted from https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/rainbow_atari.py\nimport collections\nimport math\nimport os\nimport random\nimport time\nfrom collections import deque\nfrom dataclasses import dataclass\n\nimport gymnasium as gym\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport tyro\nfrom typing import Optional, Any\nfrom torch.utils.tensorboard.writer import SummaryWriter\n\nfrom cleanrl_utils.atari_wrappers import (\n ClipRewardEnv,\n EpisodicLifeEnv,\n FireResetEnv,\n MaxAndSkipEnv,\n NoopResetEnv,\n)\n\ntry:\n from utils import save_chunks # type: ignore\nexcept Exception: # pragma: no cover\n from input_pipeline.utils import save_chunks # type: ignore\nimport json\n\n\n@dataclass\nclass Args:\n exp_name: str = os.path.basename(__file__)[: -len("".py"")]\n """"""the name of this experiment""""""\n seed: int = 1\n """"""seed of the experiment""""""\n torch_deterministic: bool = True\n """"""if toggled, `torch.backends.cudnn.deterministic=False`""""""\n cuda: bool = True\n """"""if toggled, cuda will be enabled by default""""""\n track: bool = False\n """"""if toggled, this experiment will be tracked with Weights and Biases""""""\n wandb_project_name: str = ""cleanRL""\n """"""the wandb's project name""""""\n wandb_entity: Optional[str] = None\n """"""the entity (team) of wandb's project""""""\n capture_video: bool = False\n """"""whether to capture videos of the agent performances (check out `videos` folder)""""""\n save_model: bool = False\n """"""whether to save model into the `runs/{run_name}` folder""""""\n upload_model: bool = False\n """"""whether to upload the saved model to huggingface""""""\n hf_entity: str = """"\n """"""the user or org name of the model repository from the Hugging Face Hub""""""\n\n env_id: str = ""BreakoutNoFrameskip-v4""\n """"""the id of the environment""""""\n total_timesteps: int = 10000000\n """"""total timesteps of the experiments""""""\n learning_rate: float = 0.0000625\n """"""the learning rate of the optimizer""""""\n num_envs: int = 1\n """"""the number of parallel game environments""""""\n buffer_size: int = 1000000\n """"""the replay memory buffer size""""""\n gamma: float = 0.99\n """"""the discount factor gamma""""""\n tau: float = 1.0\n """"""the target network update rate""""""\n target_network_frequency: int = 8000\n """"""the timesteps it takes to update the target network""""""\n batch_size: int = 32\n """"""the batch size of sample from the reply memory""""""\n start_e: float = 1\n """"""the starting epsilon for exploration""""""\n end_e: float = 0.01\n """"""the ending epsilon for exploration""""""\n exploration_fraction: float = 0.10\n """"""the fraction of `total-timesteps` it takes from start-e to go end-e""""""\n learning_starts: int = 80000\n """"""timestep to start learning""""""\n train_frequency: int = 4\n """"""the frequency of training""""""\n n_step: int = 3\n """"""the number of steps to look ahead for n-step Q learning""""""\n prioritized_replay_alpha: float = 0.5\n """"""alpha parameter for prioritized replay buffer""""""\n prioritized_replay_beta: float = 0.4\n """"""beta parameter for prioritized replay buffer""""""\n prioritized_replay_eps: float = 1e-6\n """"""epsilon parameter for prioritized replay buffer""""""\n n_atoms: int = 51\n """"""the number of atoms""""""\n v_min: float = -10\n """"""the return lower bound""""""\n v_max: float = 10\n """"""the return upper bound""""""\n\n # Dataset capture\n capture_dataset: bool = True\n num_episodes_train: int = 10000\n num_episodes_val: int = 500\n num_episodes_test: int = 500\n output_dir: str = ""data/atari_episodes""\n min_episode_length: int = 1\n chunk_size: int = 160\n chunks_per_file: int = 100\n stop_on_complete: bool = True\n\n\ndef make_env(env_id, seed, idx, capture_video, run_name):\n def thunk():\n if capture_video and idx == 0:\n env = gym.make(env_id, render_mode=""rgb_array"")\n env = gym.wrappers.RecordVideo(env, f""videos/{run_name}"")\n else:\n env = gym.make(env_id)\n env = gym.wrappers.RecordEpisodeStatistics(env)\n\n env = NoopResetEnv(env, noop_max=30)\n env = MaxAndSkipEnv(env, skip=4)\n env = EpisodicLifeEnv(env)\n if ""FIRE"" in env.unwrapped.get_action_meanings():\n env = FireResetEnv(env)\n env = ClipRewardEnv(env)\n env = gym.wrappers.ResizeObservation(env, (84, 84))\n env = gym.wrappers.GrayScaleObservation(env)\n env = gym.wrappers.FrameStack(env, 4)\n\n env.action_space.seed(seed)\n return env\n\n return thunk\n\n\nclass NoisyLinear(nn.Module):\n def __init__(self, in_features, out_features, std_init=0.5):\n super().__init__()\n self.in_features = in_features\n self.out_features = out_features\n self.std_init = std_init\n\n self.weight_mu = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.weight_sigma = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.register_buffer(\n ""weight_epsilon"", torch.FloatTensor(out_features, in_features)\n )\n self.bias_mu = nn.Parameter(torch.FloatTensor(out_features))\n self.bias_sigma = nn.Parameter(torch.FloatTensor(out_features))\n self.register_buffer(""bias_epsilon"", torch.FloatTensor(out_features))\n # factorized gaussian noise\n self.reset_parameters()\n self.reset_noise()\n\n def reset_parameters(self):\n mu_range = 1 / math.sqrt(self.in_features)\n self.weight_mu.data.uniform_(-mu_range, mu_range)\n self.weight_sigma.data.fill_(self.std_init / math.sqrt(self.in_features))\n self.bias_mu.data.uniform_(-mu_range, mu_range)\n self.bias_sigma.data.fill_(self.std_init / math.sqrt(self.out_features))\n\n def reset_noise(self):\n self.weight_epsilon.normal_()\n self.bias_epsilon.normal_()\n\n def forward(self, input):\n if self.training:\n weight = self.weight_mu + self.weight_sigma * self.weight_epsilon\n bias = self.bias_mu + self.bias_sigma * self.bias_epsilon\n else:\n weight = self.weight_mu\n bias = self.bias_mu\n return F.linear(input, weight, bias)\n\n\n# ALGO LOGIC: initialize agent here:\nclass NoisyDuelingDistributionalNetwork(nn.Module):\n def __init__(self, env, n_atoms, v_min, v_max):\n super().__init__()\n self.n_atoms = n_atoms\n self.v_min = v_min\n self.v_max = v_max\n self.delta_z = (v_max - v_min) / (n_atoms - 1)\n self.n_actions = env.single_action_space.n\n self.register_buffer(""support"", torch.linspace(v_min, v_max, n_atoms))\n\n self.network = nn.Sequential(\n nn.Conv2d(4, 32, 8, stride=4),\n nn.ReLU(),\n nn.Conv2d(32, 64, 4, stride=2),\n nn.ReLU(),\n nn.Conv2d(64, 64, 3, stride=1),\n nn.ReLU(),\n nn.Flatten(),\n )\n conv_output_size = 3136\n\n self.value_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512), nn.ReLU(), NoisyLinear(512, n_atoms)\n )\n\n self.advantage_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512),\n nn.ReLU(),\n NoisyLinear(512, n_atoms * self.n_actions),\n )\n\n def forward(self, x):\n h = self.network(x / 255.0)\n value = self.value_head(h).view(-1, 1, self.n_atoms)\n advantage = self.advantage_head(h).view(-1, self.n_actions, self.n_atoms)\n q_atoms = value + advantage - advantage.mean(dim=1, keepdim=True)\n q_dist = F.softmax(q_atoms, dim=2)\n return q_dist\n\n def reset_noise(self):\n for layer in self.value_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n for layer in self.advantage_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n\n\nPrioritizedBatch = collections.namedtuple(\n ""PrioritizedBatch"",\n [\n ""observations"",\n ""actions"",\n ""rewards"",\n ""next_observations"",\n ""dones"",\n ""indices"",\n ""weights"",\n ],\n)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass SumSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.zeros(self.tree_size, dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = self.tree[parent * 2 + 1] + self.tree[parent * 2 + 2]\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def total(self):\n return self.tree[0]\n\n def retrieve(self, value):\n idx = 0\n while idx * 2 + 1 < self.tree_size:\n left = idx * 2 + 1\n right = left + 1\n if value <= self.tree[left]:\n idx = left\n else:\n value -= self.tree[left]\n idx = right\n return idx - (self.capacity - 1)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass MinSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.full(self.tree_size, float(""inf""), dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = np.minimum(\n self.tree[parent * 2 + 1], self.tree[parent * 2 + 2]\n )\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def min(self):\n return self.tree[0]\n\n\nclass PrioritizedReplayBuffer:\n def __init__(\n self, capacity, obs_shape, device, n_step, gamma, alpha=0.6, beta=0.4, eps=1e-6\n ):\n self.capacity = capacity\n self.device = device\n self.n_step = n_step\n self.gamma = gamma\n self.alpha = alpha\n self.beta = beta\n self.eps = eps\n\n self.buffer_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_next_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_actions = np.zeros(capacity, dtype=np.int64)\n self.buffer_rewards = np.zeros(capacity, dtype=np.float32)\n self.buffer_dones = np.zeros(capacity, dtype=np.bool_)\n\n self.pos = 0\n self.size = 0\n self.max_priority = 1.0\n\n self.sum_tree = SumSegmentTree(capacity)\n self.min_tree = MinSegmentTree(capacity)\n\n # For n-step returns\n self.n_step_buffer = deque(maxlen=n_step)\n\n def _get_n_step_info(self):\n reward = 0.0\n next_obs = self.n_step_buffer[-1][3]\n done = self.n_step_buffer[-1][4]\n\n for i in range(len(self.n_step_buffer)):\n reward += self.gamma**i * self.n_step_buffer[i][2]\n if self.n_step_buffer[i][4]:\n next_obs = self.n_step_buffer[i][3]\n done = True\n break\n return reward, next_obs, done\n\n def add(self, obs, action, reward, next_obs, done):\n self.n_step_buffer.append((obs, action, reward, next_obs, done))\n\n if len(self.n_step_buffer) < self.n_step:\n return\n\n reward, next_obs, done = self._get_n_step_info()\n obs = self.n_step_buffer[0][0]\n action = self.n_step_buffer[0][1]\n\n idx = self.pos\n self.buffer_obs[idx] = obs\n self.buffer_next_obs[idx] = next_obs\n self.buffer_actions[idx] = action\n self.buffer_rewards[idx] = reward\n self.buffer_dones[idx] = done\n\n priority = self.max_priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n self.pos = (self.pos + 1) % self.capacity\n self.size = min(self.size + 1, self.capacity)\n\n if done:\n self.n_step_buffer.clear()\n\n def sample(self, batch_size):\n indices = []\n p_total = self.sum_tree.total()\n segment = p_total / batch_size\n\n for i in range(batch_size):\n a = segment * i\n b = segment * (i + 1)\n upperbound = np.random.uniform(a, b)\n idx = self.sum_tree.retrieve(upperbound)\n indices.append(idx)\n\n samples = {\n ""observations"": torch.from_numpy(self.buffer_obs[indices]).to(self.device),\n ""actions"": torch.from_numpy(self.buffer_actions[indices])\n .to(self.device)\n .unsqueeze(1),\n ""rewards"": torch.from_numpy(self.buffer_rewards[indices])\n .to(self.device)\n .unsqueeze(1),\n ""next_observations"": torch.from_numpy(self.buffer_next_obs[indices]).to(\n self.device\n ),\n ""dones"": torch.from_numpy(self.buffer_dones[indices])\n .to(self.device)\n .unsqueeze(1),\n }\n\n probs = np.array(\n [self.sum_tree.tree[idx + self.capacity - 1] for idx in indices]\n )\n weights = (self.size * probs / p_total) ** -self.beta\n weights = weights / weights.max()\n samples[""weights""] = torch.from_numpy(weights).to(self.device).unsqueeze(1)\n samples[""indices""] = indices\n\n return PrioritizedBatch(**samples)\n\n def update_priorities(self, indices, priorities):\n priorities = np.abs(priorities) + self.eps\n self.max_priority = max(self.max_priority, priorities.max())\n\n for idx, priority in zip(indices, priorities):\n priority = priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n\nif __name__ == ""__main__"":\n args = tyro.cli(Args)\n assert args.num_envs == 1, ""vectorized envs are not supported at the moment""\n run_name = f""{args.env_id}__{args.exp_name}__{args.seed}__{int(time.time())}""\n if args.track:\n import wandb\n\n wandb.init(\n project=args.wandb_project_name,\n entity=args.wandb_entity,\n sync_tensorboard=True,\n config=vars(args),\n name=run_name,\n monitor_gym=True,\n save_code=True,\n )\n writer = SummaryWriter(f""runs/{run_name}"")\n writer.add_text(\n ""hyperparameters"",\n ""|param|value|\n|-|-|\n%s""\n % (""\n"".join([f""|{key}|{value}|"" for key, value in vars(args).items()])),\n )\n\n # TRY NOT TO MODIFY: seeding\n random.seed(args.seed)\n np.random.seed(args.seed)\n torch.manual_seed(args.seed)\n torch.backends.cudnn.deterministic = args.torch_deterministic\n\n device = torch.device(""cuda"" if torch.cuda.is_available() and args.cuda else ""cpu"")\n\n # env setup\n envs = gym.vector.SyncVectorEnv(\n [\n make_env(args.env_id, args.seed + i, i, args.capture_video, run_name)\n for i in range(args.num_envs)\n ]\n )\n assert isinstance(\n envs.single_action_space, gym.spaces.Discrete\n ), ""only discrete action space is supported""\n\n q_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n optimizer = optim.Adam(q_network.parameters(), lr=args.learning_rate, eps=1.5e-4)\n target_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n target_network.load_state_dict(q_network.state_dict())\n\n rb = PrioritizedReplayBuffer(\n args.buffer_size,\n envs.single_observation_space.shape,\n device,\n args.n_step,\n args.gamma,\n args.prioritized_replay_alpha,\n args.prioritized_replay_beta,\n args.prioritized_replay_eps,\n )\n\n # dataset capture state\n split_targets = {\n ""train"": args.num_episodes_train,\n ""val"": args.num_episodes_val,\n ""test"": args.num_episodes_test,\n }\n # Determine splits to run (order: train -> val -> test)\n splits_in_order = [s for s in [""train"", ""val"", ""test""] if split_targets[s] > 0]\n\n episodes_captured_per_split: dict[str, int] = {\n s: 0 for s in [""train"", ""val"", ""test""]\n }\n file_idx_by_split: dict[str, int] = {s: 0 for s in [""train"", ""val"", ""test""]}\n episode_metadata_by_split: dict[str, list[dict]] = {\n s: [] for s in [""train"", ""val"", ""test""]\n }\n\n obs_chunks: list[np.ndarray] = []\n act_chunks: list[np.ndarray] = []\n\n current_split_idx = 0\n current_split = splits_in_order[0]\n split_dir = os.path.join(args.output_dir, current_split)\n if args.capture_dataset:\n os.makedirs(split_dir, exist_ok=True)\n\n start_time = time.time()\n\n # TRY NOT TO MODIFY: start the game\n obs, _ = envs.reset(seed=args.seed)\n observations_seq: list[np.ndarray] = []\n actions_seq: list[np.ndarray] = []\n for global_step in range(args.total_timesteps):\n # anneal PER beta to 1\n rb.beta = min(\n 1.0,\n args.prioritized_replay_beta\n + global_step * (1.0 - args.prioritized_replay_beta) / args.total_timesteps,\n )\n\n # ALGO LOGIC: put action logic here\n with torch.no_grad():\n q_dist = q_network(torch.Tensor(obs).to(device))\n q_values = torch.sum(q_dist * q_network.support, dim=2)\n actions = torch.argmax(q_values, dim=1).cpu().numpy()\n\n # TRY NOT TO MODIFY: execute the game and log data.\n next_obs, rewards, terminations, truncations, infos = envs.step(actions)\n\n if args.capture_dataset:\n observations_seq.append(next_obs.astype(np.uint8))\n actions_seq.append(actions.astype(np.int64))\n\n if ""final_info"" in infos:\n for info in infos[""final_info""]:\n if info and ""episode"" in info:\n print(\n f""global_step={global_step}, episodic_return={info['episode']['r']}""\n )\n writer.add_scalar(\n ""charts/episodic_return"", info[""episode""][""r""], global_step\n )\n writer.add_scalar(\n ""charts/episodic_length"", info[""episode""][""l""], global_step\n )\n\n continue_capturing_multi = any(\n episodes_captured_per_split[s] < split_targets[s]\n for s in splits_in_order\n )\n if args.capture_dataset and continue_capturing_multi:\n current_len = len(observations_seq)\n if current_len >= args.min_episode_length:\n frames = np.concatenate(observations_seq, axis=0).astype(\n np.uint8\n )\n acts = np.concatenate(actions_seq, axis=0).astype(np.int64)\n\n episode_obs_chunks = []\n episode_act_chunks = []\n start_idx = 0\n while start_idx < current_len:\n end_idx = min(start_idx + args.chunk_size, current_len)\n if end_idx - start_idx < args.chunk_size:\n print(\n f""Warning: Inconsistent chunk_sizes. Episode has {current_len} frames, ""\n f""which is smaller than the requested chunk_size: {args.chunk_size}. ""\n ""This might lead to performance degradation during training.""\n )\n episode_obs_chunks.append(frames[start_idx:end_idx])\n episode_act_chunks.append(acts[start_idx:end_idx])\n start_idx = end_idx\n\n obs_chunks_data = [\n seq.astype(np.uint8) for seq in episode_obs_chunks\n ]\n act_chunks_data = [act for act in episode_act_chunks]\n obs_chunks.extend(obs_chunks_data)\n act_chunks.extend(act_chunks_data)\n\n # Save to the active split\n ep_metadata, obs_chunks, next_file_idx, act_chunks = (\n save_chunks(\n obs_chunks,\n file_idx_by_split[current_split],\n args.chunks_per_file,\n split_dir,\n act_chunks,\n )\n )\n file_idx_by_split[current_split] = next_file_idx\n episode_metadata_by_split[current_split].extend(ep_metadata)\n\n episodes_captured_per_split[current_split] += 1\n\n if (\n episodes_captured_per_split[current_split]\n >= split_targets[current_split]\n ):\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks before switching split '"",\n {current_split},\n ""' for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n obs_chunks = []\n act_chunks = []\n if current_split_idx + 1 < len(splits_in_order):\n current_split_idx += 1\n current_split = splits_in_order[current_split_idx]\n split_dir = os.path.join(\n args.output_dir, current_split\n )\n os.makedirs(split_dir, exist_ok=True)\n else:\n print(\n f""Episode too short ({current_len}), skipping capture...""\n )\n\n observations_seq = []\n actions_seq = []\n\n # TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`\n real_next_obs = next_obs.copy()\n for idx, trunc in enumerate(truncations):\n if trunc:\n real_next_obs[idx] = infos[""final_observation""][idx]\n rb.add(obs, actions, rewards, real_next_obs, terminations)\n\n # TRY NOT TO MODIFY: CRUCIAL step easy to overlook\n obs = next_obs\n\n # ALGO LOGIC: training.\n if global_step > args.learning_starts:\n if global_step % args.train_frequency == 0:\n # reset the noise for both networks\n q_network.reset_noise()\n target_network.reset_noise()\n data = rb.sample(args.batch_size)\n\n with torch.no_grad():\n next_dist = target_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n support = target_network.support # [n_atoms]\n next_q_values = torch.sum(\n next_dist * support, dim=2\n ) # [B, num_actions]\n\n # double q-learning\n next_dist_online = q_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n next_q_online = torch.sum(\n next_dist_online * support, dim=2\n ) # [B, num_actions]\n best_actions = torch.argmax(next_q_online, dim=1) # [B]\n next_pmfs = next_dist[\n torch.arange(args.batch_size), best_actions\n ] # [B, n_atoms]\n\n # compute the n-step Bellman update.\n gamma_n = args.gamma**args.n_step\n next_atoms = data.rewards + gamma_n * support * (\n 1 - data.dones.float()\n )\n tz = next_atoms.clamp(q_network.v_min, q_network.v_max)\n\n # projection\n delta_z = q_network.delta_z\n b = (tz - q_network.v_min) / delta_z # shape: [B, n_atoms]\n l = b.floor().clamp(0, args.n_atoms - 1)\n u = b.ceil().clamp(0, args.n_atoms - 1)\n\n # (l == u).float() handles the case where bj is exactly an integer\n # example bj = 1, then the upper ceiling should be uj= 2, and lj= 1\n d_m_l = (\n u.float() + (l == b).float() - b\n ) * next_pmfs # [B, n_atoms]\n d_m_u = (b - l) * next_pmfs # [B, n_atoms]\n\n target_pmfs = torch.zeros_like(next_pmfs)\n for i in range(target_pmfs.size(0)):\n target_pmfs[i].index_add_(0, l[i].long(), d_m_l[i])\n target_pmfs[i].index_add_(0, u[i].long(), d_m_u[i])\n\n dist = q_network(data.observations) # [B, num_actions, n_atoms]\n pred_dist = dist.gather(\n 1, data.actions.unsqueeze(-1).expand(-1, -1, args.n_atoms)\n ).squeeze(1)\n log_pred = torch.log(pred_dist.clamp(min=1e-5, max=1 - 1e-5))\n\n loss_per_sample = -(target_pmfs * log_pred).sum(dim=1)\n loss = (loss_per_sample * data.weights.squeeze()).mean()\n\n # update priorities\n new_priorities = loss_per_sample.detach().cpu().numpy()\n rb.update_priorities(data.indices, new_priorities)\n\n if global_step % 100 == 0:\n writer.add_scalar(""losses/td_loss"", loss.item(), global_step)\n q_values = (pred_dist * q_network.support).sum(dim=1) # [B]\n writer.add_scalar(\n ""losses/q_values"", q_values.mean().item(), global_step\n )\n sps = int(global_step / (time.time() - start_time))\n print(""SPS:"", sps)\n writer.add_scalar(""charts/SPS"", sps, global_step)\n writer.add_scalar(""charts/beta"", rb.beta, global_step)\n\n # optimize the model\n optimizer.zero_grad()\n loss.backward()\n optimizer.step()\n\n # update target network\n if global_step % args.target_network_frequency == 0:\n for target_param, param in zip(\n target_network.parameters(), q_network.parameters()\n ):\n target_param.data.copy_(\n args.tau * param.data + (1.0 - args.tau) * target_param.data\n )\n\n # optional early stop on dataset completion\n if args.capture_dataset and args.stop_on_complete:\n all_done = (\n all(\n episodes_captured_per_split[s] >= split_targets[s]\n for s in splits_in_order\n )\n and len(splits_in_order) > 0\n )\n if all_done:\n break\n\n envs.close()\n writer.close()\n\n # write metadata for dataset\n if args.capture_dataset:\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n\n os.makedirs(args.output_dir, exist_ok=True)\n metadata_path = os.path.join(args.output_dir, ""metadata.json"")\n if os.path.exists(metadata_path):\n try:\n with open(metadata_path, ""r"") as f:\n metadata = json.load(f)\n except Exception:\n metadata = {}\n else:\n metadata = {}\n\n metadata.setdefault(""env"", args.env_id)\n metadata.setdefault(""num_actions"", int(envs.single_action_space.n))\n for split in [""train"", ""val"", ""test""]:\n metadata.setdefault(f""num_episodes_{split}"", 0)\n metadata.setdefault(f""avg_episode_len_{split}"", 0.0)\n metadata.setdefault(f""episode_metadata_{split}"", [])\n\n for split_key in splits_in_order:\n ep_meta_list = episode_metadata_by_split[split_key]\n if ep_meta_list:\n metadata[f""episode_metadata_{split_key}""].extend(ep_meta_list)\n metadata[f""num_episodes_{split_key}""] = len(\n metadata[f""episode_metadata_{split_key}""]\n )\n metadata[f""avg_episode_len_{split_key}""] = float(\n np.mean(\n [\n ep[""avg_seq_len""]\n for ep in metadata[f""episode_metadata_{split_key}""]\n ]\n )\n )\n\n with open(metadata_path, ""w"") as f:\n json.dump(metadata, f)\n",python,tab +609,3267766,"input_pipeline/generate_atari_dataset.py",17725,0,"",python,selection_mouse +610,3271974,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +611,3272346,"genie.py",0,0,"",python,tab +612,3272346,"genie.py",5992,0,"",python,selection_command +613,3281713,"genie.py",0,0,"from typing import Dict\n\nimport einops\nimport jax\nimport jax.numpy as jnp\nimport flax.nnx as nnx\nimport orbax.checkpoint as ocp\n\nfrom models.dynamics import DynamicsMaskGIT, DynamicsCausal\nfrom models.lam import LatentActionModel\nfrom models.tokenizer import TokenizerVQVAE\n\n\nclass Genie(nnx.Module):\n """"""Genie model""""""\n\n def __init__(\n self,\n in_dim: int,\n tokenizer_dim: int,\n tokenizer_ffn_dim: int,\n latent_patch_dim: int,\n num_patch_latents: int,\n patch_size: int,\n tokenizer_num_blocks: int,\n tokenizer_num_heads: int,\n lam_dim: int,\n lam_ffn_dim: int,\n latent_action_dim: int,\n num_latent_actions: int,\n lam_patch_size: int,\n lam_num_blocks: int,\n lam_num_heads: int,\n lam_co_train: bool,\n use_gt_actions: bool,\n dyna_type: str,\n dyna_dim: int,\n dyna_ffn_dim: int,\n dyna_num_blocks: int,\n dyna_num_heads: int,\n param_dtype: jnp.dtype,\n dtype: jnp.dtype,\n use_flash_attention: bool,\n decode: bool,\n rngs: nnx.Rngs,\n dropout: float = 0.0,\n mask_limit: float = 0.0,\n ):\n # --- Tokenizer ---\n self.in_dim = in_dim\n self.tokenizer_dim = tokenizer_dim\n self.tokenizer_ffn_dim = tokenizer_ffn_dim\n self.latent_patch_dim = latent_patch_dim\n self.num_patch_latents = num_patch_latents\n self.patch_size = patch_size\n self.tokenizer_num_blocks = tokenizer_num_blocks\n self.tokenizer_num_heads = tokenizer_num_heads\n # --- LAM ---\n self.lam_dim = lam_dim\n self.lam_ffn_dim = lam_ffn_dim\n self.latent_action_dim = latent_action_dim\n self.num_latent_actions = num_latent_actions\n self.lam_patch_size = lam_patch_size\n self.lam_num_blocks = lam_num_blocks\n self.lam_num_heads = lam_num_heads\n self.lam_co_train = lam_co_train\n self.use_gt_actions = use_gt_actions\n # --- Dynamics ---\n self.dyna_type = dyna_type\n self.dyna_dim = dyna_dim\n self.dyna_ffn_dim = dyna_ffn_dim\n self.dyna_num_blocks = dyna_num_blocks\n self.dyna_num_heads = dyna_num_heads\n self.param_dtype = param_dtype\n self.dtype = dtype\n self.use_flash_attention = use_flash_attention\n self.dropout = dropout\n self.mask_limit = mask_limit\n self.decode = decode\n\n self.tokenizer = TokenizerVQVAE(\n in_dim=self.in_dim,\n model_dim=self.tokenizer_dim,\n ffn_dim=self.tokenizer_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_patch_latents,\n patch_size=self.patch_size,\n num_blocks=self.tokenizer_num_blocks,\n num_heads=self.tokenizer_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n if self.use_gt_actions:\n self.action_embed = nnx.Embed(\n self.num_latent_actions, self.latent_action_dim, rngs=rngs\n )\n self.lam = None\n else:\n self.lam = LatentActionModel(\n in_dim=self.in_dim,\n model_dim=self.lam_dim,\n ffn_dim=self.lam_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_latent_actions,\n patch_size=self.lam_patch_size,\n num_blocks=self.lam_num_blocks,\n num_heads=self.lam_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n self.action_embed = None\n if self.dyna_type == ""maskgit"":\n self.dynamics = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n elif self.dyna_type == ""causal"":\n self.dynamics = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=decode,\n rngs=rngs,\n )\n else:\n raise ValueError(f""Invalid dynamics type: {self.dyna_type}"")\n\n def __call__(\n self,\n batch: Dict[str, jax.Array],\n training: bool = True,\n ) -> Dict[str, jax.Array]:\n videos_BTHWC = batch[""videos""]\n tokenizer_outputs = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_indices_BTN = tokenizer_outputs[""indices""]\n latent_actions_BTm11L = None\n action_embeddings_BTm11L = None\n if self.use_gt_actions:\n assert self.action_embed is not None\n action_indices_E = None\n action_embeddings_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n action_embeddings_BTm11L = action_embeddings_BT1L[:, 1:]\n else:\n assert self.lam is not None\n lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n outputs = dict(\n video_tokens=jax.lax.stop_gradient(token_indices_BTN),\n latent_actions=(\n action_embeddings_BTm11L\n if self.use_gt_actions\n else latent_actions_BTm11L\n ),\n )\n outputs[""mask_rng""] = batch[""rng""]\n dyna_logits_BTNV, dyna_mask = self.dynamics(outputs, training)\n outputs[""token_logits""] = dyna_logits_BTNV\n outputs[""mask""] = dyna_mask\n mle_indices_BTN = jnp.argmax(outputs[""token_logits""], axis=-1)\n H, W = batch[""videos""].shape[2:4]\n outputs[""recon""] = self.tokenizer.decode(mle_indices_BTN, (H, W))\n if action_indices_E is not None:\n outputs[""lam_indices""] = action_indices_E\n return outputs\n\n def sample(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n maskgit_steps: int = 25,\n ) -> tuple[jax.Array, jax.Array]:\n if self.dyna_type == ""maskgit"":\n return self.sample_maskgit(\n batch, seq_len, maskgit_steps, temperature, sample_argmax\n )\n elif self.dyna_type == ""causal"":\n return self.sample_causal(batch, seq_len, temperature, sample_argmax)\n else:\n raise ValueError(f""Dynamics model type unknown: {self.dyna_type}"")\n\n def sample_maskgit(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n steps: int = 25,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n P: S * N\n """"""\n assert isinstance(self.dynamics, DynamicsMaskGIT)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n init_logits_BSNV = jnp.zeros(\n shape=(*token_idxs_BSN.shape, self.num_patch_latents)\n )\n if self.use_gt_actions:\n assert self.action_embed is not None\n latent_actions_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n latent_actions_BTm11L = latent_actions_BT1L[:, 1:]\n action_tokens_EL = latent_actions_BTm11L.reshape(-1, self.latent_action_dim)\n else:\n assert self.lam is not None\n latent_actions_E = batch[""latent_actions""]\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n # --- Extract submodule state ---\n dynamics_state = nnx.state(self.dynamics)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def maskgit_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, mask_BSN, action_tokens_EL = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_maskgit = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_maskgit, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_maskgit.patch_embed(token_idxs_BSN)\n mask_token_111M = dynamics_maskgit.mask_token.value\n mask_expanded_BSN1 = mask_BSN[..., None]\n vid_embed_BSNM = jnp.where(\n mask_expanded_BSN1, mask_token_111M, vid_embed_BSNM\n )\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_maskgit.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNM += act_embed_BS1M\n unmasked_ratio = jnp.cos(jnp.pi * (step + 1) / (steps * 2))\n step_temp = temperature * (1.0 - unmasked_ratio)\n final_logits_BSNV = dynamics_maskgit.transformer(vid_embed_BSNM) / step_temp\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_BSN = jnp.argmax(final_logits_BSNV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_BSN = jax.random.categorical(_rng, final_logits_BSNV)\n gather_fn = jax.vmap(jax.vmap(jax.vmap(lambda x, y: x[y])))\n final_token_probs_BSN = gather_fn(\n jax.nn.softmax(final_logits_BSNV), sampled_token_idxs_BSN\n )\n final_token_probs_BSN += ~mask_BSN\n # Update masked tokens and logits only\n token_idxs_BSN = jnp.where(mask_BSN, sampled_token_idxs_BSN, token_idxs_BSN)\n logits_BSNV = jnp.where(\n jnp.expand_dims(mask_BSN, -1), final_logits_BSNV, logits_BSNV\n )\n\n # --- Update mask ---\n num_unmasked_tokens = jnp.round(N * (1.0 - unmasked_ratio)).astype(int)\n final_token_probs_flat_BP = einops.rearrange(\n final_token_probs_BSN, ""b s n -> b (s n)""\n )\n idx_mask_P = (\n jnp.arange(final_token_probs_flat_BP.shape[-1])\n <= N - num_unmasked_tokens\n )\n sorted_idxs_BP = jnp.argsort(final_token_probs_flat_BP, axis=-1)\n mask_update_fn = jax.vmap(lambda msk, ids: msk.at[ids].set(idx_mask_P))\n mask_flat_BP = einops.rearrange(mask_BSN, ""b s n -> b (s n)"")\n new_mask_flat_BP = mask_update_fn(mask_flat_BP, sorted_idxs_BP)\n new_mask_BSN = einops.rearrange(new_mask_flat_BP, ""b (s n) -> b s n"", n=N)\n\n new_carry = (\n rng,\n token_idxs_BSN,\n logits_BSNV,\n new_mask_BSN,\n action_tokens_EL,\n )\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # Mask current frame (i.e., t == step_t)\n mask_S = jnp.arange(seq_len) == step_t\n mask_BSN = jnp.broadcast_to(mask_S[None, :, None], (B, seq_len, N)).astype(\n bool\n )\n masked_token_idxs_BSN = current_token_idxs_BSN * ~mask_BSN\n masked_logits_BSNV = current_logits_BSNV * jnp.expand_dims(~mask_BSN, -1)\n\n # --- Initialize and run MaskGIT loop ---\n init_carry_maskgit = (\n step_rng,\n masked_token_idxs_BSN,\n masked_logits_BSNV,\n mask_BSN,\n action_tokens_EL,\n )\n final_carry_maskgit = maskgit_step_fn(init_carry_maskgit, jnp.arange(steps))\n updated_token_idxs_BSN = final_carry_maskgit[1]\n updated_logits_BSNV = final_carry_maskgit[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, init_logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def sample_causal(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n """"""\n assert isinstance(self.dynamics, DynamicsCausal)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n logits_BSNV = jnp.zeros((*token_idxs_BSN.shape, self.num_patch_latents))\n dynamics_state = nnx.state(self.dynamics)\n\n if self.use_gt_actions:\n assert self.action_embed is not None\n latent_actions_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n latent_actions_BTm11L = latent_actions_BT1L[:, 1:]\n action_tokens_EL = latent_actions_BTm11L.reshape(-1, self.latent_action_dim)\n else:\n assert self.lam is not None\n latent_actions_E = batch[""latent_actions""]\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def causal_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step_n: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_causal = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=self.decode,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_causal, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_causal.patch_embed(token_idxs_BSN)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_causal.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNp1M = jnp.concatenate([act_embed_BS1M, vid_embed_BSNM], axis=2)\n final_logits_BTNp1V = (\n dynamics_causal.transformer(vid_embed_BSNp1M, (step_t, step_n))\n / temperature\n )\n final_logits_BV = final_logits_BTNp1V[:, step_t, step_n, :]\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_B = jnp.argmax(final_logits_BV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_B = jax.random.categorical(_rng, final_logits_BV)\n # Update next tokens only\n token_idxs_BSN = token_idxs_BSN.at[:, step_t, step_n].set(\n sampled_token_idxs_B\n )\n logits_BSNV = logits_BSNV.at[:, step_t, step_n].set(final_logits_BV)\n\n new_carry = (rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t)\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # --- Initialize and run causal loop ---\n init_carry_causal = (\n step_rng,\n current_token_idxs_BSN,\n current_logits_BSNV,\n action_tokens_EL,\n step_t,\n )\n final_carry_causal = causal_step_fn(init_carry_causal, jnp.arange(N))\n updated_token_idxs_BSN = final_carry_causal[1]\n updated_logits_BSNV = final_carry_causal[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def vq_encode(self, batch: Dict[str, jax.Array], training: bool) -> jax.Array:\n # --- Preprocess videos ---\n assert self.lam is not None\n video_BTHWC = batch[""videos""]\n lam_output: Dict[str, jax.Array] = self.lam.vq_encode(\n video_BTHWC, training=training\n )\n lam_indices_E = lam_output[""indices""]\n return lam_indices_E\n\n\n# FIXME (f.srambical): add conversion script for old checkpoints\ndef restore_genie_components(\n optimizer: nnx.Optimizer,\n sharding: jax.sharding.NamedSharding,\n rng: jax.Array,\n args,\n) -> nnx.Optimizer:\n """"""Restore pre-trained Genie components""""""\n rng_tokenizer, rng_lam = jax.random.split(rng)\n rngs_tokenizer = nnx.Rngs(rng_tokenizer)\n rngs_lam = nnx.Rngs(rng_lam)\n\n tx = optimizer.tx\n model = optimizer.model\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n step_format_fixed_length=6,\n )\n tokenizer_checkpoint_manager = ocp.CheckpointManager(\n directory=args.tokenizer_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_tokenizer = TokenizerVQVAE(\n in_dim=args.image_channels,\n model_dim=args.tokenizer_dim,\n ffn_dim=args.tokenizer_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n num_blocks=args.tokenizer_num_blocks,\n num_heads=args.tokenizer_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs_tokenizer,\n )\n dummy_tokenizer_optimizer = nnx.Optimizer(dummy_tokenizer, tx)\n dummy_tokenizer_optimizer_state = nnx.state(dummy_tokenizer_optimizer)\n abstract_sharded_tokenizer_optimizer_state = _create_abstract_sharded_pytree(\n dummy_tokenizer_optimizer_state, sharding\n )\n restored_tokenizer = tokenizer_checkpoint_manager.restore(\n step=tokenizer_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_tokenizer_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_tokenizer_optimizer.model, restored_tokenizer.model)\n model.tokenizer = dummy_tokenizer_optimizer.model\n tokenizer_checkpoint_manager.close()\n\n if args.lam_checkpoint:\n lam_checkpoint_manager = ocp.CheckpointManager(\n directory=args.lam_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_lam = LatentActionModel(\n in_dim=args.image_channels,\n model_dim=args.lam_dim,\n ffn_dim=args.lam_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_latent_actions,\n patch_size=args.lam_patch_size,\n num_blocks=args.lam_num_blocks,\n num_heads=args.lam_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs_lam,\n )\n dummy_lam_optimizer = nnx.Optimizer(dummy_lam, tx)\n dummy_lam_optimizer_state = nnx.state(dummy_lam_optimizer)\n abstract_sharded_lam_optimizer_state = _create_abstract_sharded_pytree(\n dummy_lam_optimizer_state, sharding\n )\n restored_lam_optimizer = lam_checkpoint_manager.restore(\n step=lam_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_lam_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_lam_optimizer.model, restored_lam_optimizer.model)\n model.lam = dummy_lam_optimizer.model\n # Remove the LAM decoder to save memory and avoid unnecessary computation.\n del model.lam.decoder\n lam_checkpoint_manager.close()\n\n # Reinitialize the optimizer states\n optimizer = nnx.Optimizer(model, tx)\n return optimizer\n\n\ndef _create_abstract_sharded_pytree(\n pytree_template: nnx.GraphState, sharding_spec: jax.sharding.NamedSharding\n) -> jax.Array:\n """"""Replaces arrays in a pytree with ShapeDtypeStructs having the given sharding.""""""\n\n def map_fn(leaf_template):\n if hasattr(leaf_template, ""shape"") and hasattr(leaf_template, ""dtype""):\n return jax.ShapeDtypeStruct(\n leaf_template.shape, leaf_template.dtype, sharding=sharding_spec\n )\n return leaf_template\n\n return jax.tree_util.tree_map(map_fn, pytree_template)\n",python,tab +614,3281713,"genie.py",6467,0,"",python,selection_mouse +615,3281724,"genie.py",6466,0,"",python,selection_command +616,3282767,"genie.py",0,0,"",python,tab +617,3282768,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +618,3298042,"TERMINAL",0,0,"git stash",,terminal_command +619,3298043,"TERMINAL",0,0,"]633;C",,terminal_output +620,3298260,"TERMINAL",0,0,"Saved working directory and index state WIP on gt-actions: f677f49 feat: trajectory collection during rainbow training\r\n",,terminal_output +621,3298306,"TERMINAL",0,0,"]0;franz.srambical@hai-login2:~/jafar",,terminal_output +622,3299066,"input_pipeline/generate_atari_dataset.py",17759,3294," observations_seq.append(next_obs.astype(np.uint8))\n actions_seq.append(actions.astype(np.int64))\n\n if ""final_info"" in infos:\n for info in infos[""final_info""]:\n if info and ""episode"" in info:\n print(\n f""global_step={global_step}, episodic_return={info['episode']['r']}""\n )\n writer.add_scalar(\n ""charts/episodic_return"", info[""episode""][""r""], global_step\n )\n writer.add_scalar(\n ""charts/episodic_length"", info[""episode""][""l""], global_step\n )\n\n continue_capturing_multi = any(\n episodes_captured_per_split[s] < split_targets[s]\n for s in splits_in_order\n )\n if args.capture_dataset and continue_capturing_multi:\n current_len = len(observations_seq)\n if current_len >= args.min_episode_length:\n frames = np.concatenate(observations_seq, axis=0).astype(\n np.uint8\n )\n acts = np.concatenate(actions_seq, axis=0).astype(np.int64)\n\n episode_obs_chunks = []\n episode_act_chunks = []\n start_idx = 0\n while start_idx < current_len:\n end_idx = min(start_idx + args.chunk_size, current_len)\n if end_idx - start_idx < args.chunk_size:\n print(\n f""Warning: Inconsistent chunk_sizes. Episode has {current_len} frames, ""\n f""which is smaller than the requested chunk_size: {args.chunk_size}. ""\n ""This might lead to performance degradation during training.""\n )\n episode_obs_chunks.append(frames[start_idx:end_idx])\n episode_act_chunks.append(acts[start_idx:end_idx])\n start_idx = end_idx\n\n obs_chunks_data = [\n seq.astype(np.uint8) for seq in episode_obs_chunks\n ]\n act_chunks_data = [act for act in episode_act_chunks]\n obs_chunks.extend(obs_chunks_data)\n act_chunks.extend(act_chunks_data)\n\n # Save to the active split\n ep_metadata, obs_chunks, next_file_idx, act_chunks = (\n save_chunks(\n obs_chunks,\n file_idx_by_split[current_split],\n args.chunks_per_file,\n split_dir,\n act_chunks,\n )\n )\n file_idx_by_split[current_split] = next_file_idx\n",python,content +623,3303459,"input_pipeline/generate_atari_dataset.py",24085,1,"c",python,selection_command +624,3304164,"input_pipeline/generate_atari_dataset.py",24085,0,"",python,selection_command +625,3311640,"input_pipeline/generate_atari_dataset.py",24122,0,"",python,selection_command +626,3313840,"input_pipeline/generate_atari_dataset.py",0,0,"Switched from branch 'gt-actions' to 'generate-atari-dataset'",python,git_branch_checkout +627,3327704,"TERMINAL",0,0,"git stash pop",,terminal_command +628,3327705,"TERMINAL",0,0,"]633;C",,terminal_output +629,3327705,"TERMINAL",0,0,"Auto-merging genie.py\r\nCONFLICT (content): Merge conflict in genie.py\r\nOn branch generate-atari-dataset\r\nYour branch is up to date with 'origin/generate-atari-dataset'.\r\n\r\nChanges to be committed:\r\n (use ""git restore --staged ..."" to unstage)\r\n\tmodified: input_pipeline/generate_atari_dataset.py\r\n\r\nUnmerged paths:\r\n (use ""git restore --staged ..."" to unstage)\r\n (use ""git add ..."" to mark resolution)\r\n\tboth modified: genie.py\r\n\r\nUntracked files:\r\n (use ""git add ..."" to include in what will be committed)\r\n\tdata_arrayrecord/\r\n\texperiments/\r\n\tfreeze.freeze\r\n\truns/\r\n\tslurm/\r\n\ttensorboard/\r\n\ttest/\r\n\r\nThe stash entry is kept in case you need it again.\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +630,3328982,"input_pipeline/generate_atari_dataset.py",17759,3284," observations_seq.append(obs.astype(np.uint8))\n actions_seq.append(actions.astype(np.int64))\n\n if ""final_info"" in infos:\n for info in infos[""final_info""]:\n if info and ""episode"" in info:\n print(\n f""global_step={global_step}, episodic_return={info['episode']['r']}""\n )\n writer.add_scalar(\n ""charts/episodic_return"", info[""episode""][""r""], global_step\n )\n writer.add_scalar(\n ""charts/episodic_length"", info[""episode""][""l""], global_step\n )\n\n continue_capturing_multi = any(\n episodes_captured_per_split[s] < split_targets[s]\n for s in splits_in_order\n )\n if args.capture_dataset and continue_capturing_multi:\n current_len = len(observations_seq)\n if current_len >= args.min_episode_length:\n frames = np.concatenate(observations_seq, axis=0).astype(\n np.uint8\n )\n acts = np.concatenate(actions_seq, axis=0).astype(np.int64)\n\n episode_obs_chunks = []\n episode_act_chunks = []\n start_idx = 0\n while start_idx < current_len:\n end_idx = min(start_idx + args.chunk_size, current_len)\n if end_idx - start_idx < args.chunk_size:\n print(\n f""Warning: Inconsistent chunk_sizes. Episode has {current_len} frames, ""\n f""which is smaller than the requested chunk_size: {args.chunk_size}. ""\n ""This might lead to performance degradation during training.""\n )\n episode_obs_chunks.append(frames[start_idx:end_idx])\n episode_act_chunks.append(acts[start_idx:end_idx])\n start_idx = end_idx\n\n obs_chunks_data = [\n seq.astype(np.uint8) for seq in episode_obs_chunks\n ]\n act_chunks_data = [act for act in episode_act_chunks]\n obs_chunks.extend(obs_chunks_data)\n act_chunks.extend(act_chunks_data)\n\n # Save to the active split\n (\n ep_metadata,\n file_idx_by_split[current_split],\n obs_chunks,\n act_chunks,\n ) = save_chunks(\n file_idx_by_split[current_split],\n args.chunks_per_file,\n split_dir,\n obs_chunks,\n act_chunks,\n )\n",python,content +631,3338270,"genie.py",0,0,"from typing import Dict\n\nimport einops\nimport jax\nimport jax.numpy as jnp\nimport flax.nnx as nnx\nimport orbax.checkpoint as ocp\n\nfrom models.dynamics import DynamicsMaskGIT, DynamicsCausal\nfrom models.lam import LatentActionModel\nfrom models.tokenizer import TokenizerVQVAE\n\n\nclass Genie(nnx.Module):\n """"""Genie model""""""\n\n def __init__(\n self,\n in_dim: int,\n tokenizer_dim: int,\n tokenizer_ffn_dim: int,\n latent_patch_dim: int,\n num_patch_latents: int,\n patch_size: int,\n tokenizer_num_blocks: int,\n tokenizer_num_heads: int,\n lam_dim: int,\n lam_ffn_dim: int,\n latent_action_dim: int,\n num_latent_actions: int,\n lam_patch_size: int,\n lam_num_blocks: int,\n lam_num_heads: int,\n lam_co_train: bool,\n dyna_type: str,\n dyna_dim: int,\n dyna_ffn_dim: int,\n dyna_num_blocks: int,\n dyna_num_heads: int,\n param_dtype: jnp.dtype,\n dtype: jnp.dtype,\n use_flash_attention: bool,\n decode: bool,\n rngs: nnx.Rngs,\n dropout: float = 0.0,\n mask_limit: float = 0.0,\n ):\n # --- Tokenizer ---\n self.in_dim = in_dim\n self.tokenizer_dim = tokenizer_dim\n self.tokenizer_ffn_dim = tokenizer_ffn_dim\n self.latent_patch_dim = latent_patch_dim\n self.num_patch_latents = num_patch_latents\n self.patch_size = patch_size\n self.tokenizer_num_blocks = tokenizer_num_blocks\n self.tokenizer_num_heads = tokenizer_num_heads\n # --- LAM ---\n self.lam_dim = lam_dim\n self.lam_ffn_dim = lam_ffn_dim\n self.latent_action_dim = latent_action_dim\n self.num_latent_actions = num_latent_actions\n self.lam_patch_size = lam_patch_size\n self.lam_num_blocks = lam_num_blocks\n self.lam_num_heads = lam_num_heads\n self.lam_co_train = lam_co_train\n # --- Dynamics ---\n self.dyna_type = dyna_type\n self.dyna_dim = dyna_dim\n self.dyna_ffn_dim = dyna_ffn_dim\n self.dyna_num_blocks = dyna_num_blocks\n self.dyna_num_heads = dyna_num_heads\n self.param_dtype = param_dtype\n self.dtype = dtype\n self.use_flash_attention = use_flash_attention\n self.dropout = dropout\n self.mask_limit = mask_limit\n self.decode = decode\n\n self.tokenizer = TokenizerVQVAE(\n in_dim=self.in_dim,\n model_dim=self.tokenizer_dim,\n ffn_dim=self.tokenizer_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_patch_latents,\n patch_size=self.patch_size,\n num_blocks=self.tokenizer_num_blocks,\n num_heads=self.tokenizer_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n self.lam = LatentActionModel(\n in_dim=self.in_dim,\n model_dim=self.lam_dim,\n ffn_dim=self.lam_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_latent_actions,\n patch_size=self.lam_patch_size,\n num_blocks=self.lam_num_blocks,\n num_heads=self.lam_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n if self.dyna_type == ""maskgit"":\n self.dynamics = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n elif self.dyna_type == ""causal"":\n self.dynamics = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=decode,\n rngs=rngs,\n )\n else:\n raise ValueError(f""Invalid dynamics type: {self.dyna_type}"")\n\n def __call__(\n self,\n batch: Dict[str, jax.Array],\n training: bool = True,\n ) -> Dict[str, jax.Array]:\n videos_BTHWC = batch[""videos""]\n tokenizer_outputs = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_indices_BTN = tokenizer_outputs[""indices""]\n<<<<<<< Updated upstream\n lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n=======\n latent_actions_BTm11L = None\n action_embeddings_BTm11L = None\n if self.use_gt_actions:\n assert self.action_embed is not None\n action_indices_E = None\n action_embeddings_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n action_embeddings_BTm11L = action_embeddings_BT1L[:, :-1]\n else:\n assert self.lam is not None\n lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n>>>>>>> Stashed changes\n outputs = dict(\n video_tokens=jax.lax.stop_gradient(token_indices_BTN),\n latent_actions=latent_actions_BTm11L,\n )\n outputs[""mask_rng""] = batch[""rng""]\n dyna_logits_BTNV, dyna_mask = self.dynamics(outputs, training)\n outputs[""token_logits""] = dyna_logits_BTNV\n if dyna_mask is not None:\n outputs[""mask""] = dyna_mask\n mle_indices_BTN = jnp.argmax(outputs[""token_logits""], axis=-1)\n H, W = batch[""videos""].shape[2:4]\n outputs[""recon""] = self.tokenizer.decode(mle_indices_BTN, (H, W))\n outputs[""lam_indices""] = action_indices_E\n return outputs\n\n def sample(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n maskgit_steps: int = 25,\n ) -> tuple[jax.Array, jax.Array]:\n if self.dyna_type == ""maskgit"":\n return self.sample_maskgit(\n batch, seq_len, maskgit_steps, temperature, sample_argmax\n )\n elif self.dyna_type == ""causal"":\n return self.sample_causal(batch, seq_len, temperature, sample_argmax)\n else:\n raise ValueError(f""Dynamics model type unknown: {self.dyna_type}"")\n\n def sample_maskgit(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n steps: int = 25,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n P: S * N\n """"""\n assert isinstance(self.dynamics, DynamicsMaskGIT)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n latent_actions_E = batch[""latent_actions""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n init_logits_BSNV = jnp.zeros(\n shape=(*token_idxs_BSN.shape, self.num_patch_latents)\n )\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n # --- Extract submodule state ---\n dynamics_state = nnx.state(self.dynamics)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def maskgit_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, mask_BSN, action_tokens_EL = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_maskgit = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_maskgit, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_maskgit.patch_embed(token_idxs_BSN)\n mask_token_111M = dynamics_maskgit.mask_token.value\n mask_expanded_BSN1 = mask_BSN[..., None]\n vid_embed_BSNM = jnp.where(\n mask_expanded_BSN1, mask_token_111M, vid_embed_BSNM\n )\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_maskgit.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNM += act_embed_BS1M\n unmasked_ratio = jnp.cos(jnp.pi * (step + 1) / (steps * 2))\n step_temp = temperature * (1.0 - unmasked_ratio)\n final_logits_BSNV = dynamics_maskgit.transformer(vid_embed_BSNM) / step_temp\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_BSN = jnp.argmax(final_logits_BSNV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_BSN = jax.random.categorical(_rng, final_logits_BSNV)\n gather_fn = jax.vmap(jax.vmap(jax.vmap(lambda x, y: x[y])))\n final_token_probs_BSN = gather_fn(\n jax.nn.softmax(final_logits_BSNV), sampled_token_idxs_BSN\n )\n final_token_probs_BSN += ~mask_BSN\n # Update masked tokens and logits only\n token_idxs_BSN = jnp.where(mask_BSN, sampled_token_idxs_BSN, token_idxs_BSN)\n logits_BSNV = jnp.where(\n jnp.expand_dims(mask_BSN, -1), final_logits_BSNV, logits_BSNV\n )\n\n # --- Update mask ---\n num_unmasked_tokens = jnp.round(N * (1.0 - unmasked_ratio)).astype(int)\n final_token_probs_flat_BP = einops.rearrange(\n final_token_probs_BSN, ""b s n -> b (s n)""\n )\n idx_mask_P = (\n jnp.arange(final_token_probs_flat_BP.shape[-1])\n <= N - num_unmasked_tokens\n )\n sorted_idxs_BP = jnp.argsort(final_token_probs_flat_BP, axis=-1)\n mask_update_fn = jax.vmap(lambda msk, ids: msk.at[ids].set(idx_mask_P))\n mask_flat_BP = einops.rearrange(mask_BSN, ""b s n -> b (s n)"")\n new_mask_flat_BP = mask_update_fn(mask_flat_BP, sorted_idxs_BP)\n new_mask_BSN = einops.rearrange(new_mask_flat_BP, ""b (s n) -> b s n"", n=N)\n\n new_carry = (\n rng,\n token_idxs_BSN,\n logits_BSNV,\n new_mask_BSN,\n action_tokens_EL,\n )\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # Mask current frame (i.e., t == step_t)\n mask_S = jnp.arange(seq_len) == step_t\n mask_BSN = jnp.broadcast_to(mask_S[None, :, None], (B, seq_len, N)).astype(\n bool\n )\n masked_token_idxs_BSN = current_token_idxs_BSN * ~mask_BSN\n masked_logits_BSNV = current_logits_BSNV * jnp.expand_dims(~mask_BSN, -1)\n\n # --- Initialize and run MaskGIT loop ---\n init_carry_maskgit = (\n step_rng,\n masked_token_idxs_BSN,\n masked_logits_BSNV,\n mask_BSN,\n action_tokens_EL,\n )\n final_carry_maskgit = maskgit_step_fn(init_carry_maskgit, jnp.arange(steps))\n updated_token_idxs_BSN = final_carry_maskgit[1]\n updated_logits_BSNV = final_carry_maskgit[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, init_logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def sample_causal(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n """"""\n assert isinstance(self.dynamics, DynamicsCausal)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n latent_actions_E = batch[""latent_actions""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n logits_BSNV = jnp.zeros((*token_idxs_BSN.shape, self.num_patch_latents))\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n dynamics_state = nnx.state(self.dynamics)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def causal_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step_n: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_causal = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=self.decode,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_causal, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_causal.patch_embed(token_idxs_BSN)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_causal.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNp1M = jnp.concatenate([act_embed_BS1M, vid_embed_BSNM], axis=2)\n final_logits_BTNp1V = (\n dynamics_causal.transformer(vid_embed_BSNp1M, (step_t, step_n))\n / temperature\n )\n final_logits_BV = final_logits_BTNp1V[:, step_t, step_n, :]\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_B = jnp.argmax(final_logits_BV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_B = jax.random.categorical(_rng, final_logits_BV)\n # Update next tokens only\n token_idxs_BSN = token_idxs_BSN.at[:, step_t, step_n].set(\n sampled_token_idxs_B\n )\n logits_BSNV = logits_BSNV.at[:, step_t, step_n].set(final_logits_BV)\n\n new_carry = (rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t)\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # --- Initialize and run causal loop ---\n init_carry_causal = (\n step_rng,\n current_token_idxs_BSN,\n current_logits_BSNV,\n action_tokens_EL,\n step_t,\n )\n final_carry_causal = causal_step_fn(init_carry_causal, jnp.arange(N))\n updated_token_idxs_BSN = final_carry_causal[1]\n updated_logits_BSNV = final_carry_causal[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def vq_encode(self, batch: Dict[str, jax.Array], training: bool) -> jax.Array:\n # --- Preprocess videos ---\n video_BTHWC = batch[""videos""]\n lam_output: Dict[str, jax.Array] = self.lam.vq_encode(\n video_BTHWC, training=training\n )\n lam_indices_E = lam_output[""indices""]\n return lam_indices_E\n\n\n# FIXME (f.srambical): add conversion script for old checkpoints\ndef restore_genie_components(\n optimizer: nnx.Optimizer,\n sharding: jax.sharding.NamedSharding,\n rng: jax.Array,\n args,\n) -> nnx.Optimizer:\n """"""Restore pre-trained Genie components""""""\n rng_tokenizer, rng_lam = jax.random.split(rng)\n rngs_tokenizer = nnx.Rngs(rng_tokenizer)\n rngs_lam = nnx.Rngs(rng_lam)\n\n tx = optimizer.tx\n model = optimizer.model\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n step_format_fixed_length=6,\n )\n tokenizer_checkpoint_manager = ocp.CheckpointManager(\n directory=args.tokenizer_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_tokenizer = TokenizerVQVAE(\n in_dim=args.image_channels,\n model_dim=args.tokenizer_dim,\n ffn_dim=args.tokenizer_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n num_blocks=args.tokenizer_num_blocks,\n num_heads=args.tokenizer_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs_tokenizer,\n )\n dummy_tokenizer_optimizer = nnx.Optimizer(dummy_tokenizer, tx)\n dummy_tokenizer_optimizer_state = nnx.state(dummy_tokenizer_optimizer)\n abstract_sharded_tokenizer_optimizer_state = _create_abstract_sharded_pytree(\n dummy_tokenizer_optimizer_state, sharding\n )\n restored_tokenizer = tokenizer_checkpoint_manager.restore(\n step=tokenizer_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_tokenizer_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_tokenizer_optimizer.model, restored_tokenizer.model)\n model.tokenizer = dummy_tokenizer_optimizer.model\n tokenizer_checkpoint_manager.close()\n\n if args.lam_checkpoint:\n lam_checkpoint_manager = ocp.CheckpointManager(\n directory=args.lam_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_lam = LatentActionModel(\n in_dim=args.image_channels,\n model_dim=args.lam_dim,\n ffn_dim=args.lam_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_latent_actions,\n patch_size=args.lam_patch_size,\n num_blocks=args.lam_num_blocks,\n num_heads=args.lam_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs_lam,\n )\n dummy_lam_optimizer = nnx.Optimizer(dummy_lam, tx)\n dummy_lam_optimizer_state = nnx.state(dummy_lam_optimizer)\n abstract_sharded_lam_optimizer_state = _create_abstract_sharded_pytree(\n dummy_lam_optimizer_state, sharding\n )\n restored_lam_optimizer = lam_checkpoint_manager.restore(\n step=lam_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_lam_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_lam_optimizer.model, restored_lam_optimizer.model)\n model.lam = dummy_lam_optimizer.model\n # Remove the LAM decoder to save memory and avoid unnecessary computation.\n del model.lam.decoder\n lam_checkpoint_manager.close()\n\n # Reinitialize the optimizer states\n optimizer = nnx.Optimizer(model, tx)\n return optimizer\n\n\ndef _create_abstract_sharded_pytree(\n pytree_template: nnx.GraphState, sharding_spec: jax.sharding.NamedSharding\n) -> jax.Array:\n """"""Replaces arrays in a pytree with ShapeDtypeStructs having the given sharding.""""""\n\n def map_fn(leaf_template):\n if hasattr(leaf_template, ""shape"") and hasattr(leaf_template, ""dtype""):\n return jax.ShapeDtypeStruct(\n leaf_template.shape, leaf_template.dtype, sharding=sharding_spec\n )\n return leaf_template\n\n return jax.tree_util.tree_map(map_fn, pytree_template)\n",python,tab +632,3343614,"genie.py",5608,0,"",python,selection_mouse +633,3343614,"genie.py",5607,0,"",python,selection_command +634,3356072,"genie.py",5129,0,"",python,selection_mouse +635,3358150,"genie.py",5249,1244," lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n",python,content +636,3359943,"genie.py",5500,0,"",python,selection_mouse +637,3364592,"genie.py",999,0,"",python,selection_mouse +638,3381042,"genie.py",0,0,"",python,tab +639,3384655,"genie.py",0,0,"",python,tab +640,3393696,"input_pipeline/generate_atari_dataset.py",0,0,"# adapted from https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/rainbow_atari.py\nimport collections\nimport math\nimport os\nimport random\nimport time\nfrom collections import deque\nfrom dataclasses import dataclass\n\nimport gymnasium as gym\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport tyro\nfrom typing import Optional, Any\nfrom torch.utils.tensorboard.writer import SummaryWriter\n\nfrom cleanrl_utils.atari_wrappers import (\n ClipRewardEnv,\n EpisodicLifeEnv,\n FireResetEnv,\n MaxAndSkipEnv,\n NoopResetEnv,\n)\n\ntry:\n from utils import save_chunks # type: ignore\nexcept Exception: # pragma: no cover\n from input_pipeline.utils import save_chunks # type: ignore\nimport json\n\n\n@dataclass\nclass Args:\n exp_name: str = os.path.basename(__file__)[: -len("".py"")]\n """"""the name of this experiment""""""\n seed: int = 1\n """"""seed of the experiment""""""\n torch_deterministic: bool = True\n """"""if toggled, `torch.backends.cudnn.deterministic=False`""""""\n cuda: bool = True\n """"""if toggled, cuda will be enabled by default""""""\n track: bool = False\n """"""if toggled, this experiment will be tracked with Weights and Biases""""""\n wandb_project_name: str = ""cleanRL""\n """"""the wandb's project name""""""\n wandb_entity: Optional[str] = None\n """"""the entity (team) of wandb's project""""""\n capture_video: bool = False\n """"""whether to capture videos of the agent performances (check out `videos` folder)""""""\n save_model: bool = False\n """"""whether to save model into the `runs/{run_name}` folder""""""\n upload_model: bool = False\n """"""whether to upload the saved model to huggingface""""""\n hf_entity: str = """"\n """"""the user or org name of the model repository from the Hugging Face Hub""""""\n\n env_id: str = ""BreakoutNoFrameskip-v4""\n """"""the id of the environment""""""\n total_timesteps: int = 10000000\n """"""total timesteps of the experiments""""""\n learning_rate: float = 0.0000625\n """"""the learning rate of the optimizer""""""\n num_envs: int = 1\n """"""the number of parallel game environments""""""\n buffer_size: int = 1000000\n """"""the replay memory buffer size""""""\n gamma: float = 0.99\n """"""the discount factor gamma""""""\n tau: float = 1.0\n """"""the target network update rate""""""\n target_network_frequency: int = 8000\n """"""the timesteps it takes to update the target network""""""\n batch_size: int = 32\n """"""the batch size of sample from the reply memory""""""\n start_e: float = 1\n """"""the starting epsilon for exploration""""""\n end_e: float = 0.01\n """"""the ending epsilon for exploration""""""\n exploration_fraction: float = 0.10\n """"""the fraction of `total-timesteps` it takes from start-e to go end-e""""""\n learning_starts: int = 80000\n """"""timestep to start learning""""""\n train_frequency: int = 4\n """"""the frequency of training""""""\n n_step: int = 3\n """"""the number of steps to look ahead for n-step Q learning""""""\n prioritized_replay_alpha: float = 0.5\n """"""alpha parameter for prioritized replay buffer""""""\n prioritized_replay_beta: float = 0.4\n """"""beta parameter for prioritized replay buffer""""""\n prioritized_replay_eps: float = 1e-6\n """"""epsilon parameter for prioritized replay buffer""""""\n n_atoms: int = 51\n """"""the number of atoms""""""\n v_min: float = -10\n """"""the return lower bound""""""\n v_max: float = 10\n """"""the return upper bound""""""\n\n # Dataset capture\n capture_dataset: bool = True\n num_episodes_train: int = 10000\n num_episodes_val: int = 500\n num_episodes_test: int = 500\n output_dir: str = ""data/atari_episodes""\n min_episode_length: int = 1\n chunk_size: int = 160\n chunks_per_file: int = 100\n stop_on_complete: bool = True\n\n\ndef make_env(env_id, seed, idx, capture_video, run_name):\n def thunk():\n if capture_video and idx == 0:\n env = gym.make(env_id, render_mode=""rgb_array"")\n env = gym.wrappers.RecordVideo(env, f""videos/{run_name}"")\n else:\n env = gym.make(env_id)\n env = gym.wrappers.RecordEpisodeStatistics(env)\n\n env = NoopResetEnv(env, noop_max=30)\n env = MaxAndSkipEnv(env, skip=4)\n env = EpisodicLifeEnv(env)\n if ""FIRE"" in env.unwrapped.get_action_meanings():\n env = FireResetEnv(env)\n env = ClipRewardEnv(env)\n env = gym.wrappers.ResizeObservation(env, (84, 84))\n env = gym.wrappers.GrayScaleObservation(env)\n env = gym.wrappers.FrameStack(env, 4)\n\n env.action_space.seed(seed)\n return env\n\n return thunk\n\n\nclass NoisyLinear(nn.Module):\n def __init__(self, in_features, out_features, std_init=0.5):\n super().__init__()\n self.in_features = in_features\n self.out_features = out_features\n self.std_init = std_init\n\n self.weight_mu = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.weight_sigma = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.register_buffer(\n ""weight_epsilon"", torch.FloatTensor(out_features, in_features)\n )\n self.bias_mu = nn.Parameter(torch.FloatTensor(out_features))\n self.bias_sigma = nn.Parameter(torch.FloatTensor(out_features))\n self.register_buffer(""bias_epsilon"", torch.FloatTensor(out_features))\n # factorized gaussian noise\n self.reset_parameters()\n self.reset_noise()\n\n def reset_parameters(self):\n mu_range = 1 / math.sqrt(self.in_features)\n self.weight_mu.data.uniform_(-mu_range, mu_range)\n self.weight_sigma.data.fill_(self.std_init / math.sqrt(self.in_features))\n self.bias_mu.data.uniform_(-mu_range, mu_range)\n self.bias_sigma.data.fill_(self.std_init / math.sqrt(self.out_features))\n\n def reset_noise(self):\n self.weight_epsilon.normal_()\n self.bias_epsilon.normal_()\n\n def forward(self, input):\n if self.training:\n weight = self.weight_mu + self.weight_sigma * self.weight_epsilon\n bias = self.bias_mu + self.bias_sigma * self.bias_epsilon\n else:\n weight = self.weight_mu\n bias = self.bias_mu\n return F.linear(input, weight, bias)\n\n\n# ALGO LOGIC: initialize agent here:\nclass NoisyDuelingDistributionalNetwork(nn.Module):\n def __init__(self, env, n_atoms, v_min, v_max):\n super().__init__()\n self.n_atoms = n_atoms\n self.v_min = v_min\n self.v_max = v_max\n self.delta_z = (v_max - v_min) / (n_atoms - 1)\n self.n_actions = env.single_action_space.n\n self.register_buffer(""support"", torch.linspace(v_min, v_max, n_atoms))\n\n self.network = nn.Sequential(\n nn.Conv2d(4, 32, 8, stride=4),\n nn.ReLU(),\n nn.Conv2d(32, 64, 4, stride=2),\n nn.ReLU(),\n nn.Conv2d(64, 64, 3, stride=1),\n nn.ReLU(),\n nn.Flatten(),\n )\n conv_output_size = 3136\n\n self.value_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512), nn.ReLU(), NoisyLinear(512, n_atoms)\n )\n\n self.advantage_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512),\n nn.ReLU(),\n NoisyLinear(512, n_atoms * self.n_actions),\n )\n\n def forward(self, x):\n h = self.network(x / 255.0)\n value = self.value_head(h).view(-1, 1, self.n_atoms)\n advantage = self.advantage_head(h).view(-1, self.n_actions, self.n_atoms)\n q_atoms = value + advantage - advantage.mean(dim=1, keepdim=True)\n q_dist = F.softmax(q_atoms, dim=2)\n return q_dist\n\n def reset_noise(self):\n for layer in self.value_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n for layer in self.advantage_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n\n\nPrioritizedBatch = collections.namedtuple(\n ""PrioritizedBatch"",\n [\n ""observations"",\n ""actions"",\n ""rewards"",\n ""next_observations"",\n ""dones"",\n ""indices"",\n ""weights"",\n ],\n)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass SumSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.zeros(self.tree_size, dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = self.tree[parent * 2 + 1] + self.tree[parent * 2 + 2]\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def total(self):\n return self.tree[0]\n\n def retrieve(self, value):\n idx = 0\n while idx * 2 + 1 < self.tree_size:\n left = idx * 2 + 1\n right = left + 1\n if value <= self.tree[left]:\n idx = left\n else:\n value -= self.tree[left]\n idx = right\n return idx - (self.capacity - 1)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass MinSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.full(self.tree_size, float(""inf""), dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = np.minimum(\n self.tree[parent * 2 + 1], self.tree[parent * 2 + 2]\n )\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def min(self):\n return self.tree[0]\n\n\nclass PrioritizedReplayBuffer:\n def __init__(\n self, capacity, obs_shape, device, n_step, gamma, alpha=0.6, beta=0.4, eps=1e-6\n ):\n self.capacity = capacity\n self.device = device\n self.n_step = n_step\n self.gamma = gamma\n self.alpha = alpha\n self.beta = beta\n self.eps = eps\n\n self.buffer_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_next_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_actions = np.zeros(capacity, dtype=np.int64)\n self.buffer_rewards = np.zeros(capacity, dtype=np.float32)\n self.buffer_dones = np.zeros(capacity, dtype=np.bool_)\n\n self.pos = 0\n self.size = 0\n self.max_priority = 1.0\n\n self.sum_tree = SumSegmentTree(capacity)\n self.min_tree = MinSegmentTree(capacity)\n\n # For n-step returns\n self.n_step_buffer = deque(maxlen=n_step)\n\n def _get_n_step_info(self):\n reward = 0.0\n next_obs = self.n_step_buffer[-1][3]\n done = self.n_step_buffer[-1][4]\n\n for i in range(len(self.n_step_buffer)):\n reward += self.gamma**i * self.n_step_buffer[i][2]\n if self.n_step_buffer[i][4]:\n next_obs = self.n_step_buffer[i][3]\n done = True\n break\n return reward, next_obs, done\n\n def add(self, obs, action, reward, next_obs, done):\n self.n_step_buffer.append((obs, action, reward, next_obs, done))\n\n if len(self.n_step_buffer) < self.n_step:\n return\n\n reward, next_obs, done = self._get_n_step_info()\n obs = self.n_step_buffer[0][0]\n action = self.n_step_buffer[0][1]\n\n idx = self.pos\n self.buffer_obs[idx] = obs\n self.buffer_next_obs[idx] = next_obs\n self.buffer_actions[idx] = action\n self.buffer_rewards[idx] = reward\n self.buffer_dones[idx] = done\n\n priority = self.max_priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n self.pos = (self.pos + 1) % self.capacity\n self.size = min(self.size + 1, self.capacity)\n\n if done:\n self.n_step_buffer.clear()\n\n def sample(self, batch_size):\n indices = []\n p_total = self.sum_tree.total()\n segment = p_total / batch_size\n\n for i in range(batch_size):\n a = segment * i\n b = segment * (i + 1)\n upperbound = np.random.uniform(a, b)\n idx = self.sum_tree.retrieve(upperbound)\n indices.append(idx)\n\n samples = {\n ""observations"": torch.from_numpy(self.buffer_obs[indices]).to(self.device),\n ""actions"": torch.from_numpy(self.buffer_actions[indices])\n .to(self.device)\n .unsqueeze(1),\n ""rewards"": torch.from_numpy(self.buffer_rewards[indices])\n .to(self.device)\n .unsqueeze(1),\n ""next_observations"": torch.from_numpy(self.buffer_next_obs[indices]).to(\n self.device\n ),\n ""dones"": torch.from_numpy(self.buffer_dones[indices])\n .to(self.device)\n .unsqueeze(1),\n }\n\n probs = np.array(\n [self.sum_tree.tree[idx + self.capacity - 1] for idx in indices]\n )\n weights = (self.size * probs / p_total) ** -self.beta\n weights = weights / weights.max()\n samples[""weights""] = torch.from_numpy(weights).to(self.device).unsqueeze(1)\n samples[""indices""] = indices\n\n return PrioritizedBatch(**samples)\n\n def update_priorities(self, indices, priorities):\n priorities = np.abs(priorities) + self.eps\n self.max_priority = max(self.max_priority, priorities.max())\n\n for idx, priority in zip(indices, priorities):\n priority = priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n\nif __name__ == ""__main__"":\n args = tyro.cli(Args)\n assert args.num_envs == 1, ""vectorized envs are not supported at the moment""\n run_name = f""{args.env_id}__{args.exp_name}__{args.seed}__{int(time.time())}""\n if args.track:\n import wandb\n\n wandb.init(\n project=args.wandb_project_name,\n entity=args.wandb_entity,\n sync_tensorboard=True,\n config=vars(args),\n name=run_name,\n monitor_gym=True,\n save_code=True,\n )\n writer = SummaryWriter(f""runs/{run_name}"")\n writer.add_text(\n ""hyperparameters"",\n ""|param|value|\n|-|-|\n%s""\n % (""\n"".join([f""|{key}|{value}|"" for key, value in vars(args).items()])),\n )\n\n # TRY NOT TO MODIFY: seeding\n random.seed(args.seed)\n np.random.seed(args.seed)\n torch.manual_seed(args.seed)\n torch.backends.cudnn.deterministic = args.torch_deterministic\n\n device = torch.device(""cuda"" if torch.cuda.is_available() and args.cuda else ""cpu"")\n\n # env setup\n envs = gym.vector.SyncVectorEnv(\n [\n make_env(args.env_id, args.seed + i, i, args.capture_video, run_name)\n for i in range(args.num_envs)\n ]\n )\n assert isinstance(\n envs.single_action_space, gym.spaces.Discrete\n ), ""only discrete action space is supported""\n\n q_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n optimizer = optim.Adam(q_network.parameters(), lr=args.learning_rate, eps=1.5e-4)\n target_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n target_network.load_state_dict(q_network.state_dict())\n\n rb = PrioritizedReplayBuffer(\n args.buffer_size,\n envs.single_observation_space.shape,\n device,\n args.n_step,\n args.gamma,\n args.prioritized_replay_alpha,\n args.prioritized_replay_beta,\n args.prioritized_replay_eps,\n )\n\n # dataset capture state\n split_targets = {\n ""train"": args.num_episodes_train,\n ""val"": args.num_episodes_val,\n ""test"": args.num_episodes_test,\n }\n # Determine splits to run (order: train -> val -> test)\n splits_in_order = [s for s in [""train"", ""val"", ""test""] if split_targets[s] > 0]\n\n episodes_captured_per_split: dict[str, int] = {\n s: 0 for s in [""train"", ""val"", ""test""]\n }\n file_idx_by_split: dict[str, int] = {s: 0 for s in [""train"", ""val"", ""test""]}\n episode_metadata_by_split: dict[str, list[dict]] = {\n s: [] for s in [""train"", ""val"", ""test""]\n }\n\n obs_chunks: list[np.ndarray] = []\n act_chunks: list[np.ndarray] = []\n\n current_split_idx = 0\n current_split = splits_in_order[0]\n split_dir = os.path.join(args.output_dir, current_split)\n if args.capture_dataset:\n os.makedirs(split_dir, exist_ok=True)\n\n start_time = time.time()\n\n # TRY NOT TO MODIFY: start the game\n obs, _ = envs.reset(seed=args.seed)\n observations_seq: list[np.ndarray] = []\n actions_seq: list[np.ndarray] = []\n for global_step in range(args.total_timesteps):\n # anneal PER beta to 1\n rb.beta = min(\n 1.0,\n args.prioritized_replay_beta\n + global_step * (1.0 - args.prioritized_replay_beta) / args.total_timesteps,\n )\n\n # ALGO LOGIC: put action logic here\n with torch.no_grad():\n q_dist = q_network(torch.Tensor(obs).to(device))\n q_values = torch.sum(q_dist * q_network.support, dim=2)\n actions = torch.argmax(q_values, dim=1).cpu().numpy()\n\n # TRY NOT TO MODIFY: execute the game and log data.\n next_obs, rewards, terminations, truncations, infos = envs.step(actions)\n\n if args.capture_dataset:\n observations_seq.append(obs.astype(np.uint8))\n actions_seq.append(actions.astype(np.int64))\n\n if ""final_info"" in infos:\n for info in infos[""final_info""]:\n if info and ""episode"" in info:\n print(\n f""global_step={global_step}, episodic_return={info['episode']['r']}""\n )\n writer.add_scalar(\n ""charts/episodic_return"", info[""episode""][""r""], global_step\n )\n writer.add_scalar(\n ""charts/episodic_length"", info[""episode""][""l""], global_step\n )\n\n continue_capturing_multi = any(\n episodes_captured_per_split[s] < split_targets[s]\n for s in splits_in_order\n )\n if args.capture_dataset and continue_capturing_multi:\n current_len = len(observations_seq)\n if current_len >= args.min_episode_length:\n frames = np.concatenate(observations_seq, axis=0).astype(\n np.uint8\n )\n acts = np.concatenate(actions_seq, axis=0).astype(np.int64)\n\n episode_obs_chunks = []\n episode_act_chunks = []\n start_idx = 0\n while start_idx < current_len:\n end_idx = min(start_idx + args.chunk_size, current_len)\n if end_idx - start_idx < args.chunk_size:\n print(\n f""Warning: Inconsistent chunk_sizes. Episode has {current_len} frames, ""\n f""which is smaller than the requested chunk_size: {args.chunk_size}. ""\n ""This might lead to performance degradation during training.""\n )\n episode_obs_chunks.append(frames[start_idx:end_idx])\n episode_act_chunks.append(acts[start_idx:end_idx])\n start_idx = end_idx\n\n obs_chunks_data = [\n seq.astype(np.uint8) for seq in episode_obs_chunks\n ]\n act_chunks_data = [act for act in episode_act_chunks]\n obs_chunks.extend(obs_chunks_data)\n act_chunks.extend(act_chunks_data)\n\n # Save to the active split\n (\n ep_metadata,\n file_idx_by_split[current_split],\n obs_chunks,\n act_chunks,\n ) = save_chunks(\n file_idx_by_split[current_split],\n args.chunks_per_file,\n split_dir,\n obs_chunks,\n act_chunks,\n )\n episode_metadata_by_split[current_split].extend(ep_metadata)\n\n episodes_captured_per_split[current_split] += 1\n\n if (\n episodes_captured_per_split[current_split]\n >= split_targets[current_split]\n ):\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks before switching split '"",\n {current_split},\n ""' for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n obs_chunks = []\n act_chunks = []\n if current_split_idx + 1 < len(splits_in_order):\n current_split_idx += 1\n current_split = splits_in_order[current_split_idx]\n split_dir = os.path.join(\n args.output_dir, current_split\n )\n os.makedirs(split_dir, exist_ok=True)\n else:\n print(\n f""Episode too short ({current_len}), skipping capture...""\n )\n\n observations_seq = []\n actions_seq = []\n\n # TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`\n real_next_obs = next_obs.copy()\n for idx, trunc in enumerate(truncations):\n if trunc:\n real_next_obs[idx] = infos[""final_observation""][idx]\n rb.add(obs, actions, rewards, real_next_obs, terminations)\n\n # TRY NOT TO MODIFY: CRUCIAL step easy to overlook\n obs = next_obs\n\n # ALGO LOGIC: training.\n if global_step > args.learning_starts:\n if global_step % args.train_frequency == 0:\n # reset the noise for both networks\n q_network.reset_noise()\n target_network.reset_noise()\n data = rb.sample(args.batch_size)\n\n with torch.no_grad():\n next_dist = target_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n support = target_network.support # [n_atoms]\n next_q_values = torch.sum(\n next_dist * support, dim=2\n ) # [B, num_actions]\n\n # double q-learning\n next_dist_online = q_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n next_q_online = torch.sum(\n next_dist_online * support, dim=2\n ) # [B, num_actions]\n best_actions = torch.argmax(next_q_online, dim=1) # [B]\n next_pmfs = next_dist[\n torch.arange(args.batch_size), best_actions\n ] # [B, n_atoms]\n\n # compute the n-step Bellman update.\n gamma_n = args.gamma**args.n_step\n next_atoms = data.rewards + gamma_n * support * (\n 1 - data.dones.float()\n )\n tz = next_atoms.clamp(q_network.v_min, q_network.v_max)\n\n # projection\n delta_z = q_network.delta_z\n b = (tz - q_network.v_min) / delta_z # shape: [B, n_atoms]\n l = b.floor().clamp(0, args.n_atoms - 1)\n u = b.ceil().clamp(0, args.n_atoms - 1)\n\n # (l == u).float() handles the case where bj is exactly an integer\n # example bj = 1, then the upper ceiling should be uj= 2, and lj= 1\n d_m_l = (\n u.float() + (l == b).float() - b\n ) * next_pmfs # [B, n_atoms]\n d_m_u = (b - l) * next_pmfs # [B, n_atoms]\n\n target_pmfs = torch.zeros_like(next_pmfs)\n for i in range(target_pmfs.size(0)):\n target_pmfs[i].index_add_(0, l[i].long(), d_m_l[i])\n target_pmfs[i].index_add_(0, u[i].long(), d_m_u[i])\n\n dist = q_network(data.observations) # [B, num_actions, n_atoms]\n pred_dist = dist.gather(\n 1, data.actions.unsqueeze(-1).expand(-1, -1, args.n_atoms)\n ).squeeze(1)\n log_pred = torch.log(pred_dist.clamp(min=1e-5, max=1 - 1e-5))\n\n loss_per_sample = -(target_pmfs * log_pred).sum(dim=1)\n loss = (loss_per_sample * data.weights.squeeze()).mean()\n\n # update priorities\n new_priorities = loss_per_sample.detach().cpu().numpy()\n rb.update_priorities(data.indices, new_priorities)\n\n if global_step % 100 == 0:\n writer.add_scalar(""losses/td_loss"", loss.item(), global_step)\n q_values = (pred_dist * q_network.support).sum(dim=1) # [B]\n writer.add_scalar(\n ""losses/q_values"", q_values.mean().item(), global_step\n )\n sps = int(global_step / (time.time() - start_time))\n print(""SPS:"", sps)\n writer.add_scalar(""charts/SPS"", sps, global_step)\n writer.add_scalar(""charts/beta"", rb.beta, global_step)\n\n # optimize the model\n optimizer.zero_grad()\n loss.backward()\n optimizer.step()\n\n # update target network\n if global_step % args.target_network_frequency == 0:\n for target_param, param in zip(\n target_network.parameters(), q_network.parameters()\n ):\n target_param.data.copy_(\n args.tau * param.data + (1.0 - args.tau) * target_param.data\n )\n\n # optional early stop on dataset completion\n if args.capture_dataset and args.stop_on_complete:\n all_done = (\n all(\n episodes_captured_per_split[s] >= split_targets[s]\n for s in splits_in_order\n )\n and len(splits_in_order) > 0\n )\n if all_done:\n break\n\n envs.close()\n writer.close()\n\n # write metadata for dataset\n if args.capture_dataset:\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n\n os.makedirs(args.output_dir, exist_ok=True)\n metadata_path = os.path.join(args.output_dir, ""metadata.json"")\n if os.path.exists(metadata_path):\n try:\n with open(metadata_path, ""r"") as f:\n metadata = json.load(f)\n except Exception:\n metadata = {}\n else:\n metadata = {}\n\n metadata.setdefault(""env"", args.env_id)\n metadata.setdefault(""num_actions"", int(envs.single_action_space.n))\n for split in [""train"", ""val"", ""test""]:\n metadata.setdefault(f""num_episodes_{split}"", 0)\n metadata.setdefault(f""avg_episode_len_{split}"", 0.0)\n metadata.setdefault(f""episode_metadata_{split}"", [])\n\n for split_key in splits_in_order:\n ep_meta_list = episode_metadata_by_split[split_key]\n if ep_meta_list:\n metadata[f""episode_metadata_{split_key}""].extend(ep_meta_list)\n metadata[f""num_episodes_{split_key}""] = len(\n metadata[f""episode_metadata_{split_key}""]\n )\n metadata[f""avg_episode_len_{split_key}""] = float(\n np.mean(\n [\n ep[""avg_seq_len""]\n for ep in metadata[f""episode_metadata_{split_key}""]\n ]\n )\n )\n\n with open(metadata_path, ""w"") as f:\n json.dump(metadata, f)\n",python,tab +641,3393696,"input_pipeline/generate_atari_dataset.py",17759,0,"",python,selection_command +642,3609886,"TERMINAL",0,0,"ls /fast/project/HFMI_SynergyUnit/jafar_ws/data/breakout/train/",,terminal_command +643,3609887,"TERMINAL",0,0,"]633;Cdata_0000.array_record data_0007.array_record data_0014.array_record data_0021.array_record data_0028.array_record\r\ndata_0001.array_record data_0008.array_record data_0015.array_record data_0022.array_record data_0029.array_record\r\ndata_0002.array_record data_0009.array_record data_0016.array_record data_0023.array_record data_0030.array_record\r\ndata_0003.array_record data_0010.array_record data_0017.array_record data_0024.array_record data_0031.array_record\r\ndata_0004.array_record data_0011.array_record data_0018.array_record data_0025.array_record\r\ndata_0005.array_record data_0012.array_record data_0019.array_record data_0026.array_record\r\ndata_0006.array_record data_0013.array_record data_0020.array_record data_0027.array_record\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +644,3619440,"TERMINAL",0,0,"rm -r /fast/project/HFMI_SynergyUnit/jafar_ws/data/breakout/",,terminal_command +645,3619487,"TERMINAL",0,0,"]633;C",,terminal_output +646,3619499,"TERMINAL",0,0,"]0;franz.srambical@hai-login2:~/jafar",,terminal_output +647,3642472,"genie.py",0,0,"",python,tab +648,3657564,"TERMINAL",0,0,"git log",,terminal_command +649,3657565,"TERMINAL",0,0,"]633;C",,terminal_output +650,3657565,"TERMINAL",0,0,"[?1h=\rcommit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD -> generate-atari-dataset, origin/generate-atari-dataset)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/main, origin/HEAD, main)\r\nAuthor: mihir <78321484+maharajamihir@users.noreply.github.com>\r\n:",,terminal_output +651,3664132,"genie.py",0,0,"Switched from branch 'generate-atari-dataset' to 'gt-actions'",python,git_branch_checkout +652,3666976,"TERMINAL",0,0,"",,terminal_focus +653,3675313,"TERMINAL",0,0,"commit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD -> generate-atari-dataset, origin/generate-atari-dataset)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/main, origin/HEAD, main)\r\nAuthor: mihir <78321484+maharajamihir@users.noreply.github.com>\r\n:commit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD -> generate- atari-dataset, origin/generate-atari-dataset)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/main, orig :commit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD -> generate-atari-dataset, origin/generate-atari-dataset)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/main, origin/HEAD, main)\r\nAuthor: mihir <78321484+maharajamihir@users.noreply.github.com>\r\n:",,terminal_output +654,3677821,"genie.py",821,20443," use_gt_actions: bool,\n dyna_type: str,\n dyna_dim: int,\n dyna_ffn_dim: int,\n dyna_num_blocks: int,\n dyna_num_heads: int,\n param_dtype: jnp.dtype,\n dtype: jnp.dtype,\n use_flash_attention: bool,\n decode: bool,\n rngs: nnx.Rngs,\n dropout: float = 0.0,\n mask_limit: float = 0.0,\n ):\n # --- Tokenizer ---\n self.in_dim = in_dim\n self.tokenizer_dim = tokenizer_dim\n self.tokenizer_ffn_dim = tokenizer_ffn_dim\n self.latent_patch_dim = latent_patch_dim\n self.num_patch_latents = num_patch_latents\n self.patch_size = patch_size\n self.tokenizer_num_blocks = tokenizer_num_blocks\n self.tokenizer_num_heads = tokenizer_num_heads\n # --- LAM ---\n self.lam_dim = lam_dim\n self.lam_ffn_dim = lam_ffn_dim\n self.latent_action_dim = latent_action_dim\n self.num_latent_actions = num_latent_actions\n self.lam_patch_size = lam_patch_size\n self.lam_num_blocks = lam_num_blocks\n self.lam_num_heads = lam_num_heads\n self.lam_co_train = lam_co_train\n self.use_gt_actions = use_gt_actions\n # --- Dynamics ---\n self.dyna_type = dyna_type\n self.dyna_dim = dyna_dim\n self.dyna_ffn_dim = dyna_ffn_dim\n self.dyna_num_blocks = dyna_num_blocks\n self.dyna_num_heads = dyna_num_heads\n self.param_dtype = param_dtype\n self.dtype = dtype\n self.use_flash_attention = use_flash_attention\n self.dropout = dropout\n self.mask_limit = mask_limit\n self.decode = decode\n\n self.tokenizer = TokenizerVQVAE(\n in_dim=self.in_dim,\n model_dim=self.tokenizer_dim,\n ffn_dim=self.tokenizer_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_patch_latents,\n patch_size=self.patch_size,\n num_blocks=self.tokenizer_num_blocks,\n num_heads=self.tokenizer_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n if self.use_gt_actions:\n self.action_embed = nnx.Embed(\n self.num_latent_actions, self.latent_action_dim, rngs=rngs\n )\n self.lam = None\n else:\n self.lam = LatentActionModel(\n in_dim=self.in_dim,\n model_dim=self.lam_dim,\n ffn_dim=self.lam_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_latent_actions,\n patch_size=self.lam_patch_size,\n num_blocks=self.lam_num_blocks,\n num_heads=self.lam_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n self.action_embed = None\n if self.dyna_type == ""maskgit"":\n self.dynamics = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n elif self.dyna_type == ""causal"":\n self.dynamics = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=decode,\n rngs=rngs,\n )\n else:\n raise ValueError(f""Invalid dynamics type: {self.dyna_type}"")\n\n def __call__(\n self,\n batch: Dict[str, jax.Array],\n training: bool = True,\n ) -> Dict[str, jax.Array]:\n videos_BTHWC = batch[""videos""]\n tokenizer_outputs = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_indices_BTN = tokenizer_outputs[""indices""]\n latent_actions_BTm11L = None\n action_embeddings_BTm11L = None\n if self.use_gt_actions:\n assert self.action_embed is not None\n action_indices_E = None\n action_embeddings_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n action_embeddings_BTm11L = action_embeddings_BT1L[:, 1:]\n else:\n assert self.lam is not None\n lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n outputs = dict(\n video_tokens=jax.lax.stop_gradient(token_indices_BTN),\n latent_actions=(\n action_embeddings_BTm11L\n if self.use_gt_actions\n else latent_actions_BTm11L\n ),\n )\n outputs[""mask_rng""] = batch[""rng""]\n dyna_logits_BTNV, dyna_mask = self.dynamics(outputs, training)\n outputs[""token_logits""] = dyna_logits_BTNV\n outputs[""mask""] = dyna_mask\n mle_indices_BTN = jnp.argmax(outputs[""token_logits""], axis=-1)\n H, W = batch[""videos""].shape[2:4]\n outputs[""recon""] = self.tokenizer.decode(mle_indices_BTN, (H, W))\n if action_indices_E is not None:\n outputs[""lam_indices""] = action_indices_E\n return outputs\n\n def sample(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n maskgit_steps: int = 25,\n ) -> tuple[jax.Array, jax.Array]:\n if self.dyna_type == ""maskgit"":\n return self.sample_maskgit(\n batch, seq_len, maskgit_steps, temperature, sample_argmax\n )\n elif self.dyna_type == ""causal"":\n return self.sample_causal(batch, seq_len, temperature, sample_argmax)\n else:\n raise ValueError(f""Dynamics model type unknown: {self.dyna_type}"")\n\n def sample_maskgit(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n steps: int = 25,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n P: S * N\n """"""\n assert isinstance(self.dynamics, DynamicsMaskGIT)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n init_logits_BSNV = jnp.zeros(\n shape=(*token_idxs_BSN.shape, self.num_patch_latents)\n )\n if self.use_gt_actions:\n assert self.action_embed is not None\n latent_actions_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n latent_actions_BTm11L = latent_actions_BT1L[:, 1:]\n action_tokens_EL = latent_actions_BTm11L.reshape(-1, self.latent_action_dim)\n else:\n assert self.lam is not None\n latent_actions_E = batch[""latent_actions""]\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n # --- Extract submodule state ---\n dynamics_state = nnx.state(self.dynamics)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def maskgit_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, mask_BSN, action_tokens_EL = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_maskgit = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_maskgit, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_maskgit.patch_embed(token_idxs_BSN)\n mask_token_111M = dynamics_maskgit.mask_token.value\n mask_expanded_BSN1 = mask_BSN[..., None]\n vid_embed_BSNM = jnp.where(\n mask_expanded_BSN1, mask_token_111M, vid_embed_BSNM\n )\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_maskgit.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNM += act_embed_BS1M\n unmasked_ratio = jnp.cos(jnp.pi * (step + 1) / (steps * 2))\n step_temp = temperature * (1.0 - unmasked_ratio)\n final_logits_BSNV = dynamics_maskgit.transformer(vid_embed_BSNM) / step_temp\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_BSN = jnp.argmax(final_logits_BSNV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_BSN = jax.random.categorical(_rng, final_logits_BSNV)\n gather_fn = jax.vmap(jax.vmap(jax.vmap(lambda x, y: x[y])))\n final_token_probs_BSN = gather_fn(\n jax.nn.softmax(final_logits_BSNV), sampled_token_idxs_BSN\n )\n final_token_probs_BSN += ~mask_BSN\n # Update masked tokens and logits only\n token_idxs_BSN = jnp.where(mask_BSN, sampled_token_idxs_BSN, token_idxs_BSN)\n logits_BSNV = jnp.where(\n jnp.expand_dims(mask_BSN, -1), final_logits_BSNV, logits_BSNV\n )\n\n # --- Update mask ---\n num_unmasked_tokens = jnp.round(N * (1.0 - unmasked_ratio)).astype(int)\n final_token_probs_flat_BP = einops.rearrange(\n final_token_probs_BSN, ""b s n -> b (s n)""\n )\n idx_mask_P = (\n jnp.arange(final_token_probs_flat_BP.shape[-1])\n <= N - num_unmasked_tokens\n )\n sorted_idxs_BP = jnp.argsort(final_token_probs_flat_BP, axis=-1)\n mask_update_fn = jax.vmap(lambda msk, ids: msk.at[ids].set(idx_mask_P))\n mask_flat_BP = einops.rearrange(mask_BSN, ""b s n -> b (s n)"")\n new_mask_flat_BP = mask_update_fn(mask_flat_BP, sorted_idxs_BP)\n new_mask_BSN = einops.rearrange(new_mask_flat_BP, ""b (s n) -> b s n"", n=N)\n\n new_carry = (\n rng,\n token_idxs_BSN,\n logits_BSNV,\n new_mask_BSN,\n action_tokens_EL,\n )\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # Mask current frame (i.e., t == step_t)\n mask_S = jnp.arange(seq_len) == step_t\n mask_BSN = jnp.broadcast_to(mask_S[None, :, None], (B, seq_len, N)).astype(\n bool\n )\n masked_token_idxs_BSN = current_token_idxs_BSN * ~mask_BSN\n masked_logits_BSNV = current_logits_BSNV * jnp.expand_dims(~mask_BSN, -1)\n\n # --- Initialize and run MaskGIT loop ---\n init_carry_maskgit = (\n step_rng,\n masked_token_idxs_BSN,\n masked_logits_BSNV,\n mask_BSN,\n action_tokens_EL,\n )\n final_carry_maskgit = maskgit_step_fn(init_carry_maskgit, jnp.arange(steps))\n updated_token_idxs_BSN = final_carry_maskgit[1]\n updated_logits_BSNV = final_carry_maskgit[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, init_logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def sample_causal(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> tuple[jax.Array, jax.Array]:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n """"""\n assert isinstance(self.dynamics, DynamicsCausal)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n logits_BSNV = jnp.zeros((*token_idxs_BSN.shape, self.num_patch_latents))\n dynamics_state = nnx.state(self.dynamics)\n\n if self.use_gt_actions:\n assert self.action_embed is not None\n latent_actions_BT1L = self.action_embed(batch[""actions""]).reshape(\n *batch[""actions""].shape[:2], 1, self.latent_action_dim\n )\n latent_actions_BTm11L = latent_actions_BT1L[:, 1:]\n action_tokens_EL = latent_actions_BTm11L.reshape(-1, self.latent_action_dim)\n else:\n assert self.lam is not None\n latent_actions_E = batch[""latent_actions""]\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def causal_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array],\n step_n: jax.Array,\n ) -> tuple[jax.Array, jax.Array, jax.Array, jax.Array, jax.Array]:\n rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # We need to reconstruct the submodule inside scan body to prevent trace context mismatches\n dynamics_causal = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=self.decode,\n rngs=nnx.Rngs(0),\n )\n nnx.update(dynamics_causal, dynamics_state)\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_causal.patch_embed(token_idxs_BSN)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_causal.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(\n act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1])\n )\n vid_embed_BSNp1M = jnp.concatenate([act_embed_BS1M, vid_embed_BSNM], axis=2)\n final_logits_BTNp1V = (\n dynamics_causal.transformer(vid_embed_BSNp1M, (step_t, step_n))\n / temperature\n )\n final_logits_BV = final_logits_BTNp1V[:, step_t, step_n, :]\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_B = jnp.argmax(final_logits_BV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_B = jax.random.categorical(_rng, final_logits_BV)\n # Update next tokens only\n token_idxs_BSN = token_idxs_BSN.at[:, step_t, step_n].set(\n sampled_token_idxs_B\n )\n logits_BSNV = logits_BSNV.at[:, step_t, step_n].set(final_logits_BV)\n\n new_carry = (rng, token_idxs_BSN, logits_BSNV, action_tokens_EL, step_t)\n return new_carry\n\n @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[jax.Array, jax.Array, jax.Array]:\n rng, current_token_idxs_BSN, current_logits_BSNV = carry\n rng, step_rng = jax.random.split(rng)\n\n # --- Initialize and run causal loop ---\n init_carry_causal = (\n step_rng,\n current_token_idxs_BSN,\n current_logits_BSNV,\n action_tokens_EL,\n step_t,\n )\n final_carry_causal = causal_step_fn(init_carry_causal, jnp.arange(N))\n updated_token_idxs_BSN = final_carry_causal[1]\n updated_logits_BSNV = final_carry_causal[2]\n new_carry = (rng, updated_token_idxs_BSN, updated_logits_BSNV)\n return new_carry\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN, logits_BSNV)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry = generation_step_fn(initial_carry, timesteps_to_scan)\n final_token_idxs_BSN = final_carry[1]\n final_logits_BSNV = final_carry[2]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC, final_logits_BSNV\n\n def vq_encode(self, batch: Dict[str, jax.Array], training: bool) -> jax.Array:\n # --- Preprocess videos ---\n assert self.lam is not None\n",python,content +655,3682405,"TERMINAL",0,0,"resetsource /home/franz.srambical/cleanrl/.venv/bin/activate",,terminal_command +656,3682438,"TERMINAL",0,0,"]633;Cbash: resetsource: command not found\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +657,3696890,"TERMINAL",0,0,"git reset --hard HEAD~1",,terminal_command +658,3696934,"TERMINAL",0,0,"]633;CHEAD is now at 3515bca Update genie.py\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +659,3702051,"TERMINAL",0,0,"git lo",,terminal_command +660,3702051,"TERMINAL",0,0,"]633;Cgit: 'lo' is not a git command. See 'git --help'.\r\n\r\nThe most similar commands are\r\n\tlog\r\n\tclone\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +661,3708089,"TERMINAL",0,0,"git",,terminal_focus +662,3712949,"TERMINAL",0,0,"bash",,terminal_focus +663,3714879,"TERMINAL",0,0,"git cherry-pick 35e26ae50d5ac2a837e0a9670b257ca956b0ad48",,terminal_command +664,3714879,"TERMINAL",0,0,"]633;C",,terminal_output +665,3715059,"TERMINAL",0,0,"[gt-actions d62aa94] feat: trajectory collection during rainbow training\r\n Date: Sat Sep 20 17:07:07 2025 +0200\r\n 1 file changed, 782 insertions(+)\r\n create mode 100644 input_pipeline/generate_atari_dataset.py\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +666,3720711,"TERMINAL",0,0,"git",,terminal_focus +667,3725713,"TERMINAL",0,0,"bash",,terminal_focus +668,3727555,"TERMINAL",0,0,"git cherry-pick 3b83a0bbeedb6992df6b20a606b7d2a2b096de03",,terminal_command +669,3727576,"TERMINAL",0,0,"]633;C[gt-actions 7e6c5bc] fix: capture action alongside preceding obs\r\n Date: Mon Sep 22 13:46:15 2025 +0200\r\n 1 file changed, 12 insertions(+), 10 deletions(-)\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +670,3941198,"TERMINAL",0,0,"commit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD -> genera te-atari-dataset, origin/generate-atari-dataset)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/main, o :",,terminal_output +671,3942190,"TERMINAL",0,0,"commit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD ->  generate-atari-dataset, origin/generate-atari-dataset)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/ :",,terminal_output +672,3950130,"TERMINAL",0,0,"commit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD -> generate-atari-dataset, origin/generate-atari-datase t)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/main, origin/HEAD, main)\r\n:",,terminal_output +673,4000233,"genie.py",1533,0,"",python,selection_mouse +674,4007339,"input_pipeline/generate_atari_dataset.py",0,0,"# adapted from https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/rainbow_atari.py\nimport collections\nimport math\nimport os\nimport random\nimport time\nfrom collections import deque\nfrom dataclasses import dataclass\n\nimport gymnasium as gym\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport tyro\nfrom typing import Optional, Any\nfrom torch.utils.tensorboard.writer import SummaryWriter\n\nfrom cleanrl_utils.atari_wrappers import (\n ClipRewardEnv,\n EpisodicLifeEnv,\n FireResetEnv,\n MaxAndSkipEnv,\n NoopResetEnv,\n)\n\ntry:\n from utils import save_chunks # type: ignore\nexcept Exception: # pragma: no cover\n from input_pipeline.utils import save_chunks # type: ignore\nimport json\n\n\n@dataclass\nclass Args:\n exp_name: str = os.path.basename(__file__)[: -len("".py"")]\n """"""the name of this experiment""""""\n seed: int = 1\n """"""seed of the experiment""""""\n torch_deterministic: bool = True\n """"""if toggled, `torch.backends.cudnn.deterministic=False`""""""\n cuda: bool = True\n """"""if toggled, cuda will be enabled by default""""""\n track: bool = False\n """"""if toggled, this experiment will be tracked with Weights and Biases""""""\n wandb_project_name: str = ""cleanRL""\n """"""the wandb's project name""""""\n wandb_entity: Optional[str] = None\n """"""the entity (team) of wandb's project""""""\n capture_video: bool = False\n """"""whether to capture videos of the agent performances (check out `videos` folder)""""""\n save_model: bool = False\n """"""whether to save model into the `runs/{run_name}` folder""""""\n upload_model: bool = False\n """"""whether to upload the saved model to huggingface""""""\n hf_entity: str = """"\n """"""the user or org name of the model repository from the Hugging Face Hub""""""\n\n env_id: str = ""BreakoutNoFrameskip-v4""\n """"""the id of the environment""""""\n total_timesteps: int = 10000000\n """"""total timesteps of the experiments""""""\n learning_rate: float = 0.0000625\n """"""the learning rate of the optimizer""""""\n num_envs: int = 1\n """"""the number of parallel game environments""""""\n buffer_size: int = 1000000\n """"""the replay memory buffer size""""""\n gamma: float = 0.99\n """"""the discount factor gamma""""""\n tau: float = 1.0\n """"""the target network update rate""""""\n target_network_frequency: int = 8000\n """"""the timesteps it takes to update the target network""""""\n batch_size: int = 32\n """"""the batch size of sample from the reply memory""""""\n start_e: float = 1\n """"""the starting epsilon for exploration""""""\n end_e: float = 0.01\n """"""the ending epsilon for exploration""""""\n exploration_fraction: float = 0.10\n """"""the fraction of `total-timesteps` it takes from start-e to go end-e""""""\n learning_starts: int = 80000\n """"""timestep to start learning""""""\n train_frequency: int = 4\n """"""the frequency of training""""""\n n_step: int = 3\n """"""the number of steps to look ahead for n-step Q learning""""""\n prioritized_replay_alpha: float = 0.5\n """"""alpha parameter for prioritized replay buffer""""""\n prioritized_replay_beta: float = 0.4\n """"""beta parameter for prioritized replay buffer""""""\n prioritized_replay_eps: float = 1e-6\n """"""epsilon parameter for prioritized replay buffer""""""\n n_atoms: int = 51\n """"""the number of atoms""""""\n v_min: float = -10\n """"""the return lower bound""""""\n v_max: float = 10\n """"""the return upper bound""""""\n\n # Dataset capture\n capture_dataset: bool = True\n num_episodes_train: int = 10000\n num_episodes_val: int = 500\n num_episodes_test: int = 500\n output_dir: str = ""data/atari_episodes""\n min_episode_length: int = 1\n chunk_size: int = 160\n chunks_per_file: int = 100\n stop_on_complete: bool = True\n\n\ndef make_env(env_id, seed, idx, capture_video, run_name):\n def thunk():\n if capture_video and idx == 0:\n env = gym.make(env_id, render_mode=""rgb_array"")\n env = gym.wrappers.RecordVideo(env, f""videos/{run_name}"")\n else:\n env = gym.make(env_id)\n env = gym.wrappers.RecordEpisodeStatistics(env)\n\n env = NoopResetEnv(env, noop_max=30)\n env = MaxAndSkipEnv(env, skip=4)\n env = EpisodicLifeEnv(env)\n if ""FIRE"" in env.unwrapped.get_action_meanings():\n env = FireResetEnv(env)\n env = ClipRewardEnv(env)\n env = gym.wrappers.ResizeObservation(env, (84, 84))\n env = gym.wrappers.GrayScaleObservation(env)\n env = gym.wrappers.FrameStack(env, 4)\n\n env.action_space.seed(seed)\n return env\n\n return thunk\n\n\nclass NoisyLinear(nn.Module):\n def __init__(self, in_features, out_features, std_init=0.5):\n super().__init__()\n self.in_features = in_features\n self.out_features = out_features\n self.std_init = std_init\n\n self.weight_mu = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.weight_sigma = nn.Parameter(torch.FloatTensor(out_features, in_features))\n self.register_buffer(\n ""weight_epsilon"", torch.FloatTensor(out_features, in_features)\n )\n self.bias_mu = nn.Parameter(torch.FloatTensor(out_features))\n self.bias_sigma = nn.Parameter(torch.FloatTensor(out_features))\n self.register_buffer(""bias_epsilon"", torch.FloatTensor(out_features))\n # factorized gaussian noise\n self.reset_parameters()\n self.reset_noise()\n\n def reset_parameters(self):\n mu_range = 1 / math.sqrt(self.in_features)\n self.weight_mu.data.uniform_(-mu_range, mu_range)\n self.weight_sigma.data.fill_(self.std_init / math.sqrt(self.in_features))\n self.bias_mu.data.uniform_(-mu_range, mu_range)\n self.bias_sigma.data.fill_(self.std_init / math.sqrt(self.out_features))\n\n def reset_noise(self):\n self.weight_epsilon.normal_()\n self.bias_epsilon.normal_()\n\n def forward(self, input):\n if self.training:\n weight = self.weight_mu + self.weight_sigma * self.weight_epsilon\n bias = self.bias_mu + self.bias_sigma * self.bias_epsilon\n else:\n weight = self.weight_mu\n bias = self.bias_mu\n return F.linear(input, weight, bias)\n\n\n# ALGO LOGIC: initialize agent here:\nclass NoisyDuelingDistributionalNetwork(nn.Module):\n def __init__(self, env, n_atoms, v_min, v_max):\n super().__init__()\n self.n_atoms = n_atoms\n self.v_min = v_min\n self.v_max = v_max\n self.delta_z = (v_max - v_min) / (n_atoms - 1)\n self.n_actions = env.single_action_space.n\n self.register_buffer(""support"", torch.linspace(v_min, v_max, n_atoms))\n\n self.network = nn.Sequential(\n nn.Conv2d(4, 32, 8, stride=4),\n nn.ReLU(),\n nn.Conv2d(32, 64, 4, stride=2),\n nn.ReLU(),\n nn.Conv2d(64, 64, 3, stride=1),\n nn.ReLU(),\n nn.Flatten(),\n )\n conv_output_size = 3136\n\n self.value_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512), nn.ReLU(), NoisyLinear(512, n_atoms)\n )\n\n self.advantage_head = nn.Sequential(\n NoisyLinear(conv_output_size, 512),\n nn.ReLU(),\n NoisyLinear(512, n_atoms * self.n_actions),\n )\n\n def forward(self, x):\n h = self.network(x / 255.0)\n value = self.value_head(h).view(-1, 1, self.n_atoms)\n advantage = self.advantage_head(h).view(-1, self.n_actions, self.n_atoms)\n q_atoms = value + advantage - advantage.mean(dim=1, keepdim=True)\n q_dist = F.softmax(q_atoms, dim=2)\n return q_dist\n\n def reset_noise(self):\n for layer in self.value_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n for layer in self.advantage_head:\n if isinstance(layer, NoisyLinear):\n layer.reset_noise()\n\n\nPrioritizedBatch = collections.namedtuple(\n ""PrioritizedBatch"",\n [\n ""observations"",\n ""actions"",\n ""rewards"",\n ""next_observations"",\n ""dones"",\n ""indices"",\n ""weights"",\n ],\n)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass SumSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.zeros(self.tree_size, dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = self.tree[parent * 2 + 1] + self.tree[parent * 2 + 2]\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def total(self):\n return self.tree[0]\n\n def retrieve(self, value):\n idx = 0\n while idx * 2 + 1 < self.tree_size:\n left = idx * 2 + 1\n right = left + 1\n if value <= self.tree[left]:\n idx = left\n else:\n value -= self.tree[left]\n idx = right\n return idx - (self.capacity - 1)\n\n\n# adapted from: https://github.com/openai/baselines/blob/master/baselines/common/segment_tree.py\nclass MinSegmentTree:\n def __init__(self, capacity):\n self.capacity = capacity\n self.tree_size = 2 * capacity - 1\n self.tree = np.full(self.tree_size, float(""inf""), dtype=np.float32)\n\n def _propagate(self, idx):\n parent = (idx - 1) // 2\n while parent >= 0:\n self.tree[parent] = np.minimum(\n self.tree[parent * 2 + 1], self.tree[parent * 2 + 2]\n )\n parent = (parent - 1) // 2\n\n def update(self, idx, value):\n tree_idx = idx + self.capacity - 1\n self.tree[tree_idx] = value\n self._propagate(tree_idx)\n\n def min(self):\n return self.tree[0]\n\n\nclass PrioritizedReplayBuffer:\n def __init__(\n self, capacity, obs_shape, device, n_step, gamma, alpha=0.6, beta=0.4, eps=1e-6\n ):\n self.capacity = capacity\n self.device = device\n self.n_step = n_step\n self.gamma = gamma\n self.alpha = alpha\n self.beta = beta\n self.eps = eps\n\n self.buffer_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_next_obs = np.zeros((capacity,) + obs_shape, dtype=np.uint8)\n self.buffer_actions = np.zeros(capacity, dtype=np.int64)\n self.buffer_rewards = np.zeros(capacity, dtype=np.float32)\n self.buffer_dones = np.zeros(capacity, dtype=np.bool_)\n\n self.pos = 0\n self.size = 0\n self.max_priority = 1.0\n\n self.sum_tree = SumSegmentTree(capacity)\n self.min_tree = MinSegmentTree(capacity)\n\n # For n-step returns\n self.n_step_buffer = deque(maxlen=n_step)\n\n def _get_n_step_info(self):\n reward = 0.0\n next_obs = self.n_step_buffer[-1][3]\n done = self.n_step_buffer[-1][4]\n\n for i in range(len(self.n_step_buffer)):\n reward += self.gamma**i * self.n_step_buffer[i][2]\n if self.n_step_buffer[i][4]:\n next_obs = self.n_step_buffer[i][3]\n done = True\n break\n return reward, next_obs, done\n\n def add(self, obs, action, reward, next_obs, done):\n self.n_step_buffer.append((obs, action, reward, next_obs, done))\n\n if len(self.n_step_buffer) < self.n_step:\n return\n\n reward, next_obs, done = self._get_n_step_info()\n obs = self.n_step_buffer[0][0]\n action = self.n_step_buffer[0][1]\n\n idx = self.pos\n self.buffer_obs[idx] = obs\n self.buffer_next_obs[idx] = next_obs\n self.buffer_actions[idx] = action\n self.buffer_rewards[idx] = reward\n self.buffer_dones[idx] = done\n\n priority = self.max_priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n self.pos = (self.pos + 1) % self.capacity\n self.size = min(self.size + 1, self.capacity)\n\n if done:\n self.n_step_buffer.clear()\n\n def sample(self, batch_size):\n indices = []\n p_total = self.sum_tree.total()\n segment = p_total / batch_size\n\n for i in range(batch_size):\n a = segment * i\n b = segment * (i + 1)\n upperbound = np.random.uniform(a, b)\n idx = self.sum_tree.retrieve(upperbound)\n indices.append(idx)\n\n samples = {\n ""observations"": torch.from_numpy(self.buffer_obs[indices]).to(self.device),\n ""actions"": torch.from_numpy(self.buffer_actions[indices])\n .to(self.device)\n .unsqueeze(1),\n ""rewards"": torch.from_numpy(self.buffer_rewards[indices])\n .to(self.device)\n .unsqueeze(1),\n ""next_observations"": torch.from_numpy(self.buffer_next_obs[indices]).to(\n self.device\n ),\n ""dones"": torch.from_numpy(self.buffer_dones[indices])\n .to(self.device)\n .unsqueeze(1),\n }\n\n probs = np.array(\n [self.sum_tree.tree[idx + self.capacity - 1] for idx in indices]\n )\n weights = (self.size * probs / p_total) ** -self.beta\n weights = weights / weights.max()\n samples[""weights""] = torch.from_numpy(weights).to(self.device).unsqueeze(1)\n samples[""indices""] = indices\n\n return PrioritizedBatch(**samples)\n\n def update_priorities(self, indices, priorities):\n priorities = np.abs(priorities) + self.eps\n self.max_priority = max(self.max_priority, priorities.max())\n\n for idx, priority in zip(indices, priorities):\n priority = priority**self.alpha\n self.sum_tree.update(idx, priority)\n self.min_tree.update(idx, priority)\n\n\nif __name__ == ""__main__"":\n args = tyro.cli(Args)\n assert args.num_envs == 1, ""vectorized envs are not supported at the moment""\n run_name = f""{args.env_id}__{args.exp_name}__{args.seed}__{int(time.time())}""\n if args.track:\n import wandb\n\n wandb.init(\n project=args.wandb_project_name,\n entity=args.wandb_entity,\n sync_tensorboard=True,\n config=vars(args),\n name=run_name,\n monitor_gym=True,\n save_code=True,\n )\n writer = SummaryWriter(f""runs/{run_name}"")\n writer.add_text(\n ""hyperparameters"",\n ""|param|value|\n|-|-|\n%s""\n % (""\n"".join([f""|{key}|{value}|"" for key, value in vars(args).items()])),\n )\n\n # TRY NOT TO MODIFY: seeding\n random.seed(args.seed)\n np.random.seed(args.seed)\n torch.manual_seed(args.seed)\n torch.backends.cudnn.deterministic = args.torch_deterministic\n\n device = torch.device(""cuda"" if torch.cuda.is_available() and args.cuda else ""cpu"")\n\n # env setup\n envs = gym.vector.SyncVectorEnv(\n [\n make_env(args.env_id, args.seed + i, i, args.capture_video, run_name)\n for i in range(args.num_envs)\n ]\n )\n assert isinstance(\n envs.single_action_space, gym.spaces.Discrete\n ), ""only discrete action space is supported""\n\n q_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n optimizer = optim.Adam(q_network.parameters(), lr=args.learning_rate, eps=1.5e-4)\n target_network = NoisyDuelingDistributionalNetwork(\n envs, args.n_atoms, args.v_min, args.v_max\n ).to(device)\n target_network.load_state_dict(q_network.state_dict())\n\n rb = PrioritizedReplayBuffer(\n args.buffer_size,\n envs.single_observation_space.shape,\n device,\n args.n_step,\n args.gamma,\n args.prioritized_replay_alpha,\n args.prioritized_replay_beta,\n args.prioritized_replay_eps,\n )\n\n # dataset capture state\n split_targets = {\n ""train"": args.num_episodes_train,\n ""val"": args.num_episodes_val,\n ""test"": args.num_episodes_test,\n }\n # Determine splits to run (order: train -> val -> test)\n splits_in_order = [s for s in [""train"", ""val"", ""test""] if split_targets[s] > 0]\n\n episodes_captured_per_split: dict[str, int] = {\n s: 0 for s in [""train"", ""val"", ""test""]\n }\n file_idx_by_split: dict[str, int] = {s: 0 for s in [""train"", ""val"", ""test""]}\n episode_metadata_by_split: dict[str, list[dict]] = {\n s: [] for s in [""train"", ""val"", ""test""]\n }\n\n obs_chunks: list[np.ndarray] = []\n act_chunks: list[np.ndarray] = []\n\n current_split_idx = 0\n current_split = splits_in_order[0]\n split_dir = os.path.join(args.output_dir, current_split)\n if args.capture_dataset:\n os.makedirs(split_dir, exist_ok=True)\n\n start_time = time.time()\n\n # TRY NOT TO MODIFY: start the game\n obs, _ = envs.reset(seed=args.seed)\n observations_seq: list[np.ndarray] = []\n actions_seq: list[np.ndarray] = []\n for global_step in range(args.total_timesteps):\n # anneal PER beta to 1\n rb.beta = min(\n 1.0,\n args.prioritized_replay_beta\n + global_step * (1.0 - args.prioritized_replay_beta) / args.total_timesteps,\n )\n\n # ALGO LOGIC: put action logic here\n with torch.no_grad():\n q_dist = q_network(torch.Tensor(obs).to(device))\n q_values = torch.sum(q_dist * q_network.support, dim=2)\n actions = torch.argmax(q_values, dim=1).cpu().numpy()\n\n # TRY NOT TO MODIFY: execute the game and log data.\n next_obs, rewards, terminations, truncations, infos = envs.step(actions)\n\n if args.capture_dataset:\n observations_seq.append(obs.astype(np.uint8))\n actions_seq.append(actions.astype(np.int64))\n\n if ""final_info"" in infos:\n for info in infos[""final_info""]:\n if info and ""episode"" in info:\n print(\n f""global_step={global_step}, episodic_return={info['episode']['r']}""\n )\n writer.add_scalar(\n ""charts/episodic_return"", info[""episode""][""r""], global_step\n )\n writer.add_scalar(\n ""charts/episodic_length"", info[""episode""][""l""], global_step\n )\n\n continue_capturing_multi = any(\n episodes_captured_per_split[s] < split_targets[s]\n for s in splits_in_order\n )\n if args.capture_dataset and continue_capturing_multi:\n current_len = len(observations_seq)\n if current_len >= args.min_episode_length:\n frames = np.concatenate(observations_seq, axis=0).astype(\n np.uint8\n )\n acts = np.concatenate(actions_seq, axis=0).astype(np.int64)\n\n episode_obs_chunks = []\n episode_act_chunks = []\n start_idx = 0\n while start_idx < current_len:\n end_idx = min(start_idx + args.chunk_size, current_len)\n if end_idx - start_idx < args.chunk_size:\n print(\n f""Warning: Inconsistent chunk_sizes. Episode has {current_len} frames, ""\n f""which is smaller than the requested chunk_size: {args.chunk_size}. ""\n ""This might lead to performance degradation during training.""\n )\n episode_obs_chunks.append(frames[start_idx:end_idx])\n episode_act_chunks.append(acts[start_idx:end_idx])\n start_idx = end_idx\n\n obs_chunks_data = [\n seq.astype(np.uint8) for seq in episode_obs_chunks\n ]\n act_chunks_data = [act for act in episode_act_chunks]\n obs_chunks.extend(obs_chunks_data)\n act_chunks.extend(act_chunks_data)\n\n # Save to the active split\n (\n ep_metadata,\n file_idx_by_split[current_split],\n obs_chunks,\n act_chunks,\n ) = save_chunks(\n file_idx_by_split[current_split],\n args.chunks_per_file,\n split_dir,\n obs_chunks,\n act_chunks,\n )\n episode_metadata_by_split[current_split].extend(ep_metadata)\n\n episodes_captured_per_split[current_split] += 1\n\n if (\n episodes_captured_per_split[current_split]\n >= split_targets[current_split]\n ):\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks before switching split '"",\n {current_split},\n ""' for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n obs_chunks = []\n act_chunks = []\n if current_split_idx + 1 < len(splits_in_order):\n current_split_idx += 1\n current_split = splits_in_order[current_split_idx]\n split_dir = os.path.join(\n args.output_dir, current_split\n )\n os.makedirs(split_dir, exist_ok=True)\n else:\n print(\n f""Episode too short ({current_len}), skipping capture...""\n )\n\n observations_seq = []\n actions_seq = []\n\n # TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`\n real_next_obs = next_obs.copy()\n for idx, trunc in enumerate(truncations):\n if trunc:\n real_next_obs[idx] = infos[""final_observation""][idx]\n rb.add(obs, actions, rewards, real_next_obs, terminations)\n\n # TRY NOT TO MODIFY: CRUCIAL step easy to overlook\n obs = next_obs\n\n # ALGO LOGIC: training.\n if global_step > args.learning_starts:\n if global_step % args.train_frequency == 0:\n # reset the noise for both networks\n q_network.reset_noise()\n target_network.reset_noise()\n data = rb.sample(args.batch_size)\n\n with torch.no_grad():\n next_dist = target_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n support = target_network.support # [n_atoms]\n next_q_values = torch.sum(\n next_dist * support, dim=2\n ) # [B, num_actions]\n\n # double q-learning\n next_dist_online = q_network(\n data.next_observations\n ) # [B, num_actions, n_atoms]\n next_q_online = torch.sum(\n next_dist_online * support, dim=2\n ) # [B, num_actions]\n best_actions = torch.argmax(next_q_online, dim=1) # [B]\n next_pmfs = next_dist[\n torch.arange(args.batch_size), best_actions\n ] # [B, n_atoms]\n\n # compute the n-step Bellman update.\n gamma_n = args.gamma**args.n_step\n next_atoms = data.rewards + gamma_n * support * (\n 1 - data.dones.float()\n )\n tz = next_atoms.clamp(q_network.v_min, q_network.v_max)\n\n # projection\n delta_z = q_network.delta_z\n b = (tz - q_network.v_min) / delta_z # shape: [B, n_atoms]\n l = b.floor().clamp(0, args.n_atoms - 1)\n u = b.ceil().clamp(0, args.n_atoms - 1)\n\n # (l == u).float() handles the case where bj is exactly an integer\n # example bj = 1, then the upper ceiling should be uj= 2, and lj= 1\n d_m_l = (\n u.float() + (l == b).float() - b\n ) * next_pmfs # [B, n_atoms]\n d_m_u = (b - l) * next_pmfs # [B, n_atoms]\n\n target_pmfs = torch.zeros_like(next_pmfs)\n for i in range(target_pmfs.size(0)):\n target_pmfs[i].index_add_(0, l[i].long(), d_m_l[i])\n target_pmfs[i].index_add_(0, u[i].long(), d_m_u[i])\n\n dist = q_network(data.observations) # [B, num_actions, n_atoms]\n pred_dist = dist.gather(\n 1, data.actions.unsqueeze(-1).expand(-1, -1, args.n_atoms)\n ).squeeze(1)\n log_pred = torch.log(pred_dist.clamp(min=1e-5, max=1 - 1e-5))\n\n loss_per_sample = -(target_pmfs * log_pred).sum(dim=1)\n loss = (loss_per_sample * data.weights.squeeze()).mean()\n\n # update priorities\n new_priorities = loss_per_sample.detach().cpu().numpy()\n rb.update_priorities(data.indices, new_priorities)\n\n if global_step % 100 == 0:\n writer.add_scalar(""losses/td_loss"", loss.item(), global_step)\n q_values = (pred_dist * q_network.support).sum(dim=1) # [B]\n writer.add_scalar(\n ""losses/q_values"", q_values.mean().item(), global_step\n )\n sps = int(global_step / (time.time() - start_time))\n print(""SPS:"", sps)\n writer.add_scalar(""charts/SPS"", sps, global_step)\n writer.add_scalar(""charts/beta"", rb.beta, global_step)\n\n # optimize the model\n optimizer.zero_grad()\n loss.backward()\n optimizer.step()\n\n # update target network\n if global_step % args.target_network_frequency == 0:\n for target_param, param in zip(\n target_network.parameters(), q_network.parameters()\n ):\n target_param.data.copy_(\n args.tau * param.data + (1.0 - args.tau) * target_param.data\n )\n\n # optional early stop on dataset completion\n if args.capture_dataset and args.stop_on_complete:\n all_done = (\n all(\n episodes_captured_per_split[s] >= split_targets[s]\n for s in splits_in_order\n )\n and len(splits_in_order) > 0\n )\n if all_done:\n break\n\n envs.close()\n writer.close()\n\n # write metadata for dataset\n if args.capture_dataset:\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n\n os.makedirs(args.output_dir, exist_ok=True)\n metadata_path = os.path.join(args.output_dir, ""metadata.json"")\n if os.path.exists(metadata_path):\n try:\n with open(metadata_path, ""r"") as f:\n metadata = json.load(f)\n except Exception:\n metadata = {}\n else:\n metadata = {}\n\n metadata.setdefault(""env"", args.env_id)\n metadata.setdefault(""num_actions"", int(envs.single_action_space.n))\n for split in [""train"", ""val"", ""test""]:\n metadata.setdefault(f""num_episodes_{split}"", 0)\n metadata.setdefault(f""avg_episode_len_{split}"", 0.0)\n metadata.setdefault(f""episode_metadata_{split}"", [])\n\n for split_key in splits_in_order:\n ep_meta_list = episode_metadata_by_split[split_key]\n if ep_meta_list:\n metadata[f""episode_metadata_{split_key}""].extend(ep_meta_list)\n metadata[f""num_episodes_{split_key}""] = len(\n metadata[f""episode_metadata_{split_key}""]\n )\n metadata[f""avg_episode_len_{split_key}""] = float(\n np.mean(\n [\n ep[""avg_seq_len""]\n for ep in metadata[f""episode_metadata_{split_key}""]\n ]\n )\n )\n\n with open(metadata_path, ""w"") as f:\n json.dump(metadata, f)\n",python,tab +675,4009574,"input_pipeline/generate_atari_dataset.py",29774,0,"",python,selection_command +676,4009575,"input_pipeline/generate_atari_dataset.py",28060,0,"",python,selection_keyboard +677,4010495,"input_pipeline/generate_atari_dataset.py",26620,0,"",python,selection_keyboard +678,4011490,"input_pipeline/generate_atari_dataset.py",0,0,"",python,selection_command +679,4011739,"input_pipeline/generate_atari_dataset.py",947,0,"",python,selection_keyboard +680,4011887,"input_pipeline/generate_atari_dataset.py",2588,0,"",python,selection_keyboard +681,4012182,"input_pipeline/generate_atari_dataset.py",4011,0,"",python,selection_keyboard +682,4012534,"input_pipeline/generate_atari_dataset.py",5506,0,"",python,selection_keyboard +683,4013002,"input_pipeline/generate_atari_dataset.py",4011,0,"",python,selection_keyboard +684,4013296,"input_pipeline/generate_atari_dataset.py",2588,0,"",python,selection_keyboard +685,4014252,"input_pipeline/generate_atari_dataset.py",3290,0,"",python,selection_keyboard +686,4014769,"input_pipeline/generate_atari_dataset.py",3320,0,"",python,selection_command +687,4015047,"input_pipeline/generate_atari_dataset.py",3343,0,"",python,selection_command +688,4015047,"input_pipeline/generate_atari_dataset.py",3376,0,"",python,selection_command +689,4015048,"input_pipeline/generate_atari_dataset.py",3398,0,"",python,selection_command +690,4015048,"input_pipeline/generate_atari_dataset.py",3427,0,"",python,selection_command +691,4015098,"input_pipeline/generate_atari_dataset.py",3432,0,"",python,selection_command +692,4015098,"input_pipeline/generate_atari_dataset.py",3454,0,"",python,selection_command +693,4015284,"input_pipeline/generate_atari_dataset.py",3487,0,"",python,selection_command +694,4015285,"input_pipeline/generate_atari_dataset.py",3523,0,"",python,selection_command +695,4015286,"input_pipeline/generate_atari_dataset.py",3555,0,"",python,selection_command +696,4036871,"TERMINAL",0,0,"commit 3b83a0bbeedb6992df6b20a606b7d2a2b096de03 (HEAD -> generate-atari-dataset, origin/generate-atari-dataset)\r\nAuthor: emergenz \r\nDate: Mon Sep 22 13:46:15 2025 +0200\r\n\r\n fix: capture action alongside preceding obs\r\n\r\ncommit 35e26ae50d5ac2a837e0a9670b257ca956b0ad48\r\nAuthor: emergenz \r\nDate: Sat Sep 20 17:07:07 2025 +0200\r\n\r\n feat: trajectory collection during rainbow training\r\n\r\ncommit c7522f2500589c499710b23f20d3be0f37aef659 (origin/main, origin/HEAD, main)\r\nAuthor: mihir <78321484+maharajamihir@users.noreply.github.com>\r\n:",,terminal_output +697,4088212,"TERMINAL",0,0,"git",,terminal_focus +698,4090240,"TERMINAL",0,0,"\r[?1l>]0;franz.srambical@hai-login2:~/jafar",,terminal_output +699,4103376,"slurm/jobs/franz/berlin/coinrun/coinrun_lam/lam_coinrun_nan_investigation_100k_to_107k.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/lam/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/lam/%x_%j.log\n#SBATCH --job-name=lam_coinrun_nan_investigation_100k_to_107k\n\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir=/fast/project/HFMI_SynergyUnit/jafar_ws/data/coinrun/array_records_10m\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\ntags=""lam coinrun 38M nan_investigation""\n\nCHECKPOINT_PATH=""/fast/project/HFMI_SynergyUnit/jafar_ws/checkpoints/coinrun/lam/train_lam_coinrun_reproduction_20067/100000_ckpt""\n\nenv | grep SLURM\n\nsrun python train_lam.py \\n --save_ckpt \\n --restore_ckpt \\n --wandb_id $SLURM_JOB_ID \\n --ckpt_dir $CHECKPOINT_PATH \\n --batch_size=36 \\n --image_height=64 \\n --image_width=64 \\n --max_lr=3e-5 \\n --decay_end=3e-5 \\n --log_image_interval=1000 \\n --log_checkpoint_interval=1000 \\n --num_steps=107885 \\n --log \\n --name=""${job_name}_${slurm_job_id}"" \\n --tags $tags \\n --entity instant-uv \\n --project jafar \\n --data_dir $array_records_dir\n",shellscript,tab +700,4103380,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +701,4105691,"TERMINAL",0,0,"bash",,terminal_focus +702,4118579,"TERMINAL",0,0,"python input_pipeline/generate_atari_dataset.py --output_dir=/fast/project/HFMI_SynergyUnit/jafar_ws/data/breakout",,terminal_command +703,4118579,"TERMINAL",0,0,"]633;CTraceback (most recent call last):\r\n File ""/fast/home/franz.srambical/jafar/input_pipeline/generate_atari_dataset.py"", line 10, in \r\n import gymnasium as gym\r\nModuleNotFoundError: No module named 'gymnasium'\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +704,4126764,"TERMINAL",0,0,"cd ../cleanrl/",,terminal_command +705,4132879,"TERMINAL",0,0,"source .venv/bin/activate",,terminal_command +706,4135226,"TERMINAL",0,0,"cd ../jafar/",,terminal_command +707,4175184,"TERMINAL",0,0,"",,terminal_focus +708,4177233,"TERMINAL",0,0,"source /home/franz.srambical/cleanrl/.venv/bin/activate",,terminal_command +709,4185579,"TERMINAL",0,0,"bash",,terminal_focus +710,4275953,"TERMINAL",0,0,"sbatch --gpus=1 --ntasks-per-node=1 --cpus-per-task=64 --mem=100G --wrap=""python input_pipeline/generate_atari_dataset.py --output_dir=/fast/project/HFMI_SynergyUnit/jafar_ws/data/breakout""",,terminal_command +711,4275954,"TERMINAL",0,0,"]633;CSubmitted batch job 29284\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +712,4281437,"TERMINAL",0,0,"ssqueue",,terminal_command +713,4283272,"TERMINAL",0,0,"squeue",,terminal_command +714,4283278,"TERMINAL",0,0,"]633;C JOBID USER PARTITION NODES CPUS ST SUBMIT_TIME START_TIME TIME TIME_LIMIT NODELIST(REASON)\r\n 29284 franz.sram standard 1 128 R 2025-09-22T13:58:27 2025-09-22T13:58:28 0:06 1-00:00:00 hai001\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +715,4342092,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,0,"",shellscript,tab +716,4348030,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\ntags=""coinrun reproduction 10m_dataset tokenizer codebook_normal_init""\n\narray_records_dir=""/fast/project/HFMI_SynergyUnit/jafar_ws/data/coinrun/array_records_10m""\n\nCHECKPOINT_DIR=""/fast/project/HFMI_SynergyUnit/jafar_ws/checkpoints/coinrun/tokenizer/${job_name}_${slurm_job_id}""\nmkdir -p $CHECKPOINT_DIR\n\nenv | grep SLURM\n\nsrun python train_tokenizer.py \\n --wandb_id $SLURM_JOB_ID \\n --ckpt_dir $CHECKPOINT_DIR \\n --batch_size=48 \\n --image_height=64 \\n --image_width=64 \\n --init_lr=0 \\n --max_lr=3e-4 \\n --log_image_interval=1000 \\n --log_checkpoint_interval=1000 \\n --log \\n --name=""${job_name}_${slurm_job_id}"" \\n --tags $tags \\n --entity instant-uv \\n --project jafar \\n --data_dir $array_records_dir &\n\nchild_pid=$!\n\nwait $child_pid\n\n",shellscript,tab +717,4349085,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",345,0,"",shellscript,selection_mouse +718,4349878,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",329,76,"#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +719,4350507,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",234,171,"#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +720,4350728,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",138,267,"#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +721,4350754,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",117,288,"#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +722,4350802,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",91,314,"#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +723,4350846,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",67,338,"#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +724,4350874,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",39,366,"#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +725,4350899,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",21,384,"#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +726,4350919,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",20,385,"\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +727,4350974,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",0,405,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,selection_command +728,4351239,"slurm/jobs/franz/berlin/coinrun/coinrun_tokenizer/coinrun_tokenizer_repoduction.sbatch",0,0,"",shellscript,selection_command +729,4351998,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,0,"",shellscript,tab +730,4353146,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,0,"--gpus=1 --ntasks-per-node=1 --cpus-per-task=64 --mem=100G",shellscript,content +731,4353146,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",58,0,"",shellscript,selection_keyboard +732,4353428,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",57,0,"",shellscript,selection_command +733,4353920,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,58,"",shellscript,content +734,4354545,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,0,"\n#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/tokenizer/%x_%j.log\n#SBATCH --job-name=train_tokenizer_coinrun_reproduction_codebook_normal_init",shellscript,content +735,4354546,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",1,0,"",shellscript,selection_command +736,4355159,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,0,"",shellscript,selection_command +737,4355528,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,1,"",shellscript,content +738,4355974,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",20,0,"",shellscript,selection_command +739,4356043,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",21,0,"",shellscript,selection_command +740,4356430,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",22,0,"",shellscript,selection_command +741,4356598,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",29,0,"",shellscript,selection_command +742,4356858,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",31,0,"",shellscript,selection_command +743,4356886,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",36,0,"",shellscript,selection_command +744,4356902,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",37,0,"",shellscript,selection_command +745,4356957,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",39,0,"",shellscript,selection_command +746,4356975,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",40,0,"",shellscript,selection_command +747,4357014,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",47,0,"",shellscript,selection_command +748,4357044,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",49,0,"",shellscript,selection_command +749,4357093,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",55,0,"",shellscript,selection_command +750,4357129,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",56,0,"",shellscript,selection_command +751,4357158,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",59,0,"",shellscript,selection_command +752,4357180,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",60,0,"",shellscript,selection_command +753,4357201,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",64,0,"",shellscript,selection_command +754,4357250,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",65,0,"",shellscript,selection_command +755,4357291,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",67,0,"",shellscript,selection_command +756,4357344,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",68,0,"",shellscript,selection_command +757,4357344,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",75,0,"",shellscript,selection_command +758,4357412,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",77,0,"",shellscript,selection_command +759,4357428,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",81,0,"",shellscript,selection_command +760,4357470,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",82,0,"",shellscript,selection_command +761,4357476,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",84,0,"",shellscript,selection_command +762,4357512,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",85,0,"",shellscript,selection_command +763,4358057,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",84,0,"",shellscript,selection_command +764,4358231,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",82,0,"",shellscript,selection_command +765,4361612,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",106,0,"",shellscript,selection_command +766,4361612,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",132,0,"",shellscript,selection_command +767,4361613,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",106,0,"",shellscript,selection_command +768,4361613,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",132,0,"",shellscript,selection_command +769,4361922,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",153,0,"",shellscript,selection_command +770,4366439,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",154,0,"",shellscript,selection_command +771,4366648,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",156,0,"",shellscript,selection_command +772,4366713,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",160,0,"",shellscript,selection_command +773,4366714,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",161,0,"",shellscript,selection_command +774,4366735,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",168,0,"",shellscript,selection_command +775,4366815,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",169,0,"",shellscript,selection_command +776,4366815,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",185,0,"",shellscript,selection_command +777,4366901,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",186,0,"",shellscript,selection_command +778,4366901,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",194,0,"",shellscript,selection_command +779,4366958,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",195,0,"",shellscript,selection_command +780,4367077,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",199,0,"",shellscript,selection_command +781,4367301,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",200,0,"",shellscript,selection_command +782,4367408,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",205,0,"",shellscript,selection_command +783,4367541,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,0,"",shellscript,selection_command +784,4367699,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",213,0,"",shellscript,selection_command +785,4368609,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",212,0,"",shellscript,selection_command +786,4368733,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",211,0,"",shellscript,selection_command +787,4368926,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",210,0,"",shellscript,selection_command +788,4369029,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",209,0,"",shellscript,selection_command +789,4369251,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",208,0,"",shellscript,selection_command +790,4369404,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",207,0,"",shellscript,selection_command +791,4369519,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,0,"",shellscript,selection_command +792,4369945,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,1,"c",shellscript,selection_command +793,4369980,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,7,"coinrun",shellscript,selection_command +794,4370256,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,8,"coinrun/",shellscript,selection_command +795,4370387,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,17,"coinrun/tokenizer",shellscript,selection_command +796,4370684,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,19,"coinrun/tokenizer/%",shellscript,selection_command +797,4370807,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,18,"coinrun/tokenizer/",shellscript,selection_command +798,4371708,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,17,"coinrun/tokenizer",shellscript,selection_command +799,4371797,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,17,"",shellscript,content +800,4372133,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",206,0,"a",shellscript,content +801,4372133,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",207,0,"",shellscript,selection_keyboard +802,4372161,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",207,0,"t",shellscript,content +803,4372161,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",208,0,"",shellscript,selection_keyboard +804,4372224,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",208,0,"a",shellscript,content +805,4372225,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",209,0,"",shellscript,selection_keyboard +806,4372367,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",209,0,"r",shellscript,content +807,4372367,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",210,0,"",shellscript,selection_keyboard +808,4372417,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",210,0,"i",shellscript,content +809,4372418,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",211,0,"",shellscript,selection_keyboard +810,4372704,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",210,0,"",shellscript,selection_command +811,4373078,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",294,0,"",shellscript,selection_command +812,4373789,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",293,0,"",shellscript,selection_command +813,4373978,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",292,0,"",shellscript,selection_command +814,4374117,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",291,0,"",shellscript,selection_command +815,4374255,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",290,0,"",shellscript,selection_command +816,4374438,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",289,0,"",shellscript,selection_command +817,4374752,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",303,3,"",shellscript,content +818,4374752,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",302,0,"ar",shellscript,content +819,4374752,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",298,4,"",shellscript,content +820,4374752,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",297,0,"a",shellscript,content +821,4374752,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",289,8,"",shellscript,content +822,4376139,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",372,0,"",shellscript,selection_command +823,4376571,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",324,0,"",shellscript,selection_command +824,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",381,0,"/jafar_ws/data/atari/array_records_10m",shellscript,content +825,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",378,0,"gyU",shellscript,content +826,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",373,5,"",shellscript,content +827,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",372,0,"e",shellscript,content +828,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",371,1,"",shellscript,content +829,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",370,0,"Sy",shellscript,content +830,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",369,0,"ject/HFMI",shellscript,content +831,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",368,1,"",shellscript,content +832,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",367,0,"utput_dir /fast/pr",shellscript,content +833,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",366,0,"s_test 500 --",shellscript,content +834,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",365,1,"",shellscript,content +835,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",362,0,"epis",shellscript,content +836,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",361,1,"",shellscript,content +837,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",360,0,"um",shellscript,content +838,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",359,0,"des_val 500 --",shellscript,content +839,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",358,0,"s",shellscript,content +840,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",357,0,"m_ep",shellscript,content +841,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",355,2,"",shellscript,content +842,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",354,0,"es_train 10000 --n",shellscript,content +843,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",352,0,"is",shellscript,content +844,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",351,1,"",shellscript,content +845,4377942,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",348,1,"",shellscript,content +846,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",347,0,"um",shellscript,content +847,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",346,0,"tNoFrameskip-v4 --total_timesteps 10000000 --",shellscript,content +848,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",345,0,"eako",shellscript,content +849,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",344,0,"v_id B",shellscript,content +850,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",343,0,"_dataset.py --e",shellscript,content +851,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",342,0,"atar",shellscript,content +852,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",340,2,"",shellscript,content +853,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",339,0,"ate",shellscript,content +854,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",335,2,"",shellscript,content +855,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",333,0,"n g",shellscript,content +856,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",332,1,"",shellscript,content +857,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",331,0,"a\n\npyth",shellscript,content +858,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",330,0,"da",shellscript,content +859,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",328,1,"",shellscript,content +860,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",327,0,"te_atar",shellscript,content +861,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",325,0,"gene",shellscript,content +862,4377943,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",324,1,"",shellscript,content +863,4378928,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",344,0,"",shellscript,selection_command +864,4379023,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",364,0,"",shellscript,selection_command +865,4388535,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",592,0,"",shellscript,selection_mouse +866,4388535,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",591,0,"",shellscript,selection_command +867,4389979,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",592,0,"\n",shellscript,content +868,4390269,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",593,0,"\n",shellscript,content +869,4390530,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",594,0,"python input_pipeline/generate_atari_dataset.py --output_dir=/fast/project/HFMI_SynergyUnit/ja\nfar_ws/data/breakout",shellscript,content +870,4390978,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",708,0,"",shellscript,selection_command +871,4392036,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",689,0,"",shellscript,selection_command +872,4392036,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",689,0,"f",shellscript,content +873,4392036,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",690,0,"",shellscript,selection_keyboard +874,4392036,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",690,0,"f",shellscript,content +875,4392037,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",691,0,"",shellscript,selection_keyboard +876,4392091,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",691,0,"f",shellscript,content +877,4392092,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",692,0,"",shellscript,selection_keyboard +878,4392260,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",691,1,"",shellscript,content +879,4393713,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",690,1,"",shellscript,content +880,4394409,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",689,1,"",shellscript,content +881,4395441,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",688,1,"",shellscript,content +882,4396051,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",687,0,"",shellscript,selection_command +883,4396810,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",593,0,"",shellscript,selection_command +884,4396887,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",438,0,"",shellscript,selection_command +885,4403184,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",345,248,"",shellscript,content +886,4403614,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",345,1,"",shellscript,content +887,4405025,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,0,"",shellscript,selection_keyboard +888,4405512,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",459,0,"",shellscript,selection_keyboard +889,4407539,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",458,0,"",shellscript,selection_command +890,4409702,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +891,4414555,"TERMINAL",0,0,"squeue",,terminal_command +892,4414592,"TERMINAL",0,0,"]633;C JOBID USER PARTITION NODES CPUS ST SUBMIT_TIME START_TIME TIME TIME_LIMIT NODELIST(REASON)\r\n 29284 franz.sram standard 1 128 R 2025-09-22T13:58:27 2025-09-22T13:58:28 2:15 1-00:00:00 hai001\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +893,4425981,"TERMINAL",0,0,"squeue",,terminal_command +894,4425981,"TERMINAL",0,0,"]633;C JOBID USER PARTITION NODES CPUS ST SUBMIT_TIME START_TIME TIME TIME_LIMIT NODELIST(REASON)\r\n 29284 franz.sram standard 1 128 R 2025-09-22T13:58:27 2025-09-22T13:58:28 2:29 1-00:00:00 hai001\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +895,4443610,"TERMINAL",0,0,"ls /fast/project/HFMI_SynergyUnit/jafar_ws/data/breakout/train/",,terminal_command +896,4443610,"TERMINAL",0,0,"]633;Cdata_0000.array_record data_0002.array_record data_0004.array_record\r\ndata_0001.array_record data_0003.array_record data_0005.array_record\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +897,4473278,"TERMINAL",0,0,"tail -f slurm-29284.out",,terminal_command +898,4473278,"TERMINAL",0,0,"]633;CGym has been unmaintained since 2022 and does not support NumPy 2.0 amongst other critical functionality.\r\nPlease upgrade to Gymnasium, the maintained drop-in replacement of Gym, or contact the authors of your software and request that they upgrade.\r\nSee the migration guide at https://gymnasium.farama.org/introduction/migration_guide/ for additional information.\r\n/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/pygame/pkgdata.py:25: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.\r\n from pkg_resources import resource_stream, resource_exists\r\nA.L.E: Arcade Learning Environment (version 0.8.1+53f58b7)\r\n[Powered by Stella]\r\n/fast/home/franz.srambical/jafar/input_pipeline/generate_atari_dataset.py:355: DeprecationWarning: Conversion of an array with ndim > 0 to a scalar is deprecated, and will error in future. Ensure you extract a single element from your array before performing this operation. (Deprecated NumPy 1.25.)\r\n self.buffer_actions[idx] = action\r\n",,terminal_output +899,4593100,"input_pipeline/generate_atari_dataset.py",2069,0,"",python,selection_keyboard +900,4595498,"input_pipeline/generate_atari_dataset.py",2047,0,"",python,selection_command +901,4595713,"input_pipeline/generate_atari_dataset.py",2002,0,"",python,selection_command +902,4595767,"input_pipeline/generate_atari_dataset.py",1965,0,"",python,selection_command +903,4595786,"input_pipeline/generate_atari_dataset.py",1920,0,"",python,selection_command +904,4595821,"input_pipeline/generate_atari_dataset.py",1884,0,"",python,selection_command +905,4595931,"input_pipeline/generate_atari_dataset.py",1848,0,"",python,selection_command +906,4596212,"input_pipeline/generate_atari_dataset.py",1884,0,"",python,selection_command +907,4596442,"input_pipeline/generate_atari_dataset.py",1899,0,"",python,selection_command +908,4596780,"input_pipeline/generate_atari_dataset.py",1901,0,"",python,selection_command +909,4597021,"input_pipeline/generate_atari_dataset.py",1905,0,"",python,selection_command +910,4597289,"input_pipeline/generate_atari_dataset.py",1901,0,"",python,selection_command +911,4597419,"input_pipeline/generate_atari_dataset.py",1899,0,"",python,selection_command +912,4598059,"input_pipeline/generate_atari_dataset.py",3427,0,"",python,selection_keyboard +913,4598287,"input_pipeline/generate_atari_dataset.py",4700,0,"",python,selection_keyboard +914,4599329,"input_pipeline/generate_atari_dataset.py",3427,0,"",python,selection_keyboard +915,4600159,"input_pipeline/generate_atari_dataset.py",3428,0,"",python,selection_command +916,4600311,"input_pipeline/generate_atari_dataset.py",3450,0,"",python,selection_command +917,4600701,"input_pipeline/generate_atari_dataset.py",3483,0,"",python,selection_command +918,4601001,"input_pipeline/generate_atari_dataset.py",3487,0,"",python,selection_command +919,4606950,"input_pipeline/generate_atari_dataset.py",3505,0,"",python,selection_command +920,4607126,"input_pipeline/generate_atari_dataset.py",3507,0,"",python,selection_command +921,4607341,"input_pipeline/generate_atari_dataset.py",3511,0,"",python,selection_command +922,4607515,"input_pipeline/generate_atari_dataset.py",3513,0,"",python,selection_command +923,4608272,"input_pipeline/generate_atari_dataset.py",3511,0,"",python,selection_command +924,4608686,"input_pipeline/generate_atari_dataset.py",3507,0,"",python,selection_command +925,4608987,"input_pipeline/generate_atari_dataset.py",3505,0,"",python,selection_command +926,4609247,"input_pipeline/generate_atari_dataset.py",3487,0,"",python,selection_command +927,4626440,"slurm/jobs/franz/berlin/atari/atari_data_gen.sh",0,0,"",shellscript,tab +928,4627307,"input_pipeline/generate_atari_dataset.py",0,0,"",python,tab +929,4677458,"TERMINAL",0,0,"bash",,terminal_focus +930,4704233,"TERMINAL",0,0,"ls /fast/project/HFMI_SynergyUnit/jafar_ws/data/breakout/train/",,terminal_command +931,4704234,"TERMINAL",0,0,"]633;Cdata_0000.array_record data_0003.array_record data_0006.array_record data_0009.array_record\r\ndata_0001.array_record data_0004.array_record data_0007.array_record data_0010.array_record\r\ndata_0002.array_record data_0005.array_record data_0008.array_record data_0011.array_record\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output diff --git a/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-1505f3d0-0cb4-4cc0-84bf-678810d0ac8f1757148592235-2025_09_06-10.49.56.658/source.csv b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-1505f3d0-0cb4-4cc0-84bf-678810d0ac8f1757148592235-2025_09_06-10.49.56.658/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..b03180307999f21632035fbf73b5dbaaecd967a3 --- /dev/null +++ b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-1505f3d0-0cb4-4cc0-84bf-678810d0ac8f1757148592235-2025_09_06-10.49.56.658/source.csv @@ -0,0 +1,17 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,1,"train_tokenizer.py",0,0,"import os\n\nos.environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ""0.98"")\n\nfrom dataclasses import dataclass, field\nfrom typing import cast, Optional\n\nimport einops\nimport itertools\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax.checkpoint as ocp\nimport numpy as np\nimport dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\nimport grain\nimport flax.nnx as nnx\n\nfrom models.tokenizer import TokenizerVQVAE\nfrom utils.dataloader import get_dataloader\nfrom utils.train_utils import (\n get_lr_schedule,\n count_parameters_by_component,\n print_mem_stats,\n print_compiled_memory_stats,\n print_compiled_cost_analysis,\n)\n\n\n@dataclass\nclass Args:\n # Experiment\n num_steps: int = 300_000\n seed: int = 0\n seq_len: int = 16\n image_channels: int = 3\n image_height: int = 90\n image_width: int = 160\n data_dir: str = """"\n save_ckpt: bool = False\n restore_ckpt: bool = False\n # Optimization\n vq_beta: float = 0.25\n batch_size: int = 48\n init_lr: float = 0.0\n max_lr: float = 3e-4\n decay_end: float = 0.0\n wsd_decay_steps: int = (\n 20000 # NOTE: wsd_decay_steps will only be used when using a wsd-schedule\n )\n lr_schedule: str = ""wsd"" # supported options: wsd, cos\n warmup_steps: int = 10000\n # Tokenizer\n model_dim: int = 512\n ffn_dim: int = 2048\n latent_dim: int = 32\n num_latents: int = 1024\n patch_size: int = 4\n num_blocks: int = 4\n num_heads: int = 8\n dropout: float = 0.0\n codebook_dropout: float = 0.01\n param_dtype = jnp.float32\n dtype = jnp.bfloat16\n # Logging\n log: bool = False\n entity: str = """"\n project: str = """"\n name: str = ""train_tokenizer""\n tags: list[str] = field(default_factory=lambda: [""tokenizer""])\n log_interval: int = 5\n log_image_interval: int = 250\n ckpt_dir: str = """"\n log_checkpoint_interval: int = 10000\n log_checkpoint_keep_period: int = 20000\n log_gradients: bool = False\n wandb_id: str = """"\n use_flash_attention: bool = True\n\n\ndef build_model(args: Args, rng: jax.Array) -> tuple[TokenizerVQVAE, jax.Array]:\n rng, _rng = jax.random.split(rng)\n rngs = nnx.Rngs(_rng)\n return (\n TokenizerVQVAE(\n in_dim=args.image_channels,\n model_dim=args.model_dim,\n ffn_dim=args.ffn_dim,\n latent_dim=args.latent_dim,\n num_latents=args.num_latents,\n patch_size=args.patch_size,\n num_blocks=args.num_blocks,\n num_heads=args.num_heads,\n dropout=args.dropout,\n codebook_dropout=args.codebook_dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n ),\n rng,\n )\n\n\ndef build_optimizer(\n model: TokenizerVQVAE, args: Args\n) -> tuple[nnx.Optimizer, optax.Schedule]:\n lr_schedule = get_lr_schedule(\n args.lr_schedule,\n args.init_lr,\n args.max_lr,\n args.decay_end,\n args.num_steps,\n args.warmup_steps,\n args.wsd_decay_steps,\n )\n tx = optax.adamw(\n learning_rate=lr_schedule,\n b1=0.9,\n b2=0.9,\n weight_decay=1e-4,\n mu_dtype=args.param_dtype, # moments in full precision\n )\n optimizer = nnx.Optimizer(model, tx)\n return optimizer, lr_schedule\n\n\ndef build_mesh_and_sharding(\n num_devices: int,\n) -> tuple[Mesh, NamedSharding, NamedSharding]:\n device_mesh_arr = create_device_mesh((num_devices,))\n mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n replicated_sharding = NamedSharding(mesh, PartitionSpec())\n videos_sharding = NamedSharding(mesh, PartitionSpec(""data"", None, None, None, None))\n return mesh, replicated_sharding, videos_sharding\n\n\ndef shard_optimizer_states(\n optimizer: nnx.Optimizer, replicated_sharding: NamedSharding\n) -> None:\n model_state = nnx.state(optimizer.model)\n model_sharded_state = jax.lax.with_sharding_constraint(\n model_state, replicated_sharding\n )\n nnx.update(optimizer.model, model_sharded_state)\n optimizer_state = nnx.state(optimizer, nnx.optimizer.OptState)\n optimizer_sharded_state = jax.lax.with_sharding_constraint(\n optimizer_state, replicated_sharding\n )\n nnx.update(optimizer, optimizer_sharded_state)\n\n\ndef build_dataloader(args: Args) -> grain.DataLoaderIterator:\n image_shape = (args.image_height, args.image_width, args.image_channels)\n array_record_files = [\n os.path.join(args.data_dir, x)\n for x in os.listdir(args.data_dir)\n if x.endswith("".array_record"")\n ]\n grain_dataloader = get_dataloader(\n array_record_files,\n args.seq_len,\n # NOTE: We deliberately pass the global batch size\n # The dataloader shards the dataset across all processes\n args.batch_size,\n *image_shape,\n num_workers=8,\n prefetch_buffer_size=1,\n seed=args.seed,\n )\n initial_state = grain_dataloader._create_initial_state()\n grain_iterator = grain.DataLoaderIterator(grain_dataloader, initial_state)\n return grain_iterator\n\n\ndef build_checkpoint_manager(args: Args) -> ocp.CheckpointManager:\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointSave,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointRestore,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n checkpoint_options = ocp.CheckpointManagerOptions(\n save_interval_steps=args.log_checkpoint_interval,\n max_to_keep=3,\n keep_period=args.log_checkpoint_keep_period,\n step_format_fixed_length=6,\n cleanup_tmp_directories=True,\n )\n checkpoint_manager = ocp.CheckpointManager(\n args.ckpt_dir,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n return checkpoint_manager\n\n\ndef restore_checkpoint_if_needed(\n args: Args,\n checkpoint_manager: ocp.CheckpointManager,\n optimizer: nnx.Optimizer,\n grain_iterator: grain.DataLoaderIterator,\n restore_step: Optional[int] = None,\n) -> tuple[int, nnx.Optimizer, grain.DataLoaderIterator]:\n step = 0\n if restore_step is None:\n restore_step = checkpoint_manager.latest_step()\n if args.restore_ckpt:\n abstract_optimizer = nnx.eval_shape(lambda: optimizer)\n abstract_optimizer_state = nnx.state(abstract_optimizer)\n restored = checkpoint_manager.restore(\n restore_step,\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore(abstract_optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointRestore(grain_iterator), # type: ignore\n ),\n )\n restored_optimizer_state = restored[""model_state""]\n nnx.update(optimizer, restored_optimizer_state)\n grain_iterator = restored[""dataloader_state""]\n step = restore_step or 0\n print(f""Restored dataloader and model state from step {step}"")\n return step, optimizer, grain_iterator\n\n\ndef main(args: Args) -> None:\n jax.distributed.initialize()\n num_devices = jax.device_count()\n if num_devices == 0:\n raise ValueError(""No JAX devices found."")\n print(f""Running on {num_devices} devices."")\n\n if args.batch_size % num_devices != 0:\n raise ValueError(\n f""Global batch size {args.batch_size} must be divisible by ""\n f""number of devices {num_devices}.""\n )\n\n rng = jax.random.key(args.seed)\n\n # --- Initialize model ---\n tokenizer, rng = build_model(args, rng)\n\n _, params, _ = nnx.split(tokenizer, nnx.Param, ...)\n param_counts = count_parameters_by_component(params)\n\n if args.log and jax.process_index() == 0:\n wandb_init_kwargs = {\n ""entity"": args.entity,\n ""project"": args.project,\n ""name"": args.name,\n ""tags"": args.tags,\n ""group"": ""debug"",\n ""config"": args,\n }\n\n if args.wandb_id:\n wandb_init_kwargs.update(\n {\n ""id"": args.wandb_id,\n ""resume"": ""allow"",\n }\n )\n wandb.init(**wandb_init_kwargs)\n\n wandb.config.update({""model_param_count"": param_counts})\n\n print(""Parameter counts:"")\n print(param_counts)\n\n # --- Initialize optimizer ---\n optimizer, lr_schedule = build_optimizer(tokenizer, args)\n del tokenizer\n\n # FIXME: switch to create_hybrid_device_mesh for runs spanning multiple nodes\n mesh, replicated_sharding, videos_sharding = build_mesh_and_sharding(num_devices)\n\n shard_optimizer_states(optimizer, replicated_sharding)\n\n # --- Initialize checkpoint manager ---\n checkpoint_manager = build_checkpoint_manager(args)\n\n # --- Create DataLoaderIterator from dataloader ---\n grain_iterator = build_dataloader(args)\n\n # --- Restore checkpoint ---\n step, optimizer, grain_iterator = restore_checkpoint_if_needed(\n args, checkpoint_manager, optimizer, grain_iterator\n )\n\n # --- Define loss and train step (close over args) ---\n def tokenizer_loss_fn(\n model: TokenizerVQVAE, inputs: dict\n ) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n inputs[""videos""] = gt.astype(args.dtype)\n model.train()\n outputs = model(inputs, training=True)\n outputs[""recon""] = outputs[""recon""].astype(jnp.float32)\n mse = jnp.square(gt - outputs[""recon""]).mean()\n q_loss = jnp.square(jax.lax.stop_gradient(outputs[""emb""]) - outputs[""z""]).mean()\n commitment_loss = jnp.square(\n outputs[""emb""] - jax.lax.stop_gradient(outputs[""z""])\n ).mean()\n loss = mse + q_loss + args.vq_beta * commitment_loss\n\n gt_clipped = gt.clip(0, 1).reshape(-1, *gt.shape[2:])\n recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n psnr = jnp.asarray(pix.psnr(gt_clipped, recon)).mean()\n ssim = jnp.asarray(pix.ssim(gt_clipped, recon)).mean()\n _, index_counts = jnp.unique_counts(\n jnp.ravel(outputs[""indices""]), size=args.num_latents, fill_value=0\n )\n codebook_usage = (index_counts != 0).mean()\n metrics = dict(\n loss=loss,\n mse=mse,\n q_loss=q_loss,\n commitment_loss=commitment_loss,\n psnr=psnr,\n ssim=ssim,\n codebook_usage=codebook_usage,\n )\n return loss, (outputs[""recon""], metrics)\n\n @nnx.jit(donate_argnums=0)\n def train_step(\n optimizer: nnx.Optimizer, inputs: dict\n ) -> tuple[jax.Array, jax.Array, dict]:\n def loss_fn(model: TokenizerVQVAE) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n return tokenizer_loss_fn(model, inputs)\n\n (loss, (recon, metrics)), grads = nnx.value_and_grad(loss_fn, has_aux=True)(\n optimizer.model\n )\n optimizer.update(grads)\n if args.log_gradients:\n metrics[""encoder_gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""encoder""]\n )\n metrics[""vq_gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""vq""]\n )\n metrics[""decoder_gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""decoder""]\n )\n return loss, recon, metrics\n\n # --- TRAIN LOOP ---\n dataloader = (\n jax.make_array_from_process_local_data(videos_sharding, elem)\n for elem in grain_iterator\n )\n if jax.process_index() == 0:\n first_videos = next(dataloader)\n sample_inputs = dict(videos=first_videos)\n compiled = train_step.lower(optimizer, sample_inputs).compile()\n print_compiled_memory_stats(compiled.memory_analysis())\n print_compiled_cost_analysis(compiled.cost_analysis())\n # Do not skip the first batch during training\n dataloader = itertools.chain([first_videos], dataloader)\n print(f""Starting training from step {step}..."")\n first_step = step\n while step < args.num_steps:\n for videos in dataloader:\n # --- Train step ---\n inputs = dict(videos=videos)\n loss, recon, metrics = train_step(optimizer, inputs)\n if step == first_step:\n print_mem_stats(""After params initialized"")\n metrics[""lr""] = lr_schedule(step)\n print(f""Step {step}, loss: {loss}"")\n step += 1\n\n # --- Logging ---\n if args.log:\n if step % args.log_interval == 0 and jax.process_index() == 0:\n wandb.log(\n {\n ""loss"": loss,\n ""step"": step,\n **metrics,\n }\n )\n if step % args.log_image_interval == 0:\n gt_seq = inputs[""videos""][0].astype(jnp.float32) / 255.0\n recon_seq = recon[0].clip(0, 1)\n comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n comparison_seq = einops.rearrange(\n comparison_seq * 255, ""t h w c -> h (t w) c""\n )\n # NOTE: Process-dependent control flow deliberately happens\n # after indexing operation since it must not contain code\n # sections that lead to cross-accelerator communication.\n if jax.process_index() == 0:\n log_images = dict(\n image=wandb.Image(np.asarray(gt_seq[0])),\n recon=wandb.Image(np.asarray(recon_seq[0])),\n true_vs_recon=wandb.Image(\n np.asarray(comparison_seq.astype(np.uint8))\n ),\n )\n wandb.log(log_images)\n # --- Checkpointing ---\n if args.save_ckpt and step % args.log_checkpoint_interval == 0:\n optimizer_state = nnx.state(optimizer)\n checkpoint_manager.save(\n step,\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeSave(optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointSave( # type: ignore\n grain_iterator # type: ignore\n ),\n ),\n )\n print(f""Saved checkpoint at step {step}"")\n if step >= args.num_steps:\n break\n\n checkpoint_manager.close()\n\n\nif __name__ == ""__main__"":\n args = tyro.cli(Args)\n main(args)\n",python,tab +2,113,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"10:49:56 AM [info] Activating crowd-code\n10:49:56 AM [info] Recording started\n10:49:56 AM [info] Initializing git provider using file system watchers...\n10:49:56 AM [info] Git repository found\n10:49:56 AM [info] Git provider initialized successfully\n",Log,tab +3,224,"extension-output-pdoom-org.crowd-code-#1-crowd-code",250,0,"10:49:56 AM [info] Initial git state: [object Object]\n",Log,content +4,188000455,"train_tokenizer.py",0,0,"",python,tab +5,188012251,"train_tokenizer.py",0,0,"Switched from branch 'main' to 'demo-notebook'",python,git_branch_checkout +6,188015548,"README.md",0,0,"

🧞‍♀️ Jasmine: A simple, performant and scalable JAX-based world modeling codebase 🧞‍♀️

\n\n

\n \n \n \n \n

\n\nJasmine is a production-ready JAX-based world modeling codebase. It currently implements the high-level architecture of [Genie: Generative Interactive Environments](https://arxiv.org/abs/2402.15391) (Bruce et al., 2024) with [MaskGIT](https://arxiv.org/abs/2202.04200) (Chang et al., 2022), as well as an autoregressive (causal) baseline. A diffusion baseline is coming soon.\n\nJasmine scales from single hosts to hundreds of xPUs thanks to XLA and strives to be an easily hackable, batteries-included foundation for world modeling research.\n\n

Overview

\n\n- Asynchronous & distributed checkpointing thanks to [orbax.checkpoint](https://github.com/google/orbax)\n - Jasmine also supports mixing and matching hardware topologies (e.g. train on four nodes, load the checkpoint on a single node)\n- Optimized dataloading thanks to [Grain](https://github.com/google/grain)\n - Dataloading scales with the number of processes (i.e. nodes/xPUs)\n- Checkpointing of model weights, optimizer and dataloader states\n- Full reproducibility with **identical** training curves (thanks to seeded dataloading and training, and [JAX' approach to pseudo random numbers](https://docs.jax.dev/en/latest/random-numbers.html))\n- Automatic checkpoint deletion/retention according to specified retention policy thanks to `orbax.checkpoint.CheckpointManager`\n- Mixed precision training using `bfloat16`\n - `int8` training is on the roadmap via [aqt](https://github.com/google/aqt)\n- FlashAttention thanks to [cuDNN SDPA](https://github.com/jax-ml/jax/blob/a155c5a9997924170e0067d552351a9833c12c11/jax/_src/cudnn/fused_attention_stablehlo.py#L842)\n- Frame-level KV cache resets for accelerated spatiotemporal attention in causal baseline (still in PR)\n- Activation checkpointing (even onto host memory if desired)\n- DDP (changing to FSDP requires changing **a single line of code**)\n- WSD learning rate schedule\n - No need to retrain from scratch if you want to train for longer\n- Index-shuffling during dataloading\n- Google-native stack\n - https://github.com/google/orbax for checkpointing\n - https://github.com/google/grain for dataloading\n - https://github.com/google-deepmind/dm_pix for image manipulation\n - https://github.com/google/array_record as the data format\n- Easy model inspection thanks to [treescope](https://github.com/google-deepmind/treescope)\n- Easy model surgery thanks to the new [flax.nnx](https://flax.readthedocs.io/en/latest/migrating/linen_to_nnx.html) API\n- [Shape suffixes](https://medium.com/@NoamShazeer/shape-suffixes-good-coding-style-f836e72e24fd) throughout the repository\n\n

Setup 🧗

\n\nJasmine requires `python 3.10`, `jax 0.6.2`, and `flax 0.10.7`. To install the requirements, run:\n\n```bash\npip install -r requirements.txt\npre-commit install\n```\n\n---\n\n

Dataset 📂

\n\nYou can either download our preprocessed dataset from [Hugging Face](https://huggingface.co/datasets/p-doom/open_ai_minecraft_arrayrecords_chunked) or preprocess [OpenAI's VPT dataset](https://github.com/openai/Video-Pre-Training) manually.\n\n### Option 1: Use Preprocessed Dataset (Recommended)\n\nThe easiest way to get started is to download our preprocessed dataset from Hugging Face. This script will handle downloading and extracting it:\n\n```bash\nbash input_pipeline/download/download_array_records.sh\n```\n\n---\n\n### Option 2: Manual Download & Preprocessing of OpenAI's VPT Dataset\n\nIf you prefer to use the raw VPT dataset from OpenAI and preprocess it yourself, follow these steps:\n\n1. **Download index files:**\n This will download the initial index file:\n\n ```bash\n bash input_pipeline/download/openai/download_index_files.sh\n ```\n\n2. **Download from all index files:**\n This may take a long time depending on your bandwidth:\n\n ```bash\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_7xx_Apr_6.json\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_8xx_Jun_29.json\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_9xx_Jun_29.json\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_10xx_Jun_29.json\n ```\n\n3. **Preprocess videos into ArrayRecords:**\n For efficient distributed training, convert the raw videos into the arrayrecord format (make sure to have [ffmpeg](https://github.com/FFmpeg/FFmpeg) installed on your machine):\n\n ```bash\n python input_pipeline/preprocess/video_to_array_records.py\n ```\n\n> **Note:** This is a large dataset and may take considerable time and storage to download and process.\n\n\n

Quick Start 🚀

\n\nGenie has three components: a [video tokenizer](models/tokenizer.py), a [latent action model](models/lam.py), and a [dynamics model](models/dynamics.py). Each of these components are trained separately, however, the dynamics model requires a pre-trained video tokenizer (and latent action model).\n\nTo train the video tokenizer, run:\n\n```bash\npython train_tokenizer.py --ckpt_dir \n```\n\nTo train the latent action model, run:\n\n```bash\npython train_lam.py --ckpt_dir \n```\n\nOnce the tokenizer and LAM are trained, the dynamics model can be trained with:\n\n```bash\npython train_dynamics.py --tokenizer_checkpoint --lam_checkpoint \n```\n\nLogging with `wandb` is supported. To enable logging, set the `WANDB_API_KEY` environment variable or run:\n\n```bash\nwandb login\n```\n\nTraining can then be logged by setting the `--log` flag:\n\n```bash\npython train_tokenizer.py --log --entity --project \n```\n\n

Citing 📜

\n\nJasmine was built by [Mihir Mahajan](https://maharajamihir.github.io/), [Alfred Nguyen](https://avocadoali.github.io/) and [Franz Srambical](https://srambical.fr/), but started as a fork of [Jafar](https://github.com/flairox/jafar), built by [Matthew Jackson](https://matthewtjackson.com) and [Timon Willi](https://www.timonwilli.com).\n\nIf you use Jasmine in your work, please cite us, Jafar, and the original Genie paper as follows:\n\n```\n@article{\n mahajan2025jasmine,\n title={Jasmine: A simple, performant and scalable JAX-based world modeling codebase},\n author={Mihir Mahajan and Alfred Nguyen and Franz Srambical and Stefan Bauer},\n journal = {p(doom) blog},\n year={2025},\n url={https://pdoom.org/jasmine.html},\n note = {https://pdoom.org/blog.html}\n}\n```\n```\n@inproceedings{\n willi2024jafar,\n title={Jafar: An Open-Source Genie Reimplemention in Jax},\n author={Timon Willi and Matthew Thomas Jackson and Jakob Nicolaus Foerster},\n booktitle={First Workshop on Controllable Video Generation @ ICML 2024},\n year={2024},\n url={https://openreview.net/forum?id=ZZGaQHs9Jb}\n}\n```\n```\n@inproceedings{\n bruce2024genie,\n title={Genie: Generative Interactive Environments},\n author={Jake Bruce and Michael D Dennis and Ashley Edwards and Jack Parker-Holder and Yuge Shi and Edward Hughes and Matthew Lai and Aditi Mavalankar and Richie Steigerwald and Chris Apps and Yusuf Aytar and Sarah Maria Elisabeth Bechtle and Feryal Behbahani and Stephanie C.Y. Chan and Nicolas Heess and Lucy Gonzalez and Simon Osindero and Sherjil Ozair and Scott Reed and Jingwei Zhang and Konrad Zolna and Jeff Clune and Nando de Freitas and Satinder Singh and Tim Rockt{\""a}schel},\n booktitle={Forty-first International Conference on Machine Learning},\n year={2024},\n url={https://openreview.net/forum?id=bJbSbJskOS}\n}\n```\n",markdown,tab +7,188019111,"README.md",2726,0,"",markdown,selection_command +8,188019403,"README.md",2810,0,"\n",markdown,content +9,188019614,"README.md",2811,0," - Modularized training script for easy inspection using notebooks ([demo notebook](https://colab.research.google.com/drive/1zHkciFIZxXloJgue9F5LtFlA0m00rJIf?usp=sharing))",markdown,content +10,188019618,"README.md",2985,0,"",markdown,selection_keyboard +11,188020348,"README.md",2984,0,"",markdown,selection_command +12,188020659,"README.md",2815,0,"",markdown,selection_command +13,188020891,"README.md",2811,4,"",markdown,content +14,188033002,"README.md",0,0,"

🧞‍♀️ Jasmine: A simple, performant and scalable JAX-based world modeling codebase 🧞‍♀️

\n\n

\n \n \n \n \n

\n\nJasmine is a production-ready JAX-based world modeling codebase. It currently implements the high-level architecture of [Genie: Generative Interactive Environments](https://arxiv.org/abs/2402.15391) (Bruce et al., 2024) with [MaskGIT](https://arxiv.org/abs/2202.04200) (Chang et al., 2022), as well as an autoregressive (causal) baseline. A diffusion baseline is coming soon.\n\nJasmine scales from single hosts to hundreds of xPUs thanks to XLA and strives to be an easily hackable, batteries-included foundation for world modeling research.\n\n

Overview

\n\n- Asynchronous & distributed checkpointing thanks to [orbax.checkpoint](https://github.com/google/orbax)\n - Jasmine also supports mixing and matching hardware topologies (e.g. train on four nodes, load the checkpoint on a single node)\n- Optimized dataloading thanks to [Grain](https://github.com/google/grain)\n - Dataloading scales with the number of processes (i.e. nodes/xPUs)\n- Checkpointing of model weights, optimizer and dataloader states\n- Full reproducibility with **identical** training curves (thanks to seeded dataloading and training, and [JAX' approach to pseudo random numbers](https://docs.jax.dev/en/latest/random-numbers.html))\n- Automatic checkpoint deletion/retention according to specified retention policy thanks to `orbax.checkpoint.CheckpointManager`\n- Mixed precision training using `bfloat16`\n - `int8` training is on the roadmap via [aqt](https://github.com/google/aqt)\n- FlashAttention thanks to [cuDNN SDPA](https://github.com/jax-ml/jax/blob/a155c5a9997924170e0067d552351a9833c12c11/jax/_src/cudnn/fused_attention_stablehlo.py#L842)\n- Frame-level KV cache resets for accelerated spatiotemporal attention in causal baseline (still in PR)\n- Activation checkpointing (even onto host memory if desired)\n- DDP (changing to FSDP requires changing **a single line of code**)\n- WSD learning rate schedule\n - No need to retrain from scratch if you want to train for longer\n- Index-shuffling during dataloading\n- Google-native stack\n - https://github.com/google/orbax for checkpointing\n - https://github.com/google/grain for dataloading\n - https://github.com/google-deepmind/dm_pix for image manipulation\n - https://github.com/google/array_record as the data format\n- Easy model inspection thanks to [treescope](https://github.com/google-deepmind/treescope)\n- Modularized training script for easy inspection using notebooks ([demo notebook](https://colab.research.google.com/drive/1zHkciFIZxXloJgue9F5LtFlA0m00rJIf?usp=sharing))\n- Easy model surgery thanks to the new [flax.nnx](https://flax.readthedocs.io/en/latest/migrating/linen_to_nnx.html) API\n- [Shape suffixes](https://medium.com/@NoamShazeer/shape-suffixes-good-coding-style-f836e72e24fd) throughout the repository\n\n

Setup 🧗

\n\nJasmine requires `python 3.10`, `jax 0.6.2`, and `flax 0.10.7`. To install the requirements, run:\n\n```bash\npip install -r requirements.txt\npre-commit install\n```\n\n---\n\n

Dataset 📂

\n\nYou can either download our preprocessed dataset from [Hugging Face](https://huggingface.co/datasets/p-doom/open_ai_minecraft_arrayrecords_chunked) or preprocess [OpenAI's VPT dataset](https://github.com/openai/Video-Pre-Training) manually.\n\n### Option 1: Use Preprocessed Dataset (Recommended)\n\nThe easiest way to get started is to download our preprocessed dataset from Hugging Face. This script will handle downloading and extracting it:\n\n```bash\nbash input_pipeline/download/download_array_records.sh\n```\n\n---\n\n### Option 2: Manual Download & Preprocessing of OpenAI's VPT Dataset\n\nIf you prefer to use the raw VPT dataset from OpenAI and preprocess it yourself, follow these steps:\n\n1. **Download index files:**\n This will download the initial index file:\n\n ```bash\n bash input_pipeline/download/openai/download_index_files.sh\n ```\n\n2. **Download from all index files:**\n This may take a long time depending on your bandwidth:\n\n ```bash\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_7xx_Apr_6.json\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_8xx_Jun_29.json\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_9xx_Jun_29.json\n python input_pipeline/download/openai/download_videos.py --index_file_path data/open_ai_index_files/all_10xx_Jun_29.json\n ```\n\n3. **Preprocess videos into ArrayRecords:**\n For efficient distributed training, convert the raw videos into the arrayrecord format (make sure to have [ffmpeg](https://github.com/FFmpeg/FFmpeg) installed on your machine):\n\n ```bash\n python input_pipeline/preprocess/video_to_array_records.py\n ```\n\n> **Note:** This is a large dataset and may take considerable time and storage to download and process.\n\n\n

Quick Start 🚀

\n\nGenie has three components: a [video tokenizer](models/tokenizer.py), a [latent action model](models/lam.py), and a [dynamics model](models/dynamics.py). Each of these components are trained separately, however, the dynamics model requires a pre-trained video tokenizer (and latent action model).\n\nTo train the video tokenizer, run:\n\n```bash\npython train_tokenizer.py --ckpt_dir \n```\n\nTo train the latent action model, run:\n\n```bash\npython train_lam.py --ckpt_dir \n```\n\nOnce the tokenizer and LAM are trained, the dynamics model can be trained with:\n\n```bash\npython train_dynamics.py --tokenizer_checkpoint --lam_checkpoint \n```\n\nLogging with `wandb` is supported. To enable logging, set the `WANDB_API_KEY` environment variable or run:\n\n```bash\nwandb login\n```\n\nTraining can then be logged by setting the `--log` flag:\n\n```bash\npython train_tokenizer.py --log --entity --project \n```\n\n

Citing 📜

\n\nJasmine was built by [Mihir Mahajan](https://maharajamihir.github.io/), [Alfred Nguyen](https://avocadoali.github.io/) and [Franz Srambical](https://srambical.fr/), but started as a fork of [Jafar](https://github.com/flairox/jafar), built by [Matthew Jackson](https://matthewtjackson.com) and [Timon Willi](https://www.timonwilli.com).\n\nIf you use Jasmine in your work, please cite us, Jafar, and the original Genie paper as follows:\n\n```\n@article{\n mahajan2025jasmine,\n title={Jasmine: A simple, performant and scalable JAX-based world modeling codebase},\n author={Mihir Mahajan and Alfred Nguyen and Franz Srambical and Stefan Bauer},\n journal = {p(doom) blog},\n year={2025},\n url={https://pdoom.org/jasmine.html},\n note = {https://pdoom.org/blog.html}\n}\n```\n```\n@inproceedings{\n willi2024jafar,\n title={Jafar: An Open-Source Genie Reimplemention in Jax},\n author={Timon Willi and Matthew Thomas Jackson and Jakob Nicolaus Foerster},\n booktitle={First Workshop on Controllable Video Generation @ ICML 2024},\n year={2024},\n url={https://openreview.net/forum?id=ZZGaQHs9Jb}\n}\n```\n```\n@inproceedings{\n bruce2024genie,\n title={Genie: Generative Interactive Environments},\n author={Jake Bruce and Michael D Dennis and Ashley Edwards and Jack Parker-Holder and Yuge Shi and Edward Hughes and Matthew Lai and Aditi Mavalankar and Richie Steigerwald and Chris Apps and Yusuf Aytar and Sarah Maria Elisabeth Bechtle and Feryal Behbahani and Stephanie C.Y. Chan and Nicolas Heess and Lucy Gonzalez and Simon Osindero and Sherjil Ozair and Scott Reed and Jingwei Zhang and Konrad Zolna and Jeff Clune and Nando de Freitas and Satinder Singh and Tim Rockt{\""a}schel},\n booktitle={Forty-first International Conference on Machine Learning},\n year={2024},\n url={https://openreview.net/forum?id=bJbSbJskOS}\n}\n```\n",markdown,tab +15,188033071,"README.md",2811,0,"",markdown,selection_command +16,189015678,"README.md",0,0,"",markdown,tab diff --git a/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-1f52a67d-2cca-4352-b39c-6cd9f3effae01765648091845-2025_12_13-18.48.27.228/source.csv b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-1f52a67d-2cca-4352-b39c-6cd9f3effae01765648091845-2025_12_13-18.48.27.228/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..e9f0a16e54b82e99f5fcd23ea3ab65d2e8ff5484 --- /dev/null +++ b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-1f52a67d-2cca-4352-b39c-6cd9f3effae01765648091845-2025_12_13-18.48.27.228/source.csv @@ -0,0 +1,66 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,3,"crates/cli/src/main.rs",0,0,"//! CLI tool for serializing crowd-pilot IDE interaction data.\n//!\n//! This tool processes CSV session files and outputs JSONL format suitable for\n//! NeMo SFT training. It uses an embedded Python interpreter to load HuggingFace\n//! tokenizers for accurate token counting.\n\nuse std::path::PathBuf;\n\nuse clap::Parser;\nuse pyo3::prelude::*;\nuse pyo3::types::PyModule;\n\nuse crowd_pilot_serializer_core::{\n pipeline::{PipelineConfig, PipelineResult},\n process_all_sessions, write_jsonl_output, Tokenizer,\n};\n\n/// Serialize crowd-pilot CSV sessions to NeMo JSONL format.\n#[derive(Parser, Debug)]\n#[command(name = ""crowd-pilot-serialize"")]\n#[command(author, version, about, long_about = None)]\nstruct Args {\n /// Root directory containing CSV session files\n #[arg(long)]\n csv_root: PathBuf,\n\n /// Output directory for JSONL files\n #[arg(long)]\n output_dir: PathBuf,\n\n /// HuggingFace tokenizer model name or path\n #[arg(long)]\n tokenizer: String,\n\n /// Maximum tokens per conversation chunk\n #[arg(long, default_value = ""8192"")]\n max_tokens_per_conversation: usize,\n\n /// Maximum tokens per message\n #[arg(long, default_value = ""2048"")]\n max_tokens_per_message: usize,\n\n /// Minimum messages required to keep a conversation\n #[arg(long, default_value = ""5"")]\n min_conversation_messages: usize,\n\n /// Viewport radius (lines above/below cursor)\n #[arg(long, default_value = ""10"")]\n viewport_radius: usize,\n\n /// Coalesce radius for grouping nearby edits\n #[arg(long, default_value = ""5"")]\n coalesce_radius: usize,\n\n /// Fraction of sessions for validation (0.0-1.0)\n #[arg(long, default_value = ""0.1"")]\n val_ratio: f64,\n\n /// Custom system prompt (optional)\n #[arg(long)]\n system_prompt: Option,\n}\n\nconst DEFAULT_SYSTEM_PROMPT: &str = r#""You are a helpful assistant that can interact multiple times with a computer shell to solve programming tasks.\nYour response must contain exactly ONE bash code block with ONE command (or commands connected with && or ||).\n\nFormat your response as shown in .\n\n\n```bash\nyour_command_here\n```\n\n\nFailure to follow these rules will cause your response to be rejected.""#;\n\n/// Wrapper around Python tokenizer for exact token counting and truncation.\nstruct PythonTokenizer {\n tokenizer: Py,\n}\n\nimpl PythonTokenizer {\n /// Load a HuggingFace tokenizer.\n fn load(model_name: &str) -> PyResult {\n Python::with_gil(|py| {\n let transformers = PyModule::import(py, ""transformers"")?;\n let auto_tokenizer = transformers.getattr(""AutoTokenizer"")?;\n let tokenizer = auto_tokenizer.call_method1(""from_pretrained"", (model_name,))?;\n Ok(Self {\n tokenizer: tokenizer.into(),\n })\n })\n }\n}\n\nimpl Tokenizer for PythonTokenizer {\n fn count_tokens(&self, text: &str) -> usize {\n Python::with_gil(|py| {\n let tokenizer = self.tokenizer.as_ref(py);\n let tokens = tokenizer\n .call_method1(""encode"", (text,))\n .expect(""Failed to encode text with tokenizer"");\n tokens.len().unwrap()\n })\n }\n\n fn truncate_to_max_tokens(&self, text: &str, max_tokens: usize) -> String {\n Python::with_gil(|py| {\n let tokenizer = self.tokenizer.as_ref(py);\n let kwargs = pyo3::types::PyDict::new(py);\n kwargs.set_item(""max_length"", max_tokens).unwrap();\n kwargs.set_item(""truncation"", true).unwrap();\n \n let tokens = tokenizer\n .call_method(""encode"", (text,), Some(kwargs))\n .expect(""Failed to encode text with tokenizer"");\n \n tokenizer\n .call_method1(""decode"", (tokens,))\n .expect(""Failed to decode tokens"")\n .extract()\n .unwrap()\n })\n }\n}\n\nfn main() -> Result<(), Box> {\n let args = Args::parse();\n\n println!(""Loading tokenizer from {}..."", args.tokenizer);\n let tokenizer = PythonTokenizer::load(&args.tokenizer)?;\n\n let config = PipelineConfig {\n max_tokens_per_conversation: args.max_tokens_per_conversation,\n max_tokens_per_message: args.max_tokens_per_message,\n min_conversation_messages: args.min_conversation_messages,\n viewport_radius: args.viewport_radius,\n coalesce_radius: args.coalesce_radius,\n val_ratio: args.val_ratio,\n };\n\n println!(""Processing CSV files from {:?}..."", args.csv_root);\n let session_results = process_all_sessions(\n &args.csv_root,\n &tokenizer,\n &config,\n )?;\n\n let total_sessions = session_results.len();\n println!(""Processed {} sessions"", total_sessions);\n\n let system_prompt = args.system_prompt.as_deref().unwrap_or(DEFAULT_SYSTEM_PROMPT);\n\n println!(""Writing output to {:?}..."", args.output_dir);\n let result: PipelineResult = write_jsonl_output(\n session_results,\n &args.output_dir,\n args.val_ratio,\n system_prompt,\n )?;\n\n let metadata_path = args.output_dir.join(""metadata.json"");\n let metadata = serde_json::json!({\n ""config"": {\n ""csv_root"": args.csv_root.to_string_lossy(),\n ""output_dir"": args.output_dir.to_string_lossy(),\n ""tokenizer"": args.tokenizer,\n ""max_tokens_per_conversation"": args.max_tokens_per_conversation,\n ""max_tokens_per_message"": args.max_tokens_per_message,\n ""min_conversation_messages"": args.min_conversation_messages,\n ""viewport_radius"": args.viewport_radius,\n ""coalesce_radius"": args.coalesce_radius,\n ""val_ratio"": args.val_ratio,\n },\n ""counts"": {\n ""total_sessions"": result.total_sessions,\n ""total_conversations"": result.total_conversations,\n ""train_conversations"": result.train_conversations,\n ""val_conversations"": result.val_conversations,\n },\n ""stats"": {\n ""total_messages"": result.total_messages,\n ""total_tokens"": result.total_tokens,\n ""avg_messages_per_conversation"": if result.total_conversations > 0 {\n result.total_messages as f64 / result.total_conversations as f64\n } else {\n 0.0\n },\n ""avg_tokens_per_conversation"": if result.total_conversations > 0 {\n result.total_tokens as f64 / result.total_conversations as f64\n } else {\n 0.0\n },\n },\n ""files"": {\n ""train_path"": args.output_dir.join(""training.jsonl"").to_string_lossy(),\n ""val_path"": args.output_dir.join(""validation.jsonl"").to_string_lossy(),\n },\n });\n std::fs::write(&metadata_path, serde_json::to_string_pretty(&metadata)?)?;\n\n println!(""\n[summary]"");\n println!("" Total sessions processed: {}"", result.total_sessions);\n println!("" Train conversations: {}"", result.train_conversations);\n println!("" Val conversations: {}"", result.val_conversations);\n println!("" Total messages: {}"", result.total_messages);\n println!("" Total tokens: {}"", result.total_tokens);\n println!("" Output: {:?}/{{training,validation}}.jsonl"", args.output_dir);\n println!("" Metadata: {:?}"", metadata_path);\n\n Ok(())\n}\n\n",rust,tab +2,248,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"6:48:27 PM [info] Activating crowd-code\n6:48:27 PM [info] Recording started\n6:48:27 PM [info] Initializing git provider using file system watchers...\n6:48:27 PM [info] Git repository found\n",Log,tab +3,345,"extension-output-pdoom-org.crowd-code-#1-crowd-code",189,0,"6:48:27 PM [info] Git provider initialized successfully\n6:48:27 PM [info] Initial git state: [object Object]\n",Log,content +4,3836,"TERMINAL",0,0,"",,terminal_focus +5,3838,"crates/cli/src/main.rs",0,0,"",rust,tab +6,5667,"TERMINAL",0,0,"squeue",,terminal_command +7,5680,"TERMINAL",0,0,"]633;C JOBID USER PARTITION NODES CPUS ST SUBMIT_TIME START_TIME TIME TIME_LIMIT NODELIST(REASON)\r\n 36313 xiao.liu interacti 1 128 R 2025-12-13T12:56:48 2025-12-13T12:56:48 5:51:44 23:59:00 hai005\r\n 36303 xiao.liu interacti 1 128 R 2025-12-12T23:25:52 2025-12-12T23:25:52 19:22:40 23:59:00 hai006\r\n 36317 mihir.maha standard 1 10 R 2025-12-13T15:27:10 2025-12-13T15:27:10 3:21:22 1-00:00:00 hai004\r\n 36314 xiao.liu standard 1 128 R 2025-12-13T13:27:57 2025-12-13T15:13:56 3:34:36 23:59:00 hai007\r\n 36304 nishant.ku standard 3 624 R 2025-12-13T07:56:43 2025-12-13T07:56:43 10:51:49 1-00:00:00 hai[001-003]\r\n]0;franz.srambical@hai-login1:~/crowd-pilot-serializer",,terminal_output +8,8439,"TERMINAL",0,0,"bash",,terminal_focus +9,9983,"TERMINAL",0,0,"bash",,terminal_focus +10,27972,"TERMINAL",0,0,"source .venv/bin/activate",,terminal_command +11,27988,"TERMINAL",0,0,"]633;C]0;franz.srambical@hai-login1:~/crowd-pilot-serializer",,terminal_output +12,30366,"TERMINAL",0,0,"cargo clean",,terminal_command +13,30367,"TERMINAL",0,0,"]633;C",,terminal_output +14,30559,"TERMINAL",0,0," Cleaning [ ] 0.00% \r",,terminal_output +15,30651,"TERMINAL",0,0," Cleaning [ ] 0.63% \r",,terminal_output +16,30739,"TERMINAL",0,0," Cleaning [ ] 1.28% \r",,terminal_output +17,30856,"TERMINAL",0,0," Cleaning [ ] 1.91% \r",,terminal_output +18,31022,"TERMINAL",0,0," Cleaning [ ] 2.58% \r",,terminal_output +19,31127,"TERMINAL",0,0," Cleaning [ ] 3.21% \r",,terminal_output +20,31249,"TERMINAL",0,0," Cleaning [ ] 3.89% \r Cleaning [> ] 4.58% \r",,terminal_output +21,31341,"TERMINAL",0,0," Cleaning [> ] 5.30% \r",,terminal_output +22,31473,"TERMINAL",0,0," Cleaning [> ] 6.09% \r",,terminal_output +23,31558,"TERMINAL",0,0," Cleaning [> ] 6.97% \r",,terminal_output +24,31713,"TERMINAL",0,0," Cleaning [> ] 7.82% \r",,terminal_output +25,31775,"TERMINAL",0,0," Cleaning [=> ] 8.70% \r",,terminal_output +26,31883,"TERMINAL",0,0," Cleaning [=> ] 9.57% \r",,terminal_output +27,32024,"TERMINAL",0,0," Cleaning [=> ] 10.43% \r",,terminal_output +28,32160,"TERMINAL",0,0," Cleaning [=> ] 11.28% \r",,terminal_output +29,32237,"TERMINAL",0,0," Cleaning [==> ] 12.13% \r",,terminal_output +30,32307,"TERMINAL",0,0," Cleaning [==> ] 12.99% \r",,terminal_output +31,32403,"TERMINAL",0,0," Cleaning [==> ] 13.82% \r",,terminal_output +32,32476,"TERMINAL",0,0," Cleaning [==> ] 14.65% \r",,terminal_output +33,32598,"TERMINAL",0,0," Cleaning [==> ] 15.51% \r",,terminal_output +34,32705,"TERMINAL",0,0," Cleaning [===> ] 16.36% \r",,terminal_output +35,32785,"TERMINAL",0,0," Cleaning [===> ] 17.21% \r",,terminal_output +36,32907,"TERMINAL",0,0," Cleaning [===> ] 18.07% \r",,terminal_output +37,32989,"TERMINAL",0,0," Cleaning [===> ] 18.92% \r",,terminal_output +38,33122,"TERMINAL",0,0," Cleaning [===> ] 19.78% \r",,terminal_output +39,33221,"TERMINAL",0,0," Cleaning [====> ] 20.63% \r",,terminal_output +40,33352,"TERMINAL",0,0," Cleaning [====> ] 21.48% \r",,terminal_output +41,33419,"TERMINAL",0,0," Cleaning [====> ] 22.16% \r",,terminal_output +42,33518,"TERMINAL",0,0," Cleaning [====> ] 22.79% \r",,terminal_output +43,33637,"TERMINAL",0,0," Cleaning [====> ] 23.44% \r",,terminal_output +44,33812,"TERMINAL",0,0," Cleaning [=====> ] 24.07% \r",,terminal_output +45,33894,"TERMINAL",0,0," Cleaning [=====> ] 24.67% \r Cleaning [=====> ] 25.37% \r",,terminal_output +46,34019,"TERMINAL",0,0," Cleaning [=====> ] 26.02% \r",,terminal_output +47,34217,"TERMINAL",0,0," Cleaning [=====> ] 26.72% \r Cleaning [=====> ] 27.33% \r",,terminal_output +48,34297,"TERMINAL",0,0," Cleaning [======> ] 28.02% \r",,terminal_output +49,34407,"TERMINAL",0,0," Cleaning [======> ] 28.74% \r",,terminal_output +50,34526,"TERMINAL",0,0," Cleaning [======> ] 29.42% \r",,terminal_output +51,34743,"TERMINAL",0,0," Cleaning [======> ] 30.09% \r",,terminal_output +52,34746,"TERMINAL",0,0," Cleaning [======> ] 30.76% \r",,terminal_output +53,34813,"TERMINAL",0,0," Cleaning [======> ] 31.42% \r",,terminal_output +54,34910,"TERMINAL",0,0," Cleaning [=======> ] 32.07% \r",,terminal_output +55,35053,"TERMINAL",0,0," Cleaning [=======> ] 32.72% \r",,terminal_output +56,35073,"TERMINAL",0,0," Cleaning [=======> ] 33.33% \rerror: failed to remove file `/fast/home/franz.srambical/crowd-pilot-serializer/target/debug/deps/.nfs000000012d584f9d00000407`\r\n\r\nCaused by:\r\n Device or resource busy (os error 16)\r\n\r\nError: failed to remove directory `/fast/home/franz.srambical/crowd-pilot-serializer/target/debug/deps`\r\n\r\nCaused by:\r\n Device or resource busy (os error 16)\r\n]0;franz.srambical@hai-login1:~/crowd-pilot-serializer",,terminal_output +57,41510,"TERMINAL",0,0,"bash",,terminal_focus +58,43518,"TERMINAL",0,0,"exit",,terminal_command +59,44223,"TERMINAL",0,0,"]633;Cexit\r\n",,terminal_output +60,44226,"TERMINAL",0,0,"bash",,terminal_focus +61,52525,"TERMINAL",0,0,"rm -rf target/",,terminal_command +62,52574,"TERMINAL",0,0,"]633;Crm: cannot remove 'target/debug/deps/.nfs000000012d584f9d00000407': Device or resource busy\r\nrm: cannot remove 'target/debug/deps/.nfs000000012ce413230000040b': Device or resource busy\r\nrm: cannot remove 'target/debug/deps/.nfs000000012cf90e1000000408': Device or resource busy\r\nrm: cannot remove 'target/debug/deps/.nfs000000012cb3183b0000040a': Device or resource busy\r\nrm: cannot remove 'target/debug/deps/.nfs000000012c0f154900000406': Device or resource busy\r\nrm: cannot remove 'target/debug/deps/.nfs000000012c166c5400000409': Device or resource busy\r\nrm: cannot remove 'target/debug/deps/.nfs000000012d1819c400000405': Device or resource busy\r\n",,terminal_output +63,56197,"TERMINAL",0,0,"^C\r\n]0;franz.srambical@hai-login1:~/crowd-pilot-serializer",,terminal_output +64,59615,"TERMINAL",0,0,"squeue",,terminal_command +65,59616,"TERMINAL",0,0,"]633;C JOBID USER PARTITION NODES CPUS ST SUBMIT_TIME START_TIME TIME TIME_LIMIT NODELIST(REASON)\r\n 36313 xiao.liu interacti 1 128 R 2025-12-13T12:56:48 2025-12-13T12:56:48 5:52:38 23:59:00 hai005\r\n 36303 xiao.liu interacti 1 128 R 2025-12-12T23:25:52 2025-12-12T23:25:52 19:23:34 23:59:00 hai006\r\n 36317 mihir.maha standard 1 10 R 2025-12-13T15:27:10 2025-12-13T15:27:10 3:22:16 1-00:00:00 hai004\r\n 36314 xiao.liu standard 1 128 R 2025-12-13T13:27:57 2025-12-13T15:13:56 3:35:30 23:59:00 hai007\r\n 36304 nishant.ku standard 3 624 R 2025-12-13T07:56:43 2025-12-13T07:56:43 10:52:43 1-00:00:00 hai[001-003]\r\n]0;franz.srambical@hai-login1:~/crowd-pilot-serializer",,terminal_output diff --git a/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-2790c5c3-9c96-49ac-ab9e-481a8033e7cd1757945282635-2025_09_15-16.08.09.625/source.csv b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-2790c5c3-9c96-49ac-ab9e-481a8033e7cd1757945282635-2025_09_15-16.08.09.625/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..aa5137098809697ab8324c45c13c2ecf2a6ff1bf --- /dev/null +++ b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-2790c5c3-9c96-49ac-ab9e-481a8033e7cd1757945282635-2025_09_15-16.08.09.625/source.csv @@ -0,0 +1,861 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +2,193,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"4:08:09 PM [info] Activating crowd-code\n4:08:09 PM [info] Recording started\n4:08:09 PM [info] Initializing git provider using file system watchers...\n",Log,tab +3,252,"extension-output-pdoom-org.crowd-code-#1-crowd-code",150,0,"4:08:09 PM [info] Git repository found\n4:08:09 PM [info] Git provider initialized successfully\n4:08:09 PM [info] Initial git state: [object Object]\n",Log,content +4,4649,"TERMINAL",0,0,"",,terminal_command +5,11439,"TERMINAL",0,0,"",,terminal_command +6,119465,"TERMINAL",0,0,"",,terminal_command +7,1036533,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"import os\nimport math\nfrom dataclasses import dataclass\n\nimport tyro\nimport tensorflow as tf\nimport tensorflow_datasets as tfds\nimport rlds\nfrom tqdm import tqdm\n\n\n@dataclass\nclass DownloadDQNReplayPNGs:\n """"""CLI options for downloading frames from the RLU DQN Replay (Atari) dataset.\n\n The dataset name follows: rlu_atari_checkpoints_ordered/{game}_run_{run_number}\n """"""\n\n game: str = ""Pong""\n run_number: int = 1\n output_dir: str = ""data/dqn_replay_pngs""\n\n # Percentage of episodes to load per split (0 < percent <= 100)\n data_percent: float = 100.0\n\n # Save every Nth frame (1 = save all frames)\n frame_stride: int = 1\n\n # Limit total episodes processed across all splits (-1 for no limit)\n max_episodes: int = -1\n\n # Overwrite existing PNG files if they exist\n overwrite: bool = False\n\n\ndef _ensure_dir(path: str) -> None:\n os.makedirs(path, exist_ok=True)\n\n\ndef _normalize_image_to_uint8(image: tf.Tensor) -> tf.Tensor:\n """"""Converts an image tensor to uint8 [H, W, C].""""""\n # Ensure rank 2 or 3\n if image.shape.rank == 2:\n image = tf.expand_dims(image, axis=-1)\n # Some datasets might store a trailing singleton channel; keep as-is\n if image.dtype != tf.uint8:\n image = tf.image.convert_image_dtype(image, dtype=tf.uint8, saturate=True)\n return image\n\n\ndef _build_split_string(builder: tfds.core.DatasetBuilder, percent: float) -> str:\n """"""Constructs a tfds split string covering all available splits.\n\n Applies a per-split episode percentage if percent < 100.\n """"""\n splits = list(builder.info.splits.keys())\n if not splits:\n raise RuntimeError(""No splits found for the specified dataset."")\n\n if percent >= 100.0:\n return ""+"".join(splits)\n\n split_parts = []\n for split in splits:\n num_episodes = builder.info.splits[split].num_examples or 0\n take_n = max(1, int(math.floor((percent / 100.0) * num_episodes)))\n split_parts.append(f""{split}[:{take_n}]"")\n return ""+"".join(split_parts)\n\n\ndef download_pngs(\n game: str,\n run_number: int,\n output_dir: str,\n data_percent: float,\n frame_stride: int,\n max_episodes: int,\n overwrite: bool,\n) -> None:\n dataset_name = f""rlu_atari_checkpoints_ordered/{game}_run_{run_number}""\n\n # Prepare output directory path like: output_dir/{game}_run_{run_number}/episode_xxx\n base_output_dir = os.path.join(output_dir, f""{game}_run_{run_number}"")\n _ensure_dir(base_output_dir)\n\n # Discover available splits and construct split selection string\n builder = tfds.builder(dataset_name)\n split_str = _build_split_string(builder, data_percent)\n\n # Load the dataset of episodes (RLDS format)\n episodes = tfds.load(\n dataset_name,\n split=split_str,\n shuffle_files=True,\n read_config=tfds.ReadConfig(enable_ordering_guard=False),\n )\n\n episodes_processed = 0\n frames_saved = 0\n\n episode_bar = tqdm(desc=""Episodes"", unit=""ep"")\n\n for episode in episodes:\n if max_episodes > 0 and episodes_processed >= max_episodes:\n break\n\n episode_dir = os.path.join(base_output_dir, f""episode_{episodes_processed:06d}"")\n _ensure_dir(episode_dir)\n\n step_bar = tqdm(desc=f""Frames e{episodes_processed:06d}"", unit=""f"", leave=False)\n step_idx = 0\n # Iterate steps within an episode\n for step in episode[rlds.STEPS]:\n if frame_stride > 1 and (step_idx % frame_stride) != 0:\n step_idx += 1\n continue\n\n obs = step[rlds.OBSERVATION]\n img = _normalize_image_to_uint8(obs)\n\n # Ensure shape is [H, W, C]\n if img.shape.rank == 2:\n img = tf.expand_dims(img, axis=-1)\n\n file_path = os.path.join(episode_dir, f""frame_{step_idx:06d}.png"")\n if not overwrite and os.path.exists(file_path):\n step_idx += 1\n step_bar.update(1)\n continue\n\n png_bytes = tf.io.encode_png(img)\n tf.io.write_file(file_path, png_bytes)\n frames_saved += 1\n step_idx += 1\n step_bar.update(1)\n\n step_bar.close()\n episodes_processed += 1\n episode_bar.update(1)\n\n episode_bar.close()\n print(\n f""Done. Episodes processed: {episodes_processed}, frames saved: {frames_saved}. Output: {base_output_dir}""\n )\n\n\nif __name__ == ""__main__"":\n args = tyro.cli(DownloadDQNReplayPNGs)\n\n print(f""Game: {args.game}"")\n print(f""Run number: {args.run_number}"")\n print(f""Output directory: {args.output_dir}"")\n print(f""Data percent: {args.data_percent}"")\n print(f""Frame stride: {args.frame_stride}"")\n print(f""Max episodes: {args.max_episodes}"")\n print(f""Overwrite: {args.overwrite}"")\n\n download_pngs(\n game=args.game,\n run_number=args.run_number,\n output_dir=args.output_dir,\n data_percent=args.data_percent,\n frame_stride=args.frame_stride,\n max_episodes=args.max_episodes,\n overwrite=args.overwrite,\n )\n\n\n",python,tab +8,1038059,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +9,1038464,"input_pipeline/download/dqn_replay/download_pngs.py",5068,0,"",python,selection_command +10,1038741,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +11,1039231,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,selection_command +12,1041354,"input_pipeline/download/dqn_replay/download_pngs.py",905,0,"",python,selection_keyboard +13,1041923,"input_pipeline/download/dqn_replay/download_pngs.py",2149,0,"",python,selection_keyboard +14,1042558,"input_pipeline/download/dqn_replay/download_pngs.py",3473,0,"",python,selection_keyboard +15,1043049,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_keyboard +16,1043586,"input_pipeline/download/dqn_replay/download_pngs.py",5070,0,"",python,selection_keyboard +17,1044058,"input_pipeline/download/dqn_replay/download_pngs.py",3978,0,"",python,selection_keyboard +18,1044222,"input_pipeline/download/dqn_replay/download_pngs.py",2746,0,"",python,selection_keyboard +19,1044382,"input_pipeline/download/dqn_replay/download_pngs.py",1546,0,"",python,selection_keyboard +20,1044522,"input_pipeline/download/dqn_replay/download_pngs.py",372,0,"",python,selection_keyboard +21,1044664,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,selection_keyboard +22,1085419,"input_pipeline/download/dqn_replay/download_pngs.py",10,0,"",python,selection_command +23,1085670,"input_pipeline/download/dqn_replay/download_pngs.py",22,0,"",python,selection_command +24,1085687,"input_pipeline/download/dqn_replay/download_pngs.py",56,0,"",python,selection_command +25,1085839,"input_pipeline/download/dqn_replay/download_pngs.py",57,0,"",python,selection_command +26,1085983,"input_pipeline/download/dqn_replay/download_pngs.py",69,0,"",python,selection_command +27,1086108,"input_pipeline/download/dqn_replay/download_pngs.py",93,0,"",python,selection_command +28,1086480,"input_pipeline/download/dqn_replay/download_pngs.py",100,0,"",python,selection_command +29,1092804,"requirements.txt",0,0,"# core requirements\ndm_pix>=0.4.3\neinops>=0.8.0\nflax>=0.10.7\njax[cuda12]>=0.6.2\noptax>=0.2.3\ntyro>=0.8.5\nwandb>=0.17.4\ngrain>=0.2.10\narray-record>=0.7.2\n\n# data pipeline\ngsutil>=5.35\nffmpeg-python==0.2.0\nhf-transfer==0.1.9\nhuggingface-hub[cli]>=0.34.3\nprocgen>=0.10.7\ntqdm>=4.67.1\n\n# dev\npre-commit>=4.2.0",pip-requirements,tab +30,1093928,"requirements.txt",194,0,"",pip-requirements,selection_command +31,1094082,"requirements.txt",215,0,"",pip-requirements,selection_command +32,1094223,"requirements.txt",234,0,"",pip-requirements,selection_command +33,1094364,"requirements.txt",263,0,"",pip-requirements,selection_command +34,1094498,"requirements.txt",279,0,"",pip-requirements,selection_command +35,1094841,"requirements.txt",281,0,"",pip-requirements,selection_command +36,1095232,"requirements.txt",279,0,"",pip-requirements,selection_command +37,1097892,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,tab +38,1099911,"requirements.txt",0,0,"",pip-requirements,tab +39,1102774,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,tab +40,1103367,"requirements.txt",0,0,"",pip-requirements,tab +41,1108218,"requirements.txt",281,0,"",pip-requirements,selection_command +42,1108957,"requirements.txt",268,0,"",pip-requirements,selection_command +43,1110216,"requirements.txt",280,0,"\n",pip-requirements,content +44,1110313,"requirements.txt",281,0,"t",pip-requirements,content +45,1110314,"requirements.txt",282,0,"",pip-requirements,selection_keyboard +46,1110365,"requirements.txt",282,0,"e",pip-requirements,content +47,1110366,"requirements.txt",283,0,"",pip-requirements,selection_keyboard +48,1110433,"requirements.txt",283,0,"n",pip-requirements,content +49,1110434,"requirements.txt",284,0,"",pip-requirements,selection_keyboard +50,1110526,"requirements.txt",284,0,"s",pip-requirements,content +51,1110526,"requirements.txt",285,0,"",pip-requirements,selection_keyboard +52,1110669,"requirements.txt",285,0,"o",pip-requirements,content +53,1110669,"requirements.txt",286,0,"",pip-requirements,selection_keyboard +54,1110798,"requirements.txt",286,0,"r",pip-requirements,content +55,1110798,"requirements.txt",287,0,"",pip-requirements,selection_keyboard +56,1111162,"requirements.txt",287,0,"f",pip-requirements,content +57,1111162,"requirements.txt",288,0,"",pip-requirements,selection_keyboard +58,1111171,"requirements.txt",288,0,"l",pip-requirements,content +59,1111171,"requirements.txt",289,0,"",pip-requirements,selection_keyboard +60,1111264,"requirements.txt",289,0,"o",pip-requirements,content +61,1111265,"requirements.txt",290,0,"",pip-requirements,selection_keyboard +62,1111326,"requirements.txt",290,0,"w",pip-requirements,content +63,1111326,"requirements.txt",291,0,"",pip-requirements,selection_keyboard +64,1112133,"requirements.txt",291,0,">=2.19.0",pip-requirements,content +65,1112757,"requirements.txt",299,0,"\n",pip-requirements,content +66,1113355,"requirements.txt",300,0,"tensorflow-datasets>=4.12.0\nrlds>=0.1.1",pip-requirements,content +67,1114038,"requirements.txt",338,0,"",pip-requirements,selection_command +68,1114629,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,tab +69,1115320,"input_pipeline/download/dqn_replay/download_pngs.py",135,0,"",python,selection_command +70,1116495,"input_pipeline/download/dqn_replay/download_pngs.py",2685,0,"",python,selection_command +71,1120322,"input_pipeline/download/dqn_replay/download_pngs.py",3392,0,"",python,selection_command +72,1120973,"input_pipeline/download/dqn_replay/download_pngs.py",2685,0,"",python,selection_command +73,1123887,"input_pipeline/download/dqn_replay/download_pngs.py",3392,0,"",python,selection_command +74,1131647,"input_pipeline/download/dqn_replay/download_pngs.py",3552,0,"",python,selection_command +75,1132527,"input_pipeline/download/dqn_replay/download_pngs.py",3392,0,"",python,selection_command +76,1134947,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,selection_command +77,1135462,"input_pipeline/download/dqn_replay/download_pngs.py",10,0,"",python,selection_command +78,1135704,"input_pipeline/download/dqn_replay/download_pngs.py",22,0,"",python,selection_command +79,1135738,"input_pipeline/download/dqn_replay/download_pngs.py",56,0,"",python,selection_command +80,1135772,"input_pipeline/download/dqn_replay/download_pngs.py",57,0,"",python,selection_command +81,1135801,"input_pipeline/download/dqn_replay/download_pngs.py",69,0,"",python,selection_command +82,1135835,"input_pipeline/download/dqn_replay/download_pngs.py",93,0,"",python,selection_command +83,1135954,"input_pipeline/download/dqn_replay/download_pngs.py",128,0,"",python,selection_command +84,1136140,"input_pipeline/download/dqn_replay/download_pngs.py",140,0,"",python,selection_command +85,1562730,"input_pipeline/download/dqn_replay/download_pngs.py",128,0,"",python,selection_command +86,1562766,"input_pipeline/download/dqn_replay/download_pngs.py",93,0,"",python,selection_command +87,1562901,"input_pipeline/download/dqn_replay/download_pngs.py",69,0,"",python,selection_command +88,1563155,"input_pipeline/download/dqn_replay/download_pngs.py",76,0,"",python,selection_command +89,1563341,"input_pipeline/download/dqn_replay/download_pngs.py",87,0,"",python,selection_command +90,1563783,"input_pipeline/download/dqn_replay/download_pngs.py",76,0,"",python,selection_command +91,1565924,"input_pipeline/download/dqn_replay/download_pngs.py",87,0,"",python,selection_command +92,1566074,"input_pipeline/download/dqn_replay/download_pngs.py",90,0,"",python,selection_command +93,1566423,"input_pipeline/download/dqn_replay/download_pngs.py",943,0,"",python,selection_command +94,1568798,"input_pipeline/download/dqn_replay/download_pngs.py",957,0,"",python,selection_command +95,1568914,"input_pipeline/download/dqn_replay/download_pngs.py",1094,0,"",python,selection_command +96,1569063,"input_pipeline/download/dqn_replay/download_pngs.py",1220,0,"",python,selection_command +97,1569316,"input_pipeline/download/dqn_replay/download_pngs.py",1246,0,"",python,selection_command +98,1569344,"input_pipeline/download/dqn_replay/download_pngs.py",1288,0,"",python,selection_command +99,1570325,"input_pipeline/download/dqn_replay/download_pngs.py",3718,0,"",python,selection_command +100,1571085,"input_pipeline/download/dqn_replay/download_pngs.py",1288,0,"",python,selection_command +101,1571160,"input_pipeline/download/dqn_replay/download_pngs.py",1246,0,"",python,selection_command +102,1574812,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,selection_command +103,1575409,"input_pipeline/download/dqn_replay/download_pngs.py",10,0,"",python,selection_command +104,1575652,"input_pipeline/download/dqn_replay/download_pngs.py",22,0,"",python,selection_command +105,1576008,"input_pipeline/download/dqn_replay/download_pngs.py",56,0,"",python,selection_command +106,1576261,"input_pipeline/download/dqn_replay/download_pngs.py",57,0,"",python,selection_command +107,1576284,"input_pipeline/download/dqn_replay/download_pngs.py",69,0,"",python,selection_command +108,1576309,"input_pipeline/download/dqn_replay/download_pngs.py",93,0,"",python,selection_command +109,1576345,"input_pipeline/download/dqn_replay/download_pngs.py",128,0,"",python,selection_command +110,1576379,"input_pipeline/download/dqn_replay/download_pngs.py",140,0,"",python,selection_command +111,1576559,"input_pipeline/download/dqn_replay/download_pngs.py",162,0,"",python,selection_command +112,1576779,"input_pipeline/download/dqn_replay/download_pngs.py",140,0,"",python,selection_command +113,1576922,"input_pipeline/download/dqn_replay/download_pngs.py",128,0,"",python,selection_command +114,1577101,"input_pipeline/download/dqn_replay/download_pngs.py",93,0,"",python,selection_command +115,1578587,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +116,1579533,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,tab +117,1579850,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +118,1581490,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,tab +119,1581492,"TERMINAL",0,0,"",,terminal_focus +120,1582018,"TERMINAL",0,0,"source /home/franz.srambical/jafar/.venv/bin/activate",,terminal_command +121,1582020,"TERMINAL",0,0,"]633;C]0;franz.srambical@hai-login1:~/jafar",,terminal_output +122,1584479,"TERMINAL",0,0,"which python",,terminal_command +123,1584486,"TERMINAL",0,0,"]633;C/fast/home/franz.srambical/jafar/.venv/bin/python\r\n]0;franz.srambical@hai-login1:~/jafar",,terminal_output +124,1592170,"TERMINAL",0,0,"uv pip install tensorflow_datasets",,terminal_command +125,1592216,"TERMINAL",0,0,"]633;C",,terminal_output +126,1592344,"TERMINAL",0,0,"⠋ Resolving dependencies... \r⠙ Resolving dependencies... \r⠋ Resolving dependencies... \r⠙ Resolving dependencies... ",,terminal_output +127,1592543,"TERMINAL",0,0,"\r⠹ Resolving dependencies... ",,terminal_output +128,1592763,"TERMINAL",0,0,"\r⠸ Resolving dependencies... \r⠸ tensorflow-datasets==4.9.9 ",,terminal_output +129,1592887,"TERMINAL",0,0,"\r⠸ absl-py==2.3.1 \r⠸ array-record==0.7.2 \r⠸ array-record==0.7.2 ",,terminal_output +130,1592940,"TERMINAL",0,0,"\r⠼ array-record==0.7.2 ",,terminal_output +131,1593021,"TERMINAL",0,0,"\r⠼ dm-tree==0.1.9 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ etils==1.12.2 \r⠼ immutabledict==4.2.1 \r⠼ numpy==1.26.4 \r⠼ numpy==1.26.4 \r⠼ promise==2.3 ",,terminal_output +132,1593142,"TERMINAL",0,0,"\r⠼ protobuf==5.29.5 \r⠴ protobuf==5.29.5 ",,terminal_output +133,1593232,"TERMINAL",0,0,"\r⠴ psutil==7.0.0 ",,terminal_output +134,1593345,"TERMINAL",0,0,"\r⠦ psutil==7.0.0 ",,terminal_output +135,1593430,"TERMINAL",0,0,"\r⠦ pyarrow==21.0.0 ",,terminal_output +136,1593687,"TERMINAL",0,0,"\r⠦ requests==2.32.4 \r⠦ simple-parsing==0.1.7 \r⠦ tensorflow-metadata==1.17.2 \r⠦ psutil==7.0.0 \r⠦ pyarrow==21.0.0 \r⠦ requests==2.32.4 \r⠦ simple-parsing==0.1.7 \r⠦ tensorflow-metadata==1.17.1 \r⠧ protobuf==4.21.12 \rResolved 30 packages in 1.22s\r\n⠋ Preparing packages... (0/0) \r⠋ Preparing packages... (0/3) \r⠙ Preparing packages... (0/3) \r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 0 B/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 16.00 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 32.00 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 48.00 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 63.97 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 79.97 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 95.97 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 111.97 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 127.97 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\npyarrow  ------------------------------ 127.97 KiB/40.75 MiB \r\r⠙ Preparing packages... (0/3)\r\nprotobuf  ------------------------------ 0 B/400.23 KiB\r\npyarrow  ------------------------------ 127.97 KiB/40.75 MiB \r\r\r⠙ Preparing packages... (0/3)\r\nprotobuf  ------------------------------ 0 B/400.23 KiB\r\npyarrow  ------------------------------ 127.97 KiB/40.75 MiB \r\r\r⠙ Preparing packages... (0/3)\r\nprotobuf  ------------------------------ 0 B/400.23 KiB\r\npyarrow  ------------------------------ 127.97 KiB/40.75 MiB \r\r\r⠙ Preparing packages... (0/3)\r\ntensorflow-metadata  ------------------------------ 0 B/30.80 KiB\r\nprotobuf  ------------------------------ 0 B/400.23 KiB\r\npyarrow  ------------------------------ 127.97 KiB/40.75 MiB \r\r\r\r⠙ Preparing packages... (0/3)\r\ntensorflow-metadata  ------------------------------ 14.90 KiB/30.80 KiB\r\nprotobuf  ------------------------------ 0 B/400.23 KiB\r\npyarrow  ------------------------------ 127.97 KiB/40.75 MiB \r\r\r\r⠙ Preparing packages... (0/3)\r\ntensorflow-metadata  ------------------------------ 14.90 KiB/30.80 KiB\r\nprotobuf  ------------------------------ 0 B/400.23 KiB\r\npyarrow  ------------------------------ 143.97 KiB/40.75 MiB \r\r\r\r⠙ Preparing packages... (0/3)\r\ntensorflow-metadata  ------------------------------ 14.90 KiB/30.80 KiB\r\nprotobuf  ------------------------------ 0 B/400.23 KiB\r\npyarrow  ------------------------------ 159.97 KiB/40.75 MiB \r\r\r\r⠙ Preparing packages... (0/3)\r\ntensorflow-metadata  ------------------------------ 14.90 KiB/30.80 KiB\r\nprotobuf  ------------------------------ 16.00 KiB/400.23 KiB\r\npyarrow  ------------------------------ 191.97 KiB/40.75 MiB \r\r\r\r⠙ Preparing packages... (0/3)\r\nprotobuf  ------------------------------ 144.00 KiB/400.23 KiB\r\npyarrow  ------------------------------ 687.97 KiB/40.75 MiB \r\r\r⠙ Preparing packages... (0/3)\r\nprotobuf  ------------------------------ 176.00 KiB/400.23 KiB\r\npyarrow  ------------------------------ 719.97 KiB/40.75 MiB ",,terminal_output +137,1593990,"TERMINAL",0,0,"\r\r\r⠙ Preparing packages... (0/3)\r\nprotobuf  ------------------------------ 288.00 KiB/400.23 KiB\r\npyarrow  ------------------------------ 1.29 MiB/40.75 MiB \r\r\r⠹ Preparing packages... (1/3)\r\nprotobuf  ------------------------------ 383.89 KiB/400.23 KiB\r\npyarrow  ------------------------------ 1.95 MiB/40.75 MiB \r\r\r⠹ Preparing packages... (1/3)\r\npyarrow  ------------------------------ 2.36 MiB/40.75 MiB \r\r⠹ Preparing packages... (1/3)\r\npyarrow  ------------------------------ 2.56 MiB/40.75 MiB \r\r⠹ Preparing packages... (1/3)\r\npyarrow  ------------------------------ 3.08 MiB/40.75 MiB \r\r⠹ Preparing packages... (1/3)\r\npyarrow  ------------------------------ 4.44 MiB/40.75 MiB \r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 7.09 MiB/40.75 MiB ",,terminal_output +138,1594044,"TERMINAL",0,0,"\r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 10.42 MiB/40.75 MiB ",,terminal_output +139,1594190,"TERMINAL",0,0,"\r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 12.92 MiB/40.75 MiB \r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 14.99 MiB/40.75 MiB \r\r⠼ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 16.98 MiB/40.75 MiB ",,terminal_output +140,1594244,"TERMINAL",0,0,"\r\r⠼ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 18.42 MiB/40.75 MiB ",,terminal_output +141,1594343,"TERMINAL",0,0,"\r\r⠼ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 19.82 MiB/40.75 MiB \r\r⠼ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 21.81 MiB/40.75 MiB ",,terminal_output +142,1594412,"TERMINAL",0,0,"\r\r⠴ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 23.64 MiB/40.75 MiB ",,terminal_output +143,1594491,"TERMINAL",0,0,"\r\r⠴ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 24.62 MiB/40.75 MiB \r\r⠴ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 26.92 MiB/40.75 MiB ",,terminal_output +144,1594587,"TERMINAL",0,0,"\r\r⠴ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 29.53 MiB/40.75 MiB \r\r⠴ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 30.90 MiB/40.75 MiB ",,terminal_output +145,1594740,"TERMINAL",0,0,"\r\r⠦ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 32.25 MiB/40.75 MiB \r\r⠦ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 33.73 MiB/40.75 MiB \r\r⠦ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 36.14 MiB/40.75 MiB ",,terminal_output +146,1594797,"TERMINAL",0,0,"\r\r⠦ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 38.57 MiB/40.75 MiB ",,terminal_output +147,1594892,"TERMINAL",0,0,"\r\r⠧ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 38.81 MiB/40.75 MiB \r\r⠧ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 38.89 MiB/40.75 MiB ",,terminal_output +148,1594985,"TERMINAL",0,0,"\r\r⠧ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 38.97 MiB/40.75 MiB \r\r⠇ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.01 MiB/40.75 MiB ",,terminal_output +149,1595188,"TERMINAL",0,0,"\r\r⠇ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.11 MiB/40.75 MiB \r\r⠇ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.15 MiB/40.75 MiB \r\r⠇ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.20 MiB/40.75 MiB \r\r⠋ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.25 MiB/40.75 MiB ",,terminal_output +150,1595286,"TERMINAL",0,0,"\r\r⠋ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.32 MiB/40.75 MiB \r\r⠋ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.36 MiB/40.75 MiB ",,terminal_output +151,1595389,"TERMINAL",0,0,"\r\r⠋ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.40 MiB/40.75 MiB \r\r⠙ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.46 MiB/40.75 MiB ",,terminal_output +152,1595444,"TERMINAL",0,0,"\r\r⠙ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.54 MiB/40.75 MiB ",,terminal_output +153,1595536,"TERMINAL",0,0,"\r\r⠙ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.61 MiB/40.75 MiB \r\r⠙ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.67 MiB/40.75 MiB ",,terminal_output +154,1595588,"TERMINAL",0,0,"\r\r⠹ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.79 MiB/40.75 MiB ",,terminal_output +155,1595648,"TERMINAL",0,0,"\r\r⠹ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.89 MiB/40.75 MiB ",,terminal_output +156,1595787,"TERMINAL",0,0,"\r\r⠹ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 39.95 MiB/40.75 MiB \r\r⠹ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 40.03 MiB/40.75 MiB \r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 40.08 MiB/40.75 MiB ",,terminal_output +157,1595989,"TERMINAL",0,0,"\r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 40.16 MiB/40.75 MiB \r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 40.28 MiB/40.75 MiB \r\r⠸ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 40.42 MiB/40.75 MiB \r\r⠼ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 40.55 MiB/40.75 MiB ",,terminal_output +158,1596080,"TERMINAL",0,0,"\r\r⠼ Preparing packages... (2/3)\r\npyarrow  ------------------------------ 40.66 MiB/40.75 MiB \r\r⠼ Preparing packages... (2/3) \rPrepared 3 packages in 2.49s\r\n",,terminal_output +159,1596276,"TERMINAL",0,0,"Uninstalled 1 package in 192ms\r\n░░░░░░░░░░░░░░░░░░░░ [0/0] Installing wheels... \r░░░░░░░░░░░░░░░░░░░░ [0/8] Installing wheels... ",,terminal_output +160,1596419,"TERMINAL",0,0,"\r░░░░░░░░░░░░░░░░░░░░ [0/8] immutabledict==4.2.1 \r██░░░░░░░░░░░░░░░░░░ [1/8] immutabledict==4.2.1 \r██░░░░░░░░░░░░░░░░░░ [1/8] tensorflow-metadata==1.17.2 \r█████░░░░░░░░░░░░░░░ [2/8] tensorflow-metadata==1.17.2 \r█████░░░░░░░░░░░░░░░ [2/8] toml==0.10.2 \r███████░░░░░░░░░░░░░ [3/8] toml==0.10.2 \r███████░░░░░░░░░░░░░ [3/8] promise==2.3 \r██████████░░░░░░░░░░ [4/8] promise==2.3 \r██████████░░░░░░░░░░ [4/8] protobuf==4.21.12 \r████████████░░░░░░░░ [5/8] protobuf==4.21.12 \r████████████░░░░░░░░ [5/8] simple-parsing==0.1.7 \r███████████████░░░░░ [6/8] simple-parsing==0.1.7 ",,terminal_output +161,1596869,"TERMINAL",0,0,"\r███████████████░░░░░ [6/8] pyarrow==21.0.0 \r█████████████████░░░ [7/8] pyarrow==21.0.0 ",,terminal_output +162,1601381,"TERMINAL",0,0,"\r█████████████████░░░ [7/8] tensorflow-datasets==4.9.9 \r████████████████████ [8/8] tensorflow-datasets==4.9.9 \rInstalled 8 packages in 5.10s\r\n + immutabledict==4.2.1\r\n + promise==2.3\r\n - protobuf==5.29.5\r\n + protobuf==4.21.12\r\n + pyarrow==21.0.0\r\n + simple-parsing==0.1.7\r\n + tensorflow-datasets==4.9.9\r\n + tensorflow-metadata==1.17.2\r\n + toml==0.10.2\r\n]0;franz.srambical@hai-login1:~/jafar",,terminal_output +163,1601986,"input_pipeline/download/dqn_replay/download_pngs.py",128,0,"",python,selection_command +164,1602321,"input_pipeline/download/dqn_replay/download_pngs.py",93,0,"",python,selection_command +165,1602937,"input_pipeline/download/dqn_replay/download_pngs.py",128,0,"",python,selection_command +166,1603184,"input_pipeline/download/dqn_replay/download_pngs.py",140,0,"",python,selection_command +167,1603226,"input_pipeline/download/dqn_replay/download_pngs.py",162,0,"",python,selection_command +168,1603243,"input_pipeline/download/dqn_replay/download_pngs.py",163,0,"",python,selection_command +169,1603311,"input_pipeline/download/dqn_replay/download_pngs.py",164,0,"",python,selection_command +170,1603895,"input_pipeline/download/dqn_replay/download_pngs.py",175,0,"",python,selection_command +171,1604785,"input_pipeline/download/dqn_replay/download_pngs.py",1332,0,"",python,selection_keyboard +172,1605048,"input_pipeline/download/dqn_replay/download_pngs.py",2648,0,"",python,selection_keyboard +173,1605200,"input_pipeline/download/dqn_replay/download_pngs.py",3887,0,"",python,selection_keyboard +174,1605645,"input_pipeline/download/dqn_replay/download_pngs.py",5028,0,"",python,selection_keyboard +175,1605807,"input_pipeline/download/dqn_replay/download_pngs.py",5070,0,"",python,selection_keyboard +176,1606429,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +177,1606685,"input_pipeline/download/dqn_replay/download_pngs.py",5068,0,"",python,selection_command +178,1606717,"input_pipeline/download/dqn_replay/download_pngs.py",5062,0,"",python,selection_command +179,1606749,"input_pipeline/download/dqn_replay/download_pngs.py",5028,0,"",python,selection_command +180,1606784,"input_pipeline/download/dqn_replay/download_pngs.py",4988,0,"",python,selection_command +181,1606809,"input_pipeline/download/dqn_replay/download_pngs.py",4948,0,"",python,selection_command +182,1606844,"input_pipeline/download/dqn_replay/download_pngs.py",4908,0,"",python,selection_command +183,1606876,"input_pipeline/download/dqn_replay/download_pngs.py",4872,0,"",python,selection_command +184,1606909,"input_pipeline/download/dqn_replay/download_pngs.py",4836,0,"",python,selection_command +185,1606943,"input_pipeline/download/dqn_replay/download_pngs.py",4812,0,"",python,selection_command +186,1606976,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +187,1607007,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +188,1607042,"input_pipeline/download/dqn_replay/download_pngs.py",4750,0,"",python,selection_command +189,1607075,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +190,1607109,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +191,1607141,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +192,1607258,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +193,1607425,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +194,1607582,"input_pipeline/download/dqn_replay/download_pngs.py",4750,0,"",python,selection_command +195,1607742,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +196,1607935,"input_pipeline/download/dqn_replay/download_pngs.py",4750,0,"",python,selection_command +197,1608191,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +198,1608218,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +199,1608250,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +200,1608283,"input_pipeline/download/dqn_replay/download_pngs.py",4556,0,"",python,selection_command +201,1608316,"input_pipeline/download/dqn_replay/download_pngs.py",4512,0,"",python,selection_command +202,1608349,"input_pipeline/download/dqn_replay/download_pngs.py",4480,0,"",python,selection_command +203,1608383,"input_pipeline/download/dqn_replay/download_pngs.py",4479,0,"",python,selection_command +204,1608519,"input_pipeline/download/dqn_replay/download_pngs.py",4480,0,"",python,selection_command +205,1609264,"input_pipeline/download/dqn_replay/download_pngs.py",4484,0,"",python,selection_command +206,1609878,"input_pipeline/download/dqn_replay/download_pngs.py",4516,0,"",python,selection_command +207,1610664,"input_pipeline/download/dqn_replay/download_pngs.py",4484,0,"",python,selection_command +208,1610979,"input_pipeline/download/dqn_replay/download_pngs.py",4516,0,"",python,selection_command +209,1611236,"input_pipeline/download/dqn_replay/download_pngs.py",4560,0,"",python,selection_command +210,1611402,"input_pipeline/download/dqn_replay/download_pngs.py",4610,0,"",python,selection_command +211,1611798,"input_pipeline/download/dqn_replay/download_pngs.py",4658,0,"",python,selection_command +212,1612134,"input_pipeline/download/dqn_replay/download_pngs.py",4610,0,"",python,selection_command +213,1622355,"TERMINAL",0,0,"uv pip install rlds",,terminal_command +214,1622405,"TERMINAL",0,0,"]633;C",,terminal_output +215,1622483,"TERMINAL",0,0,"⠋ Resolving dependencies... \r⠙ Resolving dependencies... \r⠋ Resolving dependencies... \r⠙ Resolving dependencies... ",,terminal_output +216,1622625,"TERMINAL",0,0,"\r⠙ rlds==0.1.8 ",,terminal_output +217,1622775,"TERMINAL",0,0,"\r⠙ absl-py==2.3.1 \r⠙ numpy==1.26.4 \r⠙  \rResolved 3 packages in 197ms\r\n⠋ Preparing packages... (0/0) \r⠋ Preparing packages... (0/1) \r⠙ Preparing packages... (0/1) \r⠙ Preparing packages... (0/1)\r\nrlds  ------------------------------ 0 B/47.29 KiB \r\r⠙ Preparing packages... (0/1)\r\nrlds  ------------------------------ 14.92 KiB/47.29 KiB \r\r⠙ Preparing packages... (0/1)\r\nrlds  ------------------------------ 30.92 KiB/47.29 KiB \r\r⠙ Preparing packages... (0/1)\r\nrlds  ------------------------------ 46.92 KiB/47.29 KiB \r\r⠙ Preparing packages... (0/1)\r\nrlds  ------------------------------ 47.29 KiB/47.29 KiB \r\r⠙ Preparing packages... (0/1) \r⠙  (1/1) \rPrepared 1 package in 83ms\r\n░░░░░░░░░░░░░░░░░░░░ [0/0] Installing wheels... \r░░░░░░░░░░░░░░░░░░░░ [0/1] Installing wheels... ",,terminal_output +218,1622788,"TERMINAL",0,0,"\r░░░░░░░░░░░░░░░░░░░░ [0/1] rlds==0.1.8 \r████████████████████ [1/1] rlds==0.1.8 \rInstalled 1 package in 73ms\r\n + rlds==0.1.8\r\n]0;franz.srambical@hai-login1:~/jafar",,terminal_output +219,1632235,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,selection_command +220,1632709,"input_pipeline/download/dqn_replay/download_pngs.py",10,0,"",python,selection_command +221,1632960,"input_pipeline/download/dqn_replay/download_pngs.py",22,0,"",python,selection_command +222,1632999,"input_pipeline/download/dqn_replay/download_pngs.py",56,0,"",python,selection_command +223,1633029,"input_pipeline/download/dqn_replay/download_pngs.py",57,0,"",python,selection_command +224,1633057,"input_pipeline/download/dqn_replay/download_pngs.py",69,0,"",python,selection_command +225,1633331,"input_pipeline/download/dqn_replay/download_pngs.py",93,0,"",python,selection_command +226,1633480,"input_pipeline/download/dqn_replay/download_pngs.py",128,0,"",python,selection_command +227,1633684,"input_pipeline/download/dqn_replay/download_pngs.py",135,0,"",python,selection_command +228,1633864,"input_pipeline/download/dqn_replay/download_pngs.py",140,0,"",python,selection_command +229,1634214,"input_pipeline/download/dqn_replay/download_pngs.py",135,0,"",python,selection_command +230,1634625,".venv/lib/python3.10/site-packages/rlds/__init__.py",0,0,"# Copyright 2023 Google LLC.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# coding=utf-8\n""""""RLDS basic API.""""""\n\nfrom rlds import metadata\n\n\nfrom rlds import transformations\n\n\nfrom rlds.rlds_types import ACTION\nfrom rlds.rlds_types import ALIGNMENT\nfrom rlds.rlds_types import BatchedEpisode\nfrom rlds.rlds_types import BatchedStep\nfrom rlds.rlds_types import build_episode\nfrom rlds.rlds_types import build_step\nfrom rlds.rlds_types import CORE_STEP_FIELDS\nfrom rlds.rlds_types import DISCOUNT\nfrom rlds.rlds_types import Episode\nfrom rlds.rlds_types import IS_FIRST\nfrom rlds.rlds_types import IS_LAST\nfrom rlds.rlds_types import IS_TERMINAL\nfrom rlds.rlds_types import OBSERVATION\nfrom rlds.rlds_types import REWARD\nfrom rlds.rlds_types import Step\nfrom rlds.rlds_types import STEPS\n",python,tab +231,1635698,".venv/lib/python3.10/site-packages/rlds/__init__.py",1253,0,"",python,selection_keyboard +232,1635807,".venv/lib/python3.10/site-packages/rlds/__init__.py",1287,0,"",python,selection_keyboard +233,1636200,".venv/lib/python3.10/site-packages/rlds/__init__.py",29,0,"",python,selection_keyboard +234,1636405,".venv/lib/python3.10/site-packages/rlds/__init__.py",0,0,"",python,selection_keyboard +235,1847847,".venv/lib/python3.10/site-packages/rlds/__init__.py",1253,0,"",python,selection_keyboard +236,1847999,".venv/lib/python3.10/site-packages/rlds/__init__.py",1287,0,"",python,selection_keyboard +237,1851343,".venv/lib/python3.10/site-packages/rlds/__init__.py",1253,0,"",python,selection_command +238,1851584,".venv/lib/python3.10/site-packages/rlds/__init__.py",1220,0,"",python,selection_command +239,1851608,".venv/lib/python3.10/site-packages/rlds/__init__.py",1185,0,"",python,selection_command +240,1851649,".venv/lib/python3.10/site-packages/rlds/__init__.py",1145,0,"",python,selection_command +241,1851683,".venv/lib/python3.10/site-packages/rlds/__init__.py",1105,0,"",python,selection_command +242,1851716,".venv/lib/python3.10/site-packages/rlds/__init__.py",1069,0,"",python,selection_command +243,1851752,".venv/lib/python3.10/site-packages/rlds/__init__.py",1032,0,"",python,selection_command +244,1851783,".venv/lib/python3.10/site-packages/rlds/__init__.py",996,0,"",python,selection_command +245,1852077,".venv/lib/python3.10/site-packages/rlds/__init__.py",1032,0,"",python,selection_command +246,1852338,".venv/lib/python3.10/site-packages/rlds/__init__.py",1069,0,"",python,selection_command +247,1852365,".venv/lib/python3.10/site-packages/rlds/__init__.py",1105,0,"",python,selection_command +248,1852400,".venv/lib/python3.10/site-packages/rlds/__init__.py",1145,0,"",python,selection_command +249,1852469,".venv/lib/python3.10/site-packages/rlds/__init__.py",1105,0,"",python,selection_command +250,1852729,".venv/lib/python3.10/site-packages/rlds/__init__.py",1069,0,"",python,selection_command +251,1852754,".venv/lib/python3.10/site-packages/rlds/__init__.py",1032,0,"",python,selection_command +252,1852785,".venv/lib/python3.10/site-packages/rlds/__init__.py",996,0,"",python,selection_command +253,1852819,".venv/lib/python3.10/site-packages/rlds/__init__.py",959,0,"",python,selection_command +254,1852849,".venv/lib/python3.10/site-packages/rlds/__init__.py",914,0,"",python,selection_command +255,1852886,".venv/lib/python3.10/site-packages/rlds/__init__.py",875,0,"",python,selection_command +256,1852916,".venv/lib/python3.10/site-packages/rlds/__init__.py",833,0,"",python,selection_command +257,1853041,".venv/lib/python3.10/site-packages/rlds/__init__.py",793,0,"",python,selection_command +258,1853199,".venv/lib/python3.10/site-packages/rlds/__init__.py",750,0,"",python,selection_command +259,1853838,"input_pipeline/download/dqn_replay/download_pngs.py",0,0,"",python,tab +260,1854626,"input_pipeline/download/dqn_replay/download_pngs.py",147,0,"",python,selection_command +261,1854875,"input_pipeline/download/dqn_replay/download_pngs.py",162,0,"",python,selection_command +262,1854905,"input_pipeline/download/dqn_replay/download_pngs.py",163,0,"",python,selection_command +263,1854940,"input_pipeline/download/dqn_replay/download_pngs.py",171,0,"",python,selection_command +264,1854968,"input_pipeline/download/dqn_replay/download_pngs.py",182,0,"",python,selection_command +265,1855182,"input_pipeline/download/dqn_replay/download_pngs.py",1339,0,"",python,selection_keyboard +266,1855335,"input_pipeline/download/dqn_replay/download_pngs.py",2648,0,"",python,selection_keyboard +267,1855478,"input_pipeline/download/dqn_replay/download_pngs.py",3894,0,"",python,selection_keyboard +268,1855620,"input_pipeline/download/dqn_replay/download_pngs.py",5035,0,"",python,selection_keyboard +269,1855743,"input_pipeline/download/dqn_replay/download_pngs.py",5070,0,"",python,selection_keyboard +270,1856643,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +271,1856896,"input_pipeline/download/dqn_replay/download_pngs.py",5068,0,"",python,selection_command +272,1856925,"input_pipeline/download/dqn_replay/download_pngs.py",5062,0,"",python,selection_command +273,1856956,"input_pipeline/download/dqn_replay/download_pngs.py",5028,0,"",python,selection_command +274,1856990,"input_pipeline/download/dqn_replay/download_pngs.py",4988,0,"",python,selection_command +275,1857024,"input_pipeline/download/dqn_replay/download_pngs.py",4948,0,"",python,selection_command +276,1857051,"input_pipeline/download/dqn_replay/download_pngs.py",4908,0,"",python,selection_command +277,1857089,"input_pipeline/download/dqn_replay/download_pngs.py",4872,0,"",python,selection_command +278,1857118,"input_pipeline/download/dqn_replay/download_pngs.py",4836,0,"",python,selection_command +279,1857156,"input_pipeline/download/dqn_replay/download_pngs.py",4812,0,"",python,selection_command +280,1857191,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +281,1857225,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +282,1857258,"input_pipeline/download/dqn_replay/download_pngs.py",4750,0,"",python,selection_command +283,1857291,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +284,1857325,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +285,1857359,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +286,1857393,"input_pipeline/download/dqn_replay/download_pngs.py",4556,0,"",python,selection_command +287,1857561,"input_pipeline/download/dqn_replay/download_pngs.py",4512,0,"",python,selection_command +288,1857703,"input_pipeline/download/dqn_replay/download_pngs.py",4480,0,"",python,selection_command +289,1858272,"input_pipeline/download/dqn_replay/download_pngs.py",4512,0,"",python,selection_command +290,1858450,"input_pipeline/download/dqn_replay/download_pngs.py",4556,0,"",python,selection_command +291,1858598,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +292,1858804,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +293,1858975,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +294,1859121,"input_pipeline/download/dqn_replay/download_pngs.py",4750,0,"",python,selection_command +295,1859294,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +296,1859625,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +297,1860391,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +298,1860943,"input_pipeline/download/dqn_replay/download_pngs.py",4797,0,"",python,selection_command +299,1861832,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +300,1862295,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +301,2088811,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +302,2088956,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +303,2090353,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +304,2090598,"input_pipeline/download/dqn_replay/download_pngs.py",4750,0,"",python,selection_command +305,2090624,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +306,2090661,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +307,2090686,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +308,2090724,"input_pipeline/download/dqn_replay/download_pngs.py",4556,0,"",python,selection_command +309,2090756,"input_pipeline/download/dqn_replay/download_pngs.py",4512,0,"",python,selection_command +310,2090823,"input_pipeline/download/dqn_replay/download_pngs.py",4556,0,"",python,selection_command +311,2091077,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +312,2091118,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +313,2091149,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +314,2091223,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +315,2091480,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +316,2091503,"input_pipeline/download/dqn_replay/download_pngs.py",4556,0,"",python,selection_command +317,2091534,"input_pipeline/download/dqn_replay/download_pngs.py",4512,0,"",python,selection_command +318,2091572,"input_pipeline/download/dqn_replay/download_pngs.py",4480,0,"",python,selection_command +319,2091605,"input_pipeline/download/dqn_replay/download_pngs.py",4479,0,"",python,selection_command +320,2091737,"input_pipeline/download/dqn_replay/download_pngs.py",4436,0,"",python,selection_command +321,2091891,"input_pipeline/download/dqn_replay/download_pngs.py",4409,0,"",python,selection_command +322,2092339,"input_pipeline/download/dqn_replay/download_pngs.py",4436,0,"",python,selection_command +323,2092592,"input_pipeline/download/dqn_replay/download_pngs.py",4479,0,"",python,selection_command +324,2092620,"input_pipeline/download/dqn_replay/download_pngs.py",4480,0,"",python,selection_command +325,2092650,"input_pipeline/download/dqn_replay/download_pngs.py",4512,0,"",python,selection_command +326,2092683,"input_pipeline/download/dqn_replay/download_pngs.py",4556,0,"",python,selection_command +327,2092718,"input_pipeline/download/dqn_replay/download_pngs.py",4606,0,"",python,selection_command +328,2092745,"input_pipeline/download/dqn_replay/download_pngs.py",4654,0,"",python,selection_command +329,2092782,"input_pipeline/download/dqn_replay/download_pngs.py",4702,0,"",python,selection_command +330,2092815,"input_pipeline/download/dqn_replay/download_pngs.py",4750,0,"",python,selection_command +331,2092951,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +332,2093096,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +333,2093404,"input_pipeline/download/dqn_replay/download_pngs.py",4797,0,"",python,selection_command +334,2093748,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +335,2094599,"input_pipeline/download/dqn_replay/download_pngs.py",4797,0,"",python,selection_command +336,2095423,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +337,2203204,"input_pipeline/download/dqn_replay/download_pngs.py",5070,0,"",python,selection_keyboard +338,2203849,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +339,2204105,"input_pipeline/download/dqn_replay/download_pngs.py",5068,0,"",python,selection_command +340,2204131,"input_pipeline/download/dqn_replay/download_pngs.py",5062,0,"",python,selection_command +341,2204156,"input_pipeline/download/dqn_replay/download_pngs.py",5028,0,"",python,selection_command +342,2204195,"input_pipeline/download/dqn_replay/download_pngs.py",4988,0,"",python,selection_command +343,2204230,"input_pipeline/download/dqn_replay/download_pngs.py",4948,0,"",python,selection_command +344,2204259,"input_pipeline/download/dqn_replay/download_pngs.py",4908,0,"",python,selection_command +345,2204295,"input_pipeline/download/dqn_replay/download_pngs.py",4872,0,"",python,selection_command +346,2204330,"input_pipeline/download/dqn_replay/download_pngs.py",4836,0,"",python,selection_command +347,2204358,"input_pipeline/download/dqn_replay/download_pngs.py",4812,0,"",python,selection_command +348,2204442,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +349,2204653,"input_pipeline/download/dqn_replay/download_pngs.py",4792,0,"",python,selection_command +350,2204849,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +351,2205122,"input_pipeline/download/dqn_replay/download_pngs.py",5070,0,"",python,selection_keyboard +352,2205492,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +353,2205740,"input_pipeline/download/dqn_replay/download_pngs.py",5068,0,"",python,selection_command +354,2205774,"input_pipeline/download/dqn_replay/download_pngs.py",5062,0,"",python,selection_command +355,2205804,"input_pipeline/download/dqn_replay/download_pngs.py",5028,0,"",python,selection_command +356,2205829,"input_pipeline/download/dqn_replay/download_pngs.py",4988,0,"",python,selection_command +357,2205901,"input_pipeline/download/dqn_replay/download_pngs.py",5028,0,"",python,selection_command +358,2206288,"input_pipeline/download/dqn_replay/download_pngs.py",5062,0,"",python,selection_command +359,2206536,"input_pipeline/download/dqn_replay/download_pngs.py",5068,0,"",python,selection_command +360,2206562,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +361,2206597,"input_pipeline/download/dqn_replay/download_pngs.py",5070,0,"",python,selection_command +362,2207572,"input_pipeline/download/dqn_replay/download_pngs.py",5069,0,"",python,selection_command +363,2207809,"input_pipeline/download/dqn_replay/download_pngs.py",5068,0,"",python,selection_command +364,2207842,"input_pipeline/download/dqn_replay/download_pngs.py",5062,0,"",python,selection_command +365,2207875,"input_pipeline/download/dqn_replay/download_pngs.py",5028,0,"",python,selection_command +366,2207908,"input_pipeline/download/dqn_replay/download_pngs.py",4988,0,"",python,selection_command +367,2207943,"input_pipeline/download/dqn_replay/download_pngs.py",4948,0,"",python,selection_command +368,2207977,"input_pipeline/download/dqn_replay/download_pngs.py",4908,0,"",python,selection_command +369,2208007,"input_pipeline/download/dqn_replay/download_pngs.py",4872,0,"",python,selection_command +370,2208124,"input_pipeline/download/dqn_replay/download_pngs.py",4836,0,"",python,selection_command +371,2208287,"input_pipeline/download/dqn_replay/download_pngs.py",4812,0,"",python,selection_command +372,2208431,"input_pipeline/download/dqn_replay/download_pngs.py",4793,0,"",python,selection_command +373,2209142,"input_pipeline/download/dqn_replay/download_pngs.py",4797,0,"",python,selection_command +374,2210134,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +375,2211522,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +376,2211767,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +377,2211792,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +378,2211831,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +379,2211866,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +380,2211897,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +381,2211924,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +382,2212098,"input_pipeline/download/dqn_replay/download_pngs.py",2197,0,"",python,selection_command +383,2212256,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +384,2216645,"input_pipeline/download/dqn_replay/download_pngs.py",2221,0,"",python,selection_command +385,2216953,"input_pipeline/download/dqn_replay/download_pngs.py",2223,0,"",python,selection_command +386,2217116,"input_pipeline/download/dqn_replay/download_pngs.py",2224,0,"",python,selection_command +387,2217344,"input_pipeline/download/dqn_replay/download_pngs.py",2225,0,"",python,selection_command +388,2217597,"input_pipeline/download/dqn_replay/download_pngs.py",2254,0,"",python,selection_command +389,2217614,"input_pipeline/download/dqn_replay/download_pngs.py",2256,0,"",python,selection_command +390,2217776,"input_pipeline/download/dqn_replay/download_pngs.py",2260,0,"",python,selection_command +391,2218017,"input_pipeline/download/dqn_replay/download_pngs.py",2261,0,"",python,selection_command +392,2218201,"input_pipeline/download/dqn_replay/download_pngs.py",2266,0,"",python,selection_command +393,2218360,"input_pipeline/download/dqn_replay/download_pngs.py",2267,0,"",python,selection_command +394,2218767,"input_pipeline/download/dqn_replay/download_pngs.py",2266,0,"",python,selection_command +395,2219017,"input_pipeline/download/dqn_replay/download_pngs.py",2261,0,"",python,selection_command +396,2219037,"input_pipeline/download/dqn_replay/download_pngs.py",2260,0,"",python,selection_command +397,2219070,"input_pipeline/download/dqn_replay/download_pngs.py",2256,0,"",python,selection_command +398,2219102,"input_pipeline/download/dqn_replay/download_pngs.py",2254,0,"",python,selection_command +399,2219133,"input_pipeline/download/dqn_replay/download_pngs.py",2225,0,"",python,selection_command +400,2219173,"input_pipeline/download/dqn_replay/download_pngs.py",2224,0,"",python,selection_command +401,2219200,"input_pipeline/download/dqn_replay/download_pngs.py",2223,0,"",python,selection_command +402,2219340,"input_pipeline/download/dqn_replay/download_pngs.py",2221,0,"",python,selection_command +403,2219551,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +404,2219716,"input_pipeline/download/dqn_replay/download_pngs.py",2202,0,"",python,selection_command +405,2220414,"input_pipeline/download/dqn_replay/download_pngs.py",2213,0,"",python,selection_command +406,2255514,"input_pipeline/download/dqn_replay/download_pngs.py",2280,0,"",python,selection_command +407,2255731,"input_pipeline/download/dqn_replay/download_pngs.py",2213,0,"",python,selection_command +408,2259745,"input_pipeline/download/dqn_replay/download_pngs.py",2202,0,"",python,selection_command +409,2259984,"input_pipeline/download/dqn_replay/download_pngs.py",2181,0,"",python,selection_command +410,2260022,"input_pipeline/download/dqn_replay/download_pngs.py",2158,0,"",python,selection_command +411,2260048,"input_pipeline/download/dqn_replay/download_pngs.py",2135,0,"",python,selection_command +412,2260081,"input_pipeline/download/dqn_replay/download_pngs.py",2110,0,"",python,selection_command +413,2260113,"input_pipeline/download/dqn_replay/download_pngs.py",2089,0,"",python,selection_command +414,2260223,"input_pipeline/download/dqn_replay/download_pngs.py",2068,0,"",python,selection_command +415,2260408,"input_pipeline/download/dqn_replay/download_pngs.py",2053,0,"",python,selection_command +416,2260555,"input_pipeline/download/dqn_replay/download_pngs.py",2034,0,"",python,selection_command +417,2262481,"input_pipeline/download/dqn_replay/download_pngs.py",2053,0,"",python,selection_command +418,2262656,"input_pipeline/download/dqn_replay/download_pngs.py",2068,0,"",python,selection_command +419,2263750,"input_pipeline/download/dqn_replay/download_pngs.py",2089,0,"",python,selection_command +420,2325231,"input_pipeline/download/dqn_replay/download_pngs.py",2068,0,"",python,selection_command +421,2325329,"input_pipeline/download/dqn_replay/download_pngs.py",2053,0,"",python,selection_command +422,2325480,"input_pipeline/download/dqn_replay/download_pngs.py",2034,0,"",python,selection_command +423,2452414,"input_pipeline/download/dqn_replay/download_pngs.py",2053,0,"",python,selection_command +424,2452580,"input_pipeline/download/dqn_replay/download_pngs.py",2068,0,"",python,selection_command +425,2519096,"input_pipeline/download/dqn_replay/download_pngs.py",2089,0,"",python,selection_command +426,2519380,"input_pipeline/download/dqn_replay/download_pngs.py",2068,0,"",python,selection_command +427,2519886,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +428,2520635,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +429,2520720,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +430,2520932,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +431,2521091,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +432,2521296,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +433,2521487,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +434,2521681,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +435,2526714,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +436,2527370,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +437,2527520,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +438,2527903,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +439,2528469,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +440,2528722,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +441,2529147,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +442,2543618,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +443,2543738,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +444,2551070,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +445,2555484,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +446,2556293,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +447,2578788,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +448,2596529,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +449,2596711,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +450,2605330,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +451,2618187,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +452,2618396,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +453,2618771,"input_pipeline/download/dqn_replay/download_pngs.py",2197,0,"",python,selection_command +454,2619207,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +455,2619367,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +456,2619530,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +457,2689879,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +458,2690024,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +459,2690165,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +460,2690333,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +461,2690693,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +462,2691446,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +463,2691695,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +464,2691728,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +465,2691760,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +466,2691792,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +467,2691827,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +468,2691858,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +469,2691895,"input_pipeline/download/dqn_replay/download_pngs.py",2197,0,"",python,selection_command +470,2691927,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +471,2691963,"input_pipeline/download/dqn_replay/download_pngs.py",2280,0,"",python,selection_command +472,2692231,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +473,2695280,"input_pipeline/download/dqn_replay/download_pngs.py",2197,0,"",python,selection_command +474,2695541,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +475,2695560,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +476,2695596,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +477,2695628,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +478,2695662,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +479,2695788,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +480,2695960,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +481,2696112,"input_pipeline/download/dqn_replay/download_pngs.py",2029,0,"",python,selection_command +482,2699164,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +483,2699417,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +484,2699436,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +485,2699470,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +486,2699506,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +487,2699550,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +488,2699577,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +489,2699616,"input_pipeline/download/dqn_replay/download_pngs.py",2197,0,"",python,selection_command +490,2699659,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +491,2699672,"input_pipeline/download/dqn_replay/download_pngs.py",2280,0,"",python,selection_command +492,2699816,"input_pipeline/download/dqn_replay/download_pngs.py",2285,0,"",python,selection_command +493,2700067,"input_pipeline/download/dqn_replay/download_pngs.py",2280,0,"",python,selection_command +494,2702226,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +495,2702468,"input_pipeline/download/dqn_replay/download_pngs.py",2197,0,"",python,selection_command +496,2702491,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +497,2702530,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +498,2702555,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +499,2702592,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +500,2702621,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +501,2702762,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +502,2702933,"input_pipeline/download/dqn_replay/download_pngs.py",2048,0,"",python,selection_command +503,2703440,"input_pipeline/download/dqn_replay/download_pngs.py",2063,0,"",python,selection_command +504,2703665,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_command +505,2703915,"input_pipeline/download/dqn_replay/download_pngs.py",2105,0,"",python,selection_command +506,2703939,"input_pipeline/download/dqn_replay/download_pngs.py",2130,0,"",python,selection_command +507,2703970,"input_pipeline/download/dqn_replay/download_pngs.py",2153,0,"",python,selection_command +508,2704001,"input_pipeline/download/dqn_replay/download_pngs.py",2176,0,"",python,selection_command +509,2704032,"input_pipeline/download/dqn_replay/download_pngs.py",2197,0,"",python,selection_command +510,2704070,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +511,2704096,"input_pipeline/download/dqn_replay/download_pngs.py",2280,0,"",python,selection_command +512,2704205,"input_pipeline/download/dqn_replay/download_pngs.py",2285,0,"",python,selection_command +513,2704396,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +514,2704580,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +515,2704830,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +516,2756287,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +517,2756455,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +518,2784710,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +519,2784837,"input_pipeline/download/dqn_replay/download_pngs.py",2478,0,"",python,selection_command +520,2785036,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +521,2785199,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +522,2785487,"input_pipeline/download/dqn_replay/download_pngs.py",2285,0,"",python,selection_command +523,2785770,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +524,2824641,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +525,2824767,"input_pipeline/download/dqn_replay/download_pngs.py",2478,0,"",python,selection_command +526,2824916,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +527,2825070,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +528,2825230,"input_pipeline/download/dqn_replay/download_pngs.py",2285,0,"",python,selection_command +529,2825414,"input_pipeline/download/dqn_replay/download_pngs.py",2280,0,"",python,selection_command +530,2825628,"input_pipeline/download/dqn_replay/download_pngs.py",2285,0,"",python,selection_command +531,2825783,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +532,2847805,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +533,2847945,"input_pipeline/download/dqn_replay/download_pngs.py",2478,0,"",python,selection_command +534,2850074,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +535,2850221,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +536,2850362,"input_pipeline/download/dqn_replay/download_pngs.py",2285,0,"",python,selection_command +537,2850500,"input_pipeline/download/dqn_replay/download_pngs.py",2280,0,"",python,selection_command +538,2850777,"input_pipeline/download/dqn_replay/download_pngs.py",2285,0,"",python,selection_command +539,2850942,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +540,2853277,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +541,2853404,"input_pipeline/download/dqn_replay/download_pngs.py",2478,0,"",python,selection_command +542,2853591,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +543,2853757,"input_pipeline/download/dqn_replay/download_pngs.py",2374,0,"",python,selection_command +544,2881371,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +545,2881698,"input_pipeline/download/dqn_replay/download_pngs.py",835,0,"",python,selection_command +546,2883324,"input_pipeline/download/dqn_replay/download_pngs.py",2449,0,"",python,selection_command +547,2884453,"input_pipeline/download/dqn_replay/download_pngs.py",835,0,"",python,selection_command +548,2888463,"input_pipeline/download/dqn_replay/download_pngs.py",871,0,"",python,selection_command +549,2888607,"input_pipeline/download/dqn_replay/download_pngs.py",835,0,"",python,selection_command +550,2889485,"input_pipeline/download/dqn_replay/download_pngs.py",831,36,"",python,content +551,2889493,"input_pipeline/download/dqn_replay/download_pngs.py",835,0,"",python,selection_command +552,2889840,"input_pipeline/download/dqn_replay/download_pngs.py",831,37,"",python,content +553,2891023,"input_pipeline/download/dqn_replay/download_pngs.py",2376,0,"",python,selection_command +554,2891896,"input_pipeline/download/dqn_replay/download_pngs.py",2404,0,"\n os.makedirs(path, exist_ok=True)",python,content +555,2891902,"input_pipeline/download/dqn_replay/download_pngs.py",2409,0,"",python,selection_command +556,2892549,"input_pipeline/download/dqn_replay/download_pngs.py",2376,0,"",python,selection_command +557,2892759,"input_pipeline/download/dqn_replay/download_pngs.py",2372,33,"",python,content +558,2892769,"input_pipeline/download/dqn_replay/download_pngs.py",2376,0,"",python,selection_command +559,2894870,"input_pipeline/download/dqn_replay/download_pngs.py",2392,0,"_dir",python,content +560,2894870,"input_pipeline/download/dqn_replay/download_pngs.py",2391,1,"",python,content +561,2894870,"input_pipeline/download/dqn_replay/download_pngs.py",2390,0,"u",python,content +562,2894870,"input_pipeline/download/dqn_replay/download_pngs.py",2389,1,"",python,content +563,2894870,"input_pipeline/download/dqn_replay/download_pngs.py",2388,0,"base_out",python,content +564,2896849,"input_pipeline/download/dqn_replay/download_pngs.py",2420,0,"",python,selection_command +565,2900328,"input_pipeline/download/dqn_replay/download_pngs.py",3638,0,"",python,selection_keyboard +566,2901276,"input_pipeline/download/dqn_replay/download_pngs.py",3153,0,"",python,selection_keyboard +567,2901913,"input_pipeline/download/dqn_replay/download_pngs.py",3120,0,"",python,selection_command +568,2902091,"input_pipeline/download/dqn_replay/download_pngs.py",3031,0,"",python,selection_command +569,2902451,"input_pipeline/download/dqn_replay/download_pngs.py",3120,0,"",python,selection_command +570,2903168,"input_pipeline/download/dqn_replay/download_pngs.py",3151,0,", exist_ok=True",python,content +571,2903169,"input_pipeline/download/dqn_replay/download_pngs.py",3139,0,"s",python,content +572,2903169,"input_pipeline/download/dqn_replay/download_pngs.py",3135,1,"",python,content +573,2903169,"input_pipeline/download/dqn_replay/download_pngs.py",3134,0,".mak",python,content +574,2903169,"input_pipeline/download/dqn_replay/download_pngs.py",3132,2,"",python,content +575,2903169,"input_pipeline/download/dqn_replay/download_pngs.py",3131,0,"o",python,content +576,2903169,"input_pipeline/download/dqn_replay/download_pngs.py",3128,3,"",python,content +577,2906535,"input_pipeline/download/dqn_replay/download_pngs.py",3031,0,"",python,selection_command +578,2906785,"input_pipeline/download/dqn_replay/download_pngs.py",3030,0,"",python,selection_command +579,2906803,"input_pipeline/download/dqn_replay/download_pngs.py",3012,0,"",python,selection_command +580,2906841,"input_pipeline/download/dqn_replay/download_pngs.py",2944,0,"",python,selection_command +581,2906987,"input_pipeline/download/dqn_replay/download_pngs.py",2915,0,"",python,selection_command +582,2907246,"input_pipeline/download/dqn_replay/download_pngs.py",2919,0,"",python,selection_command +583,2907421,"input_pipeline/download/dqn_replay/download_pngs.py",2923,0,"",python,selection_command +584,2907581,"input_pipeline/download/dqn_replay/download_pngs.py",2931,0,"",python,selection_command +585,2907782,"input_pipeline/download/dqn_replay/download_pngs.py",2934,0,"",python,selection_command +586,2913152,"input_pipeline/download/dqn_replay/download_pngs.py",4227,0,"",python,selection_keyboard +587,2914737,"input_pipeline/download/dqn_replay/download_pngs.py",3151,15,"",python,content +588,2914756,"input_pipeline/download/dqn_replay/download_pngs.py",3138,1,"",python,content +589,2914757,"input_pipeline/download/dqn_replay/download_pngs.py",3128,7,"_ensure_",python,content +590,2914813,"input_pipeline/download/dqn_replay/download_pngs.py",3120,0,"",python,selection_command +591,2914861,"input_pipeline/download/dqn_replay/download_pngs.py",2388,15,"path",python,content +592,2914862,"input_pipeline/download/dqn_replay/download_pngs.py",2376,0,"",python,selection_command +593,2915374,"input_pipeline/download/dqn_replay/download_pngs.py",2376,0,"_ensure_dir(base_output_dir)\n ",python,content +594,2915381,"input_pipeline/download/dqn_replay/download_pngs.py",2376,0,"",python,selection_command +595,2915714,"input_pipeline/download/dqn_replay/download_pngs.py",2405,37,"",python,content +596,2916235,"input_pipeline/download/dqn_replay/download_pngs.py",831,0," os.makedirs(path, exist_ok=True)\n",python,content +597,2916250,"input_pipeline/download/dqn_replay/download_pngs.py",835,0,"",python,selection_command +598,2916938,"input_pipeline/download/dqn_replay/download_pngs.py",831,0,"def _ensure_dir(path: str) -> None:\n",python,content +599,2916949,"input_pipeline/download/dqn_replay/download_pngs.py",835,0,"",python,selection_command +600,2920057,"input_pipeline/download/dqn_replay/download_pngs.py",2084,0,"",python,selection_keyboard +601,2920350,"input_pipeline/download/dqn_replay/download_pngs.py",3326,0,"",python,selection_keyboard +602,2921957,"input_pipeline/download/dqn_replay/download_pngs.py",831,36,"",python,content +603,2921977,"input_pipeline/download/dqn_replay/download_pngs.py",835,0,"",python,selection_command +604,2922191,"input_pipeline/download/dqn_replay/download_pngs.py",831,37,"",python,content +605,2922212,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0," os.makedirs(path, exist_ok=True)\n",python,content +606,2922214,"input_pipeline/download/dqn_replay/download_pngs.py",2376,0,"",python,selection_command +607,2922250,"input_pipeline/download/dqn_replay/download_pngs.py",2376,33,"",python,content +608,2922280,"input_pipeline/download/dqn_replay/download_pngs.py",2388,4,"base_output_dir",python,content +609,2922313,"input_pipeline/download/dqn_replay/download_pngs.py",3128,8,"os.make",python,content +610,2922314,"input_pipeline/download/dqn_replay/download_pngs.py",3138,0,"s",python,content +611,2922317,"input_pipeline/download/dqn_replay/download_pngs.py",3151,0,", exist_ok=True",python,content +612,2922318,"input_pipeline/download/dqn_replay/download_pngs.py",3120,0,"",python,selection_command +613,2924957,"input_pipeline/download/dqn_replay/download_pngs.py",4365,0,"",python,selection_keyboard +614,2925094,"input_pipeline/download/dqn_replay/download_pngs.py",5027,0,"",python,selection_keyboard +615,2925765,"input_pipeline/download/dqn_replay/download_pngs.py",3935,0,"",python,selection_keyboard +616,2925986,"input_pipeline/download/dqn_replay/download_pngs.py",2688,0,"",python,selection_keyboard +617,2928533,"input_pipeline/download/dqn_replay/download_pngs.py",2713,0,"",python,selection_command +618,2928777,"input_pipeline/download/dqn_replay/download_pngs.py",2741,0,"",python,selection_command +619,2928799,"input_pipeline/download/dqn_replay/download_pngs.py",2807,0,"",python,selection_command +620,2928840,"input_pipeline/download/dqn_replay/download_pngs.py",2813,0,"",python,selection_command +621,2928868,"input_pipeline/download/dqn_replay/download_pngs.py",2814,0,"",python,selection_command +622,2928901,"input_pipeline/download/dqn_replay/download_pngs.py",2841,0,"",python,selection_command +623,2928930,"input_pipeline/download/dqn_replay/download_pngs.py",2862,0,"",python,selection_command +624,2928967,"input_pipeline/download/dqn_replay/download_pngs.py",2863,0,"",python,selection_command +625,2929215,"input_pipeline/download/dqn_replay/download_pngs.py",2914,0,"",python,selection_command +626,2929346,"input_pipeline/download/dqn_replay/download_pngs.py",2915,0,"",python,selection_command +627,2935615,"input_pipeline/download/dqn_replay/download_pngs.py",1724,0,"",python,selection_keyboard +628,2937299,"input_pipeline/download/dqn_replay/download_pngs.py",1792,0,"",python,selection_command +629,2937558,"input_pipeline/download/dqn_replay/download_pngs.py",1867,0,"",python,selection_command +630,2937586,"input_pipeline/download/dqn_replay/download_pngs.py",1917,0,"",python,selection_command +631,2937620,"input_pipeline/download/dqn_replay/download_pngs.py",1950,0,"",python,selection_command +632,2937650,"input_pipeline/download/dqn_replay/download_pngs.py",1951,0,"",python,selection_command +633,2937677,"input_pipeline/download/dqn_replay/download_pngs.py",1952,0,"",python,selection_command +634,2937714,"input_pipeline/download/dqn_replay/download_pngs.py",1971,0,"",python,selection_command +635,2937744,"input_pipeline/download/dqn_replay/download_pngs.py",1986,0,"",python,selection_command +636,2937784,"input_pipeline/download/dqn_replay/download_pngs.py",2007,0,"",python,selection_command +637,2937814,"input_pipeline/download/dqn_replay/download_pngs.py",2028,0,"",python,selection_command +638,2937850,"input_pipeline/download/dqn_replay/download_pngs.py",2053,0,"",python,selection_command +639,2937887,"input_pipeline/download/dqn_replay/download_pngs.py",2076,0,"",python,selection_command +640,2938144,"input_pipeline/download/dqn_replay/download_pngs.py",2099,0,"",python,selection_command +641,2938389,"input_pipeline/download/dqn_replay/download_pngs.py",2120,0,"",python,selection_command +642,2938422,"input_pipeline/download/dqn_replay/download_pngs.py",2131,0,"",python,selection_command +643,2938456,"input_pipeline/download/dqn_replay/download_pngs.py",2207,0,"",python,selection_command +644,2938560,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +645,2938745,"input_pipeline/download/dqn_replay/download_pngs.py",2297,0,"",python,selection_command +646,2938921,"input_pipeline/download/dqn_replay/download_pngs.py",2372,0,"",python,selection_command +647,2939406,"input_pipeline/download/dqn_replay/download_pngs.py",2420,0,"",python,selection_command +648,2939651,"input_pipeline/download/dqn_replay/download_pngs.py",2421,0,"",python,selection_command +649,2939676,"input_pipeline/download/dqn_replay/download_pngs.py",2490,0,"",python,selection_command +650,2939776,"input_pipeline/download/dqn_replay/download_pngs.py",2531,0,"",python,selection_command +651,2939976,"input_pipeline/download/dqn_replay/download_pngs.py",2590,0,"",python,selection_command +652,2940182,"input_pipeline/download/dqn_replay/download_pngs.py",2531,0,"",python,selection_command +653,2940354,"input_pipeline/download/dqn_replay/download_pngs.py",2490,0,"",python,selection_command +654,2942257,"input_pipeline/download/dqn_replay/download_pngs.py",2421,0,"",python,selection_command +655,2946725,"input_pipeline/download/dqn_replay/download_pngs.py",2490,0,"",python,selection_command +656,2946947,"input_pipeline/download/dqn_replay/download_pngs.py",2421,0,"",python,selection_command +657,2947157,"input_pipeline/download/dqn_replay/download_pngs.py",2490,0,"",python,selection_command +658,2948363,"input_pipeline/download/dqn_replay/download_pngs.py",2421,0,"",python,selection_command +659,2949104,"input_pipeline/download/dqn_replay/download_pngs.py",2420,0,"",python,selection_command +660,2949351,"input_pipeline/download/dqn_replay/download_pngs.py",2372,0,"",python,selection_command +661,2949377,"input_pipeline/download/dqn_replay/download_pngs.py",2297,0,"",python,selection_command +662,2949417,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +663,2949505,"input_pipeline/download/dqn_replay/download_pngs.py",2207,0,"",python,selection_command +664,2949785,"input_pipeline/download/dqn_replay/download_pngs.py",2208,0,"",python,selection_command +665,2950322,"input_pipeline/download/dqn_replay/download_pngs.py",2208,89,"",python,content +666,2950385,"input_pipeline/download/dqn_replay/download_pngs.py",2212,0,"",python,selection_command +667,2951568,"input_pipeline/download/dqn_replay/download_pngs.py",2287,0,"",python,selection_command +668,2951723,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +669,2951865,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +670,2952314,"input_pipeline/download/dqn_replay/download_pngs.py",2332,69,"",python,content +671,2952331,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +672,2954470,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"# Discover available splits and construct split selection string\n ",python,content +673,2954476,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +674,2956083,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +675,2956169,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +676,2974271,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +677,2974392,"input_pipeline/download/dqn_replay/download_pngs.py",2506,0,"",python,selection_command +678,2974594,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +679,2974760,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +680,2975094,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +681,2975238,"input_pipeline/download/dqn_replay/download_pngs.py",2506,0,"",python,selection_command +682,2975393,"input_pipeline/download/dqn_replay/download_pngs.py",2555,0,"",python,selection_command +683,2975528,"input_pipeline/download/dqn_replay/download_pngs.py",2581,0,"",python,selection_command +684,2975781,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +685,2975802,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +686,2975838,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +687,2975870,"input_pipeline/download/dqn_replay/download_pngs.py",2722,0,"",python,selection_command +688,2976005,"input_pipeline/download/dqn_replay/download_pngs.py",2724,0,"",python,selection_command +689,2976207,"input_pipeline/download/dqn_replay/download_pngs.py",2729,0,"",python,selection_command +690,2976422,"input_pipeline/download/dqn_replay/download_pngs.py",2724,0,"",python,selection_command +691,2976580,"input_pipeline/download/dqn_replay/download_pngs.py",2722,0,"",python,selection_command +692,2976745,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +693,2976914,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +694,2977107,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +695,2977280,"input_pipeline/download/dqn_replay/download_pngs.py",2581,0,"",python,selection_command +696,2978067,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +697,2978220,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +698,2978373,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +699,2978546,"input_pipeline/download/dqn_replay/download_pngs.py",2722,0,"",python,selection_command +700,2979061,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +701,2979211,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +702,2979364,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +703,2979512,"input_pipeline/download/dqn_replay/download_pngs.py",2581,0,"",python,selection_command +704,2979662,"input_pipeline/download/dqn_replay/download_pngs.py",2555,0,"",python,selection_command +705,2981554,"input_pipeline/download/dqn_replay/download_pngs.py",2506,0,"",python,selection_command +706,2981680,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +707,2982068,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +708,2982305,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +709,2982336,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +710,2982458,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +711,2982628,"input_pipeline/download/dqn_replay/download_pngs.py",2287,0,"",python,selection_command +712,2982887,"input_pipeline/download/dqn_replay/download_pngs.py",2212,0,"",python,selection_command +713,2982914,"input_pipeline/download/dqn_replay/download_pngs.py",2207,0,"",python,selection_command +714,2983075,"input_pipeline/download/dqn_replay/download_pngs.py",2135,0,"",python,selection_command +715,2983338,"input_pipeline/download/dqn_replay/download_pngs.py",2124,0,"",python,selection_command +716,2983361,"input_pipeline/download/dqn_replay/download_pngs.py",2103,0,"",python,selection_command +717,2983394,"input_pipeline/download/dqn_replay/download_pngs.py",2080,0,"",python,selection_command +718,2983424,"input_pipeline/download/dqn_replay/download_pngs.py",2057,0,"",python,selection_command +719,2983453,"input_pipeline/download/dqn_replay/download_pngs.py",2032,0,"",python,selection_command +720,2983485,"input_pipeline/download/dqn_replay/download_pngs.py",2011,0,"",python,selection_command +721,2983517,"input_pipeline/download/dqn_replay/download_pngs.py",1990,0,"",python,selection_command +722,2983551,"input_pipeline/download/dqn_replay/download_pngs.py",1975,0,"",python,selection_command +723,2983584,"input_pipeline/download/dqn_replay/download_pngs.py",1956,0,"",python,selection_command +724,2983621,"input_pipeline/download/dqn_replay/download_pngs.py",1951,0,"",python,selection_command +725,2983862,"input_pipeline/download/dqn_replay/download_pngs.py",1956,0,"",python,selection_command +726,2984113,"input_pipeline/download/dqn_replay/download_pngs.py",1951,0,"",python,selection_command +727,2984273,"input_pipeline/download/dqn_replay/download_pngs.py",1950,0,"",python,selection_command +728,2984703,"input_pipeline/download/dqn_replay/download_pngs.py",1951,0,"",python,selection_command +729,2984870,"input_pipeline/download/dqn_replay/download_pngs.py",1956,0,"",python,selection_command +730,2985196,"input_pipeline/download/dqn_replay/download_pngs.py",1951,0,"",python,selection_command +731,2985501,"input_pipeline/download/dqn_replay/download_pngs.py",1956,0,"",python,selection_command +732,2985750,"input_pipeline/download/dqn_replay/download_pngs.py",1975,0,"",python,selection_command +733,2985776,"input_pipeline/download/dqn_replay/download_pngs.py",1990,0,"",python,selection_command +734,2985810,"input_pipeline/download/dqn_replay/download_pngs.py",2011,0,"",python,selection_command +735,2985837,"input_pipeline/download/dqn_replay/download_pngs.py",2032,0,"",python,selection_command +736,2985873,"input_pipeline/download/dqn_replay/download_pngs.py",2057,0,"",python,selection_command +737,2985902,"input_pipeline/download/dqn_replay/download_pngs.py",2080,0,"",python,selection_command +738,2985938,"input_pipeline/download/dqn_replay/download_pngs.py",2103,0,"",python,selection_command +739,2985972,"input_pipeline/download/dqn_replay/download_pngs.py",2124,0,"",python,selection_command +740,2986003,"input_pipeline/download/dqn_replay/download_pngs.py",2135,0,"",python,selection_command +741,2986038,"input_pipeline/download/dqn_replay/download_pngs.py",2207,0,"",python,selection_command +742,2986068,"input_pipeline/download/dqn_replay/download_pngs.py",2212,0,"",python,selection_command +743,2986102,"input_pipeline/download/dqn_replay/download_pngs.py",2287,0,"",python,selection_command +744,2986139,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +745,2986169,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +746,2986202,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +747,2986239,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +748,2986269,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +749,2986302,"input_pipeline/download/dqn_replay/download_pngs.py",2506,0,"",python,selection_command +750,2986335,"input_pipeline/download/dqn_replay/download_pngs.py",2555,0,"",python,selection_command +751,2986374,"input_pipeline/download/dqn_replay/download_pngs.py",2581,0,"",python,selection_command +752,2986403,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +753,2986438,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +754,2986471,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +755,2986504,"input_pipeline/download/dqn_replay/download_pngs.py",2722,0,"",python,selection_command +756,2986539,"input_pipeline/download/dqn_replay/download_pngs.py",2724,0,"",python,selection_command +757,2986790,"input_pipeline/download/dqn_replay/download_pngs.py",2722,0,"",python,selection_command +758,2987041,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +759,2987059,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +760,2987095,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +761,2987133,"input_pipeline/download/dqn_replay/download_pngs.py",2581,0,"",python,selection_command +762,2987162,"input_pipeline/download/dqn_replay/download_pngs.py",2555,0,"",python,selection_command +763,2987388,"input_pipeline/download/dqn_replay/download_pngs.py",2506,0,"",python,selection_command +764,2987559,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +765,2987857,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +766,2988102,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +767,2988144,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +768,2988236,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +769,2988414,"input_pipeline/download/dqn_replay/download_pngs.py",2287,0,"",python,selection_command +770,2988586,"input_pipeline/download/dqn_replay/download_pngs.py",2212,0,"",python,selection_command +771,2988775,"input_pipeline/download/dqn_replay/download_pngs.py",2207,0,"",python,selection_command +772,2989065,"input_pipeline/download/dqn_replay/download_pngs.py",2212,0,"",python,selection_command +773,2989250,"input_pipeline/download/dqn_replay/download_pngs.py",2287,0,"",python,selection_command +774,2989412,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +775,2989579,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +776,2989889,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +777,2990138,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +778,2990163,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +779,2990217,"input_pipeline/download/dqn_replay/download_pngs.py",2506,0,"",python,selection_command +780,2990233,"input_pipeline/download/dqn_replay/download_pngs.py",2555,0,"",python,selection_command +781,2990258,"input_pipeline/download/dqn_replay/download_pngs.py",2581,0,"",python,selection_command +782,2990288,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +783,2990325,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +784,2990359,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +785,2990463,"input_pipeline/download/dqn_replay/download_pngs.py",2722,0,"",python,selection_command +786,2990717,"input_pipeline/download/dqn_replay/download_pngs.py",2724,0,"",python,selection_command +787,2990752,"input_pipeline/download/dqn_replay/download_pngs.py",2729,0,"",python,selection_command +788,2990782,"input_pipeline/download/dqn_replay/download_pngs.py",2756,0,"",python,selection_command +789,2990812,"input_pipeline/download/dqn_replay/download_pngs.py",2773,0,"",python,selection_command +790,2990844,"input_pipeline/download/dqn_replay/download_pngs.py",2778,0,"",python,selection_command +791,2990875,"input_pipeline/download/dqn_replay/download_pngs.py",2825,0,"",python,selection_command +792,2990910,"input_pipeline/download/dqn_replay/download_pngs.py",2830,0,"",python,selection_command +793,2990958,"input_pipeline/download/dqn_replay/download_pngs.py",2859,0,"",python,selection_command +794,2990981,"input_pipeline/download/dqn_replay/download_pngs.py",2927,0,"",python,selection_command +795,2991013,"input_pipeline/download/dqn_replay/download_pngs.py",2941,0,"",python,selection_command +796,2991043,"input_pipeline/download/dqn_replay/download_pngs.py",2946,0,"",python,selection_command +797,2991138,"input_pipeline/download/dqn_replay/download_pngs.py",2941,0,"",python,selection_command +798,2991408,"input_pipeline/download/dqn_replay/download_pngs.py",2927,0,"",python,selection_command +799,2991430,"input_pipeline/download/dqn_replay/download_pngs.py",2859,0,"",python,selection_command +800,2991462,"input_pipeline/download/dqn_replay/download_pngs.py",2830,0,"",python,selection_command +801,2991487,"input_pipeline/download/dqn_replay/download_pngs.py",2825,0,"",python,selection_command +802,2991522,"input_pipeline/download/dqn_replay/download_pngs.py",2778,0,"",python,selection_command +803,2991553,"input_pipeline/download/dqn_replay/download_pngs.py",2773,0,"",python,selection_command +804,2991585,"input_pipeline/download/dqn_replay/download_pngs.py",2756,0,"",python,selection_command +805,2991905,"input_pipeline/download/dqn_replay/download_pngs.py",2729,0,"",python,selection_command +806,2992143,"input_pipeline/download/dqn_replay/download_pngs.py",2724,0,"",python,selection_command +807,2992330,"input_pipeline/download/dqn_replay/download_pngs.py",2722,0,"",python,selection_command +808,2992575,"input_pipeline/download/dqn_replay/download_pngs.py",2656,0,"",python,selection_command +809,2992613,"input_pipeline/download/dqn_replay/download_pngs.py",2628,0,"",python,selection_command +810,2992648,"input_pipeline/download/dqn_replay/download_pngs.py",2603,0,"",python,selection_command +811,2992678,"input_pipeline/download/dqn_replay/download_pngs.py",2581,0,"",python,selection_command +812,2992836,"input_pipeline/download/dqn_replay/download_pngs.py",2555,0,"",python,selection_command +813,2993011,"input_pipeline/download/dqn_replay/download_pngs.py",2506,0,"",python,selection_command +814,2993184,"input_pipeline/download/dqn_replay/download_pngs.py",2501,0,"",python,selection_command +815,2993441,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +816,2993468,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +817,2993501,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +818,3032745,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +819,3032839,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +820,3050017,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +821,3050204,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +822,3050363,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +823,3050556,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +824,3052468,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +825,3052719,"input_pipeline/download/dqn_replay/download_pngs.py",2287,0,"",python,selection_command +826,3052741,"input_pipeline/download/dqn_replay/download_pngs.py",2212,0,"",python,selection_command +827,3052777,"input_pipeline/download/dqn_replay/download_pngs.py",2207,0,"",python,selection_command +828,3052812,"input_pipeline/download/dqn_replay/download_pngs.py",2135,0,"",python,selection_command +829,3052854,"input_pipeline/download/dqn_replay/download_pngs.py",2124,0,"",python,selection_command +830,3052875,"input_pipeline/download/dqn_replay/download_pngs.py",2103,0,"",python,selection_command +831,3052910,"input_pipeline/download/dqn_replay/download_pngs.py",2080,0,"",python,selection_command +832,3052948,"input_pipeline/download/dqn_replay/download_pngs.py",2057,0,"",python,selection_command +833,3052975,"input_pipeline/download/dqn_replay/download_pngs.py",2032,0,"",python,selection_command +834,3053009,"input_pipeline/download/dqn_replay/download_pngs.py",2011,0,"",python,selection_command +835,3053042,"input_pipeline/download/dqn_replay/download_pngs.py",1990,0,"",python,selection_command +836,3053075,"input_pipeline/download/dqn_replay/download_pngs.py",1975,0,"",python,selection_command +837,3053108,"input_pipeline/download/dqn_replay/download_pngs.py",1956,0,"",python,selection_command +838,3053226,"input_pipeline/download/dqn_replay/download_pngs.py",1975,0,"",python,selection_command +839,3053488,"input_pipeline/download/dqn_replay/download_pngs.py",1990,0,"",python,selection_command +840,3053511,"input_pipeline/download/dqn_replay/download_pngs.py",2011,0,"",python,selection_command +841,3053550,"input_pipeline/download/dqn_replay/download_pngs.py",2032,0,"",python,selection_command +842,3053583,"input_pipeline/download/dqn_replay/download_pngs.py",2057,0,"",python,selection_command +843,3053611,"input_pipeline/download/dqn_replay/download_pngs.py",2080,0,"",python,selection_command +844,3053640,"input_pipeline/download/dqn_replay/download_pngs.py",2103,0,"",python,selection_command +845,3053674,"input_pipeline/download/dqn_replay/download_pngs.py",2124,0,"",python,selection_command +846,3053707,"input_pipeline/download/dqn_replay/download_pngs.py",2135,0,"",python,selection_command +847,3053741,"input_pipeline/download/dqn_replay/download_pngs.py",2207,0,"",python,selection_command +848,3053847,"input_pipeline/download/dqn_replay/download_pngs.py",2212,0,"",python,selection_command +849,3054011,"input_pipeline/download/dqn_replay/download_pngs.py",2287,0,"",python,selection_command +850,3054157,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +851,3054300,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +852,3054640,"input_pipeline/download/dqn_replay/download_pngs.py",2331,0,"",python,selection_command +853,3054917,"input_pipeline/download/dqn_replay/download_pngs.py",2336,0,"",python,selection_command +854,3056908,"input_pipeline/download/dqn_replay/download_pngs.py",2405,0,"",python,selection_command +855,3056983,"input_pipeline/download/dqn_replay/download_pngs.py",2446,0,"",python,selection_command +856,3057193,"input_pipeline/download/dqn_replay/download_pngs.py",2456,0,"",python,selection_command +857,3057354,"input_pipeline/download/dqn_replay/download_pngs.py",2458,0,"",python,selection_command +858,3057650,"input_pipeline/download/dqn_replay/download_pngs.py",1263,0,"",python,selection_command +859,3063124,"input_pipeline/download/dqn_replay/download_pngs.py",2458,0,"",python,selection_command +860,3072136,"input_pipeline/download/dqn_replay/download_pngs.py",1263,0,"",python,selection_command +861,3077243,"input_pipeline/download/dqn_replay/download_pngs.py",2458,0,"",python,selection_command diff --git a/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-9543d8e2-2376-4957-873e-df7016d502961763465687199-2025_11_18-12.34.49.442/source.csv b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-9543d8e2-2376-4957-873e-df7016d502961763465687199-2025_11_18-12.34.49.442/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..8d2112025eb989c443ec8254f116ef717792dd6e --- /dev/null +++ b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-9543d8e2-2376-4957-873e-df7016d502961763465687199-2025_11_18-12.34.49.442/source.csv @@ -0,0 +1,245 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +2,67,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"12:34:49 PM [info] Activating crowd-code\n12:34:49 PM [info] Recording started\n12:34:49 PM [info] Initializing git provider using file system watchers...\n12:34:49 PM [info] Git repository found\n12:34:49 PM [info] Git provider initialized successfully\n",Log,tab +3,116,"extension-output-pdoom-org.crowd-code-#1-crowd-code",250,0,"12:34:49 PM [info] Initial git state: [object Object]\n",Log,content +4,2975,"extension-output-pdoom-org.crowd-code-#1-crowd-code",304,0,"",Log,selection_mouse +5,13030,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,tab +6,20469,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_command +7,20661,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",66,0,"",python,selection_command +8,20694,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",132,0,"",python,selection_command +9,20728,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",199,0,"",python,selection_command +10,20761,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",241,0,"",python,selection_command +11,20794,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",243,0,"",python,selection_command +12,20828,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",292,0,"",python,selection_command +13,20863,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",294,0,"",python,selection_command +14,20904,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",364,0,"",python,selection_command +15,20936,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",432,0,"",python,selection_command +16,20962,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",507,0,"",python,selection_command +17,20996,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",577,0,"",python,selection_command +18,21030,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",610,0,"",python,selection_command +19,21068,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",611,0,"",python,selection_command +20,21097,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",639,0,"",python,selection_command +21,21131,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",640,0,"",python,selection_command +22,21166,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",671,0,"",python,selection_command +23,21203,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",694,0,"",python,selection_command +24,21233,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",707,0,"",python,selection_command +25,21265,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",795,0,"",python,selection_command +26,21297,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",796,0,"",python,selection_command +27,21331,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",852,0,"",python,selection_command +28,21364,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",914,0,"",python,selection_command +29,21397,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",965,0,"",python,selection_command +30,21431,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1047,0,"",python,selection_command +31,21464,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1148,0,"",python,selection_command +32,21497,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1245,0,"",python,selection_command +33,21530,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1319,0,"",python,selection_command +34,21565,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1369,0,"",python,selection_command +35,21598,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1370,0,"",python,selection_command +36,21631,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1393,0,"",python,selection_command +37,21668,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1394,0,"",python,selection_command +38,21701,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1395,0,"",python,selection_command +39,21890,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1423,0,"",python,selection_command +40,23050,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1427,0,"",python,selection_command +41,23197,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1432,0,"",python,selection_command +42,23489,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1435,0,"",python,selection_command +43,23702,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1438,0,"",python,selection_command +44,23912,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1441,0,"",python,selection_command +45,24119,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1442,0,"",python,selection_command +46,24327,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1448,0,"",python,selection_command +47,24581,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1449,0,"",python,selection_command +48,24945,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1451,0,"",python,selection_command +49,25331,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1452,0,"",python,selection_command +50,26280,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1423,0,"",python,selection_command +51,26667,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1395,0,"",python,selection_command +52,26920,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1394,0,"",python,selection_command +53,26954,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1393,0,"",python,selection_command +54,146021,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +55,146654,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +56,203332,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2838,0,"",python,selection_keyboard +57,203923,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",5124,0,"",python,selection_keyboard +58,204398,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7126,0,"",python,selection_keyboard +59,204829,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9057,0,"",python,selection_keyboard +60,205408,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_keyboard +61,210105,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7721,0,"",python,selection_keyboard +62,210235,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",5922,0,"",python,selection_keyboard +63,210382,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3490,0,"",python,selection_keyboard +64,210771,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1906,0,"",python,selection_keyboard +65,210944,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_keyboard +66,211099,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_keyboard +67,214839,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10,0,"",python,selection_command +68,2267039,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2294,0,"",python,selection_mouse +69,2272461,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2114,0,"",python,selection_mouse +70,3052561,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +71,3052668,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +72,3145153,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_command +73,3148126,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7487,0,"",python,selection_command +74,3149770,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7876,0,"",python,selection_command +75,3150226,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",8235,0,"",python,selection_command +76,3150898,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7876,0,"",python,selection_command +77,3151966,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",8235,0,"",python,selection_command +78,3161866,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9070,0,"",python,selection_command +79,3164224,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9131,0,"",python,selection_command +80,3166810,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9194,0,"",python,selection_command +81,3167056,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9257,0,"",python,selection_command +82,3167090,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9321,0,"",python,selection_command +83,3167123,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9386,0,"",python,selection_command +84,3167156,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9443,0,"",python,selection_command +85,3167189,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9481,0,"",python,selection_command +86,3167223,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9531,0,"",python,selection_command +87,3168286,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9532,0,"",python,selection_command +88,3168527,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9538,0,"",python,selection_command +89,3168566,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9539,0,"",python,selection_command +90,3168597,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9546,0,"",python,selection_command +91,3168633,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9547,0,"",python,selection_command +92,3168887,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9555,0,"",python,selection_command +93,3169118,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +94,3169292,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9583,0,"",python,selection_command +95,3169900,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +96,3170394,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +97,3171165,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +98,3171795,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +99,3171945,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +100,3172433,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +101,3173957,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +102,3174398,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +103,3174801,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9682,0,"",python,selection_command +104,3175123,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9746,0,"",python,selection_command +105,3175686,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9682,0,"",python,selection_command +106,3175821,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +107,3175981,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +108,3240534,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_command +109,3241210,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1898,0,"",python,selection_keyboard +110,3241373,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3420,0,"",python,selection_keyboard +111,3242843,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3370,0,"",python,selection_command +112,3243091,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3298,0,"",python,selection_command +113,3243124,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3288,0,"",python,selection_command +114,3243158,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3287,0,"",python,selection_command +115,3243191,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3211,0,"",python,selection_command +116,3243225,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3137,0,"",python,selection_command +117,3243258,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3058,0,"",python,selection_command +118,3243291,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3057,0,"",python,selection_command +119,3243324,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2999,0,"",python,selection_command +120,3243357,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2991,0,"",python,selection_command +121,3243391,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2973,0,"",python,selection_command +122,3243424,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2956,0,"",python,selection_command +123,3243458,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2932,0,"",python,selection_command +124,3243491,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2915,0,"",python,selection_command +125,3243525,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2898,0,"",python,selection_command +126,3243558,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2876,0,"",python,selection_command +127,3243591,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2854,0,"",python,selection_command +128,3243632,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2838,0,"",python,selection_command +129,3243658,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2817,0,"",python,selection_command +130,3243691,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2793,0,"",python,selection_command +131,3243734,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2767,0,"",python,selection_command +132,3243758,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2756,0,"",python,selection_command +133,3243791,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2721,0,"",python,selection_command +134,3243824,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2688,0,"",python,selection_command +135,3243857,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2655,0,"",python,selection_command +136,3243891,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2621,0,"",python,selection_command +137,3243925,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2585,0,"",python,selection_command +138,3243958,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2547,0,"",python,selection_command +139,3243991,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2512,0,"",python,selection_command +140,3244025,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2483,0,"",python,selection_command +141,3244059,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2451,0,"",python,selection_command +142,3244092,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2427,0,"",python,selection_command +143,3244125,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,0,"",python,selection_command +144,3286103,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2411,0,"",python,selection_command +145,3296751,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,23,"",python,content +146,3298859,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,0," sequence_parallelis",python,content +147,3298862,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,0,"",python,selection_command +148,3483131,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +149,3484741,"TERMINAL",0,0,"",,terminal_focus +150,3484752,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +151,3788287,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2427,0,"",python,selection_command +152,3788457,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2428,0,"",python,selection_command +153,3788946,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2429,0,"",python,selection_command +154,3789583,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2427,0,"",python,selection_command +155,4549882,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +156,4550471,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +157,4559897,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +158,4560021,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +159,4560890,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,content +160,4583906,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18686,0,"",python,selection_mouse +161,4583914,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18685,0,"",python,selection_command +162,4584529,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18699,0,"",python,selection_command +163,4585394,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18743,0,"",python,selection_command +164,4585610,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18752,0,"",python,selection_command +165,4591082,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +166,4591147,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_command +167,4595172,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,10041,"",python,content +168,4597038,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +169,5454427,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9372,0,"",python,selection_mouse +170,5454796,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_command +171,5455466,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_command +172,5455793,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1906,0,"",python,selection_keyboard +173,5456269,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3490,0,"",python,selection_keyboard +174,5456884,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3420,0,"",python,selection_command +175,5457139,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3370,0,"",python,selection_command +176,5457172,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3298,0,"",python,selection_command +177,5457202,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3288,0,"",python,selection_command +178,5457232,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3287,0,"",python,selection_command +179,5457265,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3211,0,"",python,selection_command +180,5457298,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3137,0,"",python,selection_command +181,5457331,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3058,0,"",python,selection_command +182,5457367,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3057,0,"",python,selection_command +183,5457402,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2999,0,"",python,selection_command +184,5457431,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2991,0,"",python,selection_command +185,5457465,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2973,0,"",python,selection_command +186,5457497,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2956,0,"",python,selection_command +187,5457533,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2932,0,"",python,selection_command +188,5457565,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2915,0,"",python,selection_command +189,5457598,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2898,0,"",python,selection_command +190,5457631,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2876,0,"",python,selection_command +191,5457665,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2854,0,"",python,selection_command +192,5457699,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2838,0,"",python,selection_command +193,5457732,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2817,0,"",python,selection_command +194,5457764,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2793,0,"",python,selection_command +195,5457799,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2767,0,"",python,selection_command +196,5457833,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2756,0,"",python,selection_command +197,5457866,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2721,0,"",python,selection_command +198,5457899,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2688,0,"",python,selection_command +199,5457931,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2655,0,"",python,selection_command +200,5457964,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2621,0,"",python,selection_command +201,5457999,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2585,0,"",python,selection_command +202,5458033,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2547,0,"",python,selection_command +203,5458067,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2512,0,"",python,selection_command +204,5458102,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2483,0,"",python,selection_command +205,5458137,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2451,0,"",python,selection_command +206,5459402,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2483,0,"",python,selection_command +207,5459593,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2512,0,"",python,selection_command +208,5459748,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2547,0,"",python,selection_command +209,5460152,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2551,0,"",python,selection_command +210,5461248,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2589,0,"",python,selection_command +211,5461431,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2625,0,"",python,selection_command +212,5461588,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2659,0,"",python,selection_command +213,5461782,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2692,0,"",python,selection_command +214,5462187,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2725,0,"",python,selection_command +215,5462694,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2692,0,"",python,selection_command +216,5462868,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2659,0,"",python,selection_command +217,5463112,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2625,0,"",python,selection_command +218,5463146,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2589,0,"",python,selection_command +219,5463184,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2551,0,"",python,selection_command +220,5463214,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2516,0,"",python,selection_command +221,5463329,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2487,0,"",python,selection_command +222,8520791,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +223,8520860,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +224,8685736,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2797,0,"",python,selection_command +225,8686757,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2813,0,"",python,selection_command +226,8686946,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2814,0,"",python,selection_command +227,10758031,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2813,0,"",python,selection_command +228,10758522,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2797,0,"",python,selection_command +229,10759094,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2813,0,"",python,selection_command +230,10759461,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2797,0,"",python,selection_command +231,10762705,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2771,0,"",python,selection_command +232,10880161,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2797,0,"",python,selection_command +233,14746153,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9275,0,"",python,selection_mouse +234,14750022,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9465,0,"",python,selection_mouse +235,14757418,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9932,0,"",python,selection_mouse +236,14758360,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9464,0,"",python,selection_mouse +237,15922202,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_command +238,15924565,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1008,0,"",python,selection_command +239,15925176,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1016,0,"",python,selection_command +240,15925334,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1023,0,"",python,selection_command +241,15925510,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1047,0,"",python,selection_command +242,15925934,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1023,0,"",python,selection_command +243,15926409,"nemo/collections/llm/recipes/finetune_default.py",0,0,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import TYPE_CHECKING, Any, Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\n\nimport nemo.lightning as nl\nfrom nemo.collections import llm\nfrom nemo.collections.llm.gpt.data.packed_sequence import PackedSequenceSpecs\nfrom nemo.collections.llm.peft import DoRA, LoRA\nfrom nemo.collections.llm.recipes.log.default import tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.precision.mixed_precision import bf16_mixed\nfrom nemo.lightning.pytorch.callbacks import PEFT\nfrom nemo.utils.exp_manager import TimingCallback\n\nif TYPE_CHECKING:\n from lightning.pytorch.loggers import TensorBoardLogger, WandbLogger\n\nTokenizerType = Any\n\n\ndef default_finetune_recipe(\n model: run.Config[pl.LightningModule],\n resume_path: str,\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n packed_sequence: bool = False, # once packing recipe is well tested, change this default to true\n tokenizer: Optional[TokenizerType] = ""model"",\n) -> run.Partial:\n """"""\n Create a default fine-tuning recipe for any model.\n\n This function sets up a template for a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n\n Args:\n model (run.Config[pl.LightningModule]): Configuration for a NeMo model.\n resume_path (str): Path to the Huggingface model or pretrained distributed checkpoint for resume\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n packed_sequence (bool): Whether to use packed sequence.\n tokenizer (Optional[TokenizerType]): Tokenizer setting to be applied. Can be 'data' or 'model'\n or an instance of TokenizerSpec.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n See usages of this recipe for further details.\n """"""\n if packed_sequence:\n datamodule = run.Config(\n llm.SquadDataModule,\n seq_length=2048,\n global_batch_size=8,\n micro_batch_size=1,\n packed_sequence_specs=PackedSequenceSpecs(packed_sequence_size=2048),\n )\n else:\n datamodule = run.Config(llm.SquadDataModule, seq_length=2048, global_batch_size=128, micro_batch_size=1)\n recipe = run.Partial(\n llm.finetune,\n model=model,\n trainer=default_finetune_trainer(\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n ),\n data=datamodule,\n log=default_finetune_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(max_lr=1e-4, min_lr=0, warmup_steps=50, adam_beta2=0.98),\n resume=nemo_resume(resume_path),\n tokenizer=tokenizer,\n )\n\n return recipe\n\n\ndef default_finetune_trainer(\n tensor_parallelism=1,\n pipeline_parallelism=1,\n pipeline_parallelism_type=torch.bfloat16,\n virtual_pipeline_parallelism=None,\n context_parallelism=1,\n sequence_parallelism=False,\n num_nodes=1,\n num_gpus_per_node=8,\n max_steps=1000,\n limit_test_batches=None,\n limit_val_batches=None,\n val_check_interval=30,\n):\n """"""\n Create a default fine-tuning trainer for any model.\n\n This function sets up a template for strategy and trainer.\n\n Args:\n See docstrings of MegatronStrategy and Trainer.\n\n Returns:\n run.Config: Config for a finetuning trainer.\n\n See usages of this in recipes for further details.\n """"""\n strategy = run.Config(\n nl.MegatronStrategy,\n tensor_model_parallel_size=tensor_parallelism,\n pipeline_model_parallel_size=pipeline_parallelism,\n pipeline_dtype=pipeline_parallelism_type,\n virtual_pipeline_model_parallel_size=virtual_pipeline_parallelism,\n context_parallel_size=context_parallelism,\n sequence_parallel=sequence_parallelism,\n gradient_as_bucket_view=True,\n ckpt_load_strictness=""log_all"",\n )\n\n trainer = run.Config(\n nl.Trainer,\n accelerator=""gpu"",\n accumulate_grad_batches=1,\n devices=num_gpus_per_node,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=1,\n max_steps=max_steps,\n num_nodes=num_nodes,\n plugins=bf16_mixed(),\n strategy=strategy,\n use_distributed_sampler=False,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n )\n\n return trainer\n\n\ndef default_finetune_log(\n dir: Optional[str] = None,\n name: str = ""default"",\n tensorboard_logger: Optional[run.Config['TensorBoardLogger']] = None,\n wandb_logger: Optional[run.Config['WandbLogger']] = None,\n) -> run.Config[nl.NeMoLogger]:\n """"""\n Create a default fine-tuning logger for any model.\n\n This function sets up a template for ModelCheckpoint and NeMoLogger.\n\n Args:\n See docstrings of ModelCheckpoint and NeMoLogger.\n\n Returns:\n run.Config: Config for a finetuning NeMoLogger.\n\n See usages of this in recipes for further details.\n """"""\n\n ckpt = run.Config(\n nl.ModelCheckpoint,\n save_last=""link"",\n save_top_k=2,\n every_n_train_steps=50,\n filename=""{model_name}--{val_loss:.2f}-{step}-{consumed_samples}"",\n )\n\n return run.Config(\n nl.NeMoLogger,\n ckpt=ckpt,\n name=name,\n tensorboard=tensorboard_logger,\n wandb=wandb_logger,\n log_dir=dir,\n )\n\n\ndef nemo_resume(model_id: str) -> run.Config[nl.AutoResume]:\n """"""\n Configure automatic resumption from a NeMo checkpoint converted from Huggingface for\n https://huggingface.co/{model_id}.\n\n This NeMo checkpoint should be converted from Huggingface beforehand, using nemo.collections.llm.import_ckpt.\n When converting the checkpoint, the NeMo checkpoint will be saved in NEMO_HOME (set to ~/.cache/nemo by default).\n\n This function sets up the configuration to resume training from path nemo://{model_id}.\n This translates to the full path {NEMO_HOME}/models/{model_id}.\n\n Args:\n model_id (str): Path to the Huggingface model or pretrained distributed checkpoint for resume\n\n Returns:\n run.Config[nl.AutoResume]: Configuration for resuming from NeMo checkpoint.\n """"""\n return run.Config(\n nl.AutoResume,\n restore_config=run.Config(nl.RestoreConfig, path=f""nemo://{model_id}""),\n )\n\n\n@run.cli.factory(name='lora')\ndef lora() -> run.Config[PEFT]:\n """"""\n Factory function to create a LoRA configuration.\n\n Returns:\n run.Config[PEFT]: Configuration for the LoRA class.\n\n Examples:\n CLI usage:\n $ nemo llm finetune -f llama3_8b peft=lora\n\n Python API usage:\n >>> lora_config = lora()\n >>> print(lora_config)\n """"""\n return run.Config(LoRA)\n\n\n@run.cli.factory(name='dora')\ndef dora() -> run.Config[PEFT]:\n """"""\n Factory function to create a DoRA configuration.\n\n Returns:\n run.Config[PEFT]: Configuration for the DoRA class.\n\n Examples:\n CLI usage:\n $ nemo llm finetune -f llama3_8b peft=dora\n\n Python API usage:\n >>> dora_config = dora()\n >>> print(dora_config)\n """"""\n return run.Config(DoRA)\n",python,tab +244,15926417,"nemo/collections/llm/recipes/finetune_default.py",1382,0,"",python,selection_command +245,15929448,"nemo/collections/llm/recipes/finetune_default.py",3372,0,"",python,selection_keyboard diff --git a/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-ac4f665d-1bc1-467a-98b3-5da2178968731760857710663-2025_10_19-09.08.50.386/source.csv b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-ac4f665d-1bc1-467a-98b3-5da2178968731760857710663-2025_10_19-09.08.50.386/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..2b2a939d485cf70804ad85a56986997683c4d8fb --- /dev/null +++ b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-ac4f665d-1bc1-467a-98b3-5da2178968731760857710663-2025_10_19-09.08.50.386/source.csv @@ -0,0 +1,6 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,3,"slurm/dev/franz/berlin/coinrun/sample/maskgit/sample_mila_submission_case_study_vanilla.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=24:00:00\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:1\n#SBATCH --output=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/dynamics_sample/%x_%j.log\n#SBATCH --error=/fast/project/HFMI_SynergyUnit/jafar_ws/logs/franz/coinrun/dynamics_sample/%x_%j.log\n#SBATCH --job-name=coinrun_sample_maskgit_mila_submission_case_study_vanilla\n\n# Activate virtual environment\nsource .venv/bin/activate\n\narray_records_dir=""/fast/project/HFMI_SynergyUnit/jafar_ws/data/coinrun/array_records_10M_npy_arr_rec/array_record/test""\nCHECKPOINT_PATH=""/fast/project/HFMI_SynergyUnit/jafar_ws/checkpoints/coinrun/dynamics/dynamics_case_study_dataset_10M_30031""\n\ncurrent_branch=$(git rev-parse --abbrev-ref HEAD)\nif [ ""$current_branch"" != ""main"" ]; then\n echo ""This script must be run from the main branch. Current branch is $current_branch. Exiting.""\n exit 1\nfi\n\necho ""Sampling from checkpoint: $CHECKPOINT_PATH""\n\nsrun python jasmine/sample.py \\n --seed=1 \\n --maskgit_steps=1 \\n --tokenizer_ffn_dim=512 \\n --tokenizer_num_blocks=8 \\n --dyna_ffn_dim=512 \\n --dyna_num_blocks=12 \\n --output_dir=gifs/dynamics_case_study_dataset_10M_vanilla \\n --checkpoint $CHECKPOINT_PATH \\n --data_dir=$array_records_dir \\n --seq_len=16 \\n --batch_size=32 \\n --patch_size=4 \\n --start_frame=4 \\n --image_height=64 \\n --image_width=64 \\n --dyna_type=maskgit\n",shellscript,tab +2,262,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"9:08:50 AM [info] Activating crowd-code\n9:08:50 AM [info] Recording started\n9:08:50 AM [info] Initializing git provider using file system watchers...\n9:08:50 AM [info] Git repository found\n9:08:50 AM [info] Git provider initialized successfully\n9:08:50 AM [info] Initial git state: [object Object]\n",Log,tab +3,4665,"TERMINAL",0,0,"",,terminal_command +4,11496,"TERMINAL",0,0,"",,terminal_command +5,106853,"TERMINAL",0,0,"",,terminal_command diff --git a/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-c877f8c1-c8e0-4a7b-8720-40bb4df915221754138206219-2025_08_02-14.36.55.608/source.csv b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-c877f8c1-c8e0-4a7b-8720-40bb4df915221754138206219-2025_08_02-14.36.55.608/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..b458116df419c2beb541e9533bcb7962630944d7 --- /dev/null +++ b/1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-c877f8c1-c8e0-4a7b-8720-40bb4df915221754138206219-2025_08_02-14.36.55.608/source.csv @@ -0,0 +1,314 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,3,"genie.py",0,0,"from typing import Dict\nimport time\n\nimport optax\nimport jax\nimport jax.numpy as jnp\nimport flax.nnx as nnx\nimport orbax.checkpoint as ocp\n\nfrom models.dynamics import DynamicsMaskGIT, DynamicsCausal\nfrom models.lam import LatentActionModel\nfrom models.tokenizer import TokenizerVQVAE\n\n\nclass Genie(nnx.Module):\n """"""Genie model""""""\n\n def __init__(\n self,\n in_dim: int,\n tokenizer_dim: int,\n tokenizer_ffn_dim: int,\n latent_patch_dim: int,\n num_patch_latents: int,\n patch_size: int,\n tokenizer_num_blocks: int,\n tokenizer_num_heads: int,\n lam_dim: int,\n lam_ffn_dim: int,\n latent_action_dim: int,\n num_latent_actions: int,\n lam_patch_size: int,\n lam_num_blocks: int,\n lam_num_heads: int,\n lam_co_train: bool,\n dyna_type: str,\n dyna_dim: int,\n dyna_ffn_dim: int,\n dyna_num_blocks: int,\n dyna_num_heads: int,\n param_dtype: jnp.dtype,\n dtype: jnp.dtype,\n use_flash_attention: bool,\n decode: bool,\n rngs: nnx.Rngs,\n dropout: float = 0.0,\n mask_limit: float = 0.0,\n ):\n # --- Tokenizer ---\n self.in_dim = in_dim\n self.tokenizer_dim = tokenizer_dim\n self.tokenizer_ffn_dim = tokenizer_ffn_dim\n self.latent_patch_dim = latent_patch_dim\n self.num_patch_latents = num_patch_latents\n self.patch_size = patch_size\n self.tokenizer_num_blocks = tokenizer_num_blocks\n self.tokenizer_num_heads = tokenizer_num_heads\n # --- LAM ---\n self.lam_dim = lam_dim\n self.lam_ffn_dim = lam_ffn_dim\n self.latent_action_dim = latent_action_dim\n self.num_latent_actions = num_latent_actions\n self.lam_patch_size = lam_patch_size\n self.lam_num_blocks = lam_num_blocks\n self.lam_num_heads = lam_num_heads\n self.lam_co_train = lam_co_train\n # --- Dynamics ---\n self.dyna_type = dyna_type\n self.dyna_dim = dyna_dim\n self.dyna_ffn_dim = dyna_ffn_dim\n self.dyna_num_blocks = dyna_num_blocks\n self.dyna_num_heads = dyna_num_heads\n self.param_dtype = param_dtype\n self.dtype = dtype\n self.use_flash_attention = use_flash_attention\n self.dropout = dropout\n self.mask_limit = mask_limit\n\n self.tokenizer = TokenizerVQVAE(\n in_dim=self.in_dim,\n model_dim=self.tokenizer_dim,\n ffn_dim=self.tokenizer_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_patch_latents,\n patch_size=self.patch_size,\n num_blocks=self.tokenizer_num_blocks,\n num_heads=self.tokenizer_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n self.lam = LatentActionModel(\n in_dim=self.in_dim,\n model_dim=self.lam_dim,\n ffn_dim=self.lam_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_latent_actions,\n patch_size=self.lam_patch_size,\n num_blocks=self.lam_num_blocks,\n num_heads=self.lam_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n if self.dyna_type == ""maskgit"":\n self.dynamics = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n elif self.dyna_type == ""causal"":\n self.dynamics = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=decode,\n rngs=rngs,\n )\n else:\n raise ValueError(f""Invalid dynamics type: {self.dyna_type}"")\n\n def __call__(\n self, batch: Dict[str, jax.Array], training: bool = True\n ) -> Dict[str, jax.Array]:\n videos_BTHWC = batch[""videos""]\n tokenizer_outputs = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_indices_BTN = tokenizer_outputs[""indices""]\n lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n outputs = dict(\n video_tokens=jax.lax.stop_gradient(token_indices_BTN),\n latent_actions=latent_actions_BTm11L,\n )\n outputs[""mask_rng""] = batch[""mask_rng""]\n dyna_logits_BTNV, dyna_mask = self.dynamics(outputs, training)\n outputs[""token_logits""] = dyna_logits_BTNV\n if dyna_mask is not None:\n outputs[""mask""] = dyna_mask\n mle_indices_BTN = jnp.argmax(outputs[""token_logits""], axis=-1)\n H, W = batch[""videos""].shape[2:4]\n outputs[""recon""] = self.tokenizer.decode(mle_indices_BTN, (H, W))\n outputs[""lam_indices""] = action_indices_E\n return outputs\n\n # FIXME (f.srambical): sampling should be moved to the dynamics classes\n def sample(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n steps: int = 25,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> jax.Array:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n """"""\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n latent_actions_E = batch[""latent_actions""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n def maskgit_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array], step: jax.Array\n ) -> tuple[tuple[jax.Array, jax.Array, jax.Array, jax.Array], None]:\n rng, token_idxs_BSN, mask_BSN, action_tokens_EL = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # --- Construct + encode video ---\n vid_embed_BSNM = self.dynamics.patch_embed(token_idxs_BSN)\n mask_token_111M = self.dynamics.mask_token.value\n mask_expanded_BSN1 = mask_BSN[..., None]\n vid_embed_BSNM = jnp.where(mask_expanded_BSN1, mask_token_111M, vid_embed_BSNM)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = self.dynamics.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1]))\n vid_embed_BSNM += act_embed_BS1M\n unmasked_ratio = jnp.cos(jnp.pi * (step + 1) / (steps * 2))\n step_temp = temperature * (1.0 - unmasked_ratio)\n final_logits_BSNV = self.dynamics.transformer(vid_embed_BSNM) / step_temp\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_BSN = jnp.argmax(final_logits_BSNV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_BSN = jax.random.categorical(_rng, final_logits_BSNV)\n gather_fn = jax.vmap(jax.vmap(jax.vmap(lambda x, y: x[y])))\n final_token_probs_BSN = gather_fn(\n jax.nn.softmax(final_logits_BSNV), sampled_token_idxs_BSN\n )\n final_token_probs_BSN += ~mask_BSN\n # Update masked tokens only\n token_idxs_BSN = jnp.where(mask_BSN, sampled_token_idxs_BSN, token_idxs_BSN)\n\n # --- Update mask ---\n num_unmasked_tokens = jnp.round(N * (1.0 - unmasked_ratio)).astype(int)\n idx_mask_N = jnp.arange(final_token_probs_BSN.shape[-1]) > num_unmasked_tokens\n sorted_idxs_BSN = jnp.argsort(final_token_probs_BSN, axis=-1, descending=True)\n mask_update_fn = jax.vmap(lambda msk, ids: msk.at[ids].set(idx_mask_N))\n new_mask_BSN = mask_update_fn(mask_BSN, sorted_idxs_BSN)\n\n new_carry = (rng, token_idxs_BSN, new_mask_BSN, action_tokens_EL)\n return new_carry, None\n\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[tuple[jax.Array, jax.Array], None]:\n rng, current_token_idxs_BSN = carry\n rng, step_rng = jax.random.split(rng)\n\n # Mask current and future frames (i.e., t >= step_t)\n mask_S = jnp.arange(seq_len) >= step_t\n mask_BSN = jnp.broadcast_to(mask_S[None, :, None], (B, seq_len, N)).astype(\n bool\n )\n masked_token_idxs_BSN = current_token_idxs_BSN * ~mask_BSN\n\n # --- Initialize and run MaskGIT loop ---\n init_carry_maskgit = (\n step_rng,\n masked_token_idxs_BSN,\n mask_BSN,\n action_tokens_EL,\n )\n final_carry_maskgit, _ = jax.lax.scan(\n maskgit_step_fn, init_carry_maskgit, jnp.arange(steps)\n )\n updated_token_idxs_BSN = final_carry_maskgit[1]\n new_carry = (rng, updated_token_idxs_BSN)\n return new_carry, None\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry, _ = jax.lax.scan(\n generation_step_fn, initial_carry, timesteps_to_scan\n )\n final_token_idxs_BSN = final_carry[1]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC\n\n def sample_causal(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> jax.Array:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n """"""\n # FIXME (f.srambical): reset spatial kv cache after each frame\n assert isinstance(self.dynamics, DynamicsCausal)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n latent_actions_E = batch[""latent_actions""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n dynamics_causal: DynamicsCausal = self.dynamics\n\n for block in dynamics_causal.transformer.blocks:\n block.spatial_attention.init_cache((B * seq_len, (N + 1), self.dyna_dim), dtype=self.dtype)\n block.temporal_attention.init_cache((B * (N + 1), seq_len, self.dyna_dim), dtype=self.dtype)\n\n def causal_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array], step_n: jax.Array\n ) -> tuple[tuple[jax.Array, jax.Array, jax.Array, jax.Array], None]:\n rng, token_idxs_BSN, action_tokens_EL, step_t = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_causal.patch_embed(token_idxs_BSN)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_causal.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1]))\n vid_embed_BSNp1M = jnp.concatenate([act_embed_BS1M, vid_embed_BSNM], axis=2)\n final_logits_BTNp1V = dynamics_causal.transformer(vid_embed_BSNp1M, (step_t, step_n)) / temperature\n final_logits_BV = final_logits_BTNp1V[:, step_t, step_n, :]\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_B = jnp.argmax(final_logits_BV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_B = jax.random.categorical(_rng, final_logits_BV)\n # Update next tokens only\n token_idxs_BSN = token_idxs_BSN.at[:, step_t, step_n].set(sampled_token_idxs_B)\n step_t += 1\n\n new_carry = (rng, token_idxs_BSN, action_tokens_EL, step_t)\n return new_carry, None\n\n # --- Run the autoregressive generation using a for loop ---\n rng = batch[""rng""]\n current_token_idxs_BSN = token_idxs_BSN\n \n for step_t in range(T, seq_len):\n rng, step_rng = jax.random.split(rng)\n\n # --- Reset spatial KV caches before each frame ---\n for block in dynamics_causal.transformer.blocks:\n block.spatial_attention.init_cache((B * seq_len, (N + 1), self.dyna_dim), dtype=self.dtype)\n #breakpoint()\n\n # --- Initialize and run causal loop ---\n init_carry_causal = (\n step_rng,\n current_token_idxs_BSN,\n action_tokens_EL,\n jnp.array(step_t, dtype=jnp.int32),\n )\n\n # current_token_idxs_BSN.block_until_ready()\n # start = time.time()\n final_carry_causal, _ = jax.lax.scan(\n causal_step_fn, init_carry_causal, jnp.arange(N)\n )\n # final_carry_causal[1].block_until_ready()\n # elapsed = time.time() - start\n # print(f""Autoregressive generation time: {elapsed:.4f}s"")\n # breakpoint()\n current_token_idxs_BSN = final_carry_causal[1]\n \n final_token_idxs_BSN = current_token_idxs_BSN\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC\n\n def vq_encode(self, batch: Dict[str, jax.Array], training: bool) -> jax.Array:\n # --- Preprocess videos ---\n video_BTHWC = batch[""videos""]\n lam_output = self.lam.vq_encode(video_BTHWC, training=training)\n lam_indices_E = lam_output[""indices""]\n return lam_indices_E\n\n# FIXME (f.srambical): add conversion script for old checkpoints\ndef restore_genie_components(\n optimizer: nnx.Optimizer,\n sharding: jax.sharding.NamedSharding,\n rng: jax.Array,\n args,\n) -> nnx.Optimizer:\n """"""Restore pre-trained Genie components""""""\n rngs = nnx.Rngs(rng)\n\n tx = optimizer.tx\n model = optimizer.model\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n step_format_fixed_length=6,\n )\n tokenizer_checkpoint_manager = ocp.CheckpointManager(\n directory=args.tokenizer_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_tokenizer = TokenizerVQVAE(\n in_dim=args.image_channels,\n model_dim=args.tokenizer_dim,\n ffn_dim=args.tokenizer_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n num_blocks=args.tokenizer_num_blocks,\n num_heads=args.tokenizer_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n )\n dummy_tokenizer_optimizer = nnx.Optimizer(dummy_tokenizer, tx)\n dummy_tokenizer_optimizer_state = nnx.state(dummy_tokenizer_optimizer)\n abstract_sharded_tokenizer_optimizer_state = _create_abstract_sharded_pytree(\n dummy_tokenizer_optimizer_state, sharding\n )\n restored_tokenizer = tokenizer_checkpoint_manager.restore(\n step=tokenizer_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_tokenizer_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_tokenizer_optimizer.model, restored_tokenizer.model)\n model.tokenizer = dummy_tokenizer_optimizer.model\n tokenizer_checkpoint_manager.close()\n\n if args.lam_checkpoint:\n lam_checkpoint_manager = ocp.CheckpointManager(\n directory=args.lam_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_lam = LatentActionModel(\n in_dim=args.image_channels,\n model_dim=args.lam_dim,\n ffn_dim=args.lam_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_latent_actions,\n patch_size=args.lam_patch_size,\n num_blocks=args.lam_num_blocks,\n num_heads=args.lam_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n )\n dummy_lam_optimizer = nnx.Optimizer(dummy_lam, tx)\n dummy_lam_optimizer_state = nnx.state(dummy_lam_optimizer)\n abstract_sharded_lam_optimizer_state = _create_abstract_sharded_pytree(\n dummy_lam_optimizer_state, sharding\n )\n restored_lam_optimizer = lam_checkpoint_manager.restore(\n step=lam_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_lam_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_lam_optimizer.model, restored_lam_optimizer.model)\n model.lam = dummy_lam_optimizer.model\n # Remove the LAM decoder to save memory and avoid unnecessary computation.\n del model.lam.decoder\n lam_checkpoint_manager.close()\n \n # Reinitialize the optimizer states\n optimizer = nnx.Optimizer(model, tx)\n return optimizer\n\n\ndef _create_abstract_sharded_pytree(\n pytree_template: nnx.GraphState, sharding_spec: jax.sharding.NamedSharding\n) -> jax.Array:\n """"""Replaces arrays in a pytree with ShapeDtypeStructs having the given sharding.""""""\n\n def map_fn(leaf_template):\n if hasattr(leaf_template, ""shape"") and hasattr(leaf_template, ""dtype""):\n return jax.ShapeDtypeStruct(\n leaf_template.shape, leaf_template.dtype, sharding=sharding_spec\n )\n return leaf_template\n\n return jax.tree_util.tree_map(map_fn, pytree_template)\n",python,tab +2,99,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +3,156,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"2:36:55 PM [info] Activating crowd-code\n2:36:55 PM [info] Recording started\n2:36:55 PM [info] Initializing git provider using file system watchers...\n",Log,content +4,261,"extension-output-pdoom-org.crowd-code-#1-crowd-code",150,0,"2:36:55 PM [info] Git repository found\n2:36:55 PM [info] Git provider initialized successfully\n2:36:55 PM [info] Initial git state: [object Object]\n",Log,content +5,6572,"genie.py",0,0,"",python,tab +6,565308,"genie.py",17063,0,"",python,selection_command +7,565707,"genie.py",17062,0,"",python,selection_command +8,565771,"genie.py",17059,0,"",python,selection_command +9,566206,"genie.py",17059,1,"j",python,selection_command +10,566302,"genie.py",17059,3,"jax",python,selection_command +11,566508,"genie.py",17059,4,"jax.",python,selection_command +12,566690,"genie.py",17059,7,"jax.lax",python,selection_command +13,566937,"genie.py",17059,7,"",python,content +14,567181,"genie.py",17059,0,"n",python,content +15,567181,"genie.py",17060,0,"",python,selection_keyboard +16,567274,"genie.py",17060,0,"x",python,content +17,567274,"genie.py",17061,0,"",python,selection_keyboard +18,567423,"genie.py",17061,0,"x",python,content +19,567423,"genie.py",17062,0,"",python,selection_keyboard +20,567655,"genie.py",17061,1,"",python,content +21,567804,"genie.py",17060,1,"",python,content +22,567968,"genie.py",17060,0,"n",python,content +23,567969,"genie.py",17061,0,"",python,selection_keyboard +24,568083,"genie.py",17061,0,"x",python,content +25,568084,"genie.py",17062,0,"",python,selection_keyboard +26,568356,"genie.py",17061,0,"",python,selection_command +27,590947,"genie.py",17062,0,"",python,selection_command +28,591084,"genie.py",17063,0,"",python,selection_command +29,2336884,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",0,0,"# Copyright 2024 The Flax Authors.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# pytype: skip-file\n\nfrom collections import deque\nimport dataclasses\nimport functools\nimport typing as tp\n\nfrom flax import struct\nfrom flax.core.frozen_dict import FrozenDict\nfrom flax.nnx import extract, filterlib, graph, spmd, variablelib\nfrom flax.nnx import statelib\nfrom flax.nnx.module import Module\nfrom flax.nnx.statelib import State\nfrom flax.nnx.transforms.transforms import resolve_kwargs\nfrom flax.typing import Leaf, Missing, PytreeDeque\nimport jax\nimport jax.core\nimport jax.numpy as jnp\nimport jax.stages\nimport numpy as np\n\nA = tp.TypeVar('A')\nC = tp.TypeVar('C')\nB = tp.TypeVar('B')\nF = tp.TypeVar('F', bound=tp.Callable[..., tp.Any])\nG = tp.TypeVar('G', bound=tp.Callable[..., tp.Any])\nM = tp.TypeVar('M', bound=Module)\nMA = tp.TypeVar('MA', bound=Module)\nN = tp.TypeVar('N', bound=Module)\nT = tp.TypeVar('T')\nStrInt = tp.TypeVar('StrInt', str, int)\nAxisName = tp.Hashable\nLeaves = tp.List[Leaf]\nIndex = int\n\n\nclass Carry:\n pass\n\n\n# -------------------------------\n# vmap\n# -------------------------------\n\n\nclass StateAxes(extract.PrefixMapping):\n\n def __init__(\n self,\n filter_axes: (\n statelib.State\n | tp.Mapping[filterlib.Filter, Index | type[Carry] | None]\n | tp.Iterable[tuple[filterlib.Filter, Index | type[Carry] | None]]\n ),\n /,\n ):\n if isinstance(filter_axes, statelib.State):\n filter_axes = statelib.create_path_filters(filter_axes) # type: ignore\n\n iterable = tuple(\n filter_axes.items()\n if isinstance(filter_axes, tp.Mapping)\n else filter_axes\n )\n self._filters = tuple(filter for filter, _ in iterable)\n self._axes = tuple(axis for _, axis in iterable)\n\n @property\n def filters(self) -> tuple[filterlib.Filter, ...]:\n return self._filters\n\n @property\n def axes(self) -> tuple[Index | type[Carry] | None, ...]:\n return self._axes\n\n def map_prefix(\n self, path: variablelib.PathParts, variable: variablelib.Variable\n ) -> tp.Any:\n for filter, axis in zip(self.filters, self.axes):\n predicate = filterlib.to_predicate(filter)\n if predicate(path, variable):\n return axis\n raise ValueError(f'No axis found for {path=}, {variable=}')\n\n def __repr__(self):\n return f'StateAxes({dict(self.items())})'\n\n def items(self):\n return zip(self.filters, self.axes)\n\n def __eq__(self, other):\n return (\n isinstance(other, StateAxes)\n and self.filters == other.filters\n and self.axes == other.axes\n )\n\n def __hash__(self):\n return hash((self.filters, self.axes))\n\n\nAxisFn = tp.Callable[\n [graph.GraphState | variablelib.VariableState, int, tp.Mapping],\n graph.GraphState | variablelib.VariableState,\n]\n\n\ndef _update_variable_sharding_metadata(\n tree, transform_metadata, axis_fn: AxisFn\n):\n def _update_axes_fn(node_states):\n if isinstance(node_states, extract.NodeStates) and isinstance(\n node_states.metadata, (StateAxes, int)\n ):\n if isinstance(node_states.metadata, int):\n state = node_states.state\n assert isinstance(state, State | variablelib.VariableState)\n state = axis_fn(state, node_states.metadata, transform_metadata)\n return node_states.replace(states=(state,))\n else:\n states_out: list[graph.GraphState | variablelib.VariableState] = []\n for state, axis in zip(node_states.states, node_states.metadata.axes):\n assert isinstance(state, graph.State | variablelib.VariableState)\n if isinstance(axis, int):\n state = axis_fn(state, axis, transform_metadata)\n states_out.append(state)\n return node_states.replace(states=tuple(states_out))\n return node_states\n\n return jax.tree.map(\n _update_axes_fn, tree, is_leaf=lambda x: isinstance(x, extract.NodeStates)\n )\n\n\ndef _vmap_split_fn(ctx: graph.SplitContext, path, prefix, x):\n if isinstance(prefix, StateAxes):\n return extract.NodeStates.from_split(\n *ctx.split(x, *prefix.filters), metadata=prefix\n )\n return extract.NodeStates.from_split(*ctx.split(x), metadata=prefix)\n\n\n@dataclasses.dataclass(eq=False)\nclass VmapFn:\n f: tp.Callable[..., tp.Any]\n transform_metadata: tp.Mapping[str, tp.Any]\n in_axes: tp.Any\n out_axes: tp.Any\n\n def __post_init__(self):\n functools.update_wrapper(self, self.f)\n\n def __call__(self, *pure_args: tuple[tp.Any, ...]):\n if spmd.PARTITION_NAME in self.transform_metadata:\n pure_args = _update_variable_sharding_metadata(\n pure_args, self.transform_metadata, spmd.remove_axis\n )\n args = extract.from_tree(pure_args, ctxtag='vmap', is_inner=True)\n\n out = self.f(*args)\n\n args_out = extract.clear_non_graph_nodes(args)\n pure_args_out, pure_out = extract.to_tree(\n (args_out, out),\n prefix=(self.in_axes, self.out_axes),\n split_fn=_vmap_split_fn,\n ctxtag='vmap',\n )\n if spmd.PARTITION_NAME in self.transform_metadata:\n pure_args_out, pure_out = _update_variable_sharding_metadata(\n (pure_args_out, pure_out), self.transform_metadata, spmd.add_axis\n )\n return pure_args_out, pure_out\n\n\n@tp.overload\ndef vmap(\n *,\n in_axes: int | None | tp.Sequence[tp.Any] = 0,\n out_axes: tp.Any = 0,\n axis_name: AxisName | None = None,\n axis_size: int | None = None,\n spmd_axis_name: AxisName | tuple[AxisName, ...] | None = None,\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> tp.Callable[[F], F]:\n ...\n\n\n@tp.overload\ndef vmap(\n f: F,\n *,\n in_axes: int | None | tp.Sequence[tp.Any] = 0,\n out_axes: tp.Any = 0,\n axis_name: AxisName | None = None,\n axis_size: int | None = None,\n spmd_axis_name: AxisName | tuple[AxisName, ...] | None = None,\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F:\n ...\n\n\ndef vmap(\n f: F | type[Missing] = Missing,\n *,\n in_axes: int | None | tp.Sequence[tp.Any] = 0,\n out_axes: tp.Any = 0,\n axis_name: AxisName | None = None,\n axis_size: int | None = None,\n spmd_axis_name: AxisName | tuple[AxisName, ...] | None = None,\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n """"""Reference-aware version of `jax.vmap `__.\n\n Args:\n f: Function to be mapped over additional axes.\n in_axes: An integer, None, or sequence of values specifying which input\n array axes to map over (see `jax.vmap\n `__). In\n addition to integers and None, :class:`StateAxes` can be used to control\n how graph nodes like Modules are vectorized by specifying the axes to be\n applied to substates of the graph node given a `Filter\n `__.\n out_axes: An integer, None, or pytree indicating where the mapped axis\n should appear in the output (see `jax.vmap\n `__).\n axis_name: Optional, a hashable Python object used to identify the mapped\n axis so that parallel collectives can be applied.\n axis_size: Optional, an integer indicating the size of the axis to be\n mapped. If not provided, the mapped axis size is inferred from arguments.\n\n Returns:\n Batched/vectorized version of ``f`` with arguments that correspond to\n those of ``f``, but with extra array axes at positions indicated by\n ``in_axes``, and a return value that corresponds to that of ``f``, but\n with extra array axes at positions indicated by ``out_axes``.\n\n Example::\n\n >>> from flax import nnx\n >>> from jax import random, numpy as jnp\n ...\n >>> model = nnx.Linear(2, 3, rngs=nnx.Rngs(0))\n >>> x = jnp.ones((5, 2))\n ...\n >>> @nnx.vmap(in_axes=(None, 0), out_axes=0)\n ... def forward(model, x):\n ... return model(x)\n ...\n >>> y = forward(model, x)\n >>> y.shape\n (5, 3)\n\n >>> class LinearEnsemble(nnx.Module):\n ... def __init__(self, num, rngs):\n ... self.w = nnx.Param(jax.random.uniform(rngs(), (num, 2, 3)))\n ...\n >>> model = LinearEnsemble(5, rngs=nnx.Rngs(0))\n >>> x = jnp.ones((2,))\n ...\n >>> @nnx.vmap(in_axes=(0, None), out_axes=0)\n ... def forward(model, x):\n ... return jnp.dot(x, model.w.value)\n ...\n >>> y = forward(model, x)\n >>> y.shape\n (5, 3)\n\n To control control how graph node substates are vectorized, ``StateAxes``\n can be passed to ``in_axes`` and ``out_axes`` specifying the axes to be\n applied to each substate given a filter. The following example shows how to\n share the parameters between the ensemble members which keeping different\n batch statistics and dropout random state::\n\n >>> class Foo(nnx.Module):\n ... def __init__(self):\n ... self.a = nnx.Param(jnp.arange(4))\n ... self.b = nnx.BatchStat(jnp.arange(4))\n ...\n >>> state_axes = nnx.StateAxes({nnx.Param: 0, nnx.BatchStat: None})\n >>> @nnx.vmap(in_axes=(state_axes,), out_axes=0)\n ... def mul(foo):\n ... return foo.a * foo.b\n ...\n >>> foo = Foo()\n >>> y = mul(foo)\n >>> y\n Array([[0, 0, 0, 0],\n [0, 1, 2, 3],\n [0, 2, 4, 6],\n [0, 3, 6, 9]], dtype=int32)\n """"""\n if f is Missing:\n return functools.partial(\n vmap,\n in_axes=in_axes,\n out_axes=out_axes,\n axis_name=axis_name,\n axis_size=axis_size,\n spmd_axis_name=spmd_axis_name,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n jax_in_axes = jax.tree.map(\n lambda x: extract.NodeStates.from_prefixes(x.axes, metadata=x)\n if isinstance(x, StateAxes)\n else x,\n in_axes,\n )\n jax_out_axes = jax.tree.map(\n lambda x: extract.NodeStates.from_prefixes(x.axes, metadata=x)\n if isinstance(x, StateAxes)\n else x,\n out_axes,\n )\n vmapped_fn = jax.vmap(\n VmapFn(f, transform_metadata, in_axes, out_axes),\n in_axes=jax_in_axes,\n out_axes=(jax_in_axes, jax_out_axes),\n axis_name=axis_name,\n axis_size=axis_size,\n spmd_axis_name=spmd_axis_name,\n )\n\n @functools.wraps(f)\n @graph.update_context('vmap')\n def vmap_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n pure_args = extract.to_tree(\n args, prefix=in_axes, split_fn=_vmap_split_fn, ctxtag='vmap'\n )\n pure_args_out, pure_out = vmapped_fn(*pure_args)\n _args_out, out = extract.from_tree(\n (pure_args_out, pure_out), ctxtag='vmap', is_inner=False\n )\n return out\n\n return vmap_wrapper # type: ignore\n\n\n# -------------------------------\n# pmap\n# -------------------------------\n\n\n@dataclasses.dataclass(eq=False)\nclass PmapFn:\n f: tp.Callable[..., tp.Any]\n transform_metadata: tp.Mapping[str, tp.Any]\n in_axes: tp.Any\n out_axes: tp.Any\n\n def __post_init__(self):\n functools.update_wrapper(self, self.f)\n\n def __call__(self, *pure_args: tuple[tp.Any, ...]):\n if spmd.PARTITION_NAME in self.transform_metadata:\n pure_args = _update_variable_sharding_metadata(\n pure_args, self.transform_metadata, spmd.remove_axis\n )\n args = extract.from_tree(pure_args, ctxtag='pmap', is_inner=True)\n\n out = self.f(*args)\n\n args_out = extract.clear_non_graph_nodes(args)\n pure_args_out, pure_out = extract.to_tree(\n (args_out, out),\n prefix=(self.in_axes, self.out_axes),\n split_fn=_vmap_split_fn,\n ctxtag='pmap',\n )\n if spmd.PARTITION_NAME in self.transform_metadata:\n pure_args_out, pure_out = _update_variable_sharding_metadata(\n (pure_args_out, pure_out), self.transform_metadata, spmd.add_axis\n )\n return pure_args_out, pure_out\n\n\n@tp.overload\ndef pmap(\n *,\n axis_name: AxisName | None = None,\n in_axes: tp.Any = 0,\n out_axes: tp.Any = 0,\n static_broadcasted_argnums: int | tp.Iterable[int] = (),\n devices: tp.Sequence[jax.Device] | None = None, # noqa: F811\n backend: str | None = None,\n axis_size: int | None = None,\n donate_argnums: int | tp.Iterable[int] = (),\n global_arg_shapes: tuple[tuple[int, ...], ...] | None = None,\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> tp.Callable[[F], F]:\n ...\n\n\n@tp.overload\ndef pmap(\n f: F,\n *,\n axis_name: AxisName | None = None,\n in_axes: tp.Any = 0,\n out_axes: tp.Any = 0,\n static_broadcasted_argnums: int | tp.Iterable[int] = (),\n devices: tp.Sequence[jax.Device] | None = None, # noqa: F811\n backend: str | None = None,\n axis_size: int | None = None,\n donate_argnums: int | tp.Iterable[int] = (),\n global_arg_shapes: tuple[tuple[int, ...], ...] | None = None,\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F:\n ...\n\n\ndef pmap(\n f: F | type[Missing] = Missing,\n *,\n axis_name: AxisName | None = None,\n in_axes: tp.Any = 0,\n out_axes: tp.Any = 0,\n static_broadcasted_argnums: int | tp.Iterable[int] = (),\n devices: tp.Sequence[jax.Device] | None = None, # noqa: F811\n backend: str | None = None,\n axis_size: int | None = None,\n donate_argnums: int | tp.Iterable[int] = (),\n global_arg_shapes: tuple[tuple[int, ...], ...] | None = None,\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n """"""Reference-aware version of `jax.vmap `__.\n\n Args:\n f: Function to be mapped over additional axes.\n in_axes: An integer, None, or sequence of values specifying which input\n array axes to map over (see `jax.vmap\n `__). In\n addition to integers and None, :class:`StateAxes` can be used to control\n how graph nodes like Modules are vectorized by specifying the axes to be\n applied to substates of the graph node given a `Filter\n `__.\n out_axes: An integer, None, or pytree indicating where the mapped axis\n should appear in the output (see `jax.vmap\n `__).\n axis_name: Optional, a hashable Python object used to identify the mapped\n axis so that parallel collectives can be applied.\n axis_size: Optional, an integer indicating the size of the axis to be\n mapped. If not provided, the mapped axis size is inferred from arguments.\n\n Returns:\n Batched/vectorized version of ``f`` with arguments that correspond to\n those of ``f``, but with extra array axes at positions indicated by\n ``in_axes``, and a return value that corresponds to that of ``f``, but\n with extra array axes at positions indicated by ``out_axes``.\n\n Example::\n\n >>> from flax import nnx\n >>> from jax import random, numpy as jnp\n ...\n >>> model = nnx.Linear(2, 3, rngs=nnx.Rngs(0))\n >>> x = jnp.ones((5, 2))\n ...\n >>> @nnx.vmap(in_axes=(None, 0), out_axes=0)\n ... def forward(model, x):\n ... return model(x)\n ...\n >>> y = forward(model, x)\n >>> y.shape\n (5, 3)\n\n >>> class LinearEnsemble(nnx.Module):\n ... def __init__(self, num, rngs):\n ... self.w = nnx.Param(jax.random.uniform(rngs(), (num, 2, 3)))\n ...\n >>> model = LinearEnsemble(5, rngs=nnx.Rngs(0))\n >>> x = jnp.ones((2,))\n ...\n >>> @nnx.vmap(in_axes=(0, None), out_axes=0)\n ... def forward(model, x):\n ... return jnp.dot(x, model.w.value)\n ...\n >>> y = forward(model, x)\n >>> y.shape\n (5, 3)\n\n To control control how graph node substates are vectorized, ``StateAxes``\n can be passed to ``in_axes`` and ``out_axes`` specifying the axes to be\n applied to each substate given a filter. The following example shows how to\n share the parameters between the ensemble members which keeping different\n batch statistics and dropout random state::\n\n >>> class Foo(nnx.Module):\n ... def __init__(self):\n ... self.a = nnx.Param(jnp.arange(4))\n ... self.b = nnx.BatchStat(jnp.arange(4))\n ...\n >>> state_axes = nnx.StateAxes({nnx.Param: 0, nnx.BatchStat: None})\n >>> @nnx.vmap(in_axes=(state_axes,), out_axes=0)\n ... def mul(foo):\n ... return foo.a * foo.b\n ...\n >>> foo = Foo()\n >>> y = mul(foo)\n >>> y\n Array([[0, 0, 0, 0],\n [0, 1, 2, 3],\n [0, 2, 4, 6],\n [0, 3, 6, 9]], dtype=int32)\n """"""\n if f is Missing:\n return functools.partial(\n pmap,\n axis_name=axis_name,\n in_axes=in_axes,\n out_axes=out_axes,\n static_broadcasted_argnums=static_broadcasted_argnums,\n devices=devices,\n backend=backend,\n axis_size=axis_size,\n donate_argnums=donate_argnums,\n global_arg_shapes=global_arg_shapes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n jax_in_axes = jax.tree.map(\n lambda x: extract.NodeStates.from_prefixes(x.axes, metadata=x)\n if isinstance(x, StateAxes)\n else x,\n in_axes,\n )\n jax_out_axes = jax.tree.map(\n lambda x: extract.NodeStates.from_prefixes(x.axes, metadata=x)\n if isinstance(x, StateAxes)\n else x,\n out_axes,\n )\n pmapped_fn = jax.pmap(\n PmapFn(f, transform_metadata, in_axes, out_axes),\n axis_name=axis_name,\n in_axes=jax_in_axes,\n out_axes=(jax_in_axes, jax_out_axes),\n static_broadcasted_argnums=static_broadcasted_argnums,\n devices=devices,\n backend=backend,\n axis_size=axis_size,\n donate_argnums=donate_argnums,\n global_arg_shapes=global_arg_shapes,\n )\n\n @functools.wraps(f)\n @graph.update_context('pmap')\n def vmap_wrapper(*args):\n pure_args = extract.to_tree(\n args, prefix=in_axes, split_fn=_vmap_split_fn, ctxtag='pmap'\n )\n pure_args_out, pure_out = pmapped_fn(*pure_args)\n _args_out, out = extract.from_tree(\n (pure_args_out, pure_out), ctxtag='pmap', is_inner=False\n )\n return out\n\n return vmap_wrapper # type: ignore\n\n\n# -------------------------------\n# scan\n# -------------------------------\n\n\nclass Broadcasted(struct.PyTreeNode):\n data: tp.Any\n\ndef _get_carry_argnum(axes, is_in_axes: bool):\n if axes is Carry:\n return 'all'\n elif isinstance(axes, int) or axes is None:\n return None\n\n obj_repr = 'in_axes' if is_in_axes else 'out_axes'\n carry_argnum: int | None = None\n prev_key: tp.Any = None\n for key, x in jax.tree_util.tree_leaves_with_path(axes):\n if x is not Carry:\n continue\n assert isinstance(key[0], jax.tree_util.SequenceKey)\n i = key[0].idx\n if len(key) >= 2:\n raise ValueError(\n f'Carry must at the top-level, it cannot be nested. Found {axes=}'\n )\n if carry_argnum is not None:\n raise ValueError(\n f'Found multiple Carry definitions at '\n f'{obj_repr}{jax.tree_util.keystr(prev_key)} and '\n f'{obj_repr}{jax.tree_util.keystr(key)}'\n )\n carry_argnum = i\n prev_key = key\n\n return carry_argnum\n\n\ndef _check_out_axes(out_axes):\n for key, x in jax.tree_util.tree_leaves_with_path(\n out_axes, is_leaf=lambda x: x is None\n ):\n if x is None:\n raise ValueError(\n f'Cannot broadcast output state. '\n f'Got out_axes=None at: out_axes{jax.tree_util.keystr(key)}'\n )\n elif isinstance(x, StateAxes):\n for filter, value in x.items():\n if value is None:\n raise ValueError(\n f'Cannot broadcast output state. '\n f'Got StateAxes({{{filter}: None}}) at: out_axes'\n f'{jax.tree_util.keystr(key)}'\n )\n elif value is Carry:\n raise ValueError(\n f'Cannot carry output state. '\n f'Got StateAxes({{{filter}: Carry}}) at: out_axes'\n f'{jax.tree_util.keystr(key)}'\n )\ndef _check_carry_same_references(carry_arg, carry_arg_out):\n def check_carry_same_references(key_path, arg, out):\n if (\n not isinstance(arg, jax.Array) or not isinstance(out, jax.Array)\n ) and arg is not out:\n raise ValueError(\n 'Carry references must be the same between iterations. '\n f'Got {arg=} with id={id(arg)} and {out=} with id={id(out)} '\n f'at carry{jax.tree_util.keystr(key_path)}'\n )\n\n jax.tree_util.tree_map_with_path(\n check_carry_same_references, carry_arg, carry_arg_out\n )\n\ndef _extract_graphdefs(\n pure_carry_arg_out, carry_graphdefs: list[graph.GraphDef], /\n):\n def extract_index_mappings(x):\n if isinstance(x, extract.NodeStates) and isinstance(\n x._graphdef, graph.GraphDef\n ):\n graphdef = x._graphdef\n carry_graphdefs.append(graphdef)\n x = x.replace(_graphdef=graphdef.with_no_outer_index())\n return x\n\n pure_carry_arg_out = jax.tree.map(\n extract_index_mappings,\n pure_carry_arg_out,\n is_leaf=lambda x: isinstance(x, extract.NodeStates),\n )\n\n return pure_carry_arg_out\n\ndef _insert_graphdefs(\n pure_carry_arg_out,\n carry_graphdefs: deque[graph.GraphDef],\n /,\n):\n def insert_index_mappings(x):\n if isinstance(x, extract.NodeStates) and isinstance(\n x._graphdef, graph.GraphDef\n ):\n graphdef = carry_graphdefs.popleft()\n x = x.replace(_graphdef=graphdef)\n return x\n\n pure_carry_arg_out = jax.tree.map(\n insert_index_mappings,\n pure_carry_arg_out,\n is_leaf=lambda x: isinstance(x, extract.NodeStates),\n )\n return pure_carry_arg_out\n\n\ndef _scan_split_in(\n carry_deque: PytreeDeque[list[State | variablelib.VariableState]],\n broadcast_deque: PytreeDeque[list[State | variablelib.VariableState]],\n broadcast_arrays: PytreeDeque[Broadcasted],\n /,\n ctx: graph.SplitContext,\n path,\n prefix,\n x,\n):\n if graph.is_graph_node(x) or isinstance(x, variablelib.Variable):\n vectorized_states: list[State | variablelib.VariableState] = []\n carry_states: list[State | variablelib.VariableState] = []\n broadcast_states: list[State | variablelib.VariableState] = []\n if isinstance(prefix, StateAxes):\n graphdef, *states = ctx.split(x, *prefix.filters)\n\n for state, axis in zip(states, prefix.axes):\n if axis is None:\n broadcast_states.append(state)\n elif isinstance(axis, int):\n state = jax.tree.map(lambda x: jnp.moveaxis(x, axis, 0), state)\n vectorized_states.append(state)\n else: # axis is Carry\n carry_states.append(state)\n\n if not vectorized_states:\n vectorized_states.append(State({}))\n carry_deque.append(carry_states)\n broadcast_deque.append(broadcast_states)\n return extract.NodeStates.from_split(\n graphdef, *vectorized_states, metadata=prefix\n )\n elif isinstance(prefix, int):\n graphdef, state = ctx.split(x)\n state = jax.tree.map(lambda x: jnp.moveaxis(x, prefix, 0), state)\n vectorized_states.append(state)\n elif prefix is None:\n graphdef, state = ctx.split(x)\n broadcast_states.append(state)\n vectorized_states.append(State({}))\n elif prefix is Carry:\n graphdef, state = ctx.split(x)\n carry_states.append(state)\n vectorized_states.append(State({}))\n else:\n raise ValueError(\n f'Invalid axes {prefix} args{jax.tree_util.keystr(path)}'\n )\n\n if not vectorized_states:\n vectorized_states.append(State({}))\n carry_deque.append(carry_states)\n broadcast_deque.append(broadcast_states)\n return extract.NodeStates.from_split(\n graphdef, *vectorized_states, metadata=prefix\n )\n else:\n if isinstance(prefix, StateAxes):\n raise ValueError(\n 'Cannot use StateAxes on non-graph nodes, '\n f'found {prefix} args{jax.tree_util.keystr(path)}'\n )\n elif prefix is Carry:\n return x\n elif prefix is None:\n broadcast_arrays.append(Broadcasted(x))\n return Broadcasted(None)\n elif isinstance(prefix, int):\n if not isinstance(x, (jax.Array, np.ndarray)):\n raise ValueError(\n f'Expected an array, got {type(x).__name__} args'\n f'{jax.tree_util.keystr(path)}'\n )\n return jnp.moveaxis(x, prefix, 0)\n else:\n raise ValueError(\n f'Invalid axes {prefix} args{jax.tree_util.keystr(path)}'\n )\n\n\ndef _scan_split_out(\n carry_deque: PytreeDeque[list[State | variablelib.VariableState]],\n broadcast_deque: PytreeDeque[list[State | variablelib.VariableState]],\n /,\n ctx: graph.SplitContext,\n path: extract.KeyPath,\n prefix,\n x,\n):\n assert isinstance(path[0], jax.tree_util.SequenceKey)\n is_input_arg = path[0].idx == 0\n\n if graph.is_graph_node(x) or isinstance(x, variablelib.Variable):\n vectorized_states: list[State | variablelib.VariableState] = []\n carry_states: list[State | variablelib.VariableState] = []\n broadcast_states: list[State | variablelib.VariableState] = []\n if isinstance(prefix, StateAxes):\n graphdef, *states = ctx.split(x, *prefix.filters)\n\n for state, filter, axis in zip(states, prefix.filters, prefix.axes):\n if axis is None:\n assert is_input_arg # validated by _check_out_axes\n broadcast_states.append(state)\n elif isinstance(axis, int):\n vectorized_states.append(state)\n elif axis is Carry:\n assert is_input_arg # validated by _check_out_axes\n carry_states.append(state)\n else:\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n f'Invalid axes {axis} for filter {filter} at '\n f'{obj_repr}{jax.tree_util.keystr(path)}'\n )\n\n if not vectorized_states:\n vectorized_states.append(State({}))\n if is_input_arg:\n carry_deque.append(carry_states)\n broadcast_deque.append(broadcast_states)\n return extract.NodeStates.from_split(\n graphdef, *vectorized_states, metadata=prefix\n )\n elif isinstance(prefix, int):\n graphdef, state = ctx.split(x)\n vectorized_states.append(state)\n elif prefix is None:\n assert is_input_arg # validated by _check_out_axes\n graphdef, state = ctx.split(x)\n broadcast_states.append(state)\n vectorized_states.append(State({}))\n elif prefix is Carry:\n assert is_input_arg # validated by _check_out_axes\n graphdef, state = ctx.split(x)\n carry_states.append(state)\n vectorized_states.append(State({}))\n else:\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n f'Invalid axes {prefix} at {obj_repr}{jax.tree_util.keystr(path)}'\n )\n if not vectorized_states:\n vectorized_states.append(State({}))\n if is_input_arg:\n carry_deque.append(carry_states)\n broadcast_deque.append(broadcast_states)\n return extract.NodeStates.from_split(\n graphdef, *vectorized_states, metadata=prefix\n )\n else:\n if isinstance(prefix, StateAxes):\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n 'Cannot use StateAxes on non-graph nodes, '\n f'found {prefix} at {obj_repr}{jax.tree_util.keystr(path)}'\n )\n elif prefix is Carry:\n return x\n elif prefix is None:\n assert not is_input_arg # validated by _check_out_axes\n return Broadcasted(None)\n elif isinstance(prefix, int):\n return x\n else:\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n f'Invalid axes {prefix} at {obj_repr}{jax.tree_util.keystr(path)}'\n )\n\n\ndef _scan_merge_in(\n carry_deque: PytreeDeque[list[State | variablelib.VariableState]],\n broadcast_deque: PytreeDeque[list[State | variablelib.VariableState]],\n broadcast_arrays: PytreeDeque[Broadcasted],\n /,\n ctx: graph.MergeContext,\n path,\n prefix,\n x,\n):\n if isinstance(x, extract.NodeStates):\n carry_states = carry_deque.popleft()\n broadcast_states = broadcast_deque.popleft()\n return ctx.merge(x.graphdef, *x.states, *carry_states, *broadcast_states)\n elif isinstance(x, Broadcasted):\n assert x.data is None\n return broadcast_arrays.popleft().data\n else:\n return x\n\n\ndef _scan_merge_out(\n carry_deque: PytreeDeque[list[State | variablelib.VariableState]],\n broadcast_deque: PytreeDeque[list[State | variablelib.VariableState]],\n /,\n ctx: graph.MergeContext,\n path,\n prefix,\n x,\n):\n assert isinstance(path[0], jax.tree_util.SequenceKey)\n is_input_arg = path[0].idx == 0\n\n if isinstance(x, extract.NodeStates):\n states: list[State | variablelib.VariableState] = []\n if is_input_arg:\n carry_states = deque(carry_deque.popleft())\n broadcast_states = deque(broadcast_deque.popleft())\n else:\n carry_states = deque[State | variablelib.VariableState]()\n broadcast_states = deque[State | variablelib.VariableState]()\n if isinstance(prefix, StateAxes):\n vectorized_states = deque(x.states)\n for axis in prefix.axes:\n if isinstance(axis, int):\n state = vectorized_states.popleft()\n state = jax.tree.map(lambda x: jnp.moveaxis(x, 0, axis), state)\n states.append(state)\n elif axis is None:\n states.append(broadcast_states.popleft())\n else: # axis is Carry\n states.append(carry_states.popleft())\n assert not carry_states and not broadcast_states\n assert not vectorized_states or (\n len(vectorized_states) == 1 and not vectorized_states[0]\n )\n elif isinstance(prefix, int):\n state = jax.tree.map(lambda x: jnp.moveaxis(x, 0, prefix), x.state)\n states.extend((state, *carry_states, *broadcast_states))\n elif prefix is None:\n assert is_input_arg\n states.extend(broadcast_states)\n elif prefix is Carry:\n assert is_input_arg\n states.extend(carry_states)\n else:\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n f'Invalid axes {prefix} at {obj_repr}{jax.tree_util.keystr(path)}'\n )\n\n return ctx.merge(x.graphdef, *states)\n else:\n if isinstance(prefix, StateAxes):\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n 'Cannot use StateAxes on non-graph nodes, '\n f'found {prefix} at {obj_repr}{jax.tree_util.keystr(path)}'\n )\n elif prefix is Carry:\n return x\n elif prefix is None:\n return x\n elif isinstance(prefix, int):\n if not isinstance(x, (jax.Array, np.ndarray)):\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n f'Expected an array, got {type(x).__name__} at '\n f'{obj_repr}{jax.tree_util.keystr(path)}'\n )\n return jnp.moveaxis(x, 0, prefix)\n else:\n obj_repr = 'args' if is_input_arg else 'out'\n raise ValueError(\n f'Invalid axes {prefix} at {obj_repr}{jax.tree_util.keystr(path)}'\n )\n\n\n@dataclasses.dataclass(eq=False)\nclass ScanFn:\n f: tp.Callable[..., tp.Any]\n input_carry_argnum: int | None | tp.Literal['all']\n output_carry_argnum: int | None | tp.Literal['all']\n in_axes: tp.Any\n out_axes: tp.Any\n transform_metadata: tp.Mapping[str, tp.Any]\n\n def __post_init__(self):\n functools.update_wrapper(self, self.f)\n\n def __call__(\n self,\n carry: tuple[\n tp.Any, # carry_arg\n PytreeDeque[list[State | variablelib.VariableState]], # carry_deque\n PytreeDeque[list[State | variablelib.VariableState]], # broadcast_deque\n PytreeDeque[Broadcasted], # broadcast_arrays\n ],\n pure_args: tuple[tp.Any, ...],\n ):\n pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays = carry\n broadcast_deque_out = PytreeDeque(broadcast_deque)\n broadcast_arrays_out = PytreeDeque(broadcast_arrays)\n\n if self.input_carry_argnum == 'all':\n assert pure_args == ()\n pure_args = (pure_carry_arg,)\n elif isinstance(self.input_carry_argnum, int):\n assert pure_args[self.input_carry_argnum] is None\n _pure_args = list(pure_args)\n _pure_args[self.input_carry_argnum] = pure_carry_arg\n pure_args = tuple(_pure_args)\n else:\n assert self.input_carry_argnum is None\n assert pure_carry_arg is None\n\n if spmd.PARTITION_NAME in self.transform_metadata:\n pure_args = _update_variable_sharding_metadata(\n pure_args, self.transform_metadata, spmd.remove_axis\n )\n\n args: tuple = extract.from_tree(\n pure_args,\n prefix=self.in_axes,\n merge_fn=functools.partial(\n _scan_merge_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=True,\n )\n assert not carry_deque and not broadcast_deque and not broadcast_arrays\n\n out = self.f(*args)\n\n # extract the carry from the args\n if self.input_carry_argnum == 'all':\n carry_arg = args[0]\n elif isinstance(self.input_carry_argnum, int):\n carry_arg = args[self.input_carry_argnum]\n else:\n assert self.input_carry_argnum is None\n carry_arg = None\n\n # extract the carry from the output\n if self.output_carry_argnum == 'all':\n carry_arg_out = out\n out = None\n elif isinstance(self.output_carry_argnum, int):\n assert isinstance(out, tuple)\n carry_arg_out = out[self.output_carry_argnum]\n _out = list(out)\n _out[self.output_carry_argnum] = None\n out = tuple(_out)\n else:\n assert self.output_carry_argnum is None\n carry_arg_out = None\n\n # TODO(cgarciae): allowing new references might lead to inconsistencies with\n # scan's looping semantics and we would also need to propagate the input\n _check_carry_same_references(carry_arg, carry_arg_out)\n\n args_out: tuple = extract.clear_non_graph_nodes(args)\n\n # replace the carry from the input args with the carry from the output\n if self.input_carry_argnum == 'all':\n args_out = (carry_arg_out,)\n elif isinstance(self.input_carry_argnum, int):\n _args_out = list(args_out)\n _args_out[self.input_carry_argnum] = carry_arg_out\n args_out = tuple(_args_out)\n else:\n assert self.input_carry_argnum is None\n assert carry_arg_out is None\n\n carry_deque_out = PytreeDeque[list[State | variablelib.VariableState]]()\n _broadcast_deque_out_tmp = PytreeDeque[\n list[State | variablelib.VariableState]\n ]() # discarded\n pure_args_out: tuple\n pure_args_out, pure_out = extract.to_tree(\n (args_out, out),\n prefix=(self.in_axes, self.out_axes),\n split_fn=functools.partial(\n _scan_split_out, carry_deque_out, _broadcast_deque_out_tmp\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if spmd.PARTITION_NAME in self.transform_metadata:\n pure_args_out, pure_out = _update_variable_sharding_metadata(\n (pure_args_out, pure_out),\n self.transform_metadata,\n spmd.add_axis,\n )\n\n # extract the pure carry from the pure args\n if self.input_carry_argnum == 'all':\n pure_carry_arg_out = pure_args_out[0]\n pure_args_out = ()\n elif isinstance(self.input_carry_argnum, int):\n pure_carry_arg_out = pure_args_out[self.input_carry_argnum]\n _pure_args_out = list(pure_args_out)\n _pure_args_out[self.input_carry_argnum] = None\n pure_args_out = tuple(_pure_args_out)\n else:\n assert self.input_carry_argnum is None\n pure_carry_arg_out = None\n\n # next we have to remove all the index_mappings from the GraphDefs\n # in the carry outputs because they are not present in the inputs\n carry_graphdefs: list[graph.GraphDef] = []\n pure_carry_arg_out = _extract_graphdefs(pure_carry_arg_out, carry_graphdefs)\n\n carry_arg_out = (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n )\n scan_out = (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n )\n return carry_arg_out, scan_out\n\n\n@tp.overload\ndef scan(\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> tp.Callable[[F], F]:\n ...\n\n\n@tp.overload\ndef scan(\n f: F,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F:\n ...\n\n\ndef scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n\n# -------------------------------\n# while_loop\n# -------------------------------\n\n\n@dataclasses.dataclass(eq=False)\nclass WhileLoopCondFn:\n f: tp.Callable[..., tp.Any]\n\n def __post_init__(self):\n functools.update_wrapper(self, self.f)\n\n def __call__(self, pure_val):\n val = extract.from_tree(pure_val)\n out = self.f(val)\n return out\n\n\ndef _add_fake_index_mapping(tree: tp.Any):\n def per_node_state(node_state: extract.NodeStates | tp.Any):\n if not isinstance(node_state, extract.NodeStates) or not isinstance(\n node_state._graphdef, graph.GraphDef\n ):\n return node_state\n\n return dataclasses.replace(\n node_state, _graphdef=node_state._graphdef.with_same_outer_index()\n )\n\n return jax.tree.map(per_node_state, tree,\n is_leaf=lambda x: isinstance(x, extract.NodeStates))\n\n\ndef _remove_index_mapping(tree: tp.Any):\n """"""Remove a fake outer_index for the input to match that of the output.""""""\n\n def per_node_state(node_state: extract.NodeStates | tp.Any):\n if not isinstance(node_state, extract.NodeStates) or not isinstance(\n node_state._graphdef, graph.GraphDef\n ):\n return node_state\n assert isinstance(node_state._graphdef, graph.GraphDef)\n node_state = dataclasses.replace(\n node_state, _graphdef=node_state._graphdef.with_no_outer_index()\n )\n return node_state\n\n return jax.tree.map(per_node_state, tree,\n is_leaf=lambda x: isinstance(x, extract.NodeStates))\n\n\n@dataclasses.dataclass(eq=False)\nclass WhileLoopBodyFn:\n f: tp.Callable[..., tp.Any]\n\n def __post_init__(self):\n functools.update_wrapper(self, self.f)\n\n @graph.update_context('while_loop_body')\n def __call__(self, pure_val):\n # Removing the dummy index mapping being added outside of body function.\n pure_val_in = _remove_index_mapping(pure_val)\n\n val = extract.from_tree(\n pure_val_in, ctxtag='while_loop_body', is_inner=True\n )\n out = self.f(val)\n pure_out = extract.to_tree(out, ctxtag='while_loop_body')\n\n try:\n jax.tree.map(lambda a, b: None, pure_val, pure_out)\n except ValueError as e:\n msg = (\n ""nnx.while_loop requires body function's input and output to ""\n 'have the same reference and pytree structure, but they differ. '\n 'If the mismatch comes from `outer_index` field, you might '\n 'have modified reference structure within the body function, '\n 'which is not allowed.'\n f'Detail of the mismatch: \n {str(e)}'\n )\n raise ValueError(msg)\n\n return pure_out\n\n\n@graph.update_context('while_loop')\ndef while_loop(cond_fun: tp.Callable[[T], tp.Any],\n body_fun: tp.Callable[[T], T],\n init_val: T) -> T:\n """"""A Flax NNX transformation of `jax.lax.while_loop `_.\n\n Caution: for the NNX internal reference tracing mechanism to work, you cannot\n change the variable reference structure of ``init_val`` inside ``body_fun``.\n\n Example::\n\n >>> import jax\n >>> from flax import nnx\n >>> def fwd_fn(input):\n ... module, x, count = input\n ... return module, module(x), count - 1.0\n\n >>> module = nnx.Linear(10, 10, rngs=nnx.Rngs(0))\n >>> x = jax.random.normal(jax.random.key(0), (10,))\n >>> # `module` will be called three times\n >>> _, y, _ = nnx.while_loop(\n ... lambda input: input[-1] > 0, fwd_fn, (module, x, 3.0))\n\n\n Args:\n cond_fun: A function for the continue condition of the while loop, taking a\n single input of type ``T`` and outputting a boolean.\n body_fun: A function that takes an input of type ``T`` and outputs an ``T``.\n Note that both data and modules of ``T`` must have the same reference\n structure between inputs and outputs.\n init_val: The initial input for ``cond_fun`` and ``body_fun``. Must be of type ``T``.\n\n """"""\n\n pure_init_val = extract.to_tree(init_val, ctxtag='while_loop')\n\n # Adding the expected reference mapping to `pure_init_val` to match\n # `body_fun`'s output pytree structure, to make JAX while_loop happy.\n pure_init_val = _add_fake_index_mapping(pure_init_val)\n\n pure_out = jax.lax.while_loop(\n WhileLoopCondFn(cond_fun),\n WhileLoopBodyFn(body_fun),\n pure_init_val,\n )\n out = extract.from_tree(pure_out, ctxtag='while_loop', is_inner=False)\n return out\n\n\n@dataclasses.dataclass(eq=False)\nclass ForiLoopBodyFn:\n f: tp.Callable[..., tp.Any]\n\n def __post_init__(self):\n functools.update_wrapper(self, self.f)\n\n @graph.update_context('fori_loop_body')\n def __call__(self, i, pure_val):\n # Removing the dummy index mapping being added outside of body function.\n pure_val_in = _remove_index_mapping(pure_val)\n\n val = extract.from_tree(pure_val_in, ctxtag='fori_loop_body', is_inner=True)\n out = self.f(i, val)\n pure_out = extract.to_tree(out, ctxtag='fori_loop_body')\n\n try:\n jax.tree.map(lambda a, b: None, pure_val, pure_out)\n except ValueError as e:\n msg = (\n ""nnx.fori_loop requires body function's input and output to ""\n 'have the same reference and pytree structure, but they differ. '\n 'If the mismatch comes from `outer_index` field, you might '\n 'have modified reference structure within the body function, '\n 'which is not allowed. '\n f'Detail of the mismatch: \n {str(e)}'\n )\n raise ValueError(msg)\n\n return pure_out\n\n\n@graph.update_context('fori_loop')\ndef fori_loop(lower: int, upper: int,\n body_fun: tp.Callable[[int, T], T],\n init_val: T,\n *,\n unroll: int | bool | None = None) -> T:\n """"""A Flax NNX transformation of `jax.lax.fori_loop `_.\n\n Caution: for the NNX internal reference tracing mechanism to work, you cannot\n change the variable reference structure of `init_val` inside `body_fun`.\n\n Example::\n\n >>> import jax\n >>> from flax import nnx\n\n >>> def fwd_fn(i, input):\n ... m, x = input\n ... m.kernel.value = jnp.identity(10) * i\n ... return m, m(x)\n\n >>> module = nnx.Linear(10, 10, rngs=nnx.Rngs(0))\n >>> x = jax.random.normal(jax.random.key(0), (10,))\n >>> _, y = nnx.fori_loop(2, 4, fwd_fn, (module, x))\n >>> np.testing.assert_array_equal(y, x * 2 * 3)\n\n\n Args:\n lower: An integer representing the loop index lower bound (inclusive).\n upper: An integer representing the loop index upper bound (exclusive).\n body_fun: a function that takes an input of type ``T`` and outputs an ``T``.\n Note that both data and modules of ``T`` must have the same reference\n structure between inputs and outputs.\n init_val: the initial input for body_fun. Must be of type ``T``.\n unroll: An optional integer or boolean that determines how much to unroll\n the loop. If an integer is provided, it determines how many unrolled\n loop iterations to run within a single rolled iteration of the loop. If a\n boolean is provided, it will determine if the loop is competely unrolled\n (i.e. ``unroll=True``) or left completely unrolled (i.e. ``unroll=False``).\n This argument is only applicable if the loop bounds are statically known.\n\n Returns:\n A loop value from the final iteration, of type ``T``.\n\n """"""\n\n pure_init_val = extract.to_tree(init_val, ctxtag='fori_loop')\n\n # Adding the expected reference mapping to `pure_init_val` to match\n # `body_fun`'s output pytree structure, to make JAX happy.\n pure_init_val = _add_fake_index_mapping(pure_init_val)\n\n pure_out = jax.lax.fori_loop(lower, upper,\n ForiLoopBodyFn(body_fun), pure_init_val,\n unroll=unroll)\n out = extract.from_tree(pure_out, ctxtag='fori_loop', is_inner=False)\n return out\n",python,tab +30,2336884,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37013,4,"scan",python,selection_command +31,2337158,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37016,0,"",python,selection_command +32,2337737,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,9,"def scan(",python,selection_command +33,2337967,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,707,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n",python,selection_command +34,2339394,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,736,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n",python,selection_command +35,2340496,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,874,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n",python,selection_command +36,2340923,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,1176,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n",python,selection_command +37,2341423,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,1308,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n",python,selection_command +38,2341800,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,1443,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n",python,selection_command +39,2342242,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,1635,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n",python,selection_command +40,2342855,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,2373,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n",python,selection_command +41,2343360,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,2451,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n",python,selection_command +42,2343651,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,2868,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n",python,selection_command +43,2343848,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,3071,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n",python,selection_command +44,2344174,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,3480,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n",python,selection_command +45,2344396,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,3847,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n",python,selection_command +46,2344765,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,4118,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n",python,selection_command +47,2344963,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,4433,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n",python,selection_command +48,2346590,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,4449,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n",python,selection_command +49,2347045,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,4488,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n",python,selection_command +50,2348077,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,4574,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n\n# -------------------------------\n# while_loop\n# -------------------------------\n",python,selection_command +51,2349562,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37009,4492,"def scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +52,2352493,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41501,0,"",python,selection_command +53,2353639,"genie.py",0,0,"",python,tab +54,2356365,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",0,0,"",python,tab +55,2357495,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41501,82,"\n# -------------------------------\n# while_loop\n# -------------------------------\n",python,selection_command +56,2357616,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41501,170,"\n# -------------------------------\n# while_loop\n# -------------------------------\n\n\n@dataclasses.dataclass(eq=False)\nclass WhileLoopCondFn:\n f: tp.Callable[..., tp.Any]\n",python,selection_command +57,2357781,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41501,241,"\n# -------------------------------\n# while_loop\n# -------------------------------\n\n\n@dataclasses.dataclass(eq=False)\nclass WhileLoopCondFn:\n f: tp.Callable[..., tp.Any]\n\n def __post_init__(self):\n functools.update_wrapper(self, self.f)\n",python,selection_command +58,2358432,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41501,170,"\n# -------------------------------\n# while_loop\n# -------------------------------\n\n\n@dataclasses.dataclass(eq=False)\nclass WhileLoopCondFn:\n f: tp.Callable[..., tp.Any]\n",python,selection_command +59,2358702,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41501,83,"\n# -------------------------------\n# while_loop\n# -------------------------------\n\n",python,selection_command +60,2358723,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41501,0,"",python,selection_command +61,2358749,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41458,43,"\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +62,2358782,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41442,59,"\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +63,2358816,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",41127,374,"\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +64,2359007,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",40856,645,"\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +65,2359261,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",40489,1012,"\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +66,2359290,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",40080,1421,"\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +67,2359317,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",39877,1624,"\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +68,2359348,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",39460,2041,"\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +69,2359484,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",39382,2119,"\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +70,2359806,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",38644,2857,"\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +71,2359923,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",38452,3049,"\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +72,2360126,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",38317,3184,"\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +73,2360280,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",38185,3316,"\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +74,2361702,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37883,3618,"\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +75,2361842,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37745,3756,"\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +76,2362368,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37716,3785,"\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +77,2362624,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",37008,4493,"\ndef scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +78,2362806,".venv/lib/python3.10/site-packages/flax/nnx/transforms/iteration.py",36640,4861,"\n@tp.overload\ndef scan(\n f: F,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F:\n ...\n\n\ndef scan(\n f: F | type[Missing] = Missing,\n *,\n length: int | None = None,\n reverse: bool = False,\n unroll: int | bool = 1,\n _split_transpose: bool = False,\n # extended api\n in_axes: int | None | type[Carry] | tuple[tp.Any, ...] = (Carry, 0),\n out_axes: tp.Any = (Carry, 0),\n # nnx specific\n transform_metadata: tp.Mapping[str, tp.Any] = FrozenDict({}),\n) -> F | tp.Callable[[F], F]:\n if f is Missing:\n return functools.partial(\n scan,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n in_axes=in_axes,\n out_axes=out_axes,\n transform_metadata=transform_metadata,\n ) # type: ignore[return-value]\n\n _check_out_axes(out_axes)\n\n input_carry_argnum = _get_carry_argnum(in_axes, is_in_axes=True)\n output_carry_argnum = _get_carry_argnum(out_axes, is_in_axes=False)\n\n if (input_carry_argnum is None and output_carry_argnum is not None) or (\n input_carry_argnum is not None and output_carry_argnum is None\n ):\n raise ValueError(\n 'If one of in_axes or out_axes has Carry, the other must also have Carry. '\n f'Got {in_axes=!r} and {out_axes=!r}'\n )\n\n scan_fn = ScanFn(\n f,\n input_carry_argnum,\n output_carry_argnum,\n in_axes,\n out_axes,\n transform_metadata,\n )\n\n @functools.wraps(f)\n @graph.update_context('scan')\n def scan_wrapper(*args, **kwargs):\n args = resolve_kwargs(f, args, kwargs)\n\n if in_axes is Carry and len(args) != 1:\n raise ValueError(\n f'When in_axes=Carry, the function must take exactly one argument, '\n f'got {len(args)} arguments.'\n )\n\n carry_deque = PytreeDeque()\n broadcast_deque = PytreeDeque()\n broadcast_arrays = PytreeDeque()\n pure_args: tuple = extract.to_tree(\n args,\n prefix=in_axes,\n split_fn=functools.partial(\n _scan_split_in, carry_deque, broadcast_deque, broadcast_arrays\n ),\n map_non_graph_nodes=True,\n ctxtag='scan',\n )\n if isinstance(input_carry_argnum, int):\n pure_carry_arg = pure_args[input_carry_argnum]\n _pure_args = list(pure_args)\n _pure_args[input_carry_argnum] = None\n pure_args = tuple(_pure_args)\n elif input_carry_argnum == 'all':\n pure_carry_arg = pure_args[0]\n pure_args = ()\n else:\n assert input_carry_argnum is None\n pure_carry_arg = None\n\n carry = (pure_carry_arg, carry_deque, broadcast_deque, broadcast_arrays)\n\n carry_out, scan_out = jax.lax.scan(\n scan_fn,\n carry,\n pure_args,\n length=length,\n reverse=reverse,\n unroll=unroll,\n _split_transpose=_split_transpose,\n )\n (\n pure_carry_arg_out,\n carry_deque_out,\n broadcast_deque_out,\n broadcast_arrays_out,\n ) = carry_out\n (\n carry_graphdefs,\n pure_args_out,\n pure_out,\n ) = scan_out\n\n # next we have to insert all the index_mappings back into the GraphDefs\n # in the carry outputs\n pure_carry_arg_out = _insert_graphdefs(\n pure_carry_arg_out, deque(carry_graphdefs)\n )\n\n # insert pure carry into pure_args_out\n if input_carry_argnum == 'all':\n pure_args_out = (pure_carry_arg_out,)\n elif isinstance(input_carry_argnum, int):\n _pure_args_out = list(pure_args_out)\n _pure_args_out[input_carry_argnum] = pure_carry_arg_out\n pure_args_out = tuple(_pure_args_out)\n else:\n assert input_carry_argnum is None\n assert pure_carry_arg_out is None\n\n args_out, out = extract.from_tree(\n (pure_args_out, pure_out),\n prefix=(in_axes, out_axes),\n merge_fn=functools.partial(\n _scan_merge_out, carry_deque_out, broadcast_deque_out\n ),\n is_leaf=lambda x: isinstance(x, (extract.NodeStates, Broadcasted)),\n map_non_graph_nodes=True,\n ctxtag='scan',\n is_inner=False,\n )\n\n # extract the carry from args_out\n if input_carry_argnum == 'all':\n carry_arg = args_out[0]\n elif isinstance(input_carry_argnum, int):\n carry_arg = args_out[input_carry_argnum]\n else:\n assert input_carry_argnum is None\n carry_arg = None\n\n # insert carry into the output\n if output_carry_argnum == 'all':\n out = carry_arg\n elif isinstance(output_carry_argnum, int):\n _out = list(out)\n _out[output_carry_argnum] = carry_arg\n out = tuple(_out)\n else:\n assert output_carry_argnum is None\n assert carry_arg is None\n\n return out\n\n return scan_wrapper # type: ignore\n\n\n\n\n",python,selection_command +79,2366288,"genie.py",0,0,"",python,tab +80,2366995,"genie.py",17059,2,"jax.la",python,content +81,2367007,"genie.py",17059,0,"",python,selection_command +82,2369010,"genie.py",17023,49," final_carry_causal, _ = jax.lax.scan(",python,selection_command +83,2369146,"genie.py",17023,114," final_carry_causal, _ = jax.lax.scan(\n causal_step_fn, init_carry_causal, jnp.arange(N)",python,selection_command +84,2369280,"genie.py",17023,128," final_carry_causal, _ = jax.lax.scan(\n causal_step_fn, init_carry_causal, jnp.arange(N)\n )",python,selection_command +85,2829775,"genie.py",17207,0,"",python,selection_mouse +86,2829781,"genie.py",17206,0,"",python,selection_command +87,2831191,"genie.py",17059,0,"",python,selection_mouse +88,2831574,"genie.py",17059,1,"j",python,selection_command +89,2831619,"genie.py",17059,3,"jax",python,selection_command +90,2831806,"genie.py",17059,4,"jax.",python,selection_command +91,2832004,"genie.py",17059,7,"jax.lax",python,selection_command +92,2832708,"genie.py",17059,7,"",python,content +93,2832911,"genie.py",17059,0,"n",python,content +94,2832912,"genie.py",17060,0,"",python,selection_keyboard +95,2833068,"genie.py",17060,0,"n",python,content +96,2833068,"genie.py",17061,0,"",python,selection_keyboard +97,2833174,"genie.py",17061,0,"x",python,content +98,2833175,"genie.py",17062,0,"",python,selection_keyboard +99,2833804,"genie.py",17061,0,"",python,selection_command +100,2850953,"genie.py",17062,0,"",python,selection_command +101,2851327,"genie.py",17066,0,"",python,selection_command +102,2851708,"genie.py",17067,0,"",python,selection_command +103,2857745,"genie.py",17068,0,"",python,selection_command +104,2858265,"genie.py",17067,0,"",python,selection_command +105,2859026,"genie.py",17068,0,"",python,selection_command +106,2859162,"genie.py",17068,0,")",python,content +107,2859162,"genie.py",17069,0,"",python,selection_keyboard +108,2859645,"genie.py",17069,0,"()",python,content +109,2859646,"genie.py",17070,0,"",python,selection_keyboard +110,2860016,"genie.py",17069,0,"",python,selection_command +111,2860632,"genie.py",17070,0,"",python,selection_command +112,2860733,"genie.py",17070,1,"",python,content +113,2860734,"genie.py",17069,0,"",python,selection_command +114,2860958,"genie.py",17118,0,"",python,selection_command +115,2861206,"genie.py",17103,0,"",python,selection_command +116,2861464,"genie.py",17101,0,"",python,selection_command +117,2861496,"genie.py",17087,0,"",python,selection_command +118,2861920,"genie.py",17087,14,"",python,content +119,2862255,"genie.py",17039,0,"",python,selection_command +120,2865095,"genie.py",17053,0,"",python,selection_command +121,2865343,"genie.py",17055,0,"",python,selection_command +122,2865361,"genie.py",17057,0,"",python,selection_command +123,2865394,"genie.py",17059,0,"",python,selection_command +124,2865428,"genie.py",17062,0,"",python,selection_command +125,2865461,"genie.py",17063,0,"",python,selection_command +126,2865790,"genie.py",17067,0,"",python,selection_command +127,2866308,"genie.py",17068,0,"causal_step_fn",python,content +128,2866308,"genie.py",17081,0,"",python,selection_command +129,2866826,"genie.py",17082,0,"",python,selection_command +130,2866923,"genie.py",17082,0,",",python,content +131,2866924,"genie.py",17083,0,"",python,selection_keyboard +132,2867071,"genie.py",17083,0," ",python,content +133,2867072,"genie.py",17084,0,"",python,selection_keyboard +134,2867450,"genie.py",17084,0,"N",python,content +135,2867451,"genie.py",17085,0,"",python,selection_keyboard +136,2868214,"genie.py",17084,1,"",python,content +137,2868414,"genie.py",17084,0,"l",python,content +138,2868414,"genie.py",17085,0,"",python,selection_keyboard +139,2868508,"genie.py",17085,0,"e",python,content +140,2868508,"genie.py",17086,0,"",python,selection_keyboard +141,2868707,"genie.py",17086,0,"g",python,content +142,2868707,"genie.py",17087,0,"",python,selection_keyboard +143,2868724,"genie.py",17087,0,"n",python,content +144,2868724,"genie.py",17088,0,"",python,selection_keyboard +145,2869093,"genie.py",17087,1,"",python,content +146,2869229,"genie.py",17086,1,"",python,content +147,2869374,"genie.py",17086,0,"n",python,content +148,2869374,"genie.py",17087,0,"",python,selection_keyboard +149,2869471,"genie.py",17087,0,"g",python,content +150,2869471,"genie.py",17088,0,"",python,selection_keyboard +151,2869672,"genie.py",17088,0,"h",python,content +152,2869672,"genie.py",17089,0,"",python,selection_keyboard +153,2869735,"genie.py",17089,0,"t",python,content +154,2869735,"genie.py",17090,0,"",python,selection_keyboard +155,2869955,"genie.py",17090,0,"=",python,content +156,2869955,"genie.py",17091,0,"",python,selection_keyboard +157,2870996,"genie.py",17091,0,"N",python,content +158,2870997,"genie.py",17092,0,"",python,selection_keyboard +159,2871258,"genie.py",17091,0,"",python,selection_command +160,2871569,"genie.py",17090,0,"",python,selection_command +161,2871771,"genie.py",17089,0,"",python,selection_command +162,2872148,"genie.py",17089,1,"h",python,content +163,2872289,"genie.py",17088,0,"",python,selection_command +164,2873810,"genie.py",17088,1,"t",python,content +165,2875394,"genie.py",17144,0,"",python,selection_command +166,2876182,"genie.py",17111,0,"",python,selection_command +167,2877028,"genie.py",17111,1,"",python,content +168,2877190,"genie.py",17111,1,"",python,content +169,2879045,"genie.py",17095,0,"",python,selection_command +170,2885534,"genie.py",17023,0,"",python,selection_command +171,2885775,"genie.py",16989,0,"",python,selection_command +172,2885803,"genie.py",16932,0,"",python,selection_command +173,2885835,"genie.py",16931,0,"",python,selection_command +174,2885868,"genie.py",16917,0,"",python,selection_command +175,2885902,"genie.py",16865,0,"",python,selection_command +176,2885936,"genie.py",16831,0,"",python,selection_command +177,2885968,"genie.py",16791,0,"",python,selection_command +178,2886010,"genie.py",16765,0,"",python,selection_command +179,2886035,"genie.py",16731,0,"",python,selection_command +180,2886070,"genie.py",16678,0,"",python,selection_command +181,2886103,"genie.py",16677,0,"",python,selection_command +182,2886133,"genie.py",16651,0,"",python,selection_command +183,2886168,"genie.py",16543,0,"",python,selection_command +184,2886286,"genie.py",16651,0,"",python,selection_command +185,2886558,"genie.py",16677,0,"",python,selection_command +186,2886577,"genie.py",16678,0,"",python,selection_command +187,2886607,"genie.py",16731,0,"",python,selection_command +188,2886628,"genie.py",16678,0,"",python,selection_command +189,2886893,"genie.py",16677,0,"",python,selection_command +190,2888341,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +191,2889083,"genie.py",0,0,"",python,tab +192,2891868,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +193,2894619,"TERMINAL",0,0,"",,terminal_focus +194,2894620,"genie.py",0,0,"",python,tab +195,2895410,"TERMINAL",0,0,"source /home/franz.srambical/jafar/.venv/bin/activate",,terminal_command +196,2895410,"TERMINAL",0,0,"]633;C]0;franz.srambical@hai-login2:~/jafar",,terminal_output +197,2902571,"TERMINAL",0,0,"squeue --me",,terminal_command +198,2902590,"TERMINAL",0,0,"]633;C JOBID USER PARTITION NODES CPUS ST SUBMIT_TIME START_TIME TIME TIME_LIMIT NODELIST(REASON)\r\n 14857 franz.sram interacti 1 1 R 2025-08-02T09:39:03 2025-08-02T09:39:03 5:46:15 10:00:00 hai008\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +199,2907190,"TERMINAL",0,0,"scancel --me",,terminal_command +200,2907206,"TERMINAL",0,0,"]633;C]0;franz.srambical@hai-login2:~/jafar",,terminal_output +201,2908809,"TERMINAL",0,0,"salloc --gpus=1 --ntasks-per-node=1 --cpus-per-task=1--mem=100G --time=10:00:00",,terminal_command +202,2908861,"TERMINAL",0,0,"]633;Csalloc: Granted job allocation 14867\r\n",,terminal_output +203,2908965,"TERMINAL",0,0,"salloc: Nodes hai001 are ready for job\r\n",,terminal_output +204,2909335,"TERMINAL",0,0,"Running inside SLURM, Job ID 14867.\r\n",,terminal_output +205,2909417,"TERMINAL",0,0,"]0;franz.srambical@hai-login2:~/jafar[?2004h[franz.srambical@hai001.haicore.berlin:~/jafar] $ ",,terminal_output +206,2911099,"TERMINAL",0,0,"\r(reverse-i-search)`': ",,terminal_output +207,2911333,"TERMINAL",0,0,"b': source /home/franz.srambical/jafar/.venv/bin/activate",,terminal_output +208,2911403,"TERMINAL",0,0,"\ra': bash experiments/sample.sh \r",,terminal_output +209,2911467,"TERMINAL",0,0,"[1@s': bas",,terminal_output +210,2911518,"TERMINAL",0,0,"[1@h': bash",,terminal_output +211,2911968,"TERMINAL",0,0,"\r[24@[franz.srambical@hai001.haicore.berlin:~/jafar] $ bash",,terminal_output +212,2912057,"TERMINAL",0,0,"\r\n[?2004l\r",,terminal_output +213,2924964,"TERMINAL",0,0,"/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/orbax/checkpoint/_src/serialization/type_handlers.py:1256: UserWarning: Sharding info not provided when restoring. Populating sharding info from sharding file. Please note restoration time will be slightly increased due to reading from file. Note also that this option is unsafe when restoring on a different topology than the checkpoint was saved with.\r\n warnings.warn(\r\n",,terminal_output +214,2933485,"TERMINAL",0,0,"2025-08-02 15:25:48.975036: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +215,2937905,"TERMINAL",0,0,"2025-08-02 15:25:53.404109: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +216,2946099,"TERMINAL",0,0,"2025-08-02 15:26:01.593909: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +217,2957884,"TERMINAL",0,0,"---------------------------------------------------------------------------\r\nTraceContextError Traceback (most recent call last)\r\nFile /fast/home/franz.srambical/jafar/sample.py:216\r\n 207 action_batch_E = genie.vq_encode(batch, training=False)\r\n 209 # --- Sample + evaluate video ---\r\n 210 # B, S, H, W, _ = video_batch_BSHWC.shape\r\n 211 # N = math.ceil(H / args.patch_size) * math.ceil(W / args.patch_size)\r\n 212 # for block in genie.dynamics.transformer.blocks:\r\n 213 # block.spatial_attention.init_cache((B * S, 1, args.dyna_dim), dtype=args.dtype)\r\n 214 # block.temporal_attention.init_cache((B * (N + 1), 1, args.dyna_dim), dtype=args.dtype)\r\n--> 216 recon_video_BSHWC = _autoreg_sample(genie, rng, video_batch_BSHWC, action_batch_E)\r\n 217 recon_video_BSHWC = recon_video_BSHWC.astype(jnp.float32)\r\n 218 gt = gt_video[:, : recon_video_BSHWC.shape[1]].clip(0, 1).reshape(-1, *gt_video.shape[2:])\r\n\r\nFile /fast/home/franz.srambical/jafar/sample.py:180, in _autoreg_sample(genie, rng, video_batch_BSHWC, action_batch_E)\r\n 178 rng, _rng = jax.random.split(rng)\r\n 179 batch = dict(videos=input_video_BTHWC, latent_actions=action_batch_E, rng=_rng)\r\n--> 180 generated_vid_BSHWC = _sampling_fn(genie, batch)\r\n 181 return generated_vid_BSHWC\r\n\r\nFile /fast/home/franz.srambical/jafar/sample.py:164, in _sampling_fn(model, batch)\r\n 156 return model.sample(\r\n 157 batch,\r\n 158 args.seq_len,\r\n (...)\r\n 161 args.sample_argmax,\r\n 162 )\r\n 163 elif args.dyna_type == ""causal"":\r\n--> 164 return model.sample_causal(\r\n 165  batch,\r\n 166  args.seq_len,\r\n 167  args.temperature,\r\n 168  args.sample_argmax,\r\n 169  )\r\n 170 else:\r\n 171 raise ValueError(f""Invalid dynamics type: {args.dyna_type}"")\r\n\r\nFile /fast/home/franz.srambical/jafar/genie.py:397, in Genie.sample_causal(self, batch, seq_len, temperature, sample_argmax)\r\n 395 # --- Reset spatial KV caches before each frame ---\r\n 396 for block in dynamics_causal.transformer.blocks:\r\n--> 397 block.spatial_attention.init_cache((B * seq_len, (N + 1), self.dyna_dim), dtype=self.dtype)\r\n 398 #breakpoint()\r\n 399 \r\n 400 # --- Initialize and run causal loop ---\r\n 401 init_carry_causal = (\r\n 402 step_rng,\r\n 403 current_token_idxs_BSN,\r\n 404 action_tokens_EL,\r\n 405 jnp.array(step_t, dtype=jnp.int32),\r\n 406 )\r\n\r\nFile /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/nn/attention.py:652, in MultiHeadAttention.init_cache(self, input_shape, dtype)\r\n 623 """"""Initializes cache for fast autoregressive decoding. When\r\n 624 ``decode=True``, this method must be called first before performing\r\n 625 forward inference. When in decode mode, only one token must be passed\r\n (...)\r\n 649  >>> out_nnx = model_nnx(x)\r\n 650 """"""\r\n 651 cache_shape = (*input_shape[:-1], self.num_heads, self.head_dim)\r\n--> 652 self.cached_key = nnx.Cache(jnp.zeros(cache_shape, dtype))\r\n 653 self.cached_value = nnx.Cache(jnp.zeros(cache_shape, dtype))\r\n 654 self.cache_index = nnx.Cache(jnp.array(0, dtype=jnp.int32))\r\n\r\nFile /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/object.py:280, in Object.__setattr__(self, name, value)\r\n 279 def __setattr__(self, name: str, value: Any) -> None:\r\n--> 280 self._setattr(name, value)\r\n\r\nFile /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/object.py:283, in Object._setattr(self, name, value)\r\n 282 def _setattr(self, name: str, value: tp.Any) -> None:\r\n--> 283 self._check_valid_context(\r\n 284  lambda: f""Cannot mutate '{type(self).__name__}' from different trace level""\r\n 285  )\r\n 286 if (\r\n 287 type(self)._object__is_pytree\r\n 288 and isinstance(self._object__nodes, frozenset)\r\n 289 and name not in self._object__nodes\r\n 290 ):\r\n 291 for leaf in jax.tree.leaves(value):\r\n\r\nFile /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/object.py:302, in Object._check_valid_context(self, error_msg)\r\n 300 def _check_valid_context(self, error_msg: tp.Callable[[], str]) -> None:\r\n 301 if not self._object__state.trace_state.is_valid():\r\n--> 302 raise errors.TraceContextError(error_msg())\r\n\r\nTraceContextError: Cannot mutate 'MultiHeadAttention' from different trace level (https://flax.readthedocs.io/en/latest/api_reference/flax.errors.html#flax.errors.TraceContextError)\r\n> /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/object.py(302)_check_valid_context()\r\n 300  def _check_valid_context(self, error_msg: tp.Callable[[], str]) -> None:\r\n 301  if not self._object__state.trace_state.is_valid():\r\n--> 302  raise errors.TraceContextError(error_msg())\r\n 303 \r\n 304  def __deepcopy__(self: O, memo=None) -> O:\r\n\r\n",,terminal_output +218,3313179,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +219,3315299,"genie.py",0,0,"",python,tab +220,3317839,"genie.py",16678,0,"",python,selection_command +221,3318087,"genie.py",16731,0,"",python,selection_command +222,3318102,"genie.py",16765,0,"",python,selection_command +223,3318133,"genie.py",16791,0,"",python,selection_command +224,3318168,"genie.py",16831,0,"",python,selection_command +225,3318201,"genie.py",16865,0,"",python,selection_command +226,3318235,"genie.py",16917,0,"",python,selection_command +227,3318270,"genie.py",16931,0,"",python,selection_command +228,3318448,"genie.py",16932,0,"",python,selection_command +229,3318639,"genie.py",16989,0,"",python,selection_command +230,3318800,"genie.py",17023,0,"",python,selection_command +231,3323355,"genie.py",16334,0,"",python,selection_command +232,3324845,"genie.py",17035,0,"",python,selection_command +233,3326629,"genie.py",17023,71," final_carry_causal, _ = nnx.scan(causal_step_fn, length=N)(",python,selection_command +234,3326921,"genie.py",17023,120," final_carry_causal, _ = nnx.scan(causal_step_fn, length=N)(\n init_carry_causal, jnp.arange(N)",python,selection_command +235,3327293,"genie.py",17023,134," final_carry_causal, _ = nnx.scan(causal_step_fn, length=N)(\n init_carry_causal, jnp.arange(N)\n )",python,selection_command +236,3365792,"genie.py",17023,134," carry_causal = init_carry_causal\n for step_n in range(N):\n carry_causal, _ = causal_step_fn(\n carry_causal, jnp.array(step_n, dtype=jnp.int32)\n )\n final_carry_causal = carry_causal",python,content +237,3409525,"genie.py",17222,0,"",python,selection_mouse +238,3409536,"genie.py",17221,0,"",python,selection_command +239,3417630,"TERMINAL",0,0,"q",,terminal_output +240,3417759,"TERMINAL",0,0,"ui",,terminal_output +241,3417890,"TERMINAL",0,0,"t",,terminal_output +242,3418167,"TERMINAL",0,0,"()",,terminal_output +243,3418372,"TERMINAL",0,0,"\r\n",,terminal_output +244,3419369,"TERMINAL",0,0,"ipdb> ",,terminal_output +245,3419619,"TERMINAL",0,0,"srun: error: hai001: task 0: Exited with exit code 1\r\n]0;franz.srambical@hai-login2:~/jafar[?2004h[franz.srambical@hai001.haicore.berlin:~/jafar] $ ",,terminal_output +246,3419839,"TERMINAL",0,0,"[franz.srambical@hai001.haicore.berlin:~/jafar] $ ",,terminal_output +247,3419917,"TERMINAL",0,0,"bash experiments/sample.sh ",,terminal_output +248,3420254,"TERMINAL",0,0,"\r\n[?2004l\r",,terminal_output +249,3431991,"TERMINAL",0,0,"/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/orbax/checkpoint/_src/serialization/type_handlers.py:1256: UserWarning: Sharding info not provided when restoring. Populating sharding info from sharding file. Please note restoration time will be slightly increased due to reading from file. Note also that this option is unsafe when restoring on a different topology than the checkpoint was saved with.\r\n warnings.warn(\r\n",,terminal_output +250,3440414,"TERMINAL",0,0,"2025-08-02 15:34:15.909957: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +251,3444851,"TERMINAL",0,0,"2025-08-02 15:34:20.345721: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +252,3452985,"TERMINAL",0,0,"2025-08-02 15:34:28.483286: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +253,3761703,"TERMINAL",0,0,"2025-08-02 15:39:37.192132: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +254,3774083,"TERMINAL",0,0,"SSIM: 0.004261609632521868\r\n",,terminal_output +255,3775887,"TERMINAL",0,0,"]0;franz.srambical@hai-login2:~/jafar[?2004h[franz.srambical@hai001.haicore.berlin:~/jafar] $ ",,terminal_output +256,3817555,"TERMINAL",0,0,"[franz.srambical@hai001.haicore.berlin:~/jafar] $ ",,terminal_output +257,3817626,"TERMINAL",0,0,"bash experiments/sample.sh ",,terminal_output +258,3817899,"TERMINAL",0,0,"\r\n[?2004l\r",,terminal_output +259,3829644,"TERMINAL",0,0,"/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/orbax/checkpoint/_src/serialization/type_handlers.py:1256: UserWarning: Sharding info not provided when restoring. Populating sharding info from sharding file. Please note restoration time will be slightly increased due to reading from file. Note also that this option is unsafe when restoring on a different topology than the checkpoint was saved with.\r\n warnings.warn(\r\n",,terminal_output +260,3855520,"genie.py",17152,0,"",python,selection_command +261,3855717,"genie.py",17138,0,"",python,selection_command +262,3856183,"genie.py",14497,0,"",python,selection_command +263,3878064,"TERMINAL",0,0,"2025-08-02 15:41:33.552845: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +264,3882543,"TERMINAL",0,0,"2025-08-02 15:41:37.999120: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +265,3890611,"TERMINAL",0,0,"2025-08-02 15:41:46.105031: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +266,3929226,"genie.py",13497,0,"",python,selection_command +267,3934375,"genie.py",14493,0,"",python,selection_command +268,3934931,"genie.py",15594,0,"",python,selection_command +269,3937140,"genie.py",14493,0,"",python,selection_command +270,3938639,"genie.py",15594,0,"",python,selection_command +271,3938907,"genie.py",16334,0,"",python,selection_command +272,3950947,"genie.py",15594,0,"",python,selection_command +273,3951159,"genie.py",14493,0,"",python,selection_command +274,3951470,"genie.py",13497,0,"",python,selection_command +275,3953627,"genie.py",13489,71,"",python,content +276,3953654,"genie.py",13497,0,"",python,selection_command +277,3954581,"genie.py",15523,0,"",python,selection_command +278,3963561,"experiments/sample.sh",0,0,"source .venv/bin/activate\n\ndata_dir=""$PWD/data_arrayrecord/dummy""\nckpt_dir=""$PWD/checkpoints/causal_dynamics_openai_grain_tok_restore""\n\nexport PYTHONUNBUFFERED=1\nsrun ipython --pdb sample.py -- \\n --dyna_type ""causal"" \\n --batch_size 1 \\n --seq_len 3 \\n --start_frame 1 \\n --checkpoint $ckpt_dir \\n --data_dir $data_dir",shellscript,tab +279,3970600,"genie.py",0,0,"",python,tab +280,4195274,"TERMINAL",0,0,"2025-08-02 15:46:50.767433: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +281,4207619,"TERMINAL",0,0,"SSIM: 0.004261609632521868\r\n",,terminal_output +282,4209414,"TERMINAL",0,0,"]0;franz.srambical@hai-login2:~/jafar[?2004h[franz.srambical@hai001.haicore.berlin:~/jafar] $ ",,terminal_output +283,4423300,"genie.py",14422,0,"",python,selection_command +284,4423731,"genie.py",13485,0,"",python,selection_command +285,4424061,"genie.py",14422,0,"",python,selection_command +286,4424162,"genie.py",13485,0,"",python,selection_command +287,4424546,"genie.py",14422,0,"",python,selection_command +288,4425214,"genie.py",14413,0,"\n",python,content +289,4425242,"genie.py",14414,0," ",python,content +290,4426241,"genie.py",14422,0,"@",python,content +291,4426241,"genie.py",14423,0,"",python,selection_keyboard +292,4426749,"genie.py",14423,0,"n",python,content +293,4426750,"genie.py",14424,0,"",python,selection_keyboard +294,4426880,"genie.py",14424,0,"n",python,content +295,4426880,"genie.py",14425,0,"",python,selection_keyboard +296,4426946,"genie.py",14425,0,"x",python,content +297,4426946,"genie.py",14426,0,"",python,selection_keyboard +298,4427093,"genie.py",14426,0,".",python,content +299,4427094,"genie.py",14427,0,"",python,selection_keyboard +300,4427306,"genie.py",14427,0,"j",python,content +301,4427307,"genie.py",14428,0,"",python,selection_keyboard +302,4427351,"genie.py",14428,0,"i",python,content +303,4427351,"genie.py",14429,0,"",python,selection_keyboard +304,4427428,"genie.py",14429,0,"t",python,content +305,4427428,"genie.py",14430,0,"",python,selection_keyboard +306,4428139,"genie.py",14429,0,"",python,selection_command +307,4430346,"TERMINAL",0,0,"bash experiments/sample.sh ",,terminal_output +308,4431303,"TERMINAL",0,0,"\r\n[?2004l\r",,terminal_output +309,4442741,"TERMINAL",0,0,"/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/orbax/checkpoint/_src/serialization/type_handlers.py:1256: UserWarning: Sharding info not provided when restoring. Populating sharding info from sharding file. Please note restoration time will be slightly increased due to reading from file. Note also that this option is unsafe when restoring on a different topology than the checkpoint was saved with.\r\n warnings.warn(\r\n",,terminal_output +310,4451463,"TERMINAL",0,0,"2025-08-02 15:51:06.955065: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +311,4455903,"TERMINAL",0,0,"2025-08-02 15:51:11.393441: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +312,4464038,"TERMINAL",0,0,"2025-08-02 15:51:19.536637: W external/xla/xla/service/gpu/autotuning/dot_search_space.cc:200] All configs were filtered out because none of them sufficiently match the hints. Maybe the hints set does not contain a good representative set of valid configs?Working around this by using the full hints set instead.\r\n",,terminal_output +313,4477408,"TERMINAL",0,0,"---------------------------------------------------------------------------\r\nUnexpectedTracerError Traceback (most recent call last)\r\nFile /fast/home/franz.srambical/jafar/sample.py:216\r\n 207 action_batch_E = genie.vq_encode(batch, training=False)\r\n 209 # --- Sample + evaluate video ---\r\n 210 # B, S, H, W, _ = video_batch_BSHWC.shape\r\n 211 # N = math.ceil(H / args.patch_size) * math.ceil(W / args.patch_size)\r\n 212 # for block in genie.dynamics.transformer.blocks:\r\n 213 # block.spatial_attention.init_cache((B * S, 1, args.dyna_dim), dtype=args.dtype)\r\n 214 # block.temporal_attention.init_cache((B * (N + 1), 1, args.dyna_dim), dtype=args.dtype)\r\n--> 216 recon_video_BSHWC = _autoreg_sample(genie, rng, video_batch_BSHWC, action_batch_E)\r\n 217 recon_video_BSHWC = recon_video_BSHWC.astype(jnp.float32)\r\n 218 gt = gt_video[:, : recon_video_BSHWC.shape[1]].clip(0, 1).reshape(-1, *gt_video.shape[2:])\r\n\r\nFile /fast/home/franz.srambical/jafar/sample.py:180, in _autoreg_sample(genie, rng, video_batch_BSHWC, action_batch_E)\r\n 178 rng, _rng = jax.random.split(rng)\r\n 179 batch = dict(videos=input_video_BTHWC, latent_actions=action_batch_E, rng=_rng)\r\n--> 180 generated_vid_BSHWC = _sampling_fn(genie, batch)\r\n 181 return generated_vid_BSHWC\r\n\r\nFile /fast/home/franz.srambical/jafar/sample.py:164, in _sampling_fn(model, batch)\r\n 156 return model.sample(\r\n 157 batch,\r\n 158 args.seq_len,\r\n (...)\r\n 161 args.sample_argmax,\r\n 162 )\r\n 163 elif args.dyna_type == ""causal"":\r\n--> 164 return model.sample_causal(\r\n 165  batch,\r\n 166  args.seq_len,\r\n 167  args.temperature,\r\n 168  args.sample_argmax,\r\n 169  )\r\n 170 else:\r\n 171 raise ValueError(f""Invalid dynamics type: {args.dyna_type}"")\r\n\r\nFile /fast/home/franz.srambical/jafar/genie.py:412, in Genie.sample_causal(self, batch, seq_len, temperature, sample_argmax)\r\n 410 carry_causal = init_carry_causal\r\n 411 for step_n in range(N):\r\n--> 412 carry_causal, _ = causal_step_fn(\r\n 413  carry_causal, jnp.array(step_n, dtype=jnp.int32)\r\n 414  )\r\n 415 final_carry_causal = carry_causal\r\n 416 # final_carry_causal[1].block_until_ready()\r\n 417 # elapsed = time.time() - start\r\n 418 # print(f""Autoregressive generation time: {elapsed:.4f}s"")\r\n 419 # breakpoint()\r\n\r\nFile /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/transforms/compilation.py:431, in JitWrapped.__call__(self, *args, **kwargs)\r\n 429 with graph.update_context(self):\r\n 430 pure_args, pure_kwargs = self._get_pure_args_kwargs(args, kwargs)\r\n--> 431 pure_args_out, pure_kwargs_out, pure_out = self.jitted_fn(\r\n 432  *pure_args, **pure_kwargs\r\n 433  )\r\n 434 out = self._get_non_pure_out(pure_args_out, pure_kwargs_out, pure_out)\r\n 435 return out\r\n\r\n [... skipping hidden 3 frame]\r\n\r\nFile /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/jax/_src/core.py:1053, in check_eval_args(args)\r\n 1051 for arg in args:\r\n 1052 if isinstance(arg, Tracer):\r\n-> 1053 raise escaped_tracer_error(arg)\r\n\r\nUnexpectedTracerError: Encountered an unexpected tracer. A function transformed by JAX had a side effect, allowing for a reference to an intermediate value with type float32[2048] wrapped in a DynamicJaxprTracer to escape the scope of the transformation.\r\nJAX transformations require that functions explicitly return their outputs, and disallow saving intermediate values to global state.\r\nThe function being traced when the value leaked was causal_step_fn at /fast/home/franz.srambical/jafar/genie.py:355 traced for jit.\r\n------------------------------\r\nThe leaked intermediate value was created on line /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/transforms/general.py:153 (split_inputs_wrapper). \r\n------------------------------\r\nWhen the value was created, the final 5 stack frames (most recent last) excluding JAX-internal frames were:\r\n------------------------------\r\n/fast/home/franz.srambical/jafar/utils/nn.py:443 (__call__)\r\n/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/transforms/transforms.py:73 (resolve_kwargs_wrapper)\r\n/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/graph.py:2051 (update_context_manager_wrapper)\r\n/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/graph.py:2051 (update_context_manager_wrapper)\r\n/fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/flax/nnx/transforms/general.py:153 (split_inputs_wrapper)\r\n------------------------------\r\n\r\nTo catch the leak earlier, try setting the environment variable JAX_CHECK_TRACER_LEAKS or using the `jax.checking_leaks` context manager.\r\nSee https://docs.jax.dev/en/latest/errors.html#jax.errors.UnexpectedTracerError\r\n> /fast/home/franz.srambical/jafar/.venv/lib/python3.10/site-packages/jax/_src/core.py(1053)check_eval_args()\r\n 1051  for arg in args:\r\n 1052  if isinstance(arg, Tracer):\r\n-> 1053  raise escaped_tracer_error(arg)\r\n 1054 \r\n 1055 class EvalTrace(Trace):\r\n\r\n",,terminal_output diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-14f0662f-0032-43e8-be9f-8e53d6f150ad1758635869882-2025_09_23-15.57.52.616/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-14f0662f-0032-43e8-be9f-8e53d6f150ad1758635869882-2025_09_23-15.57.52.616/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..54044f6bc2bd995af4e674c869e2579e0b4ebfcf --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-14f0662f-0032-43e8-be9f-8e53d6f150ad1758635869882-2025_09_23-15.57.52.616/source.csv @@ -0,0 +1,79 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,1,"MaxText/maxtext_utils.py",0,0,"# Copyright 2023–2025 Google LLC\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# https://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# pylint: disable=line-too-long, disable=bare-except, consider-using-generator\n"""""" Utils that are only interesting to MaxText. """"""\n\nfrom typing import Optional\nimport functools\nimport pickle\n\nfrom flax import linen as nn\nfrom flax.linen import partitioning as nn_partitioning\nfrom flax.training import train_state\n\nimport numpy as np\n\nfrom collections.abc import Iterable\nfrom jax.experimental import mesh_utils\nfrom jax.experimental.serialize_executable import deserialize_and_load\nfrom jax.sharding import PartitionSpec as P\n\nimport jax\nimport jax.numpy as jnp\nimport jax.tree_util as jtu\n\nimport optax\n\nimport orbax.checkpoint.experimental.emergency.checkpoint_manager as emergency_checkpoint_manager\nimport orbax.checkpoint.experimental.emergency.replicator_checkpoint_manager as emergency_replicator_checkpoint_manager\n\nfrom MaxText import checkpointing\nfrom MaxText import max_logging\nfrom MaxText import max_utils\nfrom MaxText.common_types import DecoderBlockType, MODEL_MODE_PREFILL, MODEL_MODE_AUTOREGRESSIVE\nfrom MaxText.inference.page_manager import PageState\n\nOVERWRITE_WITH_GRADIENT = ""_overwrite_with_gradient""\n\n# Multimodal constants\nNUM_IMAGES_PER_SEQUENCE = 1\nNUM_IMAGE_CHANNELS = 3\nNUM_TILES_PER_IMAGE = 1 # Fake number of tiles for llama4, init purpose\n\n\ndef get_input_data_sharding(config, mesh):\n """"""Get the input data sharding for the model""""""\n return nn.logical_to_mesh_sharding(P(*config.input_data_sharding_logical_axes), mesh, config.logical_axis_rules)\n\n\ndef get_functional_train_with_signature(train_step, data_sharding, state_mesh_shardings, model, config):\n """"""Get the shardings (both state and data) for `train_step`.""""""\n functional_train = functools.partial(train_step, model, config, state_mesh_shardings)\n functional_train.__name__ = ""train_step""\n in_shardings = (state_mesh_shardings, data_sharding, None) # State, batch, rng\n out_shardings = (state_mesh_shardings, None) # State, metrics\n static_argnums = () # We partial out the static argnums of model and config\n donate_argnums = 0 # This is the index of the state - we allow the compiler to make use of this memory.\n return functional_train, in_shardings, out_shardings, static_argnums, donate_argnums\n\n\ndef get_functional_eval_with_signature(eval_step, data_sharding, state_mesh_shardings, model, config):\n """"""Get the shardings (both state and data) for `eval_step`.""""""\n functional_eval = functools.partial(eval_step, model, config)\n functional_eval.__name__ = ""eval_step""\n in_shardings = (state_mesh_shardings, data_sharding, None) # State, batch, rng\n out_shardings = None # metrics\n static_argnums = () # We partial out the static argnums of model, config\n donate_argnums = () # state will be kept instead of being donated in eval_step\n return functional_eval, in_shardings, out_shardings, static_argnums, donate_argnums\n\n\ndef get_shaped_batch(config):\n """"""Return the shape of the batch - this is what eval_shape would return for the\n output of create_data_iterator, but eval_shape doesn't work, see b/306901078.""""""\n batch_shape = (config.global_batch_size_to_load, config.max_target_length)\n shaped_batch = {}\n shaped_batch[""inputs""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""inputs_position""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""inputs_segmentation""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""targets""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""targets_position""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""targets_segmentation""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n if config.use_multimodal:\n image_shape = get_dummy_image_shape_for_init(config)\n shaped_batch[""images""] = jax.ShapeDtypeStruct(image_shape, jnp.int32)\n return shaped_batch\n\n\ndef get_dummy_image_shape_for_init(config):\n """"""Return the shape of the dummy image for specific model's initialization.""""""\n image_shape = ()\n if config.model_name.startswith(""gemma3""):\n image_shape = (\n config.micro_batch_size_to_train_on,\n NUM_IMAGES_PER_SEQUENCE,\n config.image_size_for_vit,\n config.image_size_for_vit,\n NUM_IMAGE_CHANNELS,\n )\n elif config.model_name.startswith(""llama4""):\n image_shape = (\n config.micro_batch_size_to_train_on,\n NUM_TILES_PER_IMAGE,\n NUM_IMAGE_CHANNELS,\n config.tile_size_for_vit,\n config.tile_size_for_vit,\n )\n return image_shape\n\n\ndef load_compiled(config, partial_train, state):\n """"""# Loading a serialized compiled train step function.""""""\n\n # Currently partial_train and state are needed to reconstruct\n # input/output shapes to construct the in_trees and out_trees for load API\n # Parker is working on a serializing these\n def load_serialized_compiled(save_name):\n with open(save_name, ""rb"") as f:\n serialized_compiled = pickle.load(f)\n return serialized_compiled\n\n def get_train_input_output_trees(func, input_args, input_kwargs):\n _, in_tree_recreated = jax.tree_util.tree_flatten((input_args, input_kwargs))\n out_shaped = jax.eval_shape(func, *input_args, **input_kwargs)\n _, out_tree_recreated = jax.tree_util.tree_flatten(out_shaped)\n return in_tree_recreated, out_tree_recreated\n\n serialized_compiled = load_serialized_compiled(config.compiled_trainstep_file)\n shaped_batch = get_shaped_batch(config)\n example_rng = jax.random.PRNGKey(0)\n shaped_input_args = (state, shaped_batch, example_rng)\n shaped_input_kwargs = {}\n in_tree, out_tree = get_train_input_output_trees(partial_train, shaped_input_args, shaped_input_kwargs)\n p_train_step = deserialize_and_load(serialized_compiled, in_tree, out_tree)\n return p_train_step\n\n\ndef calculate_tokens_training_per_device(config):\n """"""Calculate training Tokens per device""""""\n return config.max_target_length * config.per_device_batch_size * config.gradient_accumulation_steps\n\n\ndef calculate_gemma2_tflops_training_per_device(config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops):\n """"""\n Calculate training TFLOP for Gemma2 as in Gemma2 we combine [local_attention, global_attention] into one decoder\n layer and we use sliding window attention in local_attention\n """"""\n noncausal_attention_flops = (\n # global attention\n 4 * config.per_device_batch_size * config.max_target_length**2 * config.num_query_heads * config.head_dim\n +\n # local attention\n 4\n * config.per_device_batch_size\n * config.max_target_length\n * min(config.sliding_window_size, config.max_target_length)\n * config.num_query_heads\n * config.head_dim\n )\n causal_attention_flops = noncausal_attention_flops / 2\n attention_tflops = causal_attention_flops * config.num_decoder_layers * 3 / 10**12\n\n # multiply num_decoder_layers by 2 because we combine [local_attention, global_attention] into one decoder layer\n learnable_weight_tflops = (\n ((total_ffn_flops + qkv_flops + projection_flops) * config.num_decoder_layers * 2 + embedding_flops) * 3 / 10**12\n )\n\n return attention_tflops, learnable_weight_tflops\n\n\ndef calculate_gemma3_tflops_training_per_device(config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops):\n """"""\n Calculate training TFLOPs for Gemma3, which has an alternating pattern of\n 5 local attention layers and 1 global attention layer.\n """"""\n num_layers = config.num_decoder_layers\n\n num_global_layers = num_layers // 6\n num_local_layers = num_layers - num_global_layers\n\n # FLOPs for a single global attention layer (full attention)\n # Formula: 4 * batch_size * seq_len^2 * num_heads * head_dim\n global_attention_flops_per_layer = (\n 4 * config.per_device_batch_size * config.max_target_length**2 * config.num_query_heads * config.head_dim\n )\n\n # FLOPs for a single local attention layer (sliding window)\n # Formula: 4 * batch_size * seq_len * window_size * num_heads * head_dim\n local_attention_flops_per_layer = (\n 4\n * config.per_device_batch_size\n * config.max_target_length\n * min(config.sliding_window_size, config.max_target_length)\n * config.num_query_heads\n * config.head_dim\n )\n\n # Total attention FLOPs = (num_global_layers * FLOPs_per_global) + (num_local_layers * FLOPs_per_local)\n noncausal_attention_flops = (\n num_global_layers * global_attention_flops_per_layer + num_local_layers * local_attention_flops_per_layer\n )\n causal_attention_flops = noncausal_attention_flops / 2\n\n # Convert to TFLOPs and multiply by 3 for fwd/bwd pass\n attention_tflops = causal_attention_flops * 3 / 10**12\n\n # Learnable weights (FFN, QKV, Projections) are present in every layer.\n learnable_weight_tflops = ((total_ffn_flops + qkv_flops + projection_flops) * num_layers + embedding_flops) * 3 / 10**12\n\n return attention_tflops, learnable_weight_tflops\n\n\ndef _calculate_chunked_attention_flops_per_layer(config, seq_len, chunk_size):\n """"""Calculates the non-causal FLOPs for a single layer of chunked attention.""""""\n num_chunks = seq_len // chunk_size\n rem_chunk_size = seq_len % chunk_size\n # The complexity of chunked attention is the sum of squares of chunk lengths.\n chunked_complexity = (num_chunks * chunk_size**2) + (rem_chunk_size**2)\n # The formula for non-causal attention FLOPs is 4 * B * complexity * H * D,\n # where B=batch_size, H=num_heads, D=head_dim.\n return 4 * config.per_device_batch_size * chunked_complexity * config.num_query_heads * config.head_dim\n\n\ndef calculate_llama4_attention_tflops(config):\n """"""\n Calculates attention-only training TFLOPs for Llama4's specific architecture,\n which has an alternating pattern of global and chunked attention layers.\n """"""\n num_layers = config.num_decoder_layers\n seq_len = config.max_target_length\n chunk_size = config.chunk_attn_window_size\n\n # Determine number of global vs. chunked layers based on the NoPE interval.\n # A ""NoPE"" layer uses global attention.\n num_global_layers = num_layers // config.nope_layer_interval\n num_chunked_layers = num_layers - num_global_layers\n\n # FLOPs for a single global attention layer (full attention, non-causal)\n global_attention_flops_per_layer = 4 * config.per_device_batch_size * seq_len**2 * config.num_query_heads * config.head_dim\n\n # FLOPs for a single chunked attention layer (non-causal)\n chunked_attention_flops_per_layer = _calculate_chunked_attention_flops_per_layer(config, seq_len, chunk_size)\n\n # Total non-causal attention FLOPs is the sum of all global and all chunked layers\n noncausal_attention_flops = (num_global_layers * global_attention_flops_per_layer) + (\n num_chunked_layers * chunked_attention_flops_per_layer\n )\n\n # Apply causal mask and convert to TFLOPs (multiply by 3 for fwd/bwd pass)\n causal_attention_flops = noncausal_attention_flops / 2\n attention_tflops = causal_attention_flops * 3 / 10**12\n\n return attention_tflops\n\n\ndef calculate_mla_tflops_per_device(config):\n """"""Calculate Multi-Head Latent Attention TFLOP""""""\n batch_len = config.per_device_batch_size * config.max_target_length\n qk_head_dim_sum = config.qk_nope_head_dim + config.qk_rope_head_dim\n # calculate mla query projection\n if config.q_lora_rank == 0:\n q_flops = 2 * batch_len * config.emb_dim * config.num_query_heads * qk_head_dim_sum\n else:\n # calculate query down and up flops\n q_flops = (\n 2 * batch_len * (config.emb_dim * config.q_lora_rank + config.q_lora_rank * config.num_query_heads * qk_head_dim_sum)\n )\n # calculate mla kv projection with down and up flops\n kv_flops = (\n 2\n * batch_len\n * (\n config.emb_dim * (config.kv_lora_rank + config.qk_rope_head_dim)\n + config.kv_lora_rank * config.num_query_heads * (config.qk_nope_head_dim + config.v_head_dim)\n )\n )\n qkv_flops = q_flops + kv_flops\n\n attention_flops = 2 * batch_len * config.max_target_length * config.num_query_heads * (qk_head_dim_sum + config.v_head_dim)\n projection_flops = 2 * batch_len * config.emb_dim * config.num_query_heads * config.v_head_dim\n return qkv_flops, attention_flops, projection_flops\n\n\ndef calculate_ffn_mamtul_tflops_per_device(config, mlp_dim):\n """"""Helper function to calculate matmul TFLOP in ffn based on MLP dimension.\n\n Applies to:\n - Dense FFN layers (mlp_dim = config.mlp_dim).\n - MoE FFN layers (mlp_dim = config.moe_mlp_dim),\n need to scale by shared_experts or num_experts_per_tok.\n """"""\n ffn1_flops = (\n 2 * config.per_device_batch_size * config.max_target_length * mlp_dim * config.emb_dim * len(config.mlp_activations)\n )\n ffn2_flops = 2 * config.per_device_batch_size * config.max_target_length * mlp_dim * config.emb_dim\n return ffn1_flops + ffn2_flops\n\n\ndef calculate_routed_and_shared_ffn_tflops_per_device(config):\n """"""Helper function to calculate DeepSeek-style ffn TFLOP""""""\n gate_flops = 2 * config.per_device_batch_size * config.max_target_length * config.emb_dim * config.num_experts\n # Due to the mixed decoder layers, the flops is multiplied by num of layers for both dense and moe\n num_dense_layers, num_moe_layers = get_dense_moe_layers(config)\n dense_ffn_flops = calculate_ffn_mamtul_tflops_per_device(config, config.mlp_dim) * num_dense_layers\n shared_experts_flops = calculate_ffn_mamtul_tflops_per_device(config, config.moe_mlp_dim) * config.shared_experts\n routed_experts_flops = calculate_ffn_mamtul_tflops_per_device(config, config.moe_mlp_dim) * config.num_experts_per_tok\n moe_ffn_flops = (gate_flops + shared_experts_flops + routed_experts_flops) * num_moe_layers\n total_ffn_flops = dense_ffn_flops + moe_ffn_flops\n return total_ffn_flops\n\n\ndef get_dense_moe_layers(config):\n """"""Helper function to calculate number of dense and moe layers""""""\n if config.decoder_block == DecoderBlockType.DEEPSEEK:\n num_dense_layers = config.first_num_dense_layers\n num_moe_layers = config.num_decoder_layers - config.first_num_dense_layers\n return num_dense_layers, num_moe_layers\n elif config.decoder_block == DecoderBlockType.LLAMA4:\n num_moe_layers = config.num_decoder_layers // config.interleave_moe_layer_step\n num_dense_layers = config.num_decoder_layers - num_moe_layers\n else:\n raise ValueError(""Currently we only support DeepSeek and Llama4 calculation."")\n\n return num_dense_layers, num_moe_layers\n\n\ndef calculate_gemma3_vision_layers_tflops_per_device(config):\n """"""\n Estimate TFLOPs for Gemma3 vision encoder (ViT-style).\n Returns:\n total_tflops: Total TFLOPs (counts for fwd + bwd + optimizer)\n learnable_weight_tflops: TFLOPs from learnable weights (patch embedding, qkv, MLP, projections)\n attention_tflops: TFLOPs from attention multiplications\n """"""\n # Config values\n B = config.per_device_batch_size\n C = config.num_channels_for_vit\n H = W = config.image_size_for_vit # Gemma3 default 896\n embed_dim = config.emb_dim # text embedding dim after projection\n # Values below are hardcoded in Gemma3VisionEncoderLayer\n patch_size = 14\n hidden_dim = 1152\n intermediate_dim = 4304\n num_layers = 27\n vision_exit_pooling_window = 4\n\n # 1. Patch embedding (Conv2D)\n num_patches_h = H // patch_size\n num_patches_w = W // patch_size\n seq_len = num_patches_h * num_patches_w # 64*64=4096\n patch_embed_flops = 2 * B * seq_len * (C * patch_size * patch_size) * hidden_dim\n\n # 2. gemma3.Encoder: num_layers * gemma3.Encoder1DBlock\n qkv_flops_per_layer = 3 * (2 * B * seq_len * hidden_dim * hidden_dim)\n attn_flops_per_layer = 4 * B * seq_len * seq_len * hidden_dim\n projection_flops_per_layer = 2 * B * seq_len * hidden_dim * hidden_dim # projection after attention multiplication\n mlp_flops_per_layer = 2 * (2 * B * seq_len * hidden_dim * intermediate_dim) # two fc layers\n total_attn_flops = attn_flops_per_layer * num_layers\n encoder_flops = (+qkv_flops_per_layer + projection_flops_per_layer + mlp_flops_per_layer) * num_layers\n\n # 4. VisionEmbedder\n seq_len_after_pooling = (num_patches_h // vision_exit_pooling_window) * (num_patches_w // vision_exit_pooling_window)\n vision_embedder_flops = 2 * B * seq_len_after_pooling * hidden_dim * embed_dim # One linear projection\n\n # Learnable weights summation\n learnable_weight_flops = patch_embed_flops + encoder_flops + vision_embedder_flops\n\n if config.freeze_vision_encoder_params:\n learnable_weight_flops += 2 * vision_embedder_flops # only projector is learnable, add fwd+optimizer\n else:\n learnable_weight_flops *= 3 # multiply by 3 for fwd + bwd + optimizer\n\n # Convert to TFLOPs\n learnable_weight_tflops = learnable_weight_flops / 1e12\n total_attn_tflops = total_attn_flops / 1e12\n total_tflops = learnable_weight_tflops + total_attn_tflops\n\n return total_tflops, learnable_weight_tflops, total_attn_tflops\n\n\ndef calculate_llama4_vision_layers_tflops_per_device(config):\n """"""\n Estimate TFLOPs for Llama4 vision encoder (ViT-style).\n Returns:\n total_tflops: Total TFLOPs (counts for fwd + bwd + optimizer)\n learnable_weight_tflops: TFLOPs from learnable weights (patch embedding, qkv, MLP, projections)\n attention_tflops: TFLOPs from attention multiplications\n """"""\n # Config values\n B = config.per_device_batch_size\n C = config.num_channels_for_vit\n H = W = config.tile_size_for_vit\n patch_size = config.patch_size_for_vit\n hidden_dim = config.hidden_size_for_vit\n intermediate_dim = config.intermediate_size_for_vit\n num_layers = config.num_hidden_layers_for_vit\n pixel_shuffle_fc1_out_dim = config.projector_input_dim_for_vit # 4096\n pixel_shuffle_fc2_out_dim = config.projector_output_dim_for_vit # 4096\n base_emb_dim = config.base_emb_dim\n pixel_shuffle_ratio = config.pixel_shuffle_ratio_for_vit # 0.5\n num_patches = (H // patch_size) * (W // patch_size) # 24*24 = 576\n pixel_shuffle_tokens = num_patches * pixel_shuffle_ratio**2 # 144\n\n # 1. Llama4UnfoldConvolution (flops by linear projection)\n # lax.conv_general_dilated_patches extracts patches through reshaping/indexing without flops\n # Each patch: C * patch_size * patch_size -> hidden_dim\n patch_embed_flops = 2 * B * num_patches * (C * patch_size * patch_size) * hidden_dim\n\n # 2. Llama4VisionEncoder: num_layers * (qkv + att_projection + mlp)\n seq_len = num_patches + 1 # +1 for class token, so 577\n qkv_flops_per_layer = 3 * (2 * B * seq_len * hidden_dim * hidden_dim) # Q, K, V projections\n attn_flops_per_layer = 4 * B * seq_len * seq_len * hidden_dim # Attention scores and weighted sum\n projection_flops_per_layer = 2 * B * seq_len * hidden_dim * hidden_dim # projection after attention multiplication\n mlp_flops_per_layer = 2 * (2 * B * seq_len * hidden_dim * intermediate_dim) # two fc layers\n total_attn_flops = attn_flops_per_layer * num_layers\n vision_encoder_flops = (+qkv_flops_per_layer + projection_flops_per_layer + mlp_flops_per_layer) * num_layers\n\n # 3. Llama4VisionPixelShuffleMLP\n # (B, 144, 5632) -> (B, 144, 4096) -> (B, 144, 4096)\n pixel_shuffle_fc1_flops = 2 * B * pixel_shuffle_tokens * intermediate_dim * pixel_shuffle_fc1_out_dim\n pixel_shuffle_fc2_flops = 2 * B * pixel_shuffle_tokens * pixel_shuffle_fc1_out_dim * pixel_shuffle_fc2_out_dim\n pixel_shuffle_total_flops = pixel_shuffle_fc1_flops + pixel_shuffle_fc2_flops\n\n # 4. Llama4MultiModalProjector: (B, 144, 5120) x (5120, base_emb_dim)\n projector_flops = 2 * B * pixel_shuffle_tokens * pixel_shuffle_fc1_out_dim * base_emb_dim\n\n # Learnable weights: all matmuls above\n learnable_weight_flops = patch_embed_flops + vision_encoder_flops + pixel_shuffle_total_flops + projector_flops\n\n if config.freeze_vision_encoder_params:\n learnable_weight_flops += 2 * projector_flops # only projector is learnable, add fwd+optimizer\n else:\n learnable_weight_flops *= 3 # multiply by 3 for fwd + bwd + optimizer\n\n # Convert to TFLOPs\n learnable_weight_tflops = learnable_weight_flops / 1e12\n total_attn_tflops = total_attn_flops / 1e12\n total_tflops = learnable_weight_tflops + total_attn_tflops\n\n return total_tflops, learnable_weight_tflops, total_attn_tflops\n\n\ndef calculate_vision_encoder_tflops(config):\n """"""Calculate vision encoder TFLOPs per prefill step per device.""""""\n if config.model_name.startswith(""gemma3""):\n mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops = calculate_gemma3_vision_layers_tflops_per_device(\n config\n )\n elif config.model_name.startswith(""llama4""):\n mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops = calculate_llama4_vision_layers_tflops_per_device(\n config\n )\n else:\n max_logging.log(\n f""Vision encoder TFLOPs calculation not implemented for model {config.model_name}, counting as 0 for now.""\n )\n mm_total_tflops = mm_learnable_weight_tflops = mm_attention_tflops = 0\n\n return mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops\n\n\ndef calculate_tflops_training_per_device(config, log=True):\n """"""Calculate training TFLOP""""""\n # MLP flops\n if config.num_experts > 1:\n # calculation based on dropless implementation\n if config.decoder_block in (DecoderBlockType.DEEPSEEK, DecoderBlockType.LLAMA4):\n total_ffn_flops = calculate_routed_and_shared_ffn_tflops_per_device(config)\n else:\n gate_flops = 2 * config.per_device_batch_size * config.max_target_length * config.emb_dim * config.num_experts\n total_ffn_flops = (\n gate_flops + calculate_ffn_mamtul_tflops_per_device(config, config.mlp_dim) * config.num_experts_per_tok\n )\n else:\n total_ffn_flops = calculate_ffn_mamtul_tflops_per_device(config, config.mlp_dim)\n\n # Attention flops\n if config.attention_type == ""mla"":\n qkv_flops, noncausal_attention_flops, projection_flops = calculate_mla_tflops_per_device(config)\n else:\n qkv_flops = (\n 2\n * config.per_device_batch_size\n * config.max_target_length\n * config.emb_dim\n * (config.num_query_heads + 2 * config.num_kv_heads)\n * config.head_dim\n )\n noncausal_attention_flops = (\n 4 * config.per_device_batch_size * config.max_target_length**2 * config.num_query_heads * config.head_dim\n )\n projection_flops = (\n 2\n * config.per_device_batch_size\n * config.max_target_length\n * config.emb_dim\n * config.num_query_heads\n * config.head_dim\n )\n\n # Divide attantion flops by 2 due to causal mask\n # References:\n # NVIDIA/Megatron-LM (2025 March): https://github.com/NVIDIA/Megatron-LM/blob/250b79415dcc4b660521273c87f15334c804eeae/megatron/training/training.py#L361-L362\n # NVIDIA/NeMo (2025 April): https://github.com/NVIDIA/NeMo/blob/ba4d6d116463de512ff0cfc14641aa6cf4577a42/nemo/utils/flops_formulas.py#L259-L272\n causal_attention_flops = noncausal_attention_flops / 2\n\n # Embedding flops\n embedding_flops = 2 * config.per_device_batch_size * config.max_target_length * config.emb_dim * config.vocab_size\n\n # Combine flops with number of decoder layers\n if config.decoder_block == DecoderBlockType.GEMMA2:\n attention_tflops, learnable_weight_tflops = calculate_gemma2_tflops_training_per_device(\n config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops\n )\n elif config.decoder_block == DecoderBlockType.GEMMA3:\n attention_tflops, learnable_weight_tflops = calculate_gemma3_tflops_training_per_device(\n config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops\n )\n elif config.decoder_block == DecoderBlockType.LLAMA4:\n # Use the new helper to calculate attention TFLOPs correctly.\n attention_tflops = calculate_llama4_attention_tflops(config)\n # The learnable weight calculation remains the same as it correctly handles Llama4's MoE structure.\n learnable_weight_tflops = (\n (total_ffn_flops + (qkv_flops + projection_flops) * config.num_decoder_layers + embedding_flops) * 3 / 10**12\n )\n elif config.decoder_block == DecoderBlockType.DEEPSEEK:\n learnable_weight_tflops = (\n (total_ffn_flops + (qkv_flops + projection_flops) * config.num_decoder_layers + embedding_flops) * 3 / 10**12\n )\n attention_tflops = causal_attention_flops * config.num_decoder_layers * 3 / 10**12\n else:\n # multiply by 3 for both feed forward and back propagation flops\n learnable_weight_tflops = (\n ((total_ffn_flops + qkv_flops + projection_flops) * config.num_decoder_layers + embedding_flops) * 3 / 10**12\n )\n attention_tflops = causal_attention_flops * config.num_decoder_layers * 3 / 10**12\n\n learnable_weight_tflops = learnable_weight_tflops * config.gradient_accumulation_steps\n attention_tflops = attention_tflops * config.gradient_accumulation_steps\n\n # DPO includes one additional forward pass per gradient accumulation step\n if config.use_dpo:\n reference_model_tflops = learnable_weight_tflops / 3 # additional forward pass\n reference_model_attention_tflops = attention_tflops / 3\n attention_tflops = attention_tflops + reference_model_attention_tflops\n else:\n reference_model_tflops = 0\n\n total_tflops = learnable_weight_tflops + attention_tflops + reference_model_tflops\n\n if config.use_multimodal:\n # Add vision layers TFLOPs for multimodal models\n mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops = calculate_vision_encoder_tflops(config)\n if log:\n print(\n f""{config.model_name} vision layers per train step:\n"",\n f""Total TFLOPs: {mm_total_tflops:.2f} \n"",\n f""split as {100 * mm_learnable_weight_tflops/mm_total_tflops:.2f}% learnable weight flops"",\n f""and {100 * mm_attention_tflops/mm_total_tflops:.2f}% attention flops;\n"",\n f""learnable weight {mm_learnable_weight_tflops:.2f} TFLOPs, attention {mm_attention_tflops:.2f} TFLOPs"",\n )\n total_tflops += mm_total_tflops\n learnable_weight_tflops += mm_learnable_weight_tflops\n attention_tflops += mm_attention_tflops\n\n if log:\n print(\n ""Per train step:\n"",\n f""Total TFLOPs: {total_tflops:.2f} \n"",\n f""split as {100 * learnable_weight_tflops/total_tflops:.2f}% learnable weight flops"",\n f""and {100 * attention_tflops/total_tflops:.2f}% attention flops"",\n )\n return total_tflops, learnable_weight_tflops, attention_tflops\n\n\n# https://arxiv.org/pdf/2204.02311.pdf Appendix B\ndef calculate_prefill_tflops_per_device(num_model_parameters, prefill_length, config, log=True):\n """"""Calculate training TFLOP""""""\n learnable_weight_tflops = 2 * num_model_parameters * prefill_length / jax.device_count() / 1e12\n noncausal_attention_flops = (\n 4\n * config.num_query_heads\n * config.num_decoder_layers\n * config.head_dim\n * prefill_length**2\n / jax.device_count()\n / 1e12\n )\n causal_attention_tflops = noncausal_attention_flops / 2 # due to causality in attention\n total_tflops = learnable_weight_tflops + causal_attention_tflops\n\n if log:\n print(\n ""Per prefill step per device: \n"",\n f""\tTotal TFLOPs: {total_tflops:.2f} \n"",\n f""\t\tLearnable weight TFLOPs: {learnable_weight_tflops:.2f} "",\n f""({100 * learnable_weight_tflops/total_tflops:.2f})% of Total\n"",\n f""\t\tCausal attention TFLOPs: {causal_attention_tflops:.2f} "",\n f""({100 * causal_attention_tflops/total_tflops:.2f})% of Total"",\n )\n return total_tflops, learnable_weight_tflops, causal_attention_tflops\n\n\ndef get_mesh_axes_used_by_tensor_spec(tensor_sharding_spec):\n """"""\n Extracts the set of mesh axis names that a tensor's PartitionSpec uses.\n\n This function inspects a tensor's sharding specification (PartitionSpec) and\n identifies which mesh axes are actively used for sharding. If a tensor is not\n sharded (i.e., fully replicated), the resulting set will be empty.\n\n Args:\n tensor_sharding_spec: The PartitionSpec of a tensor, which defines how it's partitioned across the mesh.\n It can be None or contain strings and iterables representing the mesh axes.\n all_mesh_axis_names: A collection of all available mesh axis names in the current device mesh.\n\n Returns:\n A set of strings, where each string is a mesh axis name used by the\n tensor's sharding spec. Returns an empty set for unsharded tensors.\n """"""\n # Flatten the sharding spec, as it can contain nested iterables (e.g., ('data', 'mdl')).\n tensor_sharding_spec = sum(\n [\n [axis] if isinstance(axis, str) else list(axis) if isinstance(axis, Iterable) else []\n for axis in tensor_sharding_spec\n ],\n [],\n )\n return tensor_sharding_spec\n\n\ndef _get_nontrival_mesh_axes(mesh):\n """"""\n Returns mesh axes from config that are valid and have more than one shard.\n\n This function identifies which of the predefined potential sharding axes are\n actually present in the current device mesh and are configured with a size\n greater than one (i.e., are actually sharded).\n\n Args:\n mesh: The device mesh object, which contains information about the mesh topology, including axis names and their sizes.\n\n Returns:\n A set of strings, where each string is a mesh axis name that is both\n pre-configured as a target for sharding and has more than one shard in the mesh.\n """"""\n\n target_sharding_axes_config = [\n ""fsdp"",\n ""fsdp_transpose"",\n ""sequence"",\n ""context"",\n ""context_autoregressive"",\n ""tensor"",\n ""tensor_transpose"",\n ""tensor_sequence"",\n ""stage"",\n ""expert"",\n ]\n\n # Filter the target axes to find those that exist in the current mesh\n # and have a size greater than 1, meaning they are actually used for sharding.\n return {axis for axis in target_sharding_axes_config if axis in mesh.axis_names and mesh.shape[axis] > 1}\n\n\ndef _analyze_sharding(params, mesh, valid_target_mesh_axes):\n """"""\n Analyzes parameters to find which are unsharded on any valid mesh axis.\n\n This function iterates through all parameters in a model, checking their\n sharding specifications. It identifies parameters that are not sharded along any\n of the provided valid target axes (i.e., they are fully replicated across these axes).\n\n Args:\n params: A PyTree of model parameters.\n mesh: The device mesh object.\n valid_target_mesh_axes: A set of mesh axis names that are considered valid targets for sharding.\n\n Returns:\n A tuple containing:\n - unsharded_params_total_size (int): The total size (number of elements) of all parameters found to be\n unsharded on the target axes.\n - problematic_tensors_details (list): A list of dictionaries, where each\n dictionary contains details about a tensor that is not sharded on any of the target axes.\n """"""\n unsharded_params_total_size = 0 # Initialize a counter for the size of unsharded parameters.\n problematic_tensors_details = [] # Initialize a list to store details of problematic tensors.\n\n # Get a flattened list of all parameters (leaves) in the PyTree, along with their paths.\n all_params_leaves = jtu.tree_leaves_with_path(params)\n\n for path, p_leaf in all_params_leaves: # Iterate over each parameter leaf\n param_name_str = jtu.keystr(path) # Convert the tree path to a readable string\n\n # Check that sharding and spec exist and are valid\n sharding = getattr(p_leaf, ""sharding"", None)\n spec = getattr(sharding, ""spec"", None)\n assert sharding is not None and spec is not None and isinstance(spec, P), (\n f""Parameter '{param_name_str}' is missing a valid '.sharding.spec'.""\n ""Expected 'p_leaf.sharding.spec' to be a non-null 'partitionspec'.""\n )\n\n current_sharding_spec = p_leaf.sharding.spec # Extract the current tensor's sharding spec\n # Identify axes used for sharding\n mesh_axes_used = get_mesh_axes_used_by_tensor_spec(current_sharding_spec)\n # Check if the parameter is sharded on all the valid target axes.\n is_sharded_on_all_target_axis = all(axis in mesh_axes_used for axis in valid_target_mesh_axes)\n\n # If the parameter is not sharded on all of the target axes, it's considered ""problematic.""\n if not is_sharded_on_all_target_axis:\n unsharded_params_total_size += p_leaf.size # Add to total unsharded parameter size\n unsharded_axes = set(valid_target_mesh_axes) - set(mesh_axes_used)\n # Add detailed info to list of problematic tensors\n problematic_tensors_details.append(\n {\n ""name"": param_name_str, # Tensor name\n ""size"": p_leaf.size, # tensor size\n ""shape"": p_leaf.shape, # tensor shape\n ""spec"": str(current_sharding_spec), # Tensor sharding spec as string\n ""available_axes"": sorted(list(valid_target_mesh_axes)), # Axes that could be used for sharding\n ""unsharded_axes"": sorted(list(unsharded_axes)), # Unsharded axes\n }\n )\n # Return the total size of unsharded parameters and the list of problematic tensors.\n return unsharded_params_total_size, problematic_tensors_details # Return results\n\n\ndef _raise_if_unsharded_exceeds_tolerance(unsharded_size, total_size, tolerance, problematic_tensors_details):\n """"""\n Raises an AssertionError if the percentage of unsharded parameters exceeds the given tolerance.\n\n This function calculates the proportion of model parameters that are unsharded\n and compares it against a specified tolerance. If the tolerance is exceeded,\n it constructs and raises a detailed error message.\n\n Args:\n unsharded_size: The total size of parameters not sharded on target axes.\n total_size: The total size of all parameters in the model.\n tolerance: A float (e.g., 0.05 for 5%) representing the maximum allowed percentage of unsharded parameters.\n problematic_tensors_details: A list of details about the unsharded tensors,\n used to generate an informative error message.\n\n Raises:\n AssertionError: If the percentage of unsharded parameters is greater than the tolerance.\n """"""\n if total_size <= 0:\n raise ValueError(""Total size must be greater than zero."")\n\n # Calculate the percentage of unsharded parameters.\n unsharded_param_perc = unsharded_size / total_size\n\n # If the percentage is over the tolerance, prepare and raise an error.\n if unsharded_param_perc > tolerance:\n # Sort the problematic tensors by size to show the largest ones first.\n problematic_tensors_details.sort(key=lambda x: x[""size""], reverse=True)\n\n # Begin constructing the error message.\n error_msg_lines = [\n f""Unsharded parameter percentage ({unsharded_param_perc:.2%})"" f""exceeds tolerance ({tolerance:.2%}).""\n ]\n # Add a header explaining the issue.\n error_msg_lines.append(\n ""The following large tensors are replicated (unsharded) but could be sharded on at ""\n ""least one of the available axes:""\n )\n # Add details for the top 5 largest problematic tensors.\n for detail in problematic_tensors_details[:5]: # Show top 5 largest problematic tensors\n error_msg_lines.append(\n f"" - Name: {detail['name']}(Size: {detail['size']}, Shape: {detail['spec']}, Spec: {detail['spec']}) ""\n f"" is unsharded on axis: {detail['unsharded_axes']}""\n f"" could be sharded on: {detail['available_axes']}""\n )\n\n # Raise the assertion error with the combined, formatted message.\n raise AssertionError(""\n"".join(error_msg_lines))\n\n\ndef assert_params_sufficiently_sharded(params, mesh, tolerance):\n """"""\n Asserts that the total size of replicated parameters is within a given tolerance.\n\n This is the main function that orchestrates the sharding analysis. It determines\n the total number of parameters, identifies valid sharding axes, analyzes the\n sharding of all parameters, and then raises an error if the amount of\n unsharded parameters exceeds the specified tolerance.\n\n Args:\n params: A PyTree of model parameters.\n mesh: The device mesh object.\n tolerance: A float representing the maximum allowed percentage of unsharded parameters.\n """"""\n # Calculate the total size of all parameters in the model.\n total_num_params = max_utils.calculate_bytes_from_pytree(params)\n\n # Get the set of nontrival mesh axes that can be used for sharding.\n valid_target_mesh_axes = _get_nontrival_mesh_axes(mesh)\n # If there are no valid axes to shard along, there's nothing to check, so we can exit.\n if not valid_target_mesh_axes:\n return # Exit early\n\n # Analyze the parameters to find the total size of unsharded parameters\n # and get details on which tensors are problematic.\n unsharded_params_total_size, problematic_tensors_details = _analyze_sharding(params, mesh, valid_target_mesh_axes)\n\n # Check if the amount of unsharded parameters is within the tolerance and\n # raise an exception if it is not.\n _raise_if_unsharded_exceeds_tolerance(\n unsharded_params_total_size, total_num_params, tolerance, problematic_tensors_details\n )\n\n\ndef apply_gradient_clipping(raw_grads, state, clipping_threshold):\n """"""Applies gradient clipping to raw gradients, with special handing for FLAX fp8 stats.\n\n Args:\n raw_grads: A pytree of raw gradients.\n state: The current optimizer state.\n clipping_threshold: The gradient clipping threshold.\n\n Returns:\n A pytree of clipped gradients.\n """"""\n gradient_clip_transformation = optax.clip_by_global_norm(clipping_threshold)\n if OVERWRITE_WITH_GRADIENT in raw_grads:\n # Scales + Amax History for Delayed Tensor Scaling SHOULD NOT be clipped or affect clipping\n fp8_stats = raw_grads.pop(OVERWRITE_WITH_GRADIENT)\n grads, _ = gradient_clip_transformation.update(raw_grads, state, None)\n grads[OVERWRITE_WITH_GRADIENT] = fp8_stats # pytype: disable=unsupported-operands\n raw_grads[OVERWRITE_WITH_GRADIENT] = fp8_stats # pytype: disable=unsupported-operands\n else:\n grads, _ = gradient_clip_transformation.update(raw_grads, state, None)\n\n return grads\n\n\ndef get_nested_value(dictionary, nested_key, default=None):\n """"""\n Retrieves a value from a nested key in a dictionary.\n\n Args:\n dictionary: The dictionary to search in.\n nested_key: A tuple representing the nested key, e.g., ('level1', 'level2', 'key').\n default: The value to return if the nested key is not found.\n\n Returns:\n The value associated with the nested key, or the default value if not found.\n """"""\n current_level = dictionary\n\n for key in nested_key:\n if not isinstance(current_level, dict) or key not in current_level:\n return default\n current_level = current_level[key]\n return current_level\n\n\ndef init_decode_state(apply_fn, params) -> train_state.TrainState:\n """"""Init train state with null opt state for decode.""""""\n state = train_state.TrainState(step=0, apply_fn=apply_fn, params=params, tx=None, opt_state={}) # type: ignore\n return state\n\n\ndef init_training_state(apply_fn, params, tx):\n """"""Init train state with null opt state for decode.""""""\n state = train_state.TrainState.create(apply_fn=apply_fn, params=params, tx=tx)\n return state\n\n\ndef init_initial_state(model, tx, config, is_training, key):\n """"""\n We pass in ""static"" objects like model, tx, config as JAX compares them by\n object hash, and instantiating them inside causes pjit top-level annotations\n to fail to match as pytree prefixes if we re-instantiate.\n\n Args: model, tx, config, is_training, key\n """"""\n input_shape = (config.micro_batch_size_to_train_on, config.max_target_length)\n image_shape = get_dummy_image_shape_for_init(config)\n model_vars = model.init(\n {""params"": key, ""dropout"": key, ""aqt"": key},\n np.ones(input_shape, dtype=jnp.int32),\n np.ones(input_shape, dtype=jnp.int32),\n encoder_images=np.ones(image_shape, dtype=jnp.int32) if config.use_multimodal else None,\n )\n if is_training:\n return init_training_state(model.apply, model_vars, tx)\n return init_decode_state(model.apply, model_vars)\n\n\ndef setup_decode_state(model, config, rng, mesh, checkpoint_manager):\n """"""Setup decode state by loading params from a checkpoint.\n Args:\n model: the flax model to initialize\n config: config object\n rng: jax.prng key\n mesh: jax.devices() mesh\n checkpoint_manager: Checkpoint manager\n\n Returns:\n state: state with decode params loaded from the checkpoint\n state_mesh_annotations: the mesh annotations for the state\n """"""\n if not config.load_parameters_path:\n # generate random params\n max_logging.log(""No decode checkpoint specified - generating random weights."")\n state, state_mesh_annotations, _, _ = setup_initial_state(\n model, None, None, config, rng, mesh, checkpoint_manager, False\n )\n else:\n # Load params from checkpoint\n max_logging.log(f""Loading decode params from {config.load_parameters_path}"")\n unboxed_abstract_state, state_mesh_annotations, _ = get_abstract_state(model, None, config, rng, mesh, False)\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n params = checkpointing.load_params_from_path(\n config.load_parameters_path,\n unboxed_abstract_state.params,\n config.checkpoint_storage_concurrent_gb,\n config.checkpoint_storage_use_ocdbt,\n config.checkpoint_storage_use_zarr3,\n )\n state = init_decode_state(None, params)\n\n state = max_utils.unbox_logicallypartioned(state)\n return state, state_mesh_annotations\n\n\ndef setup_training_state(model, data_iterator, tx, config, rng, mesh, checkpoint_manager):\n is_training = True\n return setup_initial_state(\n model,\n data_iterator,\n tx,\n config,\n rng,\n mesh,\n checkpoint_manager,\n is_training,\n )\n\n\ndef setup_initial_state(\n model,\n data_iterator,\n tx,\n config,\n rng,\n mesh,\n checkpoint_manager,\n is_training=True,\n):\n """"""We initialize the model and optimizer state, and optionally load from a\n checkpoint as necessary.\n\n Args:\n model: the flax model to initialize\n tx: the optax.GradientTransformation\n config: config object\n rng: jax.prng key\n mesh: jax.devices() mesh\n checkpoint_manager: an Orbax checkpointing.CheckpointManager object\n is_training: True to initialize training state, False for decode state\n\n Returns:\n state: the initialized train state\n state_mesh_annotations: the mesh annotations for the train state\n """"""\n\n unboxed_abstract_state, state_mesh_annotations, state_mesh_shardings = get_abstract_state(\n model, tx, config, rng, mesh, is_training\n )\n\n # Initialization\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n restored, raw_params = checkpointing.load_state_if_possible(\n checkpoint_manager,\n data_iterator,\n config.load_parameters_path,\n config.load_full_state_path,\n config.checkpoint_storage_concurrent_gb,\n unboxed_abstract_state,\n config.enable_single_replica_ckpt_restoring,\n config.dataset_type,\n use_ocdbt=config.checkpoint_storage_use_ocdbt,\n use_zarr3=config.checkpoint_storage_use_zarr3,\n enable_orbax_v1=config.enable_orbax_v1,\n checkpoint_conversion_fn=config.checkpoint_conversion_fn,\n source_checkpoint_layout=config.source_checkpoint_layout,\n )\n\n if restored:\n if isinstance(\n checkpoint_manager,\n (\n emergency_checkpoint_manager.CheckpointManager,\n emergency_replicator_checkpoint_manager.ReplicatorCheckpointManager,\n ),\n ):\n state = restored\n else:\n if ""iter"" in restored and restored[""iter""] is not None:\n data_iterator.local_iterator = restored[""iter""]\n state = restored[""items""]\n else:\n init_state_partial = functools.partial(init_initial_state, model, tx, config, is_training)\n init_state_partial.__name__ = ""initialize_state""\n # pylint: disable=not-callable\n state = jax.jit(\n init_state_partial,\n in_shardings=None,\n out_shardings=state_mesh_shardings,\n )(rng)\n if raw_params: # If we loaded a partial state, we need to merge it.\n state = state.replace(params=raw_params)\n\n state = max_utils.unbox_logicallypartioned(state)\n\n return state, state_mesh_annotations, state_mesh_shardings, data_iterator\n\n\ndef get_abstract_state(model, tx, config, rng, mesh, is_training=True):\n """"""Get a shaped abstraction of the state (including optimizer)""""""\n init_state_partial = functools.partial(init_initial_state, model, tx, config, is_training, rng)\n\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n abstract_state = jax.eval_shape(init_state_partial)\n\n state_logical_annotations = nn.get_partition_spec(abstract_state)\n\n state_mesh_shardings = nn.logical_to_mesh_sharding(state_logical_annotations, mesh, config.logical_axis_rules)\n if is_training and config.optimizer_memory_host_offload:\n opt_state = jax.tree_util.tree_map(lambda x: x.with_memory_kind(kind=""pinned_host""), state_mesh_shardings.opt_state)\n state_mesh_shardings = state_mesh_shardings.replace(opt_state=opt_state)\n if is_training and config.parameter_memory_host_offload:\n assert config.param_scan_axis == 0, ""You must set the scan axis 0 to enable parameter offloading.""\n\n def move(path, x):\n max_logging.log(f""max_utils.py: Moving {path} to host"")\n return x.with_memory_kind(kind=""pinned_host"")\n\n params = jax.tree_util.tree_map_with_path(move, state_mesh_shardings.params)\n state_mesh_shardings = state_mesh_shardings.replace(params=params)\n\n abstract_sharded_state = jax.jit(init_state_partial, in_shardings=None, out_shardings=state_mesh_shardings).eval_shape()\n\n unboxed_abstract_sharded_state = max_utils.unbox_logicallypartioned(abstract_sharded_state)\n # Initialization\n with mesh, nn_partitioning.axis_rules(config.logical_axis_rules):\n state_mesh_annotations = nn.logical_to_mesh(state_logical_annotations)\n return (\n unboxed_abstract_sharded_state,\n state_mesh_annotations,\n state_mesh_shardings,\n )\n\n\ndef get_prefill_kv_cache_annotations(model, config, rng, mesh, page_state: Optional[PageState] = None):\n """"""Get a shaped abstraction of the state (including optimizer)""""""\n\n def init_kv_cache(model, config):\n input_shape = (\n config.global_batch_size_to_load,\n config.max_prefill_predict_length,\n )\n image_shape = get_dummy_image_shape_for_init(config)\n\n model_vars = model.init(\n {""params"": rng, ""dropout"": rng, ""aqt"": rng},\n jnp.ones(input_shape),\n jnp.ones(input_shape),\n encoder_images=jnp.ones(image_shape) if config.use_multimodal else None,\n model_mode=MODEL_MODE_PREFILL,\n slot=0,\n page_state=page_state,\n )\n return model_vars[""cache""]\n\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n init_kv_cache_partial = functools.partial(init_kv_cache, model, config)\n abstract_state = jax.eval_shape(init_kv_cache_partial)\n state_logical_annotations = nn.get_partition_spec(abstract_state)\n with mesh, nn_partitioning.axis_rules(config.logical_axis_rules):\n state_mesh_annotations = nn.logical_to_mesh(state_logical_annotations)\n return state_mesh_annotations\n\n\ndef get_kv_cache_annotations(model, config, rng, mesh, page_state: Optional[PageState] = None):\n """"""Get a shaped abstraction of the state (including optimizer)""""""\n\n def init_kv_cache(model, config):\n input_shape = (\n config.global_batch_size_to_load,\n 1,\n )\n image_shape = get_dummy_image_shape_for_init(config)\n\n model_vars = model.init(\n {""params"": rng, ""dropout"": rng, ""aqt"": rng},\n jnp.ones(input_shape),\n jnp.ones(input_shape),\n encoder_images=jnp.ones(image_shape) if config.use_multimodal else None,\n model_mode=MODEL_MODE_AUTOREGRESSIVE,\n slot=0,\n page_state=page_state,\n )\n return model_vars[""cache""]\n\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n init_kv_cache_partial = functools.partial(init_kv_cache, model, config)\n abstract_state = jax.eval_shape(init_kv_cache_partial)\n state_logical_annotations = nn.get_partition_spec(abstract_state)\n with mesh, nn_partitioning.axis_rules(config.logical_axis_rules):\n state_mesh_annotations = nn.logical_to_mesh(state_logical_annotations)\n return state_mesh_annotations\n\n\ndef save_quantized_checkpoint_if_configured(config, params):\n """"""Save quantized checkpoint if configured""""""\n assert config.quantization, ""quantization must be configured""\n if config.save_quantized_params_path:\n checkpointing.save_params_to_path(\n checkpoint_dir=config.save_quantized_params_path,\n params=params,\n use_ocdbt=config.checkpoint_storage_use_ocdbt,\n use_zarr3=config.checkpoint_storage_use_zarr3,\n )\n else:\n max_logging.log(""Skipping saving quantized checkpoint as save_quantized_params_path is null."")\n\n\ndef add_config_to_summary_writer(config, summary_writer):\n """"""Writes config params to tensorboard""""""\n if jax.process_index() == 0:\n for key, value in config.get_keys().items():\n max_utils.add_text_to_summary_writer(key, str(value), summary_writer)\n\n\ndef logical_axis_rules_pp_act_as_dp(logical_rules):\n """"""Add stage as a physical axes before data for each rule, so stage acts just like data instead of PP.\n This is used when we want to pipeline only a subset of layers, and leave the rest like DP.\n """"""\n new_rules = []\n for key, physical_axes in logical_rules:\n if isinstance(physical_axes, str):\n physical_axes = (physical_axes,)\n else:\n physical_axes = tuple(physical_axes)\n new_physical_axes = tuple(axis for axis in physical_axes if axis != ""stage"")\n if ""data"" in new_physical_axes:\n data_idx = new_physical_axes.index(""data"")\n new_physical_axes = new_physical_axes[0:data_idx] + (""stage"",) + new_physical_axes[data_idx:]\n new_rules.append((key, new_physical_axes))\n return tuple(new_rules)\n\n\ndef create_device_mesh(config, devices=None):\n """"""Creates a device mesh with each slice in its own data parallel group. If there is only one slice, uses two replicas""""""\n if devices is None:\n devices = jax.devices()\n if config.subslice_shape and config.enable_single_controller and config.num_slices == 1:\n max_logging.log(f""Trying to create a subslice with shape: {config.subslice_shape}"")\n subslice_shape = tuple(int(x) for x in config.subslice_shape.split("",""))\n device_coords = [device.coords for device in devices]\n device_coords_np = np.array(device_coords)\n\n # Find the minimum coordinates to start the subslice\n min_coords = device_coords_np.min(axis=0)\n\n subslice_devices = []\n for device in devices:\n coords = device.coords\n if all(min_coords[i] <= coords[i] < min_coords[i] + subslice_shape[i] for i in range(len(subslice_shape))):\n subslice_devices.append(device)\n devices = subslice_devices\n\n num_devices = len(devices)\n num_slices = 1 if config.inference_benchmark_test else config.num_slices\n num_devices_per_slice = num_devices // num_slices\n\n multi_slice_env = num_slices > 1\n\n # Find possible unspecified parallelisms\n ici_parallelism = max_utils.fill_unspecified_mesh_axes(config.ici_parallelism.copy(), num_devices_per_slice, ""ICI"")\n\n allow_split_physical_axes = config.allow_split_physical_axes if config.allow_split_physical_axes else False\n\n if multi_slice_env:\n dcn_parallelism = max_utils.fill_unspecified_mesh_axes(config.dcn_parallelism.copy(), num_slices, ""DCN"")\n if max_utils.is_valid_custom_mesh(ici_parallelism, config.custom_mesh):\n mesh = max_utils.create_custom_device_mesh(ici_parallelism, dcn_parallelism, devices, config.custom_mesh)\n else:\n mesh = mesh_utils.create_hybrid_device_mesh(\n ici_parallelism,\n dcn_parallelism,\n devices,\n allow_split_physical_axes=allow_split_physical_axes,\n )\n else:\n if allow_split_physical_axes:\n if max_utils.is_valid_custom_mesh(ici_parallelism, config.custom_mesh):\n mesh = mesh_utils.create_device_mesh(\n [16, 16],\n devices,\n contiguous_submeshes=False,\n allow_split_physical_axes=False,\n )\n mesh = max_utils.reshape_mesh_to_rings(mesh, config.custom_mesh)\n mesh = np.reshape(mesh, ici_parallelism)\n else:\n mesh = mesh_utils.create_device_mesh(\n ici_parallelism,\n devices,\n contiguous_submeshes=False,\n allow_split_physical_axes=allow_split_physical_axes,\n )\n else:\n mesh = mesh_utils.create_device_mesh(\n ici_parallelism,\n devices,\n )\n if config.optimize_mesh_for_tpu_v6e:\n mesh = max_utils.optimize_mesh_for_tpu_v6e(mesh, devices)\n\n max_logging.log(f""Num_devices: {num_devices}, shape {mesh.shape}"")\n\n return mesh\n\n\n# Learning Rate Schedule\n# -----------------------------------------------------------------------------\n\n\ndef create_learning_rate_schedule(config):\n """"""Creates a warmup and cosine decay learning rate schedule:\n We take inspiration from Llama2's learning rate (LR) schedule, see https://arxiv.org/pdf/2307.09288.pdf section 2.2\n Learning rate schedule has either two or three parts:\n 1) Linear warmup from 0 to [learning_rate] over steps 0 to [learning_rate_schedule_steps * warmup_steps_fraction]\n 2) Cosine from [learning_rate] to [learning_rate * cosine_learning_rate_final_fraction] until learning_rate_schedule_steps\n 3) Constant learning rate of 0 from learning_rate_schedule_steps to steps.\n The zero learning rate section can be used to more accurately measure the fully trained model's performance.\n """"""\n\n def make_cos_schedule(init_lr, final_lr, len_steps):\n def schedule(step):\n pct = (step) / len_steps\n a = 0.5 * (jnp.cos(jnp.pi * pct) + 1)\n lr = init_lr * a + final_lr * (1 - a)\n return lr\n\n return schedule\n\n lr = config.learning_rate\n cos_final_lr = lr * config.cosine_learning_rate_final_fraction\n\n warmup_steps = int(config.learning_rate_schedule_steps * config.warmup_steps_fraction)\n cos_steps = config.learning_rate_schedule_steps - warmup_steps\n constant_zero_steps = config.steps - config.learning_rate_schedule_steps\n\n warmup_schedule = optax.linear_schedule(init_value=0.0, end_value=lr, transition_steps=warmup_steps)\n cos_schedule = make_cos_schedule(lr, cos_final_lr, cos_steps)\n constant_schedule = optax.constant_schedule(0.0)\n\n pieces = [warmup_schedule, cos_schedule]\n boundaries = [\n warmup_steps,\n warmup_steps + cos_steps,\n ]\n\n if constant_zero_steps > 0:\n pieces.append(constant_schedule)\n boundaries.append(warmup_steps + cos_steps + constant_zero_steps)\n\n return optax.join_schedules(pieces, boundaries)\n\n\ndef get_formatted_sharding_annotations(params, mesh=None):\n """"""\n Generates a readable string report of sharding annotations for all parameters.\n\n This function iterates through a PyTree of model parameters and inspects the\n sharding information attached to each parameter (leaf). It creates a\n human-readable summary that is useful for debugging sharding configurations.\n\n Args:\n params: The PyTree of model parameters to inspect.\n mesh: (Optional) The device mesh. If provided, its axis names and shape\n are included in the report for additional context.\n\n Returns:\n A single string containing the formatted report of sharding annotations\n for every parameter, with each entry on a new line.\n """"""\n # Initialize a list to hold the lines of the report, starting with a title.\n annotation_lines = [""Comprehensice Weight Sharding Annotations:""]\n\n # If a mesh object is provided, add its details to the report header.\n if mesh:\n annotation_lines.append(f""Mesh axes: {mesh.axis_names}, Mesh shape: {mesh.shape}"")\n annotation_lines.append(""-"" * 30)\n\n # Get a flattened list of all parameters (leaves) and their corresponding paths in the PyTree.\n all_params_leaves = jtu.tree_leaves_with_path(params)\n\n # Loop through each parameter leaf in the flattened list.\n for path, p_leaf in all_params_leaves:\n # Convert the parameter's path (a sequence of keys) into a readable string name.\n param_name_str = jtu.keystr(path)\n # Get the shape of the parameter as a string.\n shape_str = str(p_leaf.shape)\n # Set a default description for sharding, in case none is found.\n sharding_desc = ""N/A""\n\n # Check if the parameter leaf has a 'sharding' attribute.\n if hasattr(p_leaf, ""sharding""):\n # Case 1: Standard JAX sharding with a PartitionSpec.\n if hasattr(p_leaf.sharding, ""spec"") and p_leaf.sharding.spec is not None:\n # The spec is a tuple (PartitionSpec), format it for readability.\n spec_parts = []\n for item in p_leaf.sharding.spec:\n # Represent None as ""Replicated"" to make it explicit.\n spec_parts.append(str(item) if item is not None else ""Replicated"")\n sharding_desc = f""PartitionSpec({', '.join(spec_parts)})""\n # Case 2: The parameter is explicitly marked as fully replicated.\n elif hasattr(p_leaf.sharding, ""spec"") and p_leaf.sharding.spec is None:\n sharding_desc = ""Fully Replicated (spec is None)""\n # Case 3: A generic fallback if a sharding object exists but has no recognized spec attribute.\n else:\n # Print the string representation of the sharding object itself.\n sharding_desc = str(p_leaf.sharding)\n # Case 4: The parameter has no .sharding attribute at all.\n else:\n sharding_desc = ""No .sharding attribute found""\n\n # Append the formatted details for the current parameter to our list of lines.\n annotation_lines.append(f"" - Param: {param_name_str}\n"" f"" Shape: {shape_str}\n"" f"" Sharding: {sharding_desc}"")\n # Join all the collected lines into a single string, separated by newlines.\n return ""\n"".join(annotation_lines)\n\n\ndef get_physical_spec_no_fsdp(full_logical, mesh, logical_axis_rules):\n """"""\n Generates a physical sharding spec for fully replicated weights.\n\n This function computes a target sharding layout where model parameters are fully\n replicated across the 'fsdp' mesh axis. It starts with the original logical\n sharding and removes any rules that shard along the 'fsdp' or\n 'fsdp_transpose' axes.\n\n Replacing a sharding axis with `None` in a PartitionSpec instructs JAX to\n replicate the array data along that physical mesh dimension. The resulting\n specification is used as a target layout for an all-gather operation.\n\n Args:\n full_logical: A PyTree of logical PartitionSpecs for the model parameters.\n mesh: The JAX device mesh.\n logical_axis_rules: Rules for converting logical axes to physical mesh axes.\n\n Returns:\n A PyTree of physical `jax.sharding.NamedSharding` objects that describe a\n layout where parameters are fully gathered (replicated) across the 'fsdp'\n mesh axis.\n """"""\n\n def remove_fsdp_sharding(sharding_tree):\n """"""Recursively traverses the sharding tree to remove fsdp axes.""""""\n\n def _remove_fsdp_from_partition_spec(named_sharding):\n """"""Removes 'fsdp' and 'fsdp_transpose' from a PartitionSpec.""""""\n if isinstance(named_sharding, jax.sharding.NamedSharding):\n new_spec = []\n # Iterate through each axis in the original PartitionSpec.\n for axis in named_sharding.spec:\n if axis is None:\n new_spec.append(None)\n elif isinstance(axis, str):\n # If the axis is 'fsdp', replace it with None to signify replication.\n if axis not in (""fsdp"", ""fsdp_transpose""):\n new_spec.append(axis)\n else:\n new_spec.append(None)\n elif isinstance(axis, (list, tuple)):\n # If the axis is a collection, filter out 'fsdp'.\n new_axis = [a for a in axis if a not in (""fsdp"", ""fsdp_transpose"")]\n new_spec.append(tuple(new_axis))\n else:\n raise ValueError(f""Unsupported_axis_type: {type(axis)}"")\n # Return a new sharding object with the modified spec.\n return jax.sharding.NamedSharding(named_sharding.mesh, jax.sharding.PartitionSpec(*new_spec))\n return named_sharding\n\n return jax.tree.map(_remove_fsdp_from_partition_spec, sharding_tree)\n\n # Convert the high-level logical spec to a physical one using default rules.\n physical = nn.logical_to_mesh_sharding(full_logical, mesh=mesh, rules=logical_axis_rules)\n # Apply the function to remove the FSDP sharding, defining our target layout.\n physical_no_fsdp = remove_fsdp_sharding(physical)\n return physical_no_fsdp\n\n\ndef all_gather_over_fsdp(variables, sharding_info, mesh, logical_axis_rules):\n """"""Performs an all-gather on FSDP-sharded variables via a sharding constraint.\n This function triggers an all-gather operation on the model's parameters.\n It does so by applying a sharding constraint that specifies a fully\n replicated layout.\n\n The JAX compiler satisfies this constraint by automatically inserting the\n necessary `all-gather` collective communication operations into the\n computation graph, effectively gathering the sharded weights.\n\n Args:\n variables: The PyTree of model parameters, currently sharded across devices.\n sharding_info: The logical partition spec of the currently sharded `variables`.\n mesh: The JAX device mesh.\n logical_axis_rules: Rules for converting logical axes to physical mesh axes.\n\n Returns:\n The model's variables with the all-gather operation applied, resulting\n in the weights being fully replicated on all devices in the 'fsdp' mesh.\n """"""\n # Get the target physical layout (weights fully replicated).\n physical_constraint_no_fsdp = get_physical_spec_no_fsdp(sharding_info, mesh, logical_axis_rules)\n # Apply the constraint to the model's current variables. This tells JAX to\n # gather the weights into this layout.\n return jax.lax.with_sharding_constraint(variables, physical_constraint_no_fsdp)\n",python,tab +2,83,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"3:57:52 PM [info] Activating crowd-code\n3:57:52 PM [info] Recording started\n3:57:52 PM [info] Initializing git provider using file system watchers...\n3:57:52 PM [info] Git repository found\n3:57:52 PM [info] Git provider initialized successfully\n",Log,tab +3,117,"extension-output-pdoom-org.crowd-code-#1-crowd-code",245,0,"3:57:52 PM [info] Initial git state: [object Object]\n",Log,content +4,5451,"MaxText/maxtext_utils.py",0,0,"",python,tab +5,17397,"MaxText/maxtext_utils.py",21489,0,"",python,selection_keyboard +6,18246,"MaxText/maxtext_utils.py",21442,0,"",python,selection_command +7,18395,"MaxText/maxtext_utils.py",21413,0,"",python,selection_command +8,18544,"MaxText/maxtext_utils.py",21399,0,"",python,selection_command +9,18679,"MaxText/maxtext_utils.py",21353,0,"",python,selection_command +10,19247,"MaxText/maxtext_utils.py",21349,0,"",python,selection_command +11,19405,"MaxText/maxtext_utils.py",21348,0,"",python,selection_command +12,19597,"MaxText/maxtext_utils.py",21312,0,"",python,selection_command +13,21678,"MaxText/maxtext_utils.py",21372,0,"",python,selection_command +14,21690,"MaxText/maxtext_utils.py",21312,0,"",python,selection_command +15,39010,"src/MaxText/get_flops.py",0,0,"# Copyright 2023–2025 Google LLC\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# https://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n"""""" A wrapper file for easily calculating training TFLOPs. """"""\n\nfrom MaxText.maxtext_utils import calculate_tflops_training_per_device\nfrom MaxText import pyconfig\nfrom MaxText.globals import MAXTEXT_PKG_DIR\nimport os\nfrom typing import Sequence, cast\nfrom absl import app\n\n\ndef main(argv: Sequence[str]):\n """"""\n Calculates and prints TFLOPs using command\n Example invocation:\n python3 -m MaxText.get_flops model_name=llama2-7b\n """"""\n pyconfig_argv = [argv[0], os.path.join(MAXTEXT_PKG_DIR, ""configs"", ""base.yml"")] + cast(list[str], argv[1:])\n config = pyconfig.initialize(pyconfig_argv)\n tflops, _, _ = calculate_tflops_training_per_device(config, log=False)\n print(f""Total TFLOPs per device per step: {tflops}"")\n\n\nif __name__ == ""__main__"":\n app.run(main)\n",python,tab +16,39015,"src/MaxText/get_flops.py",1190,16,"calculate_tflops",python,selection_command +17,39999,"src/MaxText/get_flops.py",688,0,"",python,selection_mouse +18,40997,"src/MaxText/maxtext_utils.py",0,0,"# Copyright 2023–2025 Google LLC\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# https://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# pylint: disable=line-too-long, disable=bare-except, consider-using-generator\n"""""" Utils that are only interesting to MaxText. """"""\n\nimport functools\nimport pickle\n\nfrom flax import linen as nn\nfrom flax.linen import partitioning as nn_partitioning\nfrom flax.training import train_state\n\nimport numpy as np\n\nfrom collections.abc import Iterable\nfrom jax.experimental import mesh_utils\nfrom jax.experimental.serialize_executable import deserialize_and_load\nfrom jax.sharding import PartitionSpec as P\n\nimport jax\nimport jax.numpy as jnp\nimport jax.tree_util as jtu\n\nimport optax\n\nimport orbax.checkpoint.experimental.emergency.checkpoint_manager as emergency_checkpoint_manager\nimport orbax.checkpoint.experimental.emergency.replicator_checkpoint_manager as emergency_replicator_checkpoint_manager\n\nfrom MaxText import checkpointing\nfrom MaxText import max_logging\nfrom MaxText import max_utils\nfrom MaxText import multimodal_utils\nfrom MaxText.common_types import DecoderBlockType, MODEL_MODE_PREFILL, MODEL_MODE_AUTOREGRESSIVE\nfrom MaxText.inference.page_manager import PageState\n\nOVERWRITE_WITH_GRADIENT = ""_overwrite_with_gradient""\n\n\ndef get_input_data_sharding(config, mesh):\n """"""Get the input data sharding for the model""""""\n return nn.logical_to_mesh_sharding(P(*config.input_data_sharding_logical_axes), mesh, config.logical_axis_rules)\n\n\ndef get_functional_train_with_signature(train_step, data_sharding, state_mesh_shardings, model, config):\n """"""Get the shardings (both state and data) for `train_step`.""""""\n functional_train = functools.partial(train_step, model, config, state_mesh_shardings)\n functional_train.__name__ = ""train_step""\n in_shardings = (state_mesh_shardings, data_sharding, None) # State, batch, rng\n out_shardings = (state_mesh_shardings, None) # State, metrics\n static_argnums = () # We partial out the static argnums of model and config\n donate_argnums = 0 # This is the index of the state - we allow the compiler to make use of this memory.\n return functional_train, in_shardings, out_shardings, static_argnums, donate_argnums\n\n\ndef get_functional_eval_with_signature(eval_step, data_sharding, state_mesh_shardings, model, config):\n """"""Get the shardings (both state and data) for `eval_step`.""""""\n functional_eval = functools.partial(eval_step, model, config)\n functional_eval.__name__ = ""eval_step""\n in_shardings = (state_mesh_shardings, data_sharding, None) # State, batch, rng\n out_shardings = None # metrics\n static_argnums = () # We partial out the static argnums of model, config\n donate_argnums = () # state will be kept instead of being donated in eval_step\n return functional_eval, in_shardings, out_shardings, static_argnums, donate_argnums\n\n\ndef get_shaped_batch(config):\n """"""Return the shape of the batch - this is what eval_shape would return for the\n output of create_data_iterator, but eval_shape doesn't work, see b/306901078.""""""\n batch_shape = (config.global_batch_size_to_load, config.max_target_length)\n shaped_batch = {}\n shaped_batch[""inputs""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""inputs_position""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""inputs_segmentation""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""targets""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""targets_position""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n shaped_batch[""targets_segmentation""] = jax.ShapeDtypeStruct(batch_shape, jnp.int32)\n if config.use_multimodal:\n image_shape = multimodal_utils.get_dummy_image_shape_for_init(config.model_name, batch_size=config.micro_batch_size_to_train_on)\n shaped_batch[""images""] = jax.ShapeDtypeStruct(image_shape, jnp.int32)\n return shaped_batch\n\n\ndef load_compiled(config, partial_train, state):\n """"""# Loading a serialized compiled train step function.""""""\n\n # Currently partial_train and state are needed to reconstruct\n # input/output shapes to construct the in_trees and out_trees for load API\n # Parker is working on a serializing these\n def load_serialized_compiled(save_name):\n with open(save_name, ""rb"") as f:\n serialized_compiled = pickle.load(f)\n return serialized_compiled\n\n def get_train_input_output_trees(func, input_args, input_kwargs):\n _, in_tree_recreated = jax.tree_util.tree_flatten((input_args, input_kwargs))\n out_shaped = jax.eval_shape(func, *input_args, **input_kwargs)\n _, out_tree_recreated = jax.tree_util.tree_flatten(out_shaped)\n return in_tree_recreated, out_tree_recreated\n\n serialized_compiled = load_serialized_compiled(config.compiled_trainstep_file)\n shaped_batch = get_shaped_batch(config)\n example_rng = jax.random.PRNGKey(0)\n shaped_input_args = (state, shaped_batch, example_rng)\n shaped_input_kwargs = {}\n in_tree, out_tree = get_train_input_output_trees(partial_train, shaped_input_args, shaped_input_kwargs)\n p_train_step = deserialize_and_load(serialized_compiled, in_tree, out_tree)\n return p_train_step\n\n\ndef calculate_tokens_training_per_device(config):\n """"""Calculate training Tokens per device""""""\n return config.max_target_length * config.per_device_batch_size * config.gradient_accumulation_steps\n\n\ndef calculate_gemma2_tflops_training_per_device(config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops):\n """"""\n Calculate training TFLOP for Gemma2 as in Gemma2 we combine [local_attention, global_attention] into one decoder\n layer and we use sliding window attention in local_attention\n """"""\n noncausal_attention_flops = (\n # global attention\n 4 * config.per_device_batch_size * config.max_target_length**2 * config.num_query_heads * config.head_dim\n +\n # local attention\n 4\n * config.per_device_batch_size\n * config.max_target_length\n * min(config.sliding_window_size, config.max_target_length)\n * config.num_query_heads\n * config.head_dim\n )\n causal_attention_flops = noncausal_attention_flops / 2\n attention_tflops = causal_attention_flops * config.num_decoder_layers * 3 / 10**12\n\n # multiply num_decoder_layers by 2 because we combine [local_attention, global_attention] into one decoder layer\n learnable_weight_tflops = (\n ((total_ffn_flops + qkv_flops + projection_flops) * config.num_decoder_layers * 2 + embedding_flops) * 3 / 10**12\n )\n\n return attention_tflops, learnable_weight_tflops\n\n\ndef calculate_mixed_attention_model_tflops_training_per_device(\n config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops, attention_pattern_length\n):\n """"""\n Calculate training TFLOPs for models with a mixed attention pattern of local\n and global attention layers, like Gemma3 and GPT-OSS.\n """"""\n num_layers = config.num_decoder_layers\n\n num_global_layers = num_layers // attention_pattern_length\n num_local_layers = num_layers - num_global_layers\n\n # FLOPs for a single global attention layer (full attention)\n # Formula: 4 * batch_size * seq_len^2 * num_heads * head_dim\n global_attention_flops_per_layer = (\n 4 * config.per_device_batch_size * config.max_target_length**2 * config.num_query_heads * config.head_dim\n )\n\n # FLOPs for a single local attention layer (sliding window)\n # Formula: 4 * batch_size * seq_len * window_size * num_heads * head_dim\n local_attention_flops_per_layer = (\n 4\n * config.per_device_batch_size\n * config.max_target_length\n * min(config.sliding_window_size, config.max_target_length)\n * config.num_query_heads\n * config.head_dim\n )\n\n # Total attention FLOPs = (num_global_layers * FLOPs_per_global) + (num_local_layers * FLOPs_per_local)\n noncausal_attention_flops = (\n num_global_layers * global_attention_flops_per_layer + num_local_layers * local_attention_flops_per_layer\n )\n causal_attention_flops = noncausal_attention_flops / 2\n\n # Convert to TFLOPs and multiply by 3 for fwd/bwd pass\n attention_tflops = causal_attention_flops * 3 / 10**12\n\n # Learnable weights (FFN, QKV, Projections) are present in every layer.\n learnable_weight_tflops = ((total_ffn_flops + qkv_flops + projection_flops) * num_layers + embedding_flops) * 3 / 10**12\n\n return attention_tflops, learnable_weight_tflops\n\n\ndef _calculate_chunked_attention_flops_per_layer(config, seq_len, chunk_size):\n """"""Calculates the non-causal FLOPs for a single layer of chunked attention.""""""\n num_chunks = seq_len // chunk_size\n rem_chunk_size = seq_len % chunk_size\n # The complexity of chunked attention is the sum of squares of chunk lengths.\n chunked_complexity = (num_chunks * chunk_size**2) + (rem_chunk_size**2)\n # The formula for non-causal attention FLOPs is 4 * B * complexity * H * D,\n # where B=batch_size, H=num_heads, D=head_dim.\n return 4 * config.per_device_batch_size * chunked_complexity * config.num_query_heads * config.head_dim\n\n\ndef calculate_llama4_attention_tflops(config):\n """"""\n Calculates attention-only training TFLOPs for Llama4's specific architecture,\n which has an alternating pattern of global and chunked attention layers.\n """"""\n num_layers = config.num_decoder_layers\n seq_len = config.max_target_length\n chunk_size = config.chunk_attn_window_size\n\n # Determine number of global vs. chunked layers based on the NoPE interval.\n # A ""NoPE"" layer uses global attention.\n num_global_layers = num_layers // config.nope_layer_interval\n num_chunked_layers = num_layers - num_global_layers\n\n # FLOPs for a single global attention layer (full attention, non-causal)\n global_attention_flops_per_layer = (\n 4 * config.per_device_batch_size * seq_len**2 * config.num_query_heads * config.head_dim\n )\n\n # FLOPs for a single chunked attention layer (non-causal)\n chunked_attention_flops_per_layer = _calculate_chunked_attention_flops_per_layer(config, seq_len, chunk_size)\n\n # Total non-causal attention FLOPs is the sum of all global and all chunked layers\n noncausal_attention_flops = (num_global_layers * global_attention_flops_per_layer) + (\n num_chunked_layers * chunked_attention_flops_per_layer\n )\n\n # Apply causal mask and convert to TFLOPs (multiply by 3 for fwd/bwd pass)\n causal_attention_flops = noncausal_attention_flops / 2\n attention_tflops = causal_attention_flops * 3 / 10**12\n\n return attention_tflops\n\n\ndef calculate_mla_tflops_per_device(config):\n """"""Calculate Multi-Head Latent Attention TFLOP""""""\n batch_len = config.per_device_batch_size * config.max_target_length\n qk_head_dim_sum = config.qk_nope_head_dim + config.qk_rope_head_dim\n # calculate mla query projection\n if config.q_lora_rank == 0:\n q_flops = 2 * batch_len * config.emb_dim * config.num_query_heads * qk_head_dim_sum\n else:\n # calculate query down and up flops\n q_flops = (\n 2\n * batch_len\n * (config.emb_dim * config.q_lora_rank + config.q_lora_rank * config.num_query_heads * qk_head_dim_sum)\n )\n # calculate mla kv projection with down and up flops\n kv_flops = (\n 2\n * batch_len\n * (\n config.emb_dim * (config.kv_lora_rank + config.qk_rope_head_dim)\n + config.kv_lora_rank * config.num_query_heads * (config.qk_nope_head_dim + config.v_head_dim)\n )\n )\n qkv_flops = q_flops + kv_flops\n\n attention_flops = (\n 2 * batch_len * config.max_target_length * config.num_query_heads * (qk_head_dim_sum + config.v_head_dim)\n )\n projection_flops = 2 * batch_len * config.emb_dim * config.num_query_heads * config.v_head_dim\n return qkv_flops, attention_flops, projection_flops\n\n\ndef calculate_ffn_mamtul_tflops_per_device(config, mlp_dim):\n """"""Helper function to calculate matmul TFLOP in ffn based on MLP dimension.\n\n Applies to:\n - Dense FFN layers (mlp_dim = config.mlp_dim).\n - MoE FFN layers (mlp_dim = config.moe_mlp_dim),\n need to scale by shared_experts or num_experts_per_tok.\n """"""\n ffn1_flops = (\n 2 * config.per_device_batch_size * config.max_target_length * mlp_dim * config.emb_dim * len(config.mlp_activations)\n )\n ffn2_flops = 2 * config.per_device_batch_size * config.max_target_length * mlp_dim * config.emb_dim\n return ffn1_flops + ffn2_flops\n\n\ndef calculate_routed_and_shared_ffn_tflops_per_device(config):\n """"""Helper function to calculate DeepSeek-style ffn TFLOP""""""\n gate_flops = 2 * config.per_device_batch_size * config.max_target_length * config.emb_dim * config.num_experts\n # Due to the mixed decoder layers, the flops is multiplied by num of layers for both dense and moe\n num_dense_layers, num_moe_layers = get_dense_moe_layers(config)\n dense_ffn_flops = calculate_ffn_mamtul_tflops_per_device(config, config.mlp_dim) * num_dense_layers\n shared_experts_flops = calculate_ffn_mamtul_tflops_per_device(config, config.moe_mlp_dim) * config.shared_experts\n routed_experts_flops = calculate_ffn_mamtul_tflops_per_device(config, config.moe_mlp_dim) * config.num_experts_per_tok\n moe_ffn_flops = (gate_flops + shared_experts_flops + routed_experts_flops) * num_moe_layers\n total_ffn_flops = dense_ffn_flops + moe_ffn_flops\n return total_ffn_flops\n\n\ndef get_dense_moe_layers(config):\n """"""Helper function to calculate number of dense and moe layers""""""\n if config.decoder_block == DecoderBlockType.DEEPSEEK:\n num_dense_layers = config.first_num_dense_layers\n num_moe_layers = config.num_decoder_layers - config.first_num_dense_layers\n return num_dense_layers, num_moe_layers\n elif config.decoder_block == DecoderBlockType.LLAMA4:\n num_moe_layers = config.num_decoder_layers // config.interleave_moe_layer_step\n num_dense_layers = config.num_decoder_layers - num_moe_layers\n else:\n raise ValueError(""Currently we only support DeepSeek and Llama4 calculation."")\n\n return num_dense_layers, num_moe_layers\n\n\ndef calculate_gemma3_vision_layers_tflops_per_device(config):\n """"""\n Estimate TFLOPs for Gemma3 vision encoder (ViT-style).\n Returns:\n total_tflops: Total TFLOPs (counts for fwd + bwd + optimizer)\n learnable_weight_tflops: TFLOPs from learnable weights (patch embedding, qkv, MLP, projections)\n attention_tflops: TFLOPs from attention multiplications\n """"""\n # Config values\n B = config.per_device_batch_size\n C = config.num_channels_for_vit\n H = W = config.image_size_for_vit # Gemma3 default 896\n embed_dim = config.emb_dim # text embedding dim after projection\n # Values below are hardcoded in Gemma3VisionEncoderLayer\n patch_size = 14\n hidden_dim = 1152\n intermediate_dim = 4304\n num_layers = 27\n vision_exit_pooling_window = 4\n\n # 1. Patch embedding (Conv2D)\n num_patches_h = H // patch_size\n num_patches_w = W // patch_size\n seq_len = num_patches_h * num_patches_w # 64*64=4096\n patch_embed_flops = 2 * B * seq_len * (C * patch_size * patch_size) * hidden_dim\n\n # 2. gemma3.Encoder: num_layers * gemma3.Encoder1DBlock\n qkv_flops_per_layer = 3 * (2 * B * seq_len * hidden_dim * hidden_dim)\n attn_flops_per_layer = 4 * B * seq_len * seq_len * hidden_dim\n projection_flops_per_layer = 2 * B * seq_len * hidden_dim * hidden_dim # projection after attention multiplication\n mlp_flops_per_layer = 2 * (2 * B * seq_len * hidden_dim * intermediate_dim) # two fc layers\n total_attn_flops = attn_flops_per_layer * num_layers\n encoder_flops = (+qkv_flops_per_layer + projection_flops_per_layer + mlp_flops_per_layer) * num_layers\n\n # 4. VisionEmbedder\n seq_len_after_pooling = (num_patches_h // vision_exit_pooling_window) * (num_patches_w // vision_exit_pooling_window)\n vision_embedder_flops = 2 * B * seq_len_after_pooling * hidden_dim * embed_dim # One linear projection\n\n # Learnable weights summation\n learnable_weight_flops = patch_embed_flops + encoder_flops + vision_embedder_flops\n\n if config.freeze_vision_encoder_params:\n learnable_weight_flops += 2 * vision_embedder_flops # only projector is learnable, add fwd+optimizer\n else:\n learnable_weight_flops *= 3 # multiply by 3 for fwd + bwd + optimizer\n\n # Convert to TFLOPs\n learnable_weight_tflops = learnable_weight_flops / 1e12\n total_attn_tflops = total_attn_flops / 1e12\n total_tflops = learnable_weight_tflops + total_attn_tflops\n\n return total_tflops, learnable_weight_tflops, total_attn_tflops\n\n\ndef calculate_llama4_vision_layers_tflops_per_device(config):\n """"""\n Estimate TFLOPs for Llama4 vision encoder (ViT-style).\n Returns:\n total_tflops: Total TFLOPs (counts for fwd + bwd + optimizer)\n learnable_weight_tflops: TFLOPs from learnable weights (patch embedding, qkv, MLP, projections)\n attention_tflops: TFLOPs from attention multiplications\n """"""\n # Config values\n B = config.per_device_batch_size\n C = config.num_channels_for_vit\n H = W = config.tile_size_for_vit\n patch_size = config.patch_size_for_vit\n hidden_dim = config.hidden_size_for_vit\n intermediate_dim = config.intermediate_size_for_vit\n num_layers = config.num_hidden_layers_for_vit\n pixel_shuffle_fc1_out_dim = config.projector_input_dim_for_vit # 4096\n pixel_shuffle_fc2_out_dim = config.projector_output_dim_for_vit # 4096\n base_emb_dim = config.base_emb_dim\n pixel_shuffle_ratio = config.pixel_shuffle_ratio_for_vit # 0.5\n num_patches = (H // patch_size) * (W // patch_size) # 24*24 = 576\n pixel_shuffle_tokens = num_patches * pixel_shuffle_ratio**2 # 144\n\n # 1. Llama4UnfoldConvolution (flops by linear projection)\n # lax.conv_general_dilated_patches extracts patches through reshaping/indexing without flops\n # Each patch: C * patch_size * patch_size -> hidden_dim\n patch_embed_flops = 2 * B * num_patches * (C * patch_size * patch_size) * hidden_dim\n\n # 2. Llama4VisionEncoder: num_layers * (qkv + att_projection + mlp)\n seq_len = num_patches + 1 # +1 for class token, so 577\n qkv_flops_per_layer = 3 * (2 * B * seq_len * hidden_dim * hidden_dim) # Q, K, V projections\n attn_flops_per_layer = 4 * B * seq_len * seq_len * hidden_dim # Attention scores and weighted sum\n projection_flops_per_layer = 2 * B * seq_len * hidden_dim * hidden_dim # projection after attention multiplication\n mlp_flops_per_layer = 2 * (2 * B * seq_len * hidden_dim * intermediate_dim) # two fc layers\n total_attn_flops = attn_flops_per_layer * num_layers\n vision_encoder_flops = (+qkv_flops_per_layer + projection_flops_per_layer + mlp_flops_per_layer) * num_layers\n\n # 3. Llama4VisionPixelShuffleMLP\n # (B, 144, 5632) -> (B, 144, 4096) -> (B, 144, 4096)\n pixel_shuffle_fc1_flops = 2 * B * pixel_shuffle_tokens * intermediate_dim * pixel_shuffle_fc1_out_dim\n pixel_shuffle_fc2_flops = 2 * B * pixel_shuffle_tokens * pixel_shuffle_fc1_out_dim * pixel_shuffle_fc2_out_dim\n pixel_shuffle_total_flops = pixel_shuffle_fc1_flops + pixel_shuffle_fc2_flops\n\n # 4. Llama4MultiModalProjector: (B, 144, 5120) x (5120, base_emb_dim)\n projector_flops = 2 * B * pixel_shuffle_tokens * pixel_shuffle_fc1_out_dim * base_emb_dim\n\n # Learnable weights: all matmuls above\n learnable_weight_flops = patch_embed_flops + vision_encoder_flops + pixel_shuffle_total_flops + projector_flops\n\n if config.freeze_vision_encoder_params:\n learnable_weight_flops += 2 * projector_flops # only projector is learnable, add fwd+optimizer\n else:\n learnable_weight_flops *= 3 # multiply by 3 for fwd + bwd + optimizer\n\n # Convert to TFLOPs\n learnable_weight_tflops = learnable_weight_flops / 1e12\n total_attn_tflops = total_attn_flops / 1e12\n total_tflops = learnable_weight_tflops + total_attn_tflops\n\n return total_tflops, learnable_weight_tflops, total_attn_tflops\n\n\ndef calculate_vision_encoder_tflops(config):\n """"""Calculate vision encoder TFLOPs per prefill step per device.""""""\n if config.model_name.startswith(""gemma3""):\n mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops = calculate_gemma3_vision_layers_tflops_per_device(\n config\n )\n elif config.model_name.startswith(""llama4""):\n mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops = calculate_llama4_vision_layers_tflops_per_device(\n config\n )\n else:\n max_logging.log(\n f""Vision encoder TFLOPs calculation not implemented for model {config.model_name}, counting as 0 for now.""\n )\n mm_total_tflops = mm_learnable_weight_tflops = mm_attention_tflops = 0\n\n return mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops\n\n\ndef calculate_tflops_training_per_device(config, log=True):\n """"""Calculate training TFLOP""""""\n # MLP flops\n if config.num_experts > 1:\n # calculation based on dropless implementation\n if config.decoder_block in (DecoderBlockType.DEEPSEEK, DecoderBlockType.LLAMA4):\n total_ffn_flops = calculate_routed_and_shared_ffn_tflops_per_device(config)\n else:\n gate_flops = 2 * config.per_device_batch_size * config.max_target_length * config.emb_dim * config.num_experts\n total_ffn_flops = (\n gate_flops + calculate_ffn_mamtul_tflops_per_device(config, config.mlp_dim) * config.num_experts_per_tok\n )\n else:\n total_ffn_flops = calculate_ffn_mamtul_tflops_per_device(config, config.mlp_dim)\n\n # Attention flops\n if config.attention_type == ""mla"":\n qkv_flops, noncausal_attention_flops, projection_flops = calculate_mla_tflops_per_device(config)\n else:\n qkv_flops = (\n 2\n * config.per_device_batch_size\n * config.max_target_length\n * config.emb_dim\n * (config.num_query_heads + 2 * config.num_kv_heads)\n * config.head_dim\n )\n noncausal_attention_flops = (\n 4 * config.per_device_batch_size * config.max_target_length**2 * config.num_query_heads * config.head_dim\n )\n projection_flops = (\n 2\n * config.per_device_batch_size\n * config.max_target_length\n * config.emb_dim\n * config.num_query_heads\n * config.head_dim\n )\n\n # Divide attention flops by 2 due to causal mask\n # References:\n # NVIDIA/Megatron-LM (2025 March): https://github.com/NVIDIA/Megatron-LM/blob/250b79415dcc4b660521273c87f15334c804eeae/megatron/training/training.py#L361-L362\n # NVIDIA/NeMo (2025 April): https://github.com/NVIDIA/NeMo/blob/ba4d6d116463de512ff0cfc14641aa6cf4577a42/nemo/utils/flops_formulas.py#L259-L272\n causal_attention_flops = noncausal_attention_flops / 2\n\n # Embedding flops\n embedding_flops = 2 * config.per_device_batch_size * config.max_target_length * config.emb_dim * config.vocab_size\n\n # Combine flops with number of decoder layers\n if config.decoder_block == DecoderBlockType.GEMMA2:\n attention_tflops, learnable_weight_tflops = calculate_gemma2_tflops_training_per_device(\n config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops\n )\n elif config.decoder_block == DecoderBlockType.GEMMA3:\n attention_tflops, learnable_weight_tflops = calculate_mixed_attention_model_tflops_training_per_device (\n config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops, attention_pattern_length=6\n )\n elif config.decoder_block == DecoderBlockType.GPT_OSS:\n attention_tflops, learnable_weight_tflops = calculate_mixed_attention_model_tflops_training_per_device(\n config, total_ffn_flops, qkv_flops, projection_flops, embedding_flops, attention_pattern_length=2\n )\n elif config.decoder_block == DecoderBlockType.LLAMA4:\n # Use the new helper to calculate attention TFLOPs correctly.\n attention_tflops = calculate_llama4_attention_tflops(config)\n # The learnable weight calculation remains the same as it correctly handles Llama4's MoE structure.\n learnable_weight_tflops = (\n (total_ffn_flops + (qkv_flops + projection_flops) * config.num_decoder_layers + embedding_flops) * 3 / 10**12\n )\n elif config.decoder_block == DecoderBlockType.DEEPSEEK:\n learnable_weight_tflops = (\n (total_ffn_flops + (qkv_flops + projection_flops) * config.num_decoder_layers + embedding_flops) * 3 / 10**12\n )\n attention_tflops = causal_attention_flops * config.num_decoder_layers * 3 / 10**12\n else:\n # multiply by 3 for both feed forward and back propagation flops\n learnable_weight_tflops = (\n ((total_ffn_flops + qkv_flops + projection_flops) * config.num_decoder_layers + embedding_flops) * 3 / 10**12\n )\n attention_tflops = causal_attention_flops * config.num_decoder_layers * 3 / 10**12\n\n learnable_weight_tflops = learnable_weight_tflops * config.gradient_accumulation_steps\n attention_tflops = attention_tflops * config.gradient_accumulation_steps\n\n # DPO includes one additional forward pass per gradient accumulation step\n if config.use_dpo:\n reference_model_tflops = learnable_weight_tflops / 3 # additional forward pass\n reference_model_attention_tflops = attention_tflops / 3\n attention_tflops = attention_tflops + reference_model_attention_tflops\n else:\n reference_model_tflops = 0\n\n total_tflops = learnable_weight_tflops + attention_tflops + reference_model_tflops\n\n if config.use_multimodal:\n # Add vision layers TFLOPs for multimodal models\n mm_total_tflops, mm_learnable_weight_tflops, mm_attention_tflops = calculate_vision_encoder_tflops(config)\n if log:\n print(\n f""{config.model_name} vision layers per train step:\n"",\n f""Total TFLOPs: {mm_total_tflops:.2f} \n"",\n f""split as {100 * mm_learnable_weight_tflops/mm_total_tflops:.2f}% learnable weight flops"",\n f""and {100 * mm_attention_tflops/mm_total_tflops:.2f}% attention flops;\n"",\n f""learnable weight {mm_learnable_weight_tflops:.2f} TFLOPs, attention {mm_attention_tflops:.2f} TFLOPs"",\n )\n total_tflops += mm_total_tflops\n learnable_weight_tflops += mm_learnable_weight_tflops\n attention_tflops += mm_attention_tflops\n\n if log:\n print(\n ""Per train step:\n"",\n f""Total TFLOPs: {total_tflops:.2f} \n"",\n f""split as {100 * learnable_weight_tflops/total_tflops:.2f}% learnable weight flops"",\n f""and {100 * attention_tflops/total_tflops:.2f}% attention flops"",\n )\n return total_tflops, learnable_weight_tflops, attention_tflops\n\n\n# https://arxiv.org/pdf/2204.02311.pdf Appendix B\ndef calculate_prefill_tflops_per_device(num_model_parameters, prefill_length, config, log=True):\n """"""Calculate training TFLOP""""""\n learnable_weight_tflops = 2 * num_model_parameters * prefill_length / jax.device_count() / 1e12\n noncausal_attention_flops = (\n 4\n * config.num_query_heads\n * config.num_decoder_layers\n * config.head_dim\n * prefill_length**2\n / jax.device_count()\n / 1e12\n )\n causal_attention_tflops = noncausal_attention_flops / 2 # due to causality in attention\n total_tflops = learnable_weight_tflops + causal_attention_tflops\n\n if log:\n print(\n ""Per prefill step per device: \n"",\n f""\tTotal TFLOPs: {total_tflops:.2f} \n"",\n f""\t\tLearnable weight TFLOPs: {learnable_weight_tflops:.2f} "",\n f""({100 * learnable_weight_tflops/total_tflops:.2f})% of Total\n"",\n f""\t\tCausal attention TFLOPs: {causal_attention_tflops:.2f} "",\n f""({100 * causal_attention_tflops/total_tflops:.2f})% of Total"",\n )\n return total_tflops, learnable_weight_tflops, causal_attention_tflops\n\n\ndef get_mesh_axes_used_by_tensor_spec(tensor_sharding_spec):\n """"""\n Extracts the set of mesh axis names that a tensor's PartitionSpec uses.\n\n This function inspects a tensor's sharding specification (PartitionSpec) and\n identifies which mesh axes are actively used for sharding. If a tensor is not\n sharded (i.e., fully replicated), the resulting set will be empty.\n\n Args:\n tensor_sharding_spec: The PartitionSpec of a tensor, which defines how it's partitioned across the mesh.\n It can be None or contain strings and iterables representing the mesh axes.\n all_mesh_axis_names: A collection of all available mesh axis names in the current device mesh.\n\n Returns:\n A set of strings, where each string is a mesh axis name used by the\n tensor's sharding spec. Returns an empty set for unsharded tensors.\n """"""\n # Flatten the sharding spec, as it can contain nested iterables (e.g., ('data', 'mdl')).\n tensor_sharding_spec = sum(\n [\n [axis] if isinstance(axis, str) else list(axis) if isinstance(axis, Iterable) else []\n for axis in tensor_sharding_spec\n ],\n [],\n )\n return tensor_sharding_spec\n\n\ndef _get_nontrival_mesh_axes(mesh):\n """"""\n Returns mesh axes from config that are valid and have more than one shard.\n\n This function identifies which of the predefined potential sharding axes are\n actually present in the current device mesh and are configured with a size\n greater than one (i.e., are actually sharded).\n\n Args:\n mesh: The device mesh object, which contains information about the mesh topology, including axis names and their sizes.\n\n Returns:\n A set of strings, where each string is a mesh axis name that is both\n pre-configured as a target for sharding and has more than one shard in the mesh.\n """"""\n\n target_sharding_axes_config = [\n ""fsdp"",\n ""fsdp_transpose"",\n ""sequence"",\n ""context"",\n ""context_autoregressive"",\n ""tensor"",\n ""tensor_transpose"",\n ""tensor_sequence"",\n ""stage"",\n ""expert"",\n ]\n\n # Filter the target axes to find those that exist in the current mesh\n # and have a size greater than 1, meaning they are actually used for sharding.\n return {axis for axis in target_sharding_axes_config if axis in mesh.axis_names and mesh.shape[axis] > 1}\n\n\ndef _analyze_sharding(params, mesh, valid_target_mesh_axes):\n """"""\n Analyzes parameters to find which are unsharded on any valid mesh axis.\n\n This function iterates through all parameters in a model, checking their\n sharding specifications. It identifies parameters that are not sharded along any\n of the provided valid target axes (i.e., they are fully replicated across these axes).\n\n Args:\n params: A PyTree of model parameters.\n mesh: The device mesh object.\n valid_target_mesh_axes: A set of mesh axis names that are considered valid targets for sharding.\n\n Returns:\n A tuple containing:\n - unsharded_params_total_size (int): The total size (number of elements) of all parameters found to be\n unsharded on the target axes.\n - problematic_tensors_details (list): A list of dictionaries, where each\n dictionary contains details about a tensor that is not sharded on any of the target axes.\n """"""\n unsharded_params_total_size = 0 # Initialize a counter for the size of unsharded parameters.\n problematic_tensors_details = [] # Initialize a list to store details of problematic tensors.\n\n # Get a flattened list of all parameters (leaves) in the PyTree, along with their paths.\n all_params_leaves = jtu.tree_leaves_with_path(params)\n\n for path, p_leaf in all_params_leaves: # Iterate over each parameter leaf\n param_name_str = jtu.keystr(path) # Convert the tree path to a readable string\n\n # Check that sharding and spec exist and are valid\n sharding = getattr(p_leaf, ""sharding"", None)\n spec = getattr(sharding, ""spec"", None)\n assert sharding is not None and spec is not None and isinstance(spec, P), (\n f""Parameter '{param_name_str}' is missing a valid '.sharding.spec'.""\n ""Expected 'p_leaf.sharding.spec' to be a non-null 'partitionspec'.""\n )\n\n current_sharding_spec = p_leaf.sharding.spec # Extract the current tensor's sharding spec\n # Identify axes used for sharding\n mesh_axes_used = get_mesh_axes_used_by_tensor_spec(current_sharding_spec)\n # Check if the parameter is sharded on all the valid target axes.\n is_sharded_on_all_target_axis = all(axis in mesh_axes_used for axis in valid_target_mesh_axes)\n\n # If the parameter is not sharded on all of the target axes, it's considered ""problematic.""\n if not is_sharded_on_all_target_axis:\n unsharded_params_total_size += p_leaf.size # Add to total unsharded parameter size\n unsharded_axes = set(valid_target_mesh_axes) - set(mesh_axes_used)\n # Add detailed info to list of problematic tensors\n problematic_tensors_details.append(\n {\n ""name"": param_name_str, # Tensor name\n ""size"": p_leaf.size, # tensor size\n ""shape"": p_leaf.shape, # tensor shape\n ""spec"": str(current_sharding_spec), # Tensor sharding spec as string\n ""available_axes"": sorted(list(valid_target_mesh_axes)), # Axes that could be used for sharding\n ""unsharded_axes"": sorted(list(unsharded_axes)), # Unsharded axes\n }\n )\n # Return the total size of unsharded parameters and the list of problematic tensors.\n return unsharded_params_total_size, problematic_tensors_details # Return results\n\n\ndef _raise_if_unsharded_exceeds_tolerance(unsharded_size, total_size, tolerance, problematic_tensors_details):\n """"""\n Raises an AssertionError if the percentage of unsharded parameters exceeds the given tolerance.\n\n This function calculates the proportion of model parameters that are unsharded\n and compares it against a specified tolerance. If the tolerance is exceeded,\n it constructs and raises a detailed error message.\n\n Args:\n unsharded_size: The total size of parameters not sharded on target axes.\n total_size: The total size of all parameters in the model.\n tolerance: A float (e.g., 0.05 for 5%) representing the maximum allowed percentage of unsharded parameters.\n problematic_tensors_details: A list of details about the unsharded tensors,\n used to generate an informative error message.\n\n Raises:\n AssertionError: If the percentage of unsharded parameters is greater than the tolerance.\n """"""\n if total_size <= 0:\n raise ValueError(""Total size must be greater than zero."")\n\n # Calculate the percentage of unsharded parameters.\n unsharded_param_perc = unsharded_size / total_size\n\n # If the percentage is over the tolerance, prepare and raise an error.\n if unsharded_param_perc > tolerance:\n # Sort the problematic tensors by size to show the largest ones first.\n problematic_tensors_details.sort(key=lambda x: x[""size""], reverse=True)\n\n # Begin constructing the error message.\n error_msg_lines = [\n f""Unsharded parameter percentage ({unsharded_param_perc:.2%})"" f""exceeds tolerance ({tolerance:.2%}).""\n ]\n # Add a header explaining the issue.\n error_msg_lines.append(\n ""The following large tensors are replicated (unsharded) but could be sharded on at ""\n ""least one of the available axes:""\n )\n # Add details for the top 5 largest problematic tensors.\n for detail in problematic_tensors_details[:5]: # Show top 5 largest problematic tensors\n error_msg_lines.append(\n f"" - Name: {detail['name']}(Size: {detail['size']}, Shape: {detail['spec']}, Spec: {detail['spec']}) ""\n f"" is unsharded on axis: {detail['unsharded_axes']}""\n f"" could be sharded on: {detail['available_axes']}""\n )\n\n # Raise the assertion error with the combined, formatted message.\n raise AssertionError(""\n"".join(error_msg_lines))\n\n\ndef assert_params_sufficiently_sharded(params, mesh, tolerance):\n """"""\n Asserts that the total size of replicated parameters is within a given tolerance.\n\n This is the main function that orchestrates the sharding analysis. It determines\n the total number of parameters, identifies valid sharding axes, analyzes the\n sharding of all parameters, and then raises an error if the amount of\n unsharded parameters exceeds the specified tolerance.\n\n Args:\n params: A PyTree of model parameters.\n mesh: The device mesh object.\n tolerance: A float representing the maximum allowed percentage of unsharded parameters.\n """"""\n # Calculate the total size of all parameters in the model.\n total_num_params = max_utils.calculate_bytes_from_pytree(params)\n\n # Get the set of nontrival mesh axes that can be used for sharding.\n valid_target_mesh_axes = _get_nontrival_mesh_axes(mesh)\n # If there are no valid axes to shard along, there's nothing to check, so we can exit.\n if not valid_target_mesh_axes:\n return # Exit early\n\n # Analyze the parameters to find the total size of unsharded parameters\n # and get details on which tensors are problematic.\n unsharded_params_total_size, problematic_tensors_details = _analyze_sharding(params, mesh, valid_target_mesh_axes)\n\n # Check if the amount of unsharded parameters is within the tolerance and\n # raise an exception if it is not.\n _raise_if_unsharded_exceeds_tolerance(\n unsharded_params_total_size, total_num_params, tolerance, problematic_tensors_details\n )\n\n\ndef apply_gradient_clipping(raw_grads, state, clipping_threshold):\n """"""Applies gradient clipping to raw gradients, with special handing for FLAX fp8 stats.\n\n Args:\n raw_grads: A pytree of raw gradients.\n state: The current optimizer state.\n clipping_threshold: The gradient clipping threshold.\n\n Returns:\n A pytree of clipped gradients.\n """"""\n gradient_clip_transformation = optax.clip_by_global_norm(clipping_threshold)\n if OVERWRITE_WITH_GRADIENT in raw_grads:\n # Scales + Amax History for Delayed Tensor Scaling SHOULD NOT be clipped or affect clipping\n fp8_stats = raw_grads.pop(OVERWRITE_WITH_GRADIENT)\n grads, _ = gradient_clip_transformation.update(raw_grads, state, None)\n grads[OVERWRITE_WITH_GRADIENT] = fp8_stats # pytype: disable=unsupported-operands\n raw_grads[OVERWRITE_WITH_GRADIENT] = fp8_stats # pytype: disable=unsupported-operands\n else:\n grads, _ = gradient_clip_transformation.update(raw_grads, state, None)\n\n return grads\n\n\ndef get_nested_value(dictionary, nested_key, default=None):\n """"""\n Retrieves a value from a nested key in a dictionary.\n\n Args:\n dictionary: The dictionary to search in.\n nested_key: A tuple representing the nested key, e.g., ('level1', 'level2', 'key').\n default: The value to return if the nested key is not found.\n\n Returns:\n The value associated with the nested key, or the default value if not found.\n """"""\n current_level = dictionary\n\n for key in nested_key:\n if not isinstance(current_level, dict) or key not in current_level:\n return default\n current_level = current_level[key]\n return current_level\n\n\ndef init_decode_state(apply_fn, params) -> train_state.TrainState:\n """"""Init train state with null opt state for decode.""""""\n state = train_state.TrainState(step=0, apply_fn=apply_fn, params=params, tx=None, opt_state={}) # type: ignore\n return state\n\n\ndef init_training_state(apply_fn, params, tx):\n """"""Init train state with null opt state for decode.""""""\n state = train_state.TrainState.create(apply_fn=apply_fn, params=params, tx=tx)\n return state\n\n\ndef init_initial_state(model, tx, config, is_training, key):\n """"""\n We pass in ""static"" objects like model, tx, config as JAX compares them by\n object hash, and instantiating them inside causes pjit top-level annotations\n to fail to match as pytree prefixes if we re-instantiate.\n\n Args: model, tx, config, is_training, key\n """"""\n input_shape = (config.micro_batch_size_to_train_on, config.max_target_length)\n image_shape = multimodal_utils.get_dummy_image_shape_for_init(config.model_name, batch_size=config.micro_batch_size_to_train_on)\n model_vars = model.init(\n {""params"": key, ""dropout"": key, ""aqt"": key},\n np.ones(input_shape, dtype=jnp.int32),\n np.ones(input_shape, dtype=jnp.int32),\n encoder_images=np.ones(image_shape, dtype=jnp.int32) if config.use_multimodal else None,\n # nnx_method=""no_op"",\n )\n if is_training:\n return init_training_state(model.apply, model_vars, tx)\n return init_decode_state(model.apply, model_vars)\n\n\ndef setup_decode_state(model, config, rng, mesh, checkpoint_manager):\n """"""Setup decode state by loading params from a checkpoint.\n Args:\n model: the flax model to initialize\n config: config object\n rng: jax.prng key\n mesh: jax.devices() mesh\n checkpoint_manager: Checkpoint manager\n\n Returns:\n state: state with decode params loaded from the checkpoint\n state_mesh_annotations: the mesh annotations for the state\n """"""\n if not config.load_parameters_path:\n # generate random params\n max_logging.log(""No decode checkpoint specified - generating random weights."")\n state, state_mesh_annotations, _, _ = setup_initial_state(\n model, None, None, config, rng, mesh, checkpoint_manager, False\n )\n else:\n # Load params from checkpoint\n max_logging.log(f""Loading decode params from {config.load_parameters_path}"")\n unboxed_abstract_state, state_mesh_annotations, _ = get_abstract_state(model, None, config, rng, mesh, False)\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n params = checkpointing.load_params_from_path(\n config.load_parameters_path,\n unboxed_abstract_state.params,\n config.checkpoint_storage_concurrent_gb,\n config.checkpoint_storage_use_ocdbt,\n config.checkpoint_storage_use_zarr3,\n )\n state = init_decode_state(None, params)\n\n state = max_utils.unbox_logicallypartioned(state)\n return state, state_mesh_annotations\n\n\ndef setup_training_state(model, data_iterator, tx, config, rng, mesh, checkpoint_manager):\n is_training = True\n return setup_initial_state(\n model,\n data_iterator,\n tx,\n config,\n rng,\n mesh,\n checkpoint_manager,\n is_training,\n )\n\n\ndef setup_initial_state(\n model,\n data_iterator,\n tx,\n config,\n rng,\n mesh,\n checkpoint_manager,\n is_training=True,\n):\n """"""We initialize the model and optimizer state, and optionally load from a\n checkpoint as necessary.\n\n Args:\n model: the flax model to initialize\n tx: the optax.GradientTransformation\n config: config object\n rng: jax.prng key\n mesh: jax.devices() mesh\n checkpoint_manager: an Orbax checkpointing.CheckpointManager object\n is_training: True to initialize training state, False for decode state\n\n Returns:\n state: the initialized train state\n state_mesh_annotations: the mesh annotations for the train state\n """"""\n\n unboxed_abstract_state, state_mesh_annotations, state_mesh_shardings = get_abstract_state(\n model, tx, config, rng, mesh, is_training\n )\n\n # Initialization\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n restored, raw_params = checkpointing.load_state_if_possible(\n checkpoint_manager,\n data_iterator,\n config.load_parameters_path,\n config.load_full_state_path,\n config.checkpoint_storage_concurrent_gb,\n unboxed_abstract_state,\n config.enable_single_replica_ckpt_restoring,\n config.dataset_type,\n use_ocdbt=config.checkpoint_storage_use_ocdbt,\n use_zarr3=config.checkpoint_storage_use_zarr3,\n enable_orbax_v1=config.enable_orbax_v1,\n checkpoint_conversion_fn=config.checkpoint_conversion_fn,\n source_checkpoint_layout=config.source_checkpoint_layout,\n )\n\n if restored:\n if isinstance(\n checkpoint_manager,\n (\n emergency_checkpoint_manager.CheckpointManager,\n emergency_replicator_checkpoint_manager.ReplicatorCheckpointManager,\n ),\n ):\n state = restored\n else:\n if ""iter"" in restored and restored[""iter""] is not None:\n data_iterator.local_iterator = restored[""iter""]\n state = restored[""items""]\n else:\n init_state_partial = functools.partial(init_initial_state, model, tx, config, is_training)\n init_state_partial.__name__ = ""initialize_state""\n # pylint: disable=not-callable\n state = jax.jit(\n init_state_partial,\n in_shardings=None,\n out_shardings=state_mesh_shardings,\n )(rng)\n if raw_params: # If we loaded a partial state, we need to merge it.\n state = state.replace(params=raw_params)\n\n state = max_utils.unbox_logicallypartioned(state)\n\n return state, state_mesh_annotations, state_mesh_shardings, data_iterator\n\n\ndef get_abstract_state(model, tx, config, rng, mesh, is_training=True):\n """"""Get a shaped abstraction of the state (including optimizer)""""""\n init_state_partial = functools.partial(init_initial_state, model, tx, config, is_training, rng)\n\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n abstract_state = jax.eval_shape(init_state_partial)\n\n state_logical_annotations = nn.get_partition_spec(abstract_state)\n\n state_mesh_shardings = nn.logical_to_mesh_sharding(state_logical_annotations, mesh, config.logical_axis_rules)\n if is_training and config.optimizer_memory_host_offload:\n opt_state = jax.tree_util.tree_map(lambda x: x.with_memory_kind(kind=""pinned_host""), state_mesh_shardings.opt_state)\n state_mesh_shardings = state_mesh_shardings.replace(opt_state=opt_state)\n if is_training and config.parameter_memory_host_offload:\n assert config.param_scan_axis == 0, ""You must set the scan axis 0 to enable parameter offloading.""\n\n def move(path, x):\n max_logging.log(f""max_utils.py: Moving {path} to host"")\n return x.with_memory_kind(kind=""pinned_host"")\n\n params = jax.tree_util.tree_map_with_path(move, state_mesh_shardings.params)\n state_mesh_shardings = state_mesh_shardings.replace(params=params)\n\n abstract_sharded_state = jax.jit(init_state_partial, in_shardings=None, out_shardings=state_mesh_shardings).eval_shape()\n\n unboxed_abstract_sharded_state = max_utils.unbox_logicallypartioned(abstract_sharded_state)\n # Initialization\n with mesh, nn_partitioning.axis_rules(config.logical_axis_rules):\n state_mesh_annotations = nn.logical_to_mesh(state_logical_annotations)\n return (\n unboxed_abstract_sharded_state,\n state_mesh_annotations,\n state_mesh_shardings,\n )\n\n\ndef get_prefill_kv_cache_annotations(model, config, rng, mesh, page_state: None | PageState = None):\n """"""Get a shaped abstraction of the state (including optimizer)""""""\n\n def init_kv_cache(model, config):\n input_shape = (\n config.micro_batch_size_to_train_on,\n config.max_prefill_predict_length,\n )\n image_shape = multimodal_utils.get_dummy_image_shape_for_init(config.model_name, batch_size=config.micro_batch_size_to_train_on)\n\n model_vars = model.init(\n {""params"": rng, ""dropout"": rng, ""aqt"": rng},\n jnp.ones(input_shape),\n jnp.ones(input_shape),\n encoder_images=jnp.ones(image_shape) if config.use_multimodal else None,\n model_mode=MODEL_MODE_PREFILL,\n slot=0,\n page_state=page_state,\n )\n return model_vars[""cache""]\n\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n init_kv_cache_partial = functools.partial(init_kv_cache, model, config)\n abstract_state = jax.eval_shape(init_kv_cache_partial)\n state_logical_annotations = nn.get_partition_spec(abstract_state)\n with mesh, nn_partitioning.axis_rules(config.logical_axis_rules):\n state_mesh_annotations = nn.logical_to_mesh(state_logical_annotations)\n return state_mesh_annotations\n\n\ndef get_kv_cache_annotations(model, config, rng, mesh, page_state: None | PageState = None):\n """"""Get a shaped abstraction of the state (including optimizer)""""""\n\n def init_kv_cache(model, config):\n input_shape = (config.micro_batch_size_to_train_on, 1)\n image_shape = multimodal_utils.get_dummy_image_shape_for_init(config.model_name, batch_size=config.micro_batch_size_to_train_on)\n\n model_vars = model.init(\n {""params"": rng, ""dropout"": rng, ""aqt"": rng},\n jnp.ones(input_shape),\n jnp.ones(input_shape),\n encoder_images=jnp.ones(image_shape) if config.use_multimodal else None,\n model_mode=MODEL_MODE_AUTOREGRESSIVE,\n slot=0,\n page_state=page_state,\n )\n return model_vars[""cache""]\n\n with nn_partitioning.axis_rules(config.logical_axis_rules):\n init_kv_cache_partial = functools.partial(init_kv_cache, model, config)\n abstract_state = jax.eval_shape(init_kv_cache_partial)\n state_logical_annotations = nn.get_partition_spec(abstract_state)\n with mesh, nn_partitioning.axis_rules(config.logical_axis_rules):\n state_mesh_annotations = nn.logical_to_mesh(state_logical_annotations)\n return state_mesh_annotations\n\n\ndef save_quantized_checkpoint_if_configured(config, params):\n """"""Save quantized checkpoint if configured""""""\n assert config.quantization, ""quantization must be configured""\n if config.save_quantized_params_path:\n checkpointing.save_params_to_path(\n checkpoint_dir=config.save_quantized_params_path,\n params=params,\n use_ocdbt=config.checkpoint_storage_use_ocdbt,\n use_zarr3=config.checkpoint_storage_use_zarr3,\n )\n else:\n max_logging.log(""Skipping saving quantized checkpoint as save_quantized_params_path is null."")\n\n\ndef add_config_to_summary_writer(config, summary_writer):\n """"""Writes config params to tensorboard""""""\n if jax.process_index() == 0:\n for key, value in config.get_keys().items():\n max_utils.add_text_to_summary_writer(key, str(value), summary_writer)\n\n\ndef logical_axis_rules_pp_act_as_dp(logical_rules):\n """"""Add stage as a physical axes before data for each rule, so stage acts just like data instead of PP.\n This is used when we want to pipeline only a subset of layers, and leave the rest like DP.\n """"""\n new_rules = []\n for key, physical_axes in logical_rules:\n if isinstance(physical_axes, str):\n physical_axes = (physical_axes,)\n else:\n physical_axes = tuple(physical_axes)\n new_physical_axes = tuple(axis for axis in physical_axes if axis != ""stage"")\n if ""data"" in new_physical_axes:\n data_idx = new_physical_axes.index(""data"")\n new_physical_axes = new_physical_axes[0:data_idx] + (""stage"",) + new_physical_axes[data_idx:]\n new_rules.append((key, new_physical_axes))\n return tuple(new_rules)\n\n\ndef create_device_mesh(config, devices=None):\n """"""Creates a device mesh with each slice in its own data parallel group. If there is only one slice, uses two replicas""""""\n if devices is None:\n devices = jax.devices()\n if config.subslice_shape and config.enable_single_controller and config.num_slices == 1:\n max_logging.log(f""Trying to create a subslice with shape: {config.subslice_shape}"")\n subslice_shape = tuple(int(x) for x in config.subslice_shape.split("",""))\n device_coords = [device.coords for device in devices]\n device_coords_np = np.array(device_coords)\n\n # Find the minimum coordinates to start the subslice\n min_coords = device_coords_np.min(axis=0)\n\n subslice_devices = []\n for device in devices:\n coords = device.coords\n if all(min_coords[i] <= coords[i] < min_coords[i] + subslice_shape[i] for i in range(len(subslice_shape))):\n subslice_devices.append(device)\n devices = subslice_devices\n\n num_devices = len(devices)\n num_slices = 1 if config.inference_benchmark_test else config.num_slices\n num_devices_per_slice = num_devices // num_slices\n\n multi_slice_env = num_slices > 1\n\n # Find possible unspecified parallelisms\n ici_parallelism = max_utils.fill_unspecified_mesh_axes(config.ici_parallelism.copy(), num_devices_per_slice, ""ICI"")\n\n allow_split_physical_axes = config.allow_split_physical_axes if config.allow_split_physical_axes else False\n\n if multi_slice_env:\n dcn_parallelism = max_utils.fill_unspecified_mesh_axes(config.dcn_parallelism.copy(), num_slices, ""DCN"")\n if max_utils.is_valid_custom_mesh(ici_parallelism, config.custom_mesh):\n mesh = max_utils.create_custom_device_mesh(ici_parallelism, dcn_parallelism, devices, config.custom_mesh)\n else:\n mesh = mesh_utils.create_hybrid_device_mesh(\n ici_parallelism,\n dcn_parallelism,\n devices,\n allow_split_physical_axes=allow_split_physical_axes,\n )\n else:\n if allow_split_physical_axes:\n if max_utils.is_valid_custom_mesh(ici_parallelism, config.custom_mesh):\n mesh = mesh_utils.create_device_mesh(\n [16, 16],\n devices,\n contiguous_submeshes=False,\n allow_split_physical_axes=False,\n )\n mesh = max_utils.reshape_mesh_to_rings(mesh, config.custom_mesh)\n mesh = np.reshape(mesh, ici_parallelism)\n else:\n mesh = mesh_utils.create_device_mesh(\n ici_parallelism,\n devices,\n contiguous_submeshes=False,\n allow_split_physical_axes=allow_split_physical_axes,\n )\n else:\n mesh = mesh_utils.create_device_mesh(\n ici_parallelism,\n devices,\n )\n if config.optimize_mesh_for_tpu_v6e:\n mesh = max_utils.optimize_mesh_for_tpu_v6e(mesh, devices)\n\n max_logging.log(f""Num_devices: {num_devices}, shape {mesh.shape}"")\n\n return mesh\n\n\n# Learning Rate Schedule\n# -----------------------------------------------------------------------------\n\n\ndef create_learning_rate_schedule(config):\n """"""Creates a warmup and cosine decay learning rate schedule:\n We take inspiration from Llama2's learning rate (LR) schedule, see https://arxiv.org/pdf/2307.09288.pdf section 2.2\n Learning rate schedule has either two or three parts:\n 1) Linear warmup from 0 to [learning_rate] over steps 0 to [learning_rate_schedule_steps * warmup_steps_fraction]\n 2) Cosine from [learning_rate] to [learning_rate * cosine_learning_rate_final_fraction] until learning_rate_schedule_steps\n 3) Constant learning rate of 0 from learning_rate_schedule_steps to steps.\n The zero learning rate section can be used to more accurately measure the fully trained model's performance.\n """"""\n\n def make_cos_schedule(init_lr, final_lr, len_steps):\n def schedule(step):\n pct = (step) / len_steps\n a = 0.5 * (jnp.cos(jnp.pi * pct) + 1)\n lr = init_lr * a + final_lr * (1 - a)\n return lr\n\n return schedule\n\n lr = config.learning_rate\n cos_final_lr = lr * config.cosine_learning_rate_final_fraction\n\n warmup_steps = int(config.learning_rate_schedule_steps * config.warmup_steps_fraction)\n cos_steps = config.learning_rate_schedule_steps - warmup_steps\n constant_zero_steps = config.steps - config.learning_rate_schedule_steps\n\n warmup_schedule = optax.linear_schedule(init_value=0.0, end_value=lr, transition_steps=warmup_steps)\n cos_schedule = make_cos_schedule(lr, cos_final_lr, cos_steps)\n constant_schedule = optax.constant_schedule(0.0)\n\n pieces = [warmup_schedule, cos_schedule]\n boundaries = [\n warmup_steps,\n warmup_steps + cos_steps,\n ]\n\n if constant_zero_steps > 0:\n pieces.append(constant_schedule)\n boundaries.append(warmup_steps + cos_steps + constant_zero_steps)\n\n return optax.join_schedules(pieces, boundaries)\n\n\ndef get_formatted_sharding_annotations(params, mesh=None):\n """"""\n Generates a readable string report of sharding annotations for all parameters.\n\n This function iterates through a PyTree of model parameters and inspects the\n sharding information attached to each parameter (leaf). It creates a\n human-readable summary that is useful for debugging sharding configurations.\n\n Args:\n params: The PyTree of model parameters to inspect.\n mesh: (Optional) The device mesh. If provided, its axis names and shape\n are included in the report for additional context.\n\n Returns:\n A single string containing the formatted report of sharding annotations\n for every parameter, with each entry on a new line.\n """"""\n # Initialize a list to hold the lines of the report, starting with a title.\n annotation_lines = [""Comprehensice Weight Sharding Annotations:""]\n\n # If a mesh object is provided, add its details to the report header.\n if mesh:\n annotation_lines.append(f""Mesh axes: {mesh.axis_names}, Mesh shape: {mesh.shape}"")\n annotation_lines.append(""-"" * 30)\n\n # Get a flattened list of all parameters (leaves) and their corresponding paths in the PyTree.\n all_params_leaves = jtu.tree_leaves_with_path(params)\n\n # Loop through each parameter leaf in the flattened list.\n for path, p_leaf in all_params_leaves:\n # Convert the parameter's path (a sequence of keys) into a readable string name.\n param_name_str = jtu.keystr(path)\n # Get the shape of the parameter as a string.\n shape_str = str(p_leaf.shape)\n # Set a default description for sharding, in case none is found.\n sharding_desc = ""N/A""\n\n # Check if the parameter leaf has a 'sharding' attribute.\n if hasattr(p_leaf, ""sharding""):\n # Case 1: Standard JAX sharding with a PartitionSpec.\n if hasattr(p_leaf.sharding, ""spec"") and p_leaf.sharding.spec is not None:\n # The spec is a tuple (PartitionSpec), format it for readability.\n spec_parts = []\n for item in p_leaf.sharding.spec:\n # Represent None as ""Replicated"" to make it explicit.\n spec_parts.append(str(item) if item is not None else ""Replicated"")\n sharding_desc = f""PartitionSpec({', '.join(spec_parts)})""\n # Case 2: The parameter is explicitly marked as fully replicated.\n elif hasattr(p_leaf.sharding, ""spec"") and p_leaf.sharding.spec is None:\n sharding_desc = ""Fully Replicated (spec is None)""\n # Case 3: A generic fallback if a sharding object exists but has no recognized spec attribute.\n else:\n # Print the string representation of the sharding object itself.\n sharding_desc = str(p_leaf.sharding)\n # Case 4: The parameter has no .sharding attribute at all.\n else:\n sharding_desc = ""No .sharding attribute found""\n\n # Append the formatted details for the current parameter to our list of lines.\n annotation_lines.append(f"" - Param: {param_name_str}\n"" f"" Shape: {shape_str}\n"" f"" Sharding: {sharding_desc}"")\n # Join all the collected lines into a single string, separated by newlines.\n return ""\n"".join(annotation_lines)\n\n\ndef get_physical_spec_no_fsdp(full_logical, mesh, logical_axis_rules):\n """"""\n Generates a physical sharding spec for fully replicated weights.\n\n This function computes a target sharding layout where model parameters are fully\n replicated across the 'fsdp' mesh axis. It starts with the original logical\n sharding and removes any rules that shard along the 'fsdp' or\n 'fsdp_transpose' axes.\n\n Replacing a sharding axis with `None` in a PartitionSpec instructs JAX to\n replicate the array data along that physical mesh dimension. The resulting\n specification is used as a target layout for an all-gather operation.\n\n Args:\n full_logical: A PyTree of logical PartitionSpecs for the model parameters.\n mesh: The JAX device mesh.\n logical_axis_rules: Rules for converting logical axes to physical mesh axes.\n\n Returns:\n A PyTree of physical `jax.sharding.NamedSharding` objects that describe a\n layout where parameters are fully gathered (replicated) across the 'fsdp'\n mesh axis.\n """"""\n\n def remove_fsdp_sharding(sharding_tree):\n """"""Recursively traverses the sharding tree to remove fsdp axes.""""""\n\n def _remove_fsdp_from_partition_spec(named_sharding):\n """"""Removes 'fsdp' and 'fsdp_transpose' from a PartitionSpec.""""""\n if isinstance(named_sharding, jax.sharding.NamedSharding):\n new_spec = []\n # Iterate through each axis in the original PartitionSpec.\n for axis in named_sharding.spec:\n if axis is None:\n new_spec.append(None)\n elif isinstance(axis, str):\n # If the axis is 'fsdp', replace it with None to signify replication.\n if axis not in (""fsdp"", ""fsdp_transpose""):\n new_spec.append(axis)\n else:\n new_spec.append(None)\n elif isinstance(axis, (list, tuple)):\n # If the axis is a collection, filter out 'fsdp'.\n new_axis = [a for a in axis if a not in (""fsdp"", ""fsdp_transpose"")]\n new_spec.append(tuple(new_axis))\n else:\n raise ValueError(f""Unsupported_axis_type: {type(axis)}"")\n # Return a new sharding object with the modified spec.\n return jax.sharding.NamedSharding(named_sharding.mesh, jax.sharding.PartitionSpec(*new_spec))\n return named_sharding\n\n return jax.tree.map(_remove_fsdp_from_partition_spec, sharding_tree)\n\n # Convert the high-level logical spec to a physical one using default rules.\n physical = nn.logical_to_mesh_sharding(full_logical, mesh=mesh, rules=logical_axis_rules)\n # Apply the function to remove the FSDP sharding, defining our target layout.\n physical_no_fsdp = remove_fsdp_sharding(physical)\n return physical_no_fsdp\n\n\ndef all_gather_over_fsdp(variables, sharding_info, mesh, logical_axis_rules):\n """"""Performs an all-gather on FSDP-sharded variables via a sharding constraint.\n This function triggers an all-gather operation on the model's parameters.\n It does so by applying a sharding constraint that specifies a fully\n replicated layout.\n\n The JAX compiler satisfies this constraint by automatically inserting the\n necessary `all-gather` collective communication operations into the\n computation graph, effectively gathering the sharded weights.\n\n Args:\n variables: The PyTree of model parameters, currently sharded across devices.\n sharding_info: The logical partition spec of the currently sharded `variables`.\n mesh: The JAX device mesh.\n logical_axis_rules: Rules for converting logical axes to physical mesh axes.\n\n Returns:\n The model's variables with the all-gather operation applied, resulting\n in the weights being fully replicated on all devices in the 'fsdp' mesh.\n """"""\n # Get the target physical layout (weights fully replicated).\n physical_constraint_no_fsdp = get_physical_spec_no_fsdp(sharding_info, mesh, logical_axis_rules)\n # Apply the constraint to the model's current variables. This tells JAX to\n # gather the weights into this layout.\n return jax.lax.with_sharding_constraint(variables, physical_constraint_no_fsdp)\n",python,tab +19,41000,"src/MaxText/maxtext_utils.py",20704,0,"",python,selection_command +20,51855,"src/MaxText/get_flops.py",0,0,"",python,tab +21,51857,"src/MaxText/get_flops.py",1190,36,"calculate_tflops_training_per_device",python,selection_command +22,52740,"src/MaxText/get_flops.py",1225,0,"",python,selection_command +23,53245,"src/MaxText/get_flops.py",1190,0,"",python,selection_command +24,53400,"src/MaxText/get_flops.py",1188,0,"",python,selection_command +25,53829,"src/MaxText/get_flops.py",1190,0,"",python,selection_command +26,56117,"src/MaxText/get_flops.py",1263,0,"",python,selection_command +27,56475,"src/MaxText/get_flops.py",1190,0,"",python,selection_command +28,59870,"src/MaxText/maxtext_utils.py",0,0,"",python,tab +29,59871,"src/MaxText/maxtext_utils.py",20704,0,"",python,selection_command +30,61449,"src/MaxText/get_flops.py",0,0,"",python,tab +31,61450,"src/MaxText/get_flops.py",1190,0,"",python,selection_command +32,62475,"src/MaxText/maxtext_utils.py",0,0,"",python,tab +33,62477,"src/MaxText/maxtext_utils.py",20704,0,"",python,selection_command +34,71033,"src/MaxText/get_flops.py",0,0,"",python,tab +35,71034,"src/MaxText/get_flops.py",678,36,"calculate_tflops_training_per_device",python,selection_command +36,71540,"src/MaxText/get_flops.py",713,0,"",python,selection_command +37,72628,"src/MaxText/get_flops.py",732,0,"",python,selection_command +38,72882,"src/MaxText/get_flops.py",761,0,"",python,selection_command +39,72913,"src/MaxText/get_flops.py",796,0,"",python,selection_command +40,72948,"src/MaxText/get_flops.py",815,0,"",python,selection_command +41,72980,"src/MaxText/get_flops.py",849,0,"",python,selection_command +42,73012,"src/MaxText/get_flops.py",853,0,"",python,selection_command +43,73046,"src/MaxText/get_flops.py",854,0,"",python,selection_command +44,73081,"src/MaxText/get_flops.py",872,0,"",python,selection_command +45,73113,"src/MaxText/get_flops.py",890,0,"",python,selection_command +46,73148,"src/MaxText/get_flops.py",909,0,"",python,selection_command +47,73181,"src/MaxText/get_flops.py",954,0,"",python,selection_command +48,73213,"src/MaxText/get_flops.py",976,0,"",python,selection_command +49,73247,"src/MaxText/get_flops.py",1015,0,"",python,selection_command +50,73280,"src/MaxText/get_flops.py",1034,0,"",python,selection_command +51,73466,"src/MaxText/get_flops.py",1144,0,"",python,selection_command +52,73785,"src/MaxText/get_flops.py",1190,0,"",python,selection_command +53,74199,"src/MaxText/maxtext_utils.py",0,0,"",python,tab +54,74201,"src/MaxText/maxtext_utils.py",20704,0,"",python,selection_command +55,78382,"src/MaxText/metric_logger.py",0,0,"# Copyright 2023–2025 Google LLC\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# https://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# pylint: disable=bare-except, consider-using-generator\n# pytype: disable=attribute-error\n""""""Logger that saves metrics to a local file, GCS and TensorBoard.""""""\n\nimport json\nimport os\nimport queue\n\nimport numpy as np\n\nimport jax\n\nfrom MaxText import max_logging\nfrom MaxText import max_utils\nfrom MaxText import maxtext_utils\nfrom MaxText.utils import gcs_utils\nfrom MaxText.gcp_workload_monitor import GCPWorkloadMonitor\nfrom MaxText.globals import EPS\n\nfrom collections import defaultdict\n\n\ndef _prepare_metrics_for_json(metrics, step, run_name):\n """"""Converts metric dictionary into json supported types (e.g. float)""""""\n metrics_dict = {val: float(metrics[""scalar""][val]) for val in metrics[""scalar""]}\n metrics_dict[""step""] = float(step)\n metrics_dict[""run_name""] = run_name\n return metrics_dict\n\n\ndef record_activation_metrics(output_metrics, intermediate_outputs, config):\n """"""Adds the activation metrics to the metrics dict""""""\n\n if config.scan_layers:\n metrics_dict = intermediate_outputs[""intermediates""][""decoder""][""decoder""]\n\n for layer_num in range(config.num_decoder_layers):\n output_metrics[""scalar""][f""activ_fraction_zero/layer_{layer_num:03d}""] = metrics_dict[""activation_fraction_zero""][0][\n layer_num\n ]\n output_metrics[""scalar""][f""activ_mean/layer_{layer_num:03d}""] = metrics_dict[""activation_mean""][0][layer_num]\n output_metrics[""scalar""][f""activ_stdev/layer_{layer_num:03d}""] = metrics_dict[""activation_stdev""][0][layer_num]\n else:\n for layer_num in range(config.num_decoder_layers):\n layer = intermediate_outputs[""intermediates""][""decoder""][f""layers_{layer_num}""]\n output_metrics[""scalar""][f""activ_fraction_zero/layer_{layer_num:03d}""] = layer[""activation_fraction_zero""][0]\n output_metrics[""scalar""][f""activ_mean/layer_{layer_num:03d}""] = layer[""activation_mean""][0]\n output_metrics[""scalar""][f""activ_stdev/layer_{layer_num:03d}""] = layer[""activation_stdev""][0]\n\n\nclass MetricLogger:\n """"""\n Logger for saving metrics to a local file, GCS and TensorBoard.\n """"""\n\n def __init__(self, config, learning_rate_schedule):\n self.writer = max_utils.initialize_summary_writer(config.tensorboard_dir, config.run_name)\n self.config = config\n self.metadata = {}\n self.running_gcs_metrics = [] if config.gcs_metrics else None\n self.performance_metric_queue = self.get_performance_metric_queue(config)\n self.learning_rate_schedule = learning_rate_schedule\n self.cumulative_eval_metrics = {""scalar"": defaultdict(float)}\n self.buffered_train_metrics = None\n\n def reset_eval_metrics(self):\n """"""Resets the cumulative metrics dictionary for a new evaluation run.""""""\n self.cumulative_eval_metrics = {""scalar"": defaultdict(float)}\n\n def write_metrics(self, metrics, step, is_training=True):\n """"""Entry point for all metrics writing in Train's Main.""""""\n if metrics:\n self.log_metrics(metrics, step, is_training)\n\n if self.config.enable_tensorboard:\n self.write_metrics_to_tensorboard(metrics, step, is_training)\n\n if self.config.metrics_file:\n self.write_metrics_locally(metrics, step)\n\n if self.config.gcs_metrics and jax.process_index() == 0:\n self.write_metrics_for_gcs(metrics, step, is_training)\n\n def log_metrics(self, metrics, step, is_training):\n """"""Logs metrics via max_logging.""""""\n if is_training:\n loss = metrics[""scalar""][""learning/loss""]\n log_message = (\n f""completed step: {step}, seconds: {metrics['scalar']['perf/step_time_seconds']:.3f}, ""\n f""TFLOP/s/device: {metrics['scalar']['perf/per_device_tflops_per_sec']:.3f}, ""\n f""Tokens/s/device: {metrics['scalar']['perf/per_device_tokens_per_sec']:.3f}, ""\n f""total_weights: {metrics['scalar']['learning/total_weights']}, ""\n f""loss: {loss:.3f}""\n )\n\n if self.config.mtp_num_layers > 0:\n mtp_loss = metrics[""scalar""].get(""learning/mtp_loss"", 0.0)\n main_model_loss = loss - mtp_loss\n log_message += f"", main_model_loss: {main_model_loss:.3f}, mtp_loss: {mtp_loss:.3f}""\n\n else:\n log_message = (\n f""eval metrics after step: {step},""\n f"" loss={metrics['scalar']['eval/avg_loss']:.3f},""\n f"" total_weights={metrics['scalar']['eval/total_weights']}""\n )\n\n if self.config.mtp_num_layers > 0:\n log_message += (\n f"", avg_mtp_loss={metrics['scalar']['eval/avg_mtp_loss']:.3f},""\n f"" avg_mtp_acceptance_rate={metrics['scalar']['eval/avg_mtp_acceptance_rate_percent']:.2f}%""\n )\n\n max_logging.log(log_message)\n\n def write_metrics_locally(self, metrics, step):\n """"""Writes metrics locally for testing.""""""\n with open(self.config.metrics_file, ""a"", encoding=""utf8"") as local_metrics_file:\n if step == 0:\n local_metrics_file.truncate(0)\n\n metrics_dict = _prepare_metrics_for_json(metrics, step, self.config.run_name)\n local_metrics_file.write(str(json.dumps(metrics_dict)) + ""\n"")\n\n def write_metrics_for_gcs(self, metrics, step, is_training):\n """"""Writes metrics to GCS.""""""\n metrics_dict_step = _prepare_metrics_for_json(metrics, step, self.config.run_name)\n self.running_gcs_metrics.append(metrics_dict_step)\n if is_training and (step + 1) % self.config.log_period == 0 or step == self.config.steps - 1:\n start_step = (step // self.config.log_period) * self.config.log_period\n metrics_filename = f""metrics_step_{start_step:06}_to_step_{step:06}.txt""\n with open(metrics_filename, ""wt"", encoding=""utf8"") as metrics_for_gcs:\n for metrics_step in self.running_gcs_metrics:\n metrics_for_gcs.write(str(json.dumps(metrics_step)) + ""\n"")\n\n gcs_filename = os.path.join(self.config.metrics_dir, metrics_filename)\n max_logging.log(f""Moving file {metrics_filename} to GCS..."")\n gcs_utils.upload_blob(gcs_filename, metrics_filename)\n max_logging.log(f""File {metrics_filename} moved successfully!"")\n self.running_gcs_metrics = [] # reset running_metrics to empty list\n\n def write_metrics_to_tensorboard(self, metrics, step, is_training):\n """"""Writes metrics to TensorBoard.""""""\n if jax.process_index() == 0:\n for metric_name in metrics.get(""scalar"", []):\n self.writer.add_scalar(metric_name, np.array(metrics[""scalar""][metric_name]), step)\n for metric_name in metrics.get(""scalars"", []):\n self.writer.add_scalars(metric_name, metrics[""scalars""][metric_name], step)\n\n if is_training:\n full_log = step % self.config.log_period == 0\n\n if full_log and jax.process_index() == 0:\n max_logging.log(f""To see full metrics 'tensorboard --logdir={self.config.tensorboard_dir}'"")\n self.writer.flush()\n\n def write_setup_info_to_tensorboard(self, params):\n """"""Writes setup information like train config params, num model params, and XLA flags to TensorBoard.""""""\n num_model_parameters = max_utils.calculate_num_params_from_pytree(params)\n self.metadata[""per_device_tflops""], _, _ = maxtext_utils.calculate_tflops_training_per_device(self.config)\n self.metadata[""per_device_tokens""] = maxtext_utils.calculate_tokens_training_per_device(self.config)\n max_logging.log(f""number parameters: {num_model_parameters/1e9:.3f} billion"")\n max_utils.add_text_to_summary_writer(""num_model_parameters"", str(num_model_parameters), self.writer)\n max_utils.add_text_to_summary_writer(""libtpu_init_args"", os.environ[""LIBTPU_INIT_ARGS""], self.writer)\n maxtext_utils.add_config_to_summary_writer(self.config, self.writer)\n\n def get_performance_metric_queue(self, config):\n """"""Records heartbeat metrics and performance metrics to GCP.""""""\n performance_metric_queue = None\n if config.report_heartbeat_metric_for_gcp_monitoring or config.report_performance_metric_for_gcp_monitoring:\n gcp_workload_monitor = GCPWorkloadMonitor(config.run_name)\n if config.report_heartbeat_metric_for_gcp_monitoring:\n gcp_workload_monitor.start_heartbeat_reporting_thread(config.heartbeat_reporting_interval_in_seconds)\n if config.report_performance_metric_for_gcp_monitoring:\n performance_metric_queue = queue.Queue()\n gcp_workload_monitor.start_performance_reporting_thread(performance_metric_queue)\n return performance_metric_queue\n\n def buffer_and_write_train_metrics(self, metrics, step, step_time_delta):\n """"""\n Buffers metrics for the current training step and simultaneously writes the training metrics\n for the previous step to GCS and/or TensorBoard. This buffering strategy allows for back-to-back\n execution of training steps, by overlapping data loading for step n with the execution of step n−1.\n This significantly boosts training efficiency.\n """"""\n if self.buffered_train_metrics is not None:\n (step_to_write, metrics_to_write) = self.buffered_train_metrics\n self.write_metrics(metrics_to_write, step_to_write)\n\n self.record_train_metrics(metrics, step, step_time_delta.total_seconds())\n self.buffered_train_metrics = (step, metrics)\n\n def record_train_metrics(self, metrics, step, step_time):\n """"""Records training metrics for the current step.""""""\n metrics[""scalar""].update({""perf/step_time_seconds"": step_time})\n metrics[""scalar""].update({""perf/per_device_tflops"": self.metadata[""per_device_tflops""]})\n metrics[""scalar""].update(\n {""perf/per_device_tflops_per_sec"": self.metadata[""per_device_tflops""] / step_time}\n )\n metrics[""scalar""].update({""perf/per_device_tokens"": self.metadata[""per_device_tokens""]})\n metrics[""scalar""].update(\n {""perf/per_device_tokens_per_sec"": self.metadata[""per_device_tokens""] / step_time}\n )\n metrics[""scalar""].update({""learning/current_learning_rate"": self.learning_rate_schedule(step)})\n if self.performance_metric_queue:\n self.performance_metric_queue.put(step_time)\n\n def record_eval_metrics(self, step, metrics=None, eval_step_count=None):\n """"""Records eval metrics and writes the metrics to GCS and/or to TensorBoard.""""""\n if metrics:\n self.cumulative_eval_metrics[""scalar""][""eval/total_loss""] += float(\n metrics[""scalar""].get(""evaluation/total_loss"", 0.0)\n )\n self.cumulative_eval_metrics[""scalar""][""eval/total_weights""] += float(\n metrics[""scalar""].get(""evaluation/total_weights"", 0.0)\n )\n self.cumulative_eval_metrics[""scalar""][""eval/moe_lb_loss""] += float(\n metrics[""scalar""].get(""evaluation/moe_lb_loss"", 0.0)\n )\n self.cumulative_eval_metrics[""scalar""][""eval/mtp_loss""] += float(metrics[""scalar""].get(""evaluation/mtp_loss"", 0.0))\n self.cumulative_eval_metrics[""scalar""][""eval/mtp_acceptance_rate_percent""] += float(\n metrics[""scalar""].get(""evaluation/mtp_acceptance_rate_percent"", 0.0)\n )\n if self.config.use_dpo:\n self.cumulative_eval_metrics[""scalar""][""eval/dpo_reward_accuracy""] += float(\n metrics[""scalar""].get(""evaluation/dpo_reward_accuracy"", 0.0)\n )\n\n if eval_step_count:\n eval_loss = self.cumulative_eval_metrics[""scalar""][""eval/total_loss""] / (\n self.cumulative_eval_metrics[""scalar""][""eval/total_weights""] + EPS\n )\n self.cumulative_eval_metrics[""scalar""][""eval/avg_loss""] = eval_loss\n self.cumulative_eval_metrics[""scalar""][""eval/avg_moe_lb_loss""] = (\n self.cumulative_eval_metrics[""scalar""][""eval/moe_lb_loss""] / eval_step_count\n )\n self.cumulative_eval_metrics[""scalar""][""eval/avg_mtp_loss""] = (\n self.cumulative_eval_metrics[""scalar""][""eval/mtp_loss""] / eval_step_count\n )\n self.cumulative_eval_metrics[""scalar""][""eval/avg_mtp_acceptance_rate_percent""] = (\n self.cumulative_eval_metrics[""scalar""][""eval/mtp_acceptance_rate_percent""] / eval_step_count\n )\n if self.config.use_dpo:\n self.cumulative_eval_metrics[""scalar""][""eval/dpo_reward_accuracy""] = (\n self.cumulative_eval_metrics[""scalar""][""eval/dpo_reward_accuracy""] / eval_step_count\n )\n\n self.write_metrics(self.cumulative_eval_metrics, step, is_training=False)\n\n def flush_metrics_and_cleanup(self):\n """"""\n This is a terminal operation that uploads any buffered metrics to GCS\n and/or TensorBoard before closing the writer objects. Once called, the\n logger instance should not be used to add or write more metrics as the\n underlying writer objects (e.g., TensorBoard SummaryWriter) will be closed.\n """"""\n if self.buffered_train_metrics is not None:\n (step_to_write, metrics_to_write) = self.buffered_train_metrics\n self.write_metrics(metrics_to_write, step_to_write)\n\n max_utils.close_summary_writer(self.writer)\n",python,tab +56,78386,"src/MaxText/metric_logger.py",7570,36,"calculate_tflops_training_per_device",python,selection_command +57,78605,"src/MaxText/metric_logger.py",7605,0,"",python,selection_command +58,79663,"src/MaxText/metric_logger.py",7570,0,"",python,selection_command +59,80462,"src/MaxText/metric_logger.py",7569,0,"",python,selection_command +60,80706,"src/MaxText/metric_logger.py",7556,0,"",python,selection_command +61,80734,"src/MaxText/metric_logger.py",7554,0,"",python,selection_command +62,80765,"src/MaxText/metric_logger.py",7552,0,"",python,selection_command +63,80972,"src/MaxText/metric_logger.py",7550,0,"",python,selection_command +64,81121,"src/MaxText/metric_logger.py",7549,0,"",python,selection_command +65,81381,"src/MaxText/metric_logger.py",7545,0,"",python,selection_command +66,81402,"src/MaxText/metric_logger.py",7528,0,"",python,selection_command +67,81556,"src/MaxText/metric_logger.py",7526,0,"",python,selection_command +68,81747,"src/MaxText/metric_logger.py",7518,0,"",python,selection_command +69,82608,"src/MaxText/metric_logger.py",7440,0,"",python,selection_command +70,82687,"src/MaxText/metric_logger.py",7331,0,"",python,selection_command +71,83491,"src/MaxText/metric_logger.py",7278,0,"",python,selection_command +72,157484,"src/MaxText/metric_logger.py",7331,0,"",python,selection_command +73,157663,"src/MaxText/metric_logger.py",7440,0,"",python,selection_command +74,157804,"src/MaxText/metric_logger.py",7518,0,"",python,selection_command +75,158070,"src/MaxText/metric_logger.py",7440,0,"",python,selection_command +76,170618,"src/MaxText/metric_logger.py",7518,0,"",python,selection_command +77,170707,"src/MaxText/metric_logger.py",7629,0,"",python,selection_command +78,170916,"src/MaxText/metric_logger.py",7518,0,"",python,selection_command diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-1fa9b85d-3794-4f3b-b7a0-5170b7d2faaa1762362332596-2025_11_05-18.05.39.648/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-1fa9b85d-3794-4f3b-b7a0-5170b7d2faaa1762362332596-2025_11_05-18.05.39.648/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..45a6a61797fa0ebe721d55bfb346126436a269bc --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-1fa9b85d-3794-4f3b-b7a0-5170b7d2faaa1762362332596-2025_11_05-18.05.39.648/source.csv @@ -0,0 +1,19 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,3,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",0,0,"/*---------------------------------------------------------------------------------------------\n * Copyright (c) Microsoft Corporation. All rights reserved.\n * Licensed under the MIT License. See License.txt in the project root for license information.\n *--------------------------------------------------------------------------------------------*/\n\n/** @jsxRuntime automatic */\n/** @jsxImportSource ../../../../prompt/jsx-runtime/ */\nimport { CopilotContentExclusionManager, StatusBarEvent } from '../../contentExclusion/contentExclusionManager';\nimport { ICompletionsContextService } from '../../context';\nimport { logger, LogTarget } from '../../logger';\n\nimport { IInstantiationService, ServicesAccessor } from '../../../../../../../util/vs/platform/instantiation/common/instantiation';\nimport { ICompletionsTelemetryService } from '../../../../bridge/src/completionsTelemetryServiceBridge';\nimport { DataPipe, VirtualPrompt } from '../../../../prompt/src/components/virtualPrompt';\nimport { TokenizerName } from '../../../../prompt/src/tokenization';\nimport { CancellationToken, Position } from '../../../../types/src';\nimport { CompletionState } from '../../completionState';\nimport { telemetryException, TelemetryWithExp } from '../../telemetry';\nimport { TextDocumentContents } from '../../textDocument';\nimport { CodeSnippets } from '../components/codeSnippets';\nimport { CompletionsContext } from '../components/completionsContext';\nimport { CompletionsPromptOk, CompletionsPromptRenderer } from '../components/completionsPromptRenderer';\nimport { ContextProviderBridge } from '../components/contextProviderBridge';\nimport { CurrentFile } from '../components/currentFile';\nimport { DocumentMarker } from '../components/marker';\nimport { RecentEdits } from '../components/recentEdits';\nimport { SimilarFiles } from '../components/similarFiles';\nimport { splitContextCompletionsPrompt } from '../components/splitContextPrompt';\nimport { SplitContextPromptRenderer } from '../components/splitContextPromptRenderer';\nimport { Traits } from '../components/traits';\nimport {\n\tContextProviderTelemetry,\n\tmatchContextItems,\n\tResolvedContextItem,\n\ttelemetrizeContextItems,\n\tuseContextProviderAPI,\n} from '../contextProviderRegistry';\nimport { getCodeSnippetsFromContextItems } from '../contextProviders/codeSnippets';\nimport {\n\tCodeSnippetWithId,\n\tSupportedContextItemWithId,\n\tTraitWithId,\n} from '../contextProviders/contextItemSchemas';\nimport { getTraitsFromContextItems, ReportTraitsTelemetry } from '../contextProviders/traits';\nimport { componentStatisticsToPromptMatcher, ContextProviderStatistics } from '../contextProviderStatistics';\nimport {\n\t_contextTooShort,\n\t_copilotContentExclusion,\n\t_promptCancelled,\n\t_promptError,\n\tgetPromptOptions,\n\tMIN_PROMPT_CHARS,\n\tPromptResponse,\n\ttrimLastLine,\n} from '../prompt';\nimport { isIncludeNeighborFilesActive } from '../similarFiles/neighborFiles';\nimport {\n\tCompletionsPromptFactory,\n\tCompletionsPromptOptions,\n\tPromptOpts,\n} from './completionsPromptFactory';\n\nexport type CompletionRequestDocument = TextDocumentContents;\n\nexport type CompletionRequestData = {\n\tdocument: CompletionRequestDocument;\n\tposition: Position;\n\ttelemetryData: TelemetryWithExp;\n\tcancellationToken?: CancellationToken;\n\t// see inlineCompletions data param\n\tdata?: unknown;\n\t// Context provider items\n\ttraits?: TraitWithId[];\n\tcodeSnippets?: CodeSnippetWithId[];\n\tturnOffSimilarFiles?: boolean;\n\tsuffixMatchThreshold?: number;\n\tmaxPromptTokens: number;\n\ttokenizer?: TokenizerName;\n};\n\nexport function isCompletionRequestData(data: unknown): data is CompletionRequestData {\n\tif (!data || typeof data !== 'object') { return false; }\n\n\tconst req = data as Partial;\n\n\t// Check document\n\tif (!req.document) { return false; }\n\n\t// Check position\n\tif (!req.position) { return false; }\n\tif (req.position.line === undefined) { return false; }\n\tif (req.position.character === undefined) { return false; }\n\n\t// Check telemetryData\n\tif (!req.telemetryData) { return false; }\n\n\treturn true;\n}\n\nexport enum PromptOrdering {\n\tDefault = 'default',\n\tSplitContext = 'splitContext',\n}\n\ntype DeclarativePromptFunction = typeof defaultCompletionsPrompt;\ntype AvailableDeclarativePrompts = {\n\t[K in PromptOrdering]: {\n\t\tpromptFunction: DeclarativePromptFunction;\n\t\trenderer: typeof CompletionsPromptRenderer;\n\t};\n};\n\nconst availableDeclarativePrompts: AvailableDeclarativePrompts = {\n\t[PromptOrdering.Default]: {\n\t\tpromptFunction: defaultCompletionsPrompt,\n\t\trenderer: CompletionsPromptRenderer,\n\t},\n\t[PromptOrdering.SplitContext]: {\n\t\tpromptFunction: splitContextCompletionsPrompt,\n\t\trenderer: SplitContextPromptRenderer,\n\t},\n};\n\n// The weights mimic the PromptPriorityList from prompt/src/wishlist.ts\nfunction defaultCompletionsPrompt(accessor: ServicesAccessor) {\n\tconst ctx = accessor.get(ICompletionsContextService);\n\treturn (\n\t\t<>\n\t\t\t\n\t\t\t\t\n\t\t\t\t\n\t\t\t\t\n\t\t\t\t\n\t\t\t\t\n\t\t\t\n\t\t\t\n\t\t\n\t);\n}\n\n// Exported for testing\nexport class ComponentsCompletionsPromptFactory implements CompletionsPromptFactory {\n\tprivate virtualPrompt: VirtualPrompt;\n\tprivate pipe: DataPipe;\n\tprivate renderer: CompletionsPromptRenderer;\n\tprivate promptOrdering: PromptOrdering;\n\tprivate logTarget;\n\n\tconstructor(\n\t\tvirtualPrompt: VirtualPrompt | undefined = undefined,\n\t\tordering: PromptOrdering | undefined = undefined,\n\t\t@ICompletionsContextService private readonly ctx: ICompletionsContextService,\n\t\t@IInstantiationService private readonly instantiationService: IInstantiationService,\n\t\t@ICompletionsTelemetryService private readonly completionsTelemetryService: ICompletionsTelemetryService,\n\t) {\n\t\tthis.logTarget = this.ctx.get(LogTarget);\n\t\tthis.promptOrdering = ordering ?? PromptOrdering.Default;\n\t\tthis.virtualPrompt = virtualPrompt ?? new VirtualPrompt(this.completionsPrompt());\n\t\tthis.pipe = this.virtualPrompt.createPipe();\n\t\tthis.renderer = this.getRenderer();\n\t}\n\n\tasync prompt(opts: CompletionsPromptOptions, cancellationToken?: CancellationToken): Promise {\n\t\ttry {\n\t\t\treturn await this.createPromptUnsafe(opts, cancellationToken);\n\t\t} catch (e) {\n\t\t\treturn this.errorPrompt(e as Error);\n\t\t}\n\t}\n\n\tasync createPromptUnsafe(\n\t\t{ completionId, completionState, telemetryData, promptOpts }: CompletionsPromptOptions,\n\t\tcancellationToken?: CancellationToken\n\t): Promise {\n\t\tconst { maxPromptLength, suffixPercent, suffixMatchThreshold } = this.instantiationService.invokeFunction(getPromptOptions,\n\t\t\ttelemetryData,\n\t\t\tcompletionState.textDocument.detectedLanguageId\n\t\t);\n\n\t\tconst failFastPrompt = await this.failFastPrompt(\n\t\t\tcompletionState.textDocument,\n\t\t\tcompletionState.position,\n\t\t\tsuffixPercent,\n\t\t\tcancellationToken\n\t\t);\n\t\tif (failFastPrompt) {\n\t\t\treturn failFastPrompt;\n\t\t}\n\n\t\t// TODO: Prompt ordering changes are triggered by ExP changes.\n\t\t// TODO@benibenj remove this as its always true (except in tests)\n\t\tconst promptOrdering = promptOpts?.separateContext ? PromptOrdering.SplitContext : PromptOrdering.Default;\n\t\tthis.setPromptOrdering(promptOrdering);\n\n\t\tconst start = performance.now();\n\n\t\tconst { traits, codeSnippets, turnOffSimilarFiles, resolvedContextItems } = await this.resolveContext(\n\t\t\tcompletionId,\n\t\t\tcompletionState,\n\t\t\ttelemetryData,\n\t\t\tcancellationToken,\n\t\t\tpromptOpts\n\t\t);\n\n\t\tawait this.updateComponentData(\n\t\t\tcompletionState.textDocument,\n\t\t\tcompletionState.position,\n\t\t\ttraits,\n\t\t\tcodeSnippets,\n\t\t\ttelemetryData,\n\t\t\tturnOffSimilarFiles,\n\t\t\tmaxPromptLength,\n\t\t\tcancellationToken,\n\t\t\tpromptOpts,\n\t\t\tsuffixMatchThreshold,\n\t\t\tpromptOpts?.tokenizer\n\t\t);\n\n\t\tif (cancellationToken?.isCancellationRequested) {\n\t\t\treturn _promptCancelled;\n\t\t}\n\n\t\tconst snapshot = this.virtualPrompt.snapshot(cancellationToken);\n\t\tconst snapshotStatus = snapshot.status;\n\t\tif (snapshotStatus === 'cancelled') {\n\t\t\treturn _promptCancelled;\n\t\t} else if (snapshotStatus === 'error') {\n\t\t\treturn this.errorPrompt(snapshot.error);\n\t\t}\n\n\t\tconst rendered = this.renderer.render(\n\t\t\tsnapshot.snapshot!,\n\t\t\t{\n\t\t\t\tdelimiter: '\n',\n\t\t\t\ttokenizer: promptOpts?.tokenizer,\n\t\t\t\tpromptTokenLimit: maxPromptLength,\n\t\t\t\tsuffixPercent: suffixPercent,\n\t\t\t\tlanguageId: completionState.textDocument.detectedLanguageId,\n\t\t\t},\n\t\t\tcancellationToken\n\t\t);\n\t\tif (rendered.status === 'cancelled') {\n\t\t\treturn _promptCancelled;\n\t\t} else if (rendered.status === 'error') {\n\t\t\treturn this.errorPrompt(rendered.error);\n\t\t}\n\n\t\tconst [prefix, trailingWs] = trimLastLine(rendered.prefix);\n\t\tconst renderedTrimmed = { ...rendered, prefix };\n\n\t\tlet contextProvidersTelemetry: ContextProviderTelemetry[] | undefined = undefined;\n\t\tconst languageId = completionState.textDocument.detectedLanguageId;\n\t\tif (this.instantiationService.invokeFunction(useContextProviderAPI, languageId, telemetryData)) {\n\t\t\tconst promptMatcher = componentStatisticsToPromptMatcher(rendered.metadata.componentStatistics);\n\t\t\tthis.ctx\n\t\t\t\t.get(ContextProviderStatistics)\n\t\t\t\t.getStatisticsForCompletion(completionId)\n\t\t\t\t.computeMatch(promptMatcher);\n\t\t\tcontextProvidersTelemetry = telemetrizeContextItems(this.ctx, completionId, resolvedContextItems);\n\t\t\t// To support generating context provider metrics of completion in COffE.\n\t\t\tlogger.debug(this.logTarget, `Context providers telemetry: '${JSON.stringify(contextProvidersTelemetry)}'`);\n\t\t}\n\t\tconst end = performance.now();\n\t\tthis.resetIfEmpty(rendered);\n\t\treturn this.successPrompt(renderedTrimmed, end, start, trailingWs, contextProvidersTelemetry);\n\t}\n\n\tprivate async updateComponentData(\n\t\ttextDocument: CompletionRequestDocument,\n\t\tposition: Position,\n\t\ttraits: TraitWithId[] | undefined,\n\t\tcodeSnippets: CodeSnippetWithId[] | undefined,\n\t\ttelemetryData: TelemetryWithExp,\n\t\tturnOffSimilarFiles: boolean,\n\t\tmaxPromptLength: number,\n\t\tcancellationToken?: CancellationToken,\n\t\topts: PromptOpts = {},\n\t\tsuffixMatchThreshold?: number,\n\t\ttokenizer?: TokenizerName\n\t) {\n\t\tconst completionRequestData = this.createRequestData(\n\t\t\ttextDocument,\n\t\t\tposition,\n\t\t\ttelemetryData,\n\t\t\tcancellationToken,\n\t\t\topts,\n\t\t\tmaxPromptLength,\n\t\t\ttraits,\n\t\t\tcodeSnippets,\n\t\t\tturnOffSimilarFiles,\n\t\t\tsuffixMatchThreshold,\n\t\t\ttokenizer\n\t\t);\n\t\tawait this.pipe.pump(completionRequestData);\n\t}\n\n\tprivate async resolveContext(\n\t\tcompletionId: string,\n\t\tcompletionState: CompletionState,\n\t\ttelemetryData: TelemetryWithExp,\n\t\tcancellationToken?: CancellationToken,\n\t\topts: PromptOpts = {}\n\t): Promise<{\n\t\ttraits: TraitWithId[] | undefined;\n\t\tcodeSnippets: CodeSnippetWithId[] | undefined;\n\t\tturnOffSimilarFiles: boolean;\n\t\tresolvedContextItems: ResolvedContextItem[];\n\t}> {\n\t\tlet resolvedContextItems: ResolvedContextItem[] = [];\n\t\tlet traits: TraitWithId[] | undefined;\n\t\tlet codeSnippets: CodeSnippetWithId[] | undefined;\n\t\tlet turnOffSimilarFiles = false;\n\t\tif (this.instantiationService.invokeFunction(useContextProviderAPI, completionState.textDocument.detectedLanguageId, telemetryData)) {\n\t\t\tresolvedContextItems = await this.ctx.get(ContextProviderBridge).resolution(completionId);\n\t\t\tconst { textDocument } = completionState;\n\t\t\t// Turn off neighboring files if:\n\t\t\t// - it's not explicitly enabled via EXP flag\n\t\t\t// - there are matched context providers\n\t\t\tconst matchedContextItems = resolvedContextItems.filter(matchContextItems);\n\t\t\tif (!this.instantiationService.invokeFunction(similarFilesEnabled, textDocument.detectedLanguageId, matchedContextItems, telemetryData)) {\n\t\t\t\tturnOffSimilarFiles = true;\n\t\t\t}\n\n\t\t\ttraits = await this.instantiationService.invokeFunction(getTraitsFromContextItems, completionId, matchedContextItems);\n\t\t\tvoid this.instantiationService.invokeFunction(ReportTraitsTelemetry,\n\t\t\t\t`contextProvider.traits`,\n\t\t\t\ttraits,\n\t\t\t\ttextDocument.detectedLanguageId,\n\t\t\t\ttextDocument.detectedLanguageId, // TextDocumentContext does not have clientLanguageId\n\t\t\t\ttelemetryData\n\t\t\t);\n\n\t\t\tcodeSnippets = await this.instantiationService.invokeFunction(getCodeSnippetsFromContextItems,\n\t\t\t\tcompletionId,\n\t\t\t\tmatchedContextItems,\n\t\t\t\ttextDocument.detectedLanguageId\n\t\t\t);\n\t\t}\n\t\treturn { traits, codeSnippets, turnOffSimilarFiles, resolvedContextItems };\n\t}\n\n\tprivate async failFastPrompt(\n\t\ttextDocument: TextDocumentContents,\n\t\tposition: Position,\n\t\tsuffixPercent: number,\n\t\tcancellationToken: CancellationToken | undefined\n\t) {\n\t\tif (cancellationToken?.isCancellationRequested) {\n\t\t\treturn _promptCancelled;\n\t\t}\n\t\tif (\n\t\t\t(\n\t\t\t\tawait this.ctx\n\t\t\t\t\t.get(CopilotContentExclusionManager)\n\t\t\t\t\t.evaluate(textDocument.uri, textDocument.getText(), StatusBarEvent.UPDATE)\n\t\t\t).isBlocked\n\t\t) {\n\t\t\treturn _copilotContentExclusion;\n\t\t}\n\n\t\tconst eligibleChars = suffixPercent > 0 ? textDocument.getText().length : textDocument.offsetAt(position);\n\t\tif (eligibleChars < MIN_PROMPT_CHARS) {\n\t\t\t// Too short context\n\t\t\treturn _contextTooShort;\n\t\t}\n\t}\n\n\tprivate createRequestData(\n\t\ttextDocument: CompletionRequestDocument,\n\t\tposition: Position,\n\t\ttelemetryData: TelemetryWithExp,\n\t\tcancellationToken: CancellationToken | undefined,\n\t\topts: PromptOpts,\n\t\tmaxPromptLength: number,\n\t\ttraits?: TraitWithId[],\n\t\tcodeSnippets?: CodeSnippetWithId[],\n\t\tturnOffSimilarFiles?: boolean,\n\t\tsuffixMatchThreshold?: number,\n\t\ttokenizer?: TokenizerName\n\t): CompletionRequestData {\n\t\treturn {\n\t\t\tdocument: textDocument,\n\t\t\tposition,\n\t\t\ttelemetryData,\n\t\t\tcancellationToken,\n\t\t\tdata: opts.data,\n\t\t\ttraits,\n\t\t\tcodeSnippets,\n\t\t\tturnOffSimilarFiles,\n\t\t\tsuffixMatchThreshold,\n\t\t\tmaxPromptTokens: maxPromptLength,\n\t\t\ttokenizer,\n\t\t};\n\t}\n\n\tprivate resetIfEmpty(rendered: CompletionsPromptOk) {\n\t\tif (rendered.prefix.length === 0 && rendered.suffix.length === 0) {\n\t\t\tthis.reset();\n\t\t}\n\t}\n\n\tprivate successPrompt(\n\t\trendered: CompletionsPromptOk,\n\t\tend: number,\n\t\tstart: number,\n\t\ttrailingWs: string,\n\t\tcontextProvidersTelemetry?: ContextProviderTelemetry[]\n\t): PromptResponse {\n\t\treturn {\n\t\t\ttype: 'prompt',\n\t\t\tprompt: {\n\t\t\t\tprefix: rendered.prefix,\n\t\t\t\tprefixTokens: rendered.prefixTokens,\n\t\t\t\tsuffix: rendered.suffix,\n\t\t\t\tsuffixTokens: rendered.suffixTokens,\n\t\t\t\tcontext: rendered.context,\n\t\t\t\tisFimEnabled: rendered.suffix.length > 0,\n\t\t\t},\n\t\t\tcomputeTimeMs: end - start,\n\t\t\ttrailingWs,\n\t\t\tneighborSource: new Map(),\n\t\t\tmetadata: rendered.metadata,\n\t\t\tcontextProvidersTelemetry,\n\t\t};\n\t}\n\n\tprivate errorPrompt(error: Error): PromptResponse {\n\t\ttelemetryException(this.completionsTelemetryService, error, 'PromptComponents.CompletionsPromptFactory');\n\t\tthis.reset();\n\t\treturn _promptError;\n\t}\n\n\tprivate reset() {\n\t\tthis.renderer = this.getRenderer();\n\t\tthis.virtualPrompt = new VirtualPrompt(this.completionsPrompt());\n\t\tthis.pipe = this.virtualPrompt.createPipe();\n\t}\n\n\tprivate setPromptOrdering(ordering: PromptOrdering) {\n\t\tif (this.promptOrdering !== ordering) {\n\t\t\tthis.promptOrdering = ordering;\n\t\t\tthis.reset();\n\t\t}\n\t}\n\n\tprivate completionsPrompt() {\n\t\tconst promptFunction =\n\t\t\tavailableDeclarativePrompts[this.promptOrdering]?.promptFunction ?? defaultCompletionsPrompt;\n\t\treturn this.instantiationService.invokeFunction(promptFunction);\n\t}\n\n\tprivate getRenderer() {\n\t\tconst promptInfo =\n\t\t\tavailableDeclarativePrompts[this.promptOrdering] ?? availableDeclarativePrompts[PromptOrdering.Default];\n\t\treturn new promptInfo.renderer();\n\t}\n}\n\n// Similar files is enabled if:\n// - the languageId is C/C++.\n// - it's explicitly enabled via EXP flag or config.\n// - no code snippets are provided (which includes the case when all providers error).\nfunction similarFilesEnabled(\n\taccessor: ServicesAccessor,\n\tdetectedLanguageId: string,\n\tmatchedContextItems: ResolvedContextItem[],\n\ttelemetryData: TelemetryWithExp\n) {\n\tconst cppLanguageIds = ['cpp', 'c'];\n\tconst includeNeighboringFiles =\n\t\tisIncludeNeighborFilesActive(accessor, detectedLanguageId, telemetryData) || cppLanguageIds.includes(detectedLanguageId);\n\treturn (\n\t\tincludeNeighboringFiles || !matchedContextItems.some(ci => ci.data.some(item => item.type === 'CodeSnippet'))\n\t);\n}\n",typescriptreact,tab +2,289,"tasks",0,0,"",Log,tab +3,290,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",0,0,"",typescriptreact,tab +4,343,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"6:05:39 PM [info] Activating crowd-code\n6:05:39 PM [info] Recording started\n6:05:39 PM [info] Initializing git provider using file system watchers...\n6:05:39 PM [info] Git repository found\n6:05:39 PM [info] Git provider initialized successfully\n6:05:39 PM [info] Initial git state: [object Object]\n",Log,tab +5,12297,"extension-output-pdoom-org.crowd-code-#1-crowd-code",298,0,"",Log,selection_mouse +6,13219,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",0,0,"",typescriptreact,tab +7,347453,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",5114,0,"",typescriptreact,selection_command +8,347571,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",5070,0,"",typescriptreact,selection_command +9,1004602,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",5026,0,"",typescriptreact,selection_command +10,1004729,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",4982,0,"",typescriptreact,selection_command +11,1004853,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",4954,0,"",typescriptreact,selection_command +12,1005293,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",4908,0,"",typescriptreact,selection_command +13,2090202,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +14,2092221,"TERMINAL",0,0,"",,terminal_focus +15,2092222,"src/extension/completions-core/vscode-node/lib/src/prompt/completionsPromptFactory/componentsCompletionsPromptFactory.tsx",0,0,"",typescriptreact,tab +16,2093995,"TERMINAL",0,0,"cd ..",,terminal_command +17,2094451,"TERMINAL",0,0,"ls",,terminal_command +18,2094466,"TERMINAL",0,0,"]633;Ccleanrl crowd-code-player crowd-pilot jafar jax_cache maxtext npm-global oai-compatible-copilot sbatch-runner Stoix vscode-crowd-pilot-chat\r\n]0;franz.srambical@hai-login2:~",,terminal_output diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-4b7a193b-6fd0-48b6-a605-a1ce6ba179221764439942282-2025_11_29-19.12.25.371/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-4b7a193b-6fd0-48b6-a605-a1ce6ba179221764439942282-2025_11_29-19.12.25.371/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..e4d6dc2d110eef6b56ebddd11993850d363df6b2 --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-4b7a193b-6fd0-48b6-a605-a1ce6ba179221764439942282-2025_11_29-19.12.25.371/source.csv @@ -0,0 +1,10 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +2,106,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"7:12:25 PM [info] Activating crowd-code\n7:12:25 PM [info] Recording started\n7:12:25 PM [info] Initializing git provider using file system watchers...\n7:12:25 PM [info] No workspace folder found\n",Log,tab +3,2027,"extension-output-pdoom-org.crowd-code-#1-crowd-code",194,0,"7:12:27 PM [info] Retrying git provider initialization...\n7:12:27 PM [info] No workspace folder found\n",Log,content +4,9498,"Untitled-1",0,0,"",plaintext,tab +5,11149,"TERMINAL",0,0,"Test",,terminal_focus +6,11153,"Untitled-1",0,0,"/* crowd-pilot: insert start */\nline A\nline B\n/* crowd-pilot: insert end */\n",plaintext,content +7,18176,"Untitled-1",46,0,"",plaintext,selection_command +8,18275,"Untitled-1",39,0,"",plaintext,selection_command +9,18422,"Untitled-1",32,0,"",plaintext,selection_command +10,18611,"Untitled-1",0,0,"",plaintext,selection_command diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-549b7320-9591-428d-ba6a-d3d8fb65a0001764416891089-2025_11_29-12.48.19.667/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-549b7320-9591-428d-ba6a-d3d8fb65a0001764416891089-2025_11_29-12.48.19.667/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..b8a2051e4518b4086abf964af8b9ba1275f47d5d --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-549b7320-9591-428d-ba6a-d3d8fb65a0001764416891089-2025_11_29-12.48.19.667/source.csv @@ -0,0 +1,1006 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +2,103,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"12:48:19 PM [info] Activating crowd-code\n12:48:19 PM [info] Recording started\n12:48:19 PM [info] Initializing git provider using file system watchers...\n12:48:19 PM [info] Git repository found\n12:48:19 PM [info] Git provider initialized successfully\n12:48:19 PM [info] Initial git state: [object Object]\n",Log,tab +3,1736,"TERMINAL",0,0,"",,terminal_focus +4,4333,"TERMINAL",0,0,"zsh",,terminal_focus +5,20864,"TERMINAL",0,0,"cd webview-ui/sto",,terminal_command +6,20866,"TERMINAL",0,0,"]633;Ccd: no such file or directory: webview-ui/sto\r\n% \r \r",,terminal_output +7,22604,"TERMINAL",0,0,"cd webview-ui/sto",,terminal_command +8,22605,"TERMINAL",0,0,"]633;Ccd: no such file or directory: webview-ui/sto\r\n% \r \r",,terminal_output +9,28427,"TERMINAL",0,0,"cd webview-ui/.storybook",,terminal_command +10,28428,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +11,31719,"TERMINAL",0,0,"npm run storybook",,terminal_command +12,31770,"TERMINAL",0,0,"]633;C",,terminal_output +13,32548,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\nsh: storybook: command not found\r\n⠙% \r \r",,terminal_output +14,201178,"TERMINAL",0,0,"cd ..",,terminal_command +15,201179,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +16,202166,"TERMINAL",0,0,"cd ..",,terminal_command +17,202171,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +18,208467,"TERMINAL",0,0,"npm install",,terminal_command +19,208518,"TERMINAL",0,0,"]633;C",,terminal_output +20,211999,"TERMINAL",0,0,"⠙",,terminal_output +21,212258,"TERMINAL",0,0,"npm warn EBADENGINE Unsupported engine {\r\nnpm warn EBADENGINE package: '@sap/xsenv@6.0.0',\r\nnpm warn EBADENGINE required: { node: '^20.0.0 || ^22.0.0 || ^24.0.0' },\r\nnpm warn EBADENGINE current: { node: 'v23.11.0', npm: '10.9.2' }\r\nnpm warn EBADENGINE }\r\n⠹npm warn EBADENGINE Unsupported engine {\r\nnpm warn EBADENGINE package: 'opossum@9.0.0',\r\nnpm warn EBADENGINE required: { node: '^24 || ^22 || ^20' },\r\nnpm warn EBADENGINE current: { node: 'v23.11.0', npm: '10.9.2' }\r\nnpm warn EBADENGINE }\r\n⠹",,terminal_output +22,212340,"TERMINAL",0,0,"⠸",,terminal_output +23,212476,"TERMINAL",0,0,"⠼",,terminal_output +24,212566,"TERMINAL",0,0,"⠴",,terminal_output +25,212680,"TERMINAL",0,0,"⠦",,terminal_output +26,212761,"TERMINAL",0,0,"⠧",,terminal_output +27,212842,"TERMINAL",0,0,"⠇",,terminal_output +28,212922,"TERMINAL",0,0,"⠏",,terminal_output +29,213003,"TERMINAL",0,0,"⠋",,terminal_output +30,213084,"TERMINAL",0,0,"⠙",,terminal_output +31,213163,"TERMINAL",0,0,"⠹",,terminal_output +32,213244,"TERMINAL",0,0,"⠸",,terminal_output +33,213325,"TERMINAL",0,0,"⠼",,terminal_output +34,213403,"TERMINAL",0,0,"⠴",,terminal_output +35,213484,"TERMINAL",0,0,"⠦",,terminal_output +36,213784,"TERMINAL",0,0,"⠧",,terminal_output +37,213862,"TERMINAL",0,0,"⠇",,terminal_output +38,213949,"TERMINAL",0,0,"⠏",,terminal_output +39,214040,"TERMINAL",0,0,"⠋npm warn deprecated rimraf@3.0.2: Rimraf versions prior to v4 are no longer supported\r\n⠋npm warn deprecated rimraf@3.0.2: Rimraf versions prior to v4 are no longer supported\r\n⠋npm warn deprecated rimraf@3.0.2: Rimraf versions prior to v4 are no longer supported\r\n⠋",,terminal_output +40,214240,"TERMINAL",0,0,"⠙npm warn deprecated glob@7.2.3: Glob versions prior to v9 are no longer supported\r\n⠙npm warn deprecated glob@7.2.3: Glob versions prior to v9 are no longer supported\r\n⠙npm warn deprecated glob@7.2.3: Glob versions prior to v9 are no longer supported\r\n⠙npm warn deprecated glob@7.2.3: Glob versions prior to v9 are no longer supported\r\n⠙",,terminal_output +41,214321,"TERMINAL",0,0,"⠹",,terminal_output +42,214401,"TERMINAL",0,0,"⠸",,terminal_output +43,214483,"TERMINAL",0,0,"⠼",,terminal_output +44,214563,"TERMINAL",0,0,"⠴",,terminal_output +45,214643,"TERMINAL",0,0,"⠦",,terminal_output +46,214722,"TERMINAL",0,0,"⠧",,terminal_output +47,214803,"TERMINAL",0,0,"⠇",,terminal_output +48,214882,"TERMINAL",0,0,"⠏",,terminal_output +49,214963,"TERMINAL",0,0,"⠋",,terminal_output +50,215043,"TERMINAL",0,0,"⠙",,terminal_output +51,215122,"TERMINAL",0,0,"⠹",,terminal_output +52,215200,"TERMINAL",0,0,"⠸",,terminal_output +53,215288,"TERMINAL",0,0,"⠼",,terminal_output +54,215364,"TERMINAL",0,0,"⠴",,terminal_output +55,215443,"TERMINAL",0,0,"⠦",,terminal_output +56,215524,"TERMINAL",0,0,"⠧",,terminal_output +57,215604,"TERMINAL",0,0,"⠇",,terminal_output +58,215684,"TERMINAL",0,0,"⠏",,terminal_output +59,215765,"TERMINAL",0,0,"⠋",,terminal_output +60,215845,"TERMINAL",0,0,"⠙",,terminal_output +61,215925,"TERMINAL",0,0,"⠹",,terminal_output +62,216004,"TERMINAL",0,0,"⠸",,terminal_output +63,216085,"TERMINAL",0,0,"⠼",,terminal_output +64,216164,"TERMINAL",0,0,"⠴",,terminal_output +65,216254,"TERMINAL",0,0,"⠦",,terminal_output +66,216326,"TERMINAL",0,0,"⠧",,terminal_output +67,216408,"TERMINAL",0,0,"⠇",,terminal_output +68,216487,"TERMINAL",0,0,"⠏",,terminal_output +69,216567,"TERMINAL",0,0,"⠋",,terminal_output +70,216647,"TERMINAL",0,0,"⠙",,terminal_output +71,216728,"TERMINAL",0,0,"⠹",,terminal_output +72,216809,"TERMINAL",0,0,"⠸",,terminal_output +73,216888,"TERMINAL",0,0,"⠼",,terminal_output +74,216971,"TERMINAL",0,0,"⠴",,terminal_output +75,217050,"TERMINAL",0,0,"⠦",,terminal_output +76,217132,"TERMINAL",0,0,"⠧",,terminal_output +77,217211,"TERMINAL",0,0,"⠇",,terminal_output +78,217294,"TERMINAL",0,0,"⠏",,terminal_output +79,217371,"TERMINAL",0,0,"⠋",,terminal_output +80,217452,"TERMINAL",0,0,"⠙",,terminal_output +81,217532,"TERMINAL",0,0,"⠹",,terminal_output +82,217613,"TERMINAL",0,0,"⠸",,terminal_output +83,217694,"TERMINAL",0,0,"⠼",,terminal_output +84,217774,"TERMINAL",0,0,"⠴",,terminal_output +85,217856,"TERMINAL",0,0,"⠦",,terminal_output +86,217936,"TERMINAL",0,0,"⠧",,terminal_output +87,218018,"TERMINAL",0,0,"⠇",,terminal_output +88,218098,"TERMINAL",0,0,"⠏",,terminal_output +89,218176,"TERMINAL",0,0,"⠋",,terminal_output +90,218258,"TERMINAL",0,0,"⠙",,terminal_output +91,218339,"TERMINAL",0,0,"⠹",,terminal_output +92,218427,"TERMINAL",0,0,"⠸",,terminal_output +93,218501,"TERMINAL",0,0,"⠼",,terminal_output +94,218582,"TERMINAL",0,0,"⠴",,terminal_output +95,218662,"TERMINAL",0,0,"⠦",,terminal_output +96,218742,"TERMINAL",0,0,"⠧",,terminal_output +97,218821,"TERMINAL",0,0,"⠇",,terminal_output +98,218900,"TERMINAL",0,0,"⠏",,terminal_output +99,218982,"TERMINAL",0,0,"⠋",,terminal_output +100,219063,"TERMINAL",0,0,"⠙",,terminal_output +101,219142,"TERMINAL",0,0,"⠹",,terminal_output +102,219223,"TERMINAL",0,0,"⠸",,terminal_output +103,219304,"TERMINAL",0,0,"⠼",,terminal_output +104,219385,"TERMINAL",0,0,"⠴",,terminal_output +105,219464,"TERMINAL",0,0,"⠦",,terminal_output +106,219544,"TERMINAL",0,0,"⠧",,terminal_output +107,219625,"TERMINAL",0,0,"⠇",,terminal_output +108,219704,"TERMINAL",0,0,"⠏",,terminal_output +109,219785,"TERMINAL",0,0,"⠋",,terminal_output +110,219866,"TERMINAL",0,0,"⠙",,terminal_output +111,219946,"TERMINAL",0,0,"⠹",,terminal_output +112,220024,"TERMINAL",0,0,"⠸",,terminal_output +113,220107,"TERMINAL",0,0,"⠼",,terminal_output +114,220188,"TERMINAL",0,0,"⠴",,terminal_output +115,220268,"TERMINAL",0,0,"⠦",,terminal_output +116,220349,"TERMINAL",0,0,"⠧",,terminal_output +117,220431,"TERMINAL",0,0,"⠇",,terminal_output +118,220509,"TERMINAL",0,0,"⠏",,terminal_output +119,220591,"TERMINAL",0,0,"⠋",,terminal_output +120,220668,"TERMINAL",0,0,"⠙",,terminal_output +121,220751,"TERMINAL",0,0,"⠹",,terminal_output +122,220831,"TERMINAL",0,0,"⠸",,terminal_output +123,220911,"TERMINAL",0,0,"⠼",,terminal_output +124,220991,"TERMINAL",0,0,"⠴",,terminal_output +125,221072,"TERMINAL",0,0,"⠦",,terminal_output +126,221151,"TERMINAL",0,0,"⠧",,terminal_output +127,221232,"TERMINAL",0,0,"⠇",,terminal_output +128,221314,"TERMINAL",0,0,"⠏",,terminal_output +129,221396,"TERMINAL",0,0,"⠋",,terminal_output +130,221474,"TERMINAL",0,0,"⠙",,terminal_output +131,221556,"TERMINAL",0,0,"⠹",,terminal_output +132,221634,"TERMINAL",0,0,"⠸",,terminal_output +133,221715,"TERMINAL",0,0,"⠼",,terminal_output +134,221794,"TERMINAL",0,0,"⠴",,terminal_output +135,221881,"TERMINAL",0,0,"⠦",,terminal_output +136,221958,"TERMINAL",0,0,"⠧",,terminal_output +137,222038,"TERMINAL",0,0,"⠇",,terminal_output +138,222119,"TERMINAL",0,0,"⠏",,terminal_output +139,222202,"TERMINAL",0,0,"⠋",,terminal_output +140,222281,"TERMINAL",0,0,"⠙",,terminal_output +141,222361,"TERMINAL",0,0,"⠹",,terminal_output +142,222441,"TERMINAL",0,0,"⠸",,terminal_output +143,222522,"TERMINAL",0,0,"⠼",,terminal_output +144,222603,"TERMINAL",0,0,"⠴",,terminal_output +145,222683,"TERMINAL",0,0,"⠦",,terminal_output +146,222763,"TERMINAL",0,0,"⠧",,terminal_output +147,222847,"TERMINAL",0,0,"⠇",,terminal_output +148,222944,"TERMINAL",0,0,"⠏",,terminal_output +149,223014,"TERMINAL",0,0,"⠋",,terminal_output +150,223095,"TERMINAL",0,0,"⠙",,terminal_output +151,223175,"TERMINAL",0,0,"⠹",,terminal_output +152,223255,"TERMINAL",0,0,"⠸",,terminal_output +153,223335,"TERMINAL",0,0,"⠼",,terminal_output +154,223417,"TERMINAL",0,0,"⠴",,terminal_output +155,223497,"TERMINAL",0,0,"⠦",,terminal_output +156,223577,"TERMINAL",0,0,"⠧",,terminal_output +157,223664,"TERMINAL",0,0,"⠇",,terminal_output +158,223739,"TERMINAL",0,0,"⠏",,terminal_output +159,223820,"TERMINAL",0,0,"⠋",,terminal_output +160,223900,"TERMINAL",0,0,"⠙",,terminal_output +161,223982,"TERMINAL",0,0,"⠹",,terminal_output +162,224062,"TERMINAL",0,0,"⠸",,terminal_output +163,224142,"TERMINAL",0,0,"⠼",,terminal_output +164,224221,"TERMINAL",0,0,"⠴",,terminal_output +165,224304,"TERMINAL",0,0,"⠦",,terminal_output +166,224385,"TERMINAL",0,0,"⠧",,terminal_output +167,224465,"TERMINAL",0,0,"⠇",,terminal_output +168,224545,"TERMINAL",0,0,"⠏",,terminal_output +169,224624,"TERMINAL",0,0,"⠋",,terminal_output +170,224703,"TERMINAL",0,0,"⠙",,terminal_output +171,224796,"TERMINAL",0,0,"⠹",,terminal_output +172,224874,"TERMINAL",0,0,"⠸",,terminal_output +173,224954,"TERMINAL",0,0,"⠼",,terminal_output +174,225033,"TERMINAL",0,0,"⠴",,terminal_output +175,225113,"TERMINAL",0,0,"⠦",,terminal_output +176,225193,"TERMINAL",0,0,"⠧",,terminal_output +177,225275,"TERMINAL",0,0,"⠇",,terminal_output +178,225354,"TERMINAL",0,0,"⠏",,terminal_output +179,225434,"TERMINAL",0,0,"⠋",,terminal_output +180,225514,"TERMINAL",0,0,"⠙",,terminal_output +181,225594,"TERMINAL",0,0,"⠹",,terminal_output +182,225677,"TERMINAL",0,0,"⠸",,terminal_output +183,225757,"TERMINAL",0,0,"⠼",,terminal_output +184,225837,"TERMINAL",0,0,"⠴",,terminal_output +185,225918,"TERMINAL",0,0,"⠦",,terminal_output +186,225996,"TERMINAL",0,0,"⠧",,terminal_output +187,226078,"TERMINAL",0,0,"⠇",,terminal_output +188,226158,"TERMINAL",0,0,"⠏",,terminal_output +189,226237,"TERMINAL",0,0,"⠋",,terminal_output +190,226319,"TERMINAL",0,0,"⠙",,terminal_output +191,226396,"TERMINAL",0,0,"⠹",,terminal_output +192,226478,"TERMINAL",0,0,"⠸",,terminal_output +193,226559,"TERMINAL",0,0,"⠼",,terminal_output +194,226637,"TERMINAL",0,0,"⠴",,terminal_output +195,226721,"TERMINAL",0,0,"⠦",,terminal_output +196,226800,"TERMINAL",0,0,"⠧",,terminal_output +197,226880,"TERMINAL",0,0,"⠇",,terminal_output +198,226960,"TERMINAL",0,0,"⠏",,terminal_output +199,227039,"TERMINAL",0,0,"⠋",,terminal_output +200,227137,"TERMINAL",0,0,"⠙",,terminal_output +201,227200,"TERMINAL",0,0,"⠹",,terminal_output +202,227279,"TERMINAL",0,0,"⠸",,terminal_output +203,227362,"TERMINAL",0,0,"⠼",,terminal_output +204,227442,"TERMINAL",0,0,"⠴",,terminal_output +205,227532,"TERMINAL",0,0,"⠦",,terminal_output +206,227612,"TERMINAL",0,0,"⠧",,terminal_output +207,227696,"TERMINAL",0,0,"⠇",,terminal_output +208,227774,"TERMINAL",0,0,"⠏",,terminal_output +209,227859,"TERMINAL",0,0,"⠋",,terminal_output +210,227946,"TERMINAL",0,0,"⠙",,terminal_output +211,228025,"TERMINAL",0,0,"⠹",,terminal_output +212,228107,"TERMINAL",0,0,"⠸",,terminal_output +213,228186,"TERMINAL",0,0,"⠼",,terminal_output +214,228267,"TERMINAL",0,0,"⠴",,terminal_output +215,228347,"TERMINAL",0,0,"⠦",,terminal_output +216,228427,"TERMINAL",0,0,"⠧",,terminal_output +217,228508,"TERMINAL",0,0,"⠇",,terminal_output +218,228586,"TERMINAL",0,0,"⠏",,terminal_output +219,228667,"TERMINAL",0,0,"⠋",,terminal_output +220,228747,"TERMINAL",0,0,"⠙",,terminal_output +221,228827,"TERMINAL",0,0,"⠹",,terminal_output +222,228906,"TERMINAL",0,0,"⠸",,terminal_output +223,228987,"TERMINAL",0,0,"⠼",,terminal_output +224,229067,"TERMINAL",0,0,"⠴",,terminal_output +225,229147,"TERMINAL",0,0,"⠦",,terminal_output +226,229226,"TERMINAL",0,0,"⠧",,terminal_output +227,229306,"TERMINAL",0,0,"⠇",,terminal_output +228,229452,"TERMINAL",0,0,"⠏",,terminal_output +229,229562,"TERMINAL",0,0,"⠋",,terminal_output +230,229645,"TERMINAL",0,0,"⠙",,terminal_output +231,229725,"TERMINAL",0,0,"⠹",,terminal_output +232,229802,"TERMINAL",0,0,"⠸",,terminal_output +233,229888,"TERMINAL",0,0,"⠼",,terminal_output +234,229968,"TERMINAL",0,0,"⠴",,terminal_output +235,230048,"TERMINAL",0,0,"⠦",,terminal_output +236,230129,"TERMINAL",0,0,"⠧",,terminal_output +237,230211,"TERMINAL",0,0,"⠇",,terminal_output +238,230294,"TERMINAL",0,0,"⠏",,terminal_output +239,230373,"TERMINAL",0,0,"⠋",,terminal_output +240,230452,"TERMINAL",0,0,"⠙",,terminal_output +241,230532,"TERMINAL",0,0,"⠹",,terminal_output +242,230615,"TERMINAL",0,0,"⠸",,terminal_output +243,230693,"TERMINAL",0,0,"⠼",,terminal_output +244,230774,"TERMINAL",0,0,"⠴",,terminal_output +245,230854,"TERMINAL",0,0,"⠦",,terminal_output +246,230933,"TERMINAL",0,0,"⠧",,terminal_output +247,231014,"TERMINAL",0,0,"⠇",,terminal_output +248,231093,"TERMINAL",0,0,"⠏",,terminal_output +249,231174,"TERMINAL",0,0,"⠋",,terminal_output +250,231253,"TERMINAL",0,0,"⠙",,terminal_output +251,231333,"TERMINAL",0,0,"⠹",,terminal_output +252,231413,"TERMINAL",0,0,"⠸",,terminal_output +253,231492,"TERMINAL",0,0,"⠼",,terminal_output +254,231573,"TERMINAL",0,0,"⠴",,terminal_output +255,231653,"TERMINAL",0,0,"⠦",,terminal_output +256,231732,"TERMINAL",0,0,"⠧",,terminal_output +257,231813,"TERMINAL",0,0,"⠇",,terminal_output +258,231894,"TERMINAL",0,0,"⠏",,terminal_output +259,231973,"TERMINAL",0,0,"⠋",,terminal_output +260,232054,"TERMINAL",0,0,"⠙",,terminal_output +261,232134,"TERMINAL",0,0,"⠹",,terminal_output +262,232213,"TERMINAL",0,0,"⠸",,terminal_output +263,232293,"TERMINAL",0,0,"⠼",,terminal_output +264,232373,"TERMINAL",0,0,"⠴",,terminal_output +265,232453,"TERMINAL",0,0,"⠦",,terminal_output +266,232534,"TERMINAL",0,0,"⠧",,terminal_output +267,232613,"TERMINAL",0,0,"⠇",,terminal_output +268,232694,"TERMINAL",0,0,"⠏",,terminal_output +269,232773,"TERMINAL",0,0,"⠋",,terminal_output +270,232854,"TERMINAL",0,0,"⠙",,terminal_output +271,232939,"TERMINAL",0,0,"⠹",,terminal_output +272,233013,"TERMINAL",0,0,"⠸",,terminal_output +273,233094,"TERMINAL",0,0,"⠼",,terminal_output +274,233173,"TERMINAL",0,0,"⠴",,terminal_output +275,233253,"TERMINAL",0,0,"⠦",,terminal_output +276,233334,"TERMINAL",0,0,"⠧",,terminal_output +277,233415,"TERMINAL",0,0,"⠇",,terminal_output +278,233494,"TERMINAL",0,0,"⠏",,terminal_output +279,233574,"TERMINAL",0,0,"⠋",,terminal_output +280,233655,"TERMINAL",0,0,"⠙",,terminal_output +281,233735,"TERMINAL",0,0,"⠹",,terminal_output +282,233816,"TERMINAL",0,0,"⠸",,terminal_output +283,233895,"TERMINAL",0,0,"⠼",,terminal_output +284,233974,"TERMINAL",0,0,"⠴",,terminal_output +285,234055,"TERMINAL",0,0,"⠦",,terminal_output +286,234134,"TERMINAL",0,0,"⠧",,terminal_output +287,234219,"TERMINAL",0,0,"⠇",,terminal_output +288,234299,"TERMINAL",0,0,"⠏",,terminal_output +289,234378,"TERMINAL",0,0,"⠋",,terminal_output +290,234458,"TERMINAL",0,0,"⠙",,terminal_output +291,234539,"TERMINAL",0,0,"⠹",,terminal_output +292,234618,"TERMINAL",0,0,"⠸",,terminal_output +293,234716,"TERMINAL",0,0,"⠼",,terminal_output +294,234793,"TERMINAL",0,0,"⠴",,terminal_output +295,234872,"TERMINAL",0,0,"⠦",,terminal_output +296,234952,"TERMINAL",0,0,"⠧",,terminal_output +297,235033,"TERMINAL",0,0,"⠇",,terminal_output +298,235112,"TERMINAL",0,0,"⠏",,terminal_output +299,235194,"TERMINAL",0,0,"⠋",,terminal_output +300,235274,"TERMINAL",0,0,"⠙",,terminal_output +301,235352,"TERMINAL",0,0,"⠹",,terminal_output +302,235434,"TERMINAL",0,0,"⠸",,terminal_output +303,235513,"TERMINAL",0,0,"⠼",,terminal_output +304,235594,"TERMINAL",0,0,"⠴",,terminal_output +305,235673,"TERMINAL",0,0,"⠦",,terminal_output +306,235752,"TERMINAL",0,0,"⠧",,terminal_output +307,235833,"TERMINAL",0,0,"⠇",,terminal_output +308,235914,"TERMINAL",0,0,"⠏",,terminal_output +309,235994,"TERMINAL",0,0,"⠋",,terminal_output +310,236073,"TERMINAL",0,0,"⠙",,terminal_output +311,236154,"TERMINAL",0,0,"⠹",,terminal_output +312,236260,"TERMINAL",0,0,"⠸",,terminal_output +313,236339,"TERMINAL",0,0,"⠼",,terminal_output +314,236419,"TERMINAL",0,0,"⠴",,terminal_output +315,236500,"TERMINAL",0,0,"⠦",,terminal_output +316,236581,"TERMINAL",0,0,"⠧",,terminal_output +317,236668,"TERMINAL",0,0,"⠇",,terminal_output +318,236741,"TERMINAL",0,0,"⠏",,terminal_output +319,236824,"TERMINAL",0,0,"⠋",,terminal_output +320,236902,"TERMINAL",0,0,"⠙",,terminal_output +321,236984,"TERMINAL",0,0,"⠹",,terminal_output +322,237064,"TERMINAL",0,0,"⠸",,terminal_output +323,237145,"TERMINAL",0,0,"⠼",,terminal_output +324,237225,"TERMINAL",0,0,"⠴",,terminal_output +325,237305,"TERMINAL",0,0,"⠦",,terminal_output +326,237386,"TERMINAL",0,0,"⠧",,terminal_output +327,237467,"TERMINAL",0,0,"⠇",,terminal_output +328,237549,"TERMINAL",0,0,"⠏",,terminal_output +329,237628,"TERMINAL",0,0,"⠋",,terminal_output +330,237709,"TERMINAL",0,0,"⠙",,terminal_output +331,237788,"TERMINAL",0,0,"⠹",,terminal_output +332,237871,"TERMINAL",0,0,"⠸",,terminal_output +333,237953,"TERMINAL",0,0,"⠼",,terminal_output +334,238063,"TERMINAL",0,0,"⠴",,terminal_output +335,238116,"TERMINAL",0,0,"⠦",,terminal_output +336,238198,"TERMINAL",0,0,"⠧",,terminal_output +337,238280,"TERMINAL",0,0,"⠇",,terminal_output +338,238359,"TERMINAL",0,0,"⠏",,terminal_output +339,238440,"TERMINAL",0,0,"⠋",,terminal_output +340,238522,"TERMINAL",0,0,"⠙",,terminal_output +341,238604,"TERMINAL",0,0,"⠹",,terminal_output +342,238685,"TERMINAL",0,0,"⠸",,terminal_output +343,238765,"TERMINAL",0,0,"⠼",,terminal_output +344,238845,"TERMINAL",0,0,"⠴",,terminal_output +345,238928,"TERMINAL",0,0,"⠦",,terminal_output +346,239008,"TERMINAL",0,0,"⠧",,terminal_output +347,239088,"TERMINAL",0,0,"⠇",,terminal_output +348,239171,"TERMINAL",0,0,"⠏",,terminal_output +349,239252,"TERMINAL",0,0,"⠋",,terminal_output +350,239333,"TERMINAL",0,0,"⠙",,terminal_output +351,239413,"TERMINAL",0,0,"⠹",,terminal_output +352,239494,"TERMINAL",0,0,"⠸",,terminal_output +353,239573,"TERMINAL",0,0,"⠼",,terminal_output +354,239655,"TERMINAL",0,0,"⠴",,terminal_output +355,239736,"TERMINAL",0,0,"⠦",,terminal_output +356,239817,"TERMINAL",0,0,"⠧",,terminal_output +357,239898,"TERMINAL",0,0,"⠇",,terminal_output +358,239978,"TERMINAL",0,0,"⠏",,terminal_output +359,240059,"TERMINAL",0,0,"⠋",,terminal_output +360,240140,"TERMINAL",0,0,"⠙",,terminal_output +361,240222,"TERMINAL",0,0,"⠹",,terminal_output +362,240303,"TERMINAL",0,0,"⠸",,terminal_output +363,240384,"TERMINAL",0,0,"⠼",,terminal_output +364,240464,"TERMINAL",0,0,"⠴",,terminal_output +365,240544,"TERMINAL",0,0,"⠦",,terminal_output +366,240627,"TERMINAL",0,0,"⠧",,terminal_output +367,240707,"TERMINAL",0,0,"⠇",,terminal_output +368,240789,"TERMINAL",0,0,"⠏",,terminal_output +369,240867,"TERMINAL",0,0,"⠋",,terminal_output +370,240950,"TERMINAL",0,0,"⠙",,terminal_output +371,241029,"TERMINAL",0,0,"⠹",,terminal_output +372,241110,"TERMINAL",0,0,"⠸",,terminal_output +373,241192,"TERMINAL",0,0,"⠼",,terminal_output +374,241273,"TERMINAL",0,0,"⠴",,terminal_output +375,241356,"TERMINAL",0,0,"⠦",,terminal_output +376,241434,"TERMINAL",0,0,"⠧",,terminal_output +377,241515,"TERMINAL",0,0,"⠇",,terminal_output +378,241595,"TERMINAL",0,0,"⠏",,terminal_output +379,241676,"TERMINAL",0,0,"⠋",,terminal_output +380,241759,"TERMINAL",0,0,"⠙",,terminal_output +381,241841,"TERMINAL",0,0,"⠹",,terminal_output +382,241921,"TERMINAL",0,0,"⠸",,terminal_output +383,242003,"TERMINAL",0,0,"⠼",,terminal_output +384,242082,"TERMINAL",0,0,"⠴",,terminal_output +385,242163,"TERMINAL",0,0,"⠦",,terminal_output +386,242244,"TERMINAL",0,0,"⠧",,terminal_output +387,242325,"TERMINAL",0,0,"⠇",,terminal_output +388,242406,"TERMINAL",0,0,"⠏",,terminal_output +389,242487,"TERMINAL",0,0,"⠋",,terminal_output +390,242568,"TERMINAL",0,0,"⠙",,terminal_output +391,242648,"TERMINAL",0,0,"⠹",,terminal_output +392,242732,"TERMINAL",0,0,"⠸",,terminal_output +393,242813,"TERMINAL",0,0,"⠼",,terminal_output +394,242892,"TERMINAL",0,0,"⠴",,terminal_output +395,242971,"TERMINAL",0,0,"⠦",,terminal_output +396,243051,"TERMINAL",0,0,"⠧",,terminal_output +397,243135,"TERMINAL",0,0,"⠇",,terminal_output +398,243215,"TERMINAL",0,0,"⠏",,terminal_output +399,243295,"TERMINAL",0,0,"⠋",,terminal_output +400,243377,"TERMINAL",0,0,"⠙",,terminal_output +401,243457,"TERMINAL",0,0,"⠹",,terminal_output +402,243540,"TERMINAL",0,0,"⠸",,terminal_output +403,243621,"TERMINAL",0,0,"⠼",,terminal_output +404,243700,"TERMINAL",0,0,"⠴",,terminal_output +405,243781,"TERMINAL",0,0,"⠦",,terminal_output +406,243864,"TERMINAL",0,0,"⠧",,terminal_output +407,243946,"TERMINAL",0,0,"⠇",,terminal_output +408,244025,"TERMINAL",0,0,"⠏",,terminal_output +409,244106,"TERMINAL",0,0,"⠋",,terminal_output +410,244186,"TERMINAL",0,0,"⠙",,terminal_output +411,244267,"TERMINAL",0,0,"⠹",,terminal_output +412,244348,"TERMINAL",0,0,"⠸",,terminal_output +413,244436,"TERMINAL",0,0,"⠼",,terminal_output +414,244515,"TERMINAL",0,0,"⠴",,terminal_output +415,244597,"TERMINAL",0,0,"⠦",,terminal_output +416,244678,"TERMINAL",0,0,"⠧",,terminal_output +417,244768,"TERMINAL",0,0,"⠇",,terminal_output +418,244846,"TERMINAL",0,0,"⠏",,terminal_output +419,244927,"TERMINAL",0,0,"⠋",,terminal_output +420,245007,"TERMINAL",0,0,"⠙",,terminal_output +421,245089,"TERMINAL",0,0,"⠹",,terminal_output +422,245170,"TERMINAL",0,0,"⠸",,terminal_output +423,245251,"TERMINAL",0,0,"⠼",,terminal_output +424,245332,"TERMINAL",0,0,"⠴",,terminal_output +425,245411,"TERMINAL",0,0,"⠦",,terminal_output +426,245495,"TERMINAL",0,0,"⠧",,terminal_output +427,245573,"TERMINAL",0,0,"⠇",,terminal_output +428,245656,"TERMINAL",0,0,"⠏",,terminal_output +429,245735,"TERMINAL",0,0,"⠋",,terminal_output +430,245818,"TERMINAL",0,0,"⠙",,terminal_output +431,245898,"TERMINAL",0,0,"⠹",,terminal_output +432,245979,"TERMINAL",0,0,"⠸",,terminal_output +433,246061,"TERMINAL",0,0,"⠼",,terminal_output +434,246142,"TERMINAL",0,0,"⠴",,terminal_output +435,246222,"TERMINAL",0,0,"⠦",,terminal_output +436,246312,"TERMINAL",0,0,"⠧",,terminal_output +437,246384,"TERMINAL",0,0,"⠇",,terminal_output +438,246467,"TERMINAL",0,0,"⠏",,terminal_output +439,246549,"TERMINAL",0,0,"⠋",,terminal_output +440,246635,"TERMINAL",0,0,"⠙",,terminal_output +441,246716,"TERMINAL",0,0,"⠹",,terminal_output +442,246798,"TERMINAL",0,0,"⠸",,terminal_output +443,246879,"TERMINAL",0,0,"⠼",,terminal_output +444,246961,"TERMINAL",0,0,"⠴",,terminal_output +445,247041,"TERMINAL",0,0,"⠦",,terminal_output +446,247122,"TERMINAL",0,0,"⠧",,terminal_output +447,247201,"TERMINAL",0,0,"⠇",,terminal_output +448,247282,"TERMINAL",0,0,"⠏",,terminal_output +449,247362,"TERMINAL",0,0,"⠋",,terminal_output +450,247444,"TERMINAL",0,0,"⠙",,terminal_output +451,247525,"TERMINAL",0,0,"⠹",,terminal_output +452,247607,"TERMINAL",0,0,"⠸",,terminal_output +453,247686,"TERMINAL",0,0,"⠼",,terminal_output +454,247768,"TERMINAL",0,0,"⠴",,terminal_output +455,247846,"TERMINAL",0,0,"⠦",,terminal_output +456,247929,"TERMINAL",0,0,"⠧",,terminal_output +457,248011,"TERMINAL",0,0,"⠇",,terminal_output +458,248096,"TERMINAL",0,0,"⠏",,terminal_output +459,248177,"TERMINAL",0,0,"⠋",,terminal_output +460,248257,"TERMINAL",0,0,"⠙",,terminal_output +461,248340,"TERMINAL",0,0,"⠹",,terminal_output +462,248420,"TERMINAL",0,0,"⠸",,terminal_output +463,248499,"TERMINAL",0,0,"⠼",,terminal_output +464,248580,"TERMINAL",0,0,"⠴",,terminal_output +465,248663,"TERMINAL",0,0,"⠦",,terminal_output +466,248743,"TERMINAL",0,0,"⠧",,terminal_output +467,248824,"TERMINAL",0,0,"⠇",,terminal_output +468,248906,"TERMINAL",0,0,"⠏",,terminal_output +469,248986,"TERMINAL",0,0,"⠋",,terminal_output +470,249068,"TERMINAL",0,0,"⠙",,terminal_output +471,249149,"TERMINAL",0,0,"⠹",,terminal_output +472,249244,"TERMINAL",0,0,"⠸",,terminal_output +473,249326,"TERMINAL",0,0,"⠼",,terminal_output +474,249407,"TERMINAL",0,0,"⠴",,terminal_output +475,249488,"TERMINAL",0,0,"⠦",,terminal_output +476,249568,"TERMINAL",0,0,"⠧",,terminal_output +477,249655,"TERMINAL",0,0,"⠇",,terminal_output +478,249767,"TERMINAL",0,0,"⠏",,terminal_output +479,249840,"TERMINAL",0,0,"⠋",,terminal_output +480,249912,"TERMINAL",0,0,"⠙",,terminal_output +481,249993,"TERMINAL",0,0,"⠹",,terminal_output +482,250073,"TERMINAL",0,0,"⠸",,terminal_output +483,250155,"TERMINAL",0,0,"⠼",,terminal_output +484,250235,"TERMINAL",0,0,"⠴",,terminal_output +485,250316,"TERMINAL",0,0,"⠦",,terminal_output +486,250398,"TERMINAL",0,0,"⠧",,terminal_output +487,250479,"TERMINAL",0,0,"⠇",,terminal_output +488,250558,"TERMINAL",0,0,"⠏",,terminal_output +489,250642,"TERMINAL",0,0,"⠋",,terminal_output +490,250720,"TERMINAL",0,0,"⠙",,terminal_output +491,250800,"TERMINAL",0,0,"⠹",,terminal_output +492,250881,"TERMINAL",0,0,"⠸",,terminal_output +493,250963,"TERMINAL",0,0,"⠼",,terminal_output +494,251048,"TERMINAL",0,0,"⠴",,terminal_output +495,251123,"TERMINAL",0,0,"⠦",,terminal_output +496,251205,"TERMINAL",0,0,"⠧",,terminal_output +497,251286,"TERMINAL",0,0,"⠇",,terminal_output +498,251367,"TERMINAL",0,0,"⠏",,terminal_output +499,251447,"TERMINAL",0,0,"⠋",,terminal_output +500,251528,"TERMINAL",0,0,"⠙",,terminal_output +501,251609,"TERMINAL",0,0,"⠹",,terminal_output +502,251689,"TERMINAL",0,0,"⠸",,terminal_output +503,251771,"TERMINAL",0,0,"⠼",,terminal_output +504,251850,"TERMINAL",0,0,"⠴",,terminal_output +505,251932,"TERMINAL",0,0,"⠦",,terminal_output +506,252013,"TERMINAL",0,0,"⠧",,terminal_output +507,252091,"TERMINAL",0,0,"⠇",,terminal_output +508,252173,"TERMINAL",0,0,"⠏",,terminal_output +509,252255,"TERMINAL",0,0,"⠋",,terminal_output +510,252335,"TERMINAL",0,0,"⠙",,terminal_output +511,252415,"TERMINAL",0,0,"⠹",,terminal_output +512,252497,"TERMINAL",0,0,"⠸",,terminal_output +513,252579,"TERMINAL",0,0,"⠼",,terminal_output +514,252659,"TERMINAL",0,0,"⠴",,terminal_output +515,252740,"TERMINAL",0,0,"⠦",,terminal_output +516,252819,"TERMINAL",0,0,"⠧",,terminal_output +517,252899,"TERMINAL",0,0,"⠇",,terminal_output +518,252981,"TERMINAL",0,0,"⠏",,terminal_output +519,253062,"TERMINAL",0,0,"⠋",,terminal_output +520,253144,"TERMINAL",0,0,"⠙",,terminal_output +521,253224,"TERMINAL",0,0,"⠹",,terminal_output +522,253305,"TERMINAL",0,0,"⠸",,terminal_output +523,253386,"TERMINAL",0,0,"⠼",,terminal_output +524,253469,"TERMINAL",0,0,"⠴",,terminal_output +525,253549,"TERMINAL",0,0,"⠦",,terminal_output +526,253628,"TERMINAL",0,0,"⠧",,terminal_output +527,253710,"TERMINAL",0,0,"⠇",,terminal_output +528,253791,"TERMINAL",0,0,"⠏",,terminal_output +529,253870,"TERMINAL",0,0,"⠋",,terminal_output +530,253953,"TERMINAL",0,0,"⠙",,terminal_output +531,254033,"TERMINAL",0,0,"⠹",,terminal_output +532,254113,"TERMINAL",0,0,"⠸",,terminal_output +533,254194,"TERMINAL",0,0,"⠼",,terminal_output +534,254276,"TERMINAL",0,0,"⠴",,terminal_output +535,254357,"TERMINAL",0,0,"⠦",,terminal_output +536,254435,"TERMINAL",0,0,"⠧",,terminal_output +537,254516,"TERMINAL",0,0,"⠇",,terminal_output +538,254598,"TERMINAL",0,0,"⠏",,terminal_output +539,254678,"TERMINAL",0,0,"⠋",,terminal_output +540,254759,"TERMINAL",0,0,"⠙",,terminal_output +541,254840,"TERMINAL",0,0,"⠹",,terminal_output +542,254920,"TERMINAL",0,0,"⠸",,terminal_output +543,255002,"TERMINAL",0,0,"⠼",,terminal_output +544,255083,"TERMINAL",0,0,"⠴",,terminal_output +545,255162,"TERMINAL",0,0,"⠦",,terminal_output +546,255248,"TERMINAL",0,0,"⠧",,terminal_output +547,255326,"TERMINAL",0,0,"⠇",,terminal_output +548,255408,"TERMINAL",0,0,"⠏",,terminal_output +549,255487,"TERMINAL",0,0,"⠋",,terminal_output +550,255572,"TERMINAL",0,0,"⠙",,terminal_output +551,255648,"TERMINAL",0,0,"⠹",,terminal_output +552,255729,"TERMINAL",0,0,"⠸",,terminal_output +553,255811,"TERMINAL",0,0,"⠼",,terminal_output +554,255894,"TERMINAL",0,0,"⠴",,terminal_output +555,255976,"TERMINAL",0,0,"⠦",,terminal_output +556,256056,"TERMINAL",0,0,"⠧",,terminal_output +557,256141,"TERMINAL",0,0,"⠇",,terminal_output +558,256218,"TERMINAL",0,0,"⠏",,terminal_output +559,256297,"TERMINAL",0,0,"⠋",,terminal_output +560,256380,"TERMINAL",0,0,"⠙",,terminal_output +561,256475,"TERMINAL",0,0,"⠹",,terminal_output +562,256559,"TERMINAL",0,0,"⠸",,terminal_output +563,256632,"TERMINAL",0,0,"⠼",,terminal_output +564,256712,"TERMINAL",0,0,"⠴",,terminal_output +565,256795,"TERMINAL",0,0,"⠦",,terminal_output +566,256874,"TERMINAL",0,0,"⠧",,terminal_output +567,256956,"TERMINAL",0,0,"⠇",,terminal_output +568,257036,"TERMINAL",0,0,"⠏",,terminal_output +569,257117,"TERMINAL",0,0,"⠋",,terminal_output +570,257205,"TERMINAL",0,0,"⠙",,terminal_output +571,257281,"TERMINAL",0,0,"⠹",,terminal_output +572,257360,"TERMINAL",0,0,"⠸",,terminal_output +573,257441,"TERMINAL",0,0,"⠼",,terminal_output +574,257521,"TERMINAL",0,0,"⠴",,terminal_output +575,257601,"TERMINAL",0,0,"⠦",,terminal_output +576,257682,"TERMINAL",0,0,"⠧",,terminal_output +577,257764,"TERMINAL",0,0,"⠇",,terminal_output +578,257845,"TERMINAL",0,0,"⠏",,terminal_output +579,257927,"TERMINAL",0,0,"⠋",,terminal_output +580,258008,"TERMINAL",0,0,"⠙",,terminal_output +581,258091,"TERMINAL",0,0,"⠹",,terminal_output +582,258171,"TERMINAL",0,0,"⠸",,terminal_output +583,258249,"TERMINAL",0,0,"⠼",,terminal_output +584,258334,"TERMINAL",0,0,"⠴",,terminal_output +585,258431,"TERMINAL",0,0,"⠦",,terminal_output +586,258506,"TERMINAL",0,0,"⠧",,terminal_output +587,258584,"TERMINAL",0,0,"⠇",,terminal_output +588,258666,"TERMINAL",0,0,"⠏",,terminal_output +589,258758,"TERMINAL",0,0,"⠋",,terminal_output +590,258826,"TERMINAL",0,0,"⠙",,terminal_output +591,258906,"TERMINAL",0,0,"⠹",,terminal_output +592,258991,"TERMINAL",0,0,"⠸",,terminal_output +593,259069,"TERMINAL",0,0,"⠼",,terminal_output +594,259150,"TERMINAL",0,0,"⠴",,terminal_output +595,259231,"TERMINAL",0,0,"⠦",,terminal_output +596,259296,"TERMINAL",0,0,"",,terminal_focus +597,259404,"TERMINAL",0,0,"⠧⠇",,terminal_output +598,259475,"TERMINAL",0,0,"⠏",,terminal_output +599,259553,"TERMINAL",0,0,"⠋",,terminal_output +600,259636,"TERMINAL",0,0,"⠙",,terminal_output +601,259715,"TERMINAL",0,0,"⠹",,terminal_output +602,259796,"TERMINAL",0,0,"⠸",,terminal_output +603,259878,"TERMINAL",0,0,"⠼",,terminal_output +604,259959,"TERMINAL",0,0,"⠴",,terminal_output +605,260043,"TERMINAL",0,0,"⠦",,terminal_output +606,260122,"TERMINAL",0,0,"⠧",,terminal_output +607,260203,"TERMINAL",0,0,"⠇",,terminal_output +608,260283,"TERMINAL",0,0,"⠏",,terminal_output +609,260364,"TERMINAL",0,0,"⠋",,terminal_output +610,260445,"TERMINAL",0,0,"⠙",,terminal_output +611,260525,"TERMINAL",0,0,"⠹",,terminal_output +612,260608,"TERMINAL",0,0,"⠸",,terminal_output +613,260688,"TERMINAL",0,0,"⠼",,terminal_output +614,260772,"TERMINAL",0,0,"⠴",,terminal_output +615,260852,"TERMINAL",0,0,"⠦",,terminal_output +616,260943,"TERMINAL",0,0,"⠧",,terminal_output +617,261012,"TERMINAL",0,0,"⠇",,terminal_output +618,261091,"TERMINAL",0,0,"⠏",,terminal_output +619,261171,"TERMINAL",0,0,"⠋",,terminal_output +620,261253,"TERMINAL",0,0,"⠙",,terminal_output +621,261331,"TERMINAL",0,0,"⠹",,terminal_output +622,261413,"TERMINAL",0,0,"⠸",,terminal_output +623,261492,"TERMINAL",0,0,"⠼",,terminal_output +624,261572,"TERMINAL",0,0,"⠴",,terminal_output +625,261652,"TERMINAL",0,0,"⠦",,terminal_output +626,261733,"TERMINAL",0,0,"⠧",,terminal_output +627,261813,"TERMINAL",0,0,"⠇",,terminal_output +628,261894,"TERMINAL",0,0,"⠏",,terminal_output +629,261973,"TERMINAL",0,0,"⠋",,terminal_output +630,262055,"TERMINAL",0,0,"⠙",,terminal_output +631,262085,"TERMINAL",0,0,"cd webview-ui",,terminal_command +632,262088,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +633,262136,"TERMINAL",0,0,"⠹",,terminal_output +634,262216,"TERMINAL",0,0,"⠸",,terminal_output +635,262297,"TERMINAL",0,0,"⠼",,terminal_output +636,262377,"TERMINAL",0,0,"⠴",,terminal_output +637,262460,"TERMINAL",0,0,"⠦",,terminal_output +638,262540,"TERMINAL",0,0,"⠧",,terminal_output +639,262625,"TERMINAL",0,0,"⠇",,terminal_output +640,262701,"TERMINAL",0,0,"⠏",,terminal_output +641,262783,"TERMINAL",0,0,"⠋",,terminal_output +642,262862,"TERMINAL",0,0,"⠙",,terminal_output +643,262946,"TERMINAL",0,0,"⠹",,terminal_output +644,263027,"TERMINAL",0,0,"⠸",,terminal_output +645,263108,"TERMINAL",0,0,"⠼",,terminal_output +646,263188,"TERMINAL",0,0,"⠴",,terminal_output +647,263270,"TERMINAL",0,0,"⠦",,terminal_output +648,263350,"TERMINAL",0,0,"⠧",,terminal_output +649,263431,"TERMINAL",0,0,"⠇",,terminal_output +650,263513,"TERMINAL",0,0,"⠏",,terminal_output +651,263593,"TERMINAL",0,0,"⠋",,terminal_output +652,263674,"TERMINAL",0,0,"⠙",,terminal_output +653,263755,"TERMINAL",0,0,"⠹",,terminal_output +654,263836,"TERMINAL",0,0,"⠸",,terminal_output +655,263917,"TERMINAL",0,0,"⠼",,terminal_output +656,263999,"TERMINAL",0,0,"⠴",,terminal_output +657,264076,"TERMINAL",0,0,"⠦",,terminal_output +658,264158,"TERMINAL",0,0,"⠧",,terminal_output +659,264238,"TERMINAL",0,0,"⠇",,terminal_output +660,264320,"TERMINAL",0,0,"⠏",,terminal_output +661,264400,"TERMINAL",0,0,"⠋",,terminal_output +662,264481,"TERMINAL",0,0,"⠙",,terminal_output +663,264563,"TERMINAL",0,0,"⠹",,terminal_output +664,264644,"TERMINAL",0,0,"⠸",,terminal_output +665,264725,"TERMINAL",0,0,"⠼",,terminal_output +666,264806,"TERMINAL",0,0,"⠴",,terminal_output +667,264886,"TERMINAL",0,0,"⠦",,terminal_output +668,264968,"TERMINAL",0,0,"⠧",,terminal_output +669,265048,"TERMINAL",0,0,"⠇",,terminal_output +670,265132,"TERMINAL",0,0,"⠏",,terminal_output +671,265212,"TERMINAL",0,0,"⠋",,terminal_output +672,265293,"TERMINAL",0,0,"⠙",,terminal_output +673,265340,"TERMINAL",0,0,"npm run storybook",,terminal_command +674,265373,"TERMINAL",0,0,"⠹",,terminal_output +675,265391,"TERMINAL",0,0,"]633;C",,terminal_output +676,265456,"TERMINAL",0,0,"⠸",,terminal_output +677,265535,"TERMINAL",0,0,"⠼",,terminal_output +678,265582,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\nsh: storybook: command not found\r\n⠙% \r \r",,terminal_output +679,265617,"TERMINAL",0,0,"⠴",,terminal_output +680,265697,"TERMINAL",0,0,"⠦",,terminal_output +681,265780,"TERMINAL",0,0,"⠧",,terminal_output +682,265860,"TERMINAL",0,0,"⠇",,terminal_output +683,265943,"TERMINAL",0,0,"⠏",,terminal_output +684,266022,"TERMINAL",0,0,"⠋",,terminal_output +685,266104,"TERMINAL",0,0,"⠙",,terminal_output +686,266184,"TERMINAL",0,0,"⠹",,terminal_output +687,266264,"TERMINAL",0,0,"⠸",,terminal_output +688,266344,"TERMINAL",0,0,"⠼",,terminal_output +689,266427,"TERMINAL",0,0,"⠴",,terminal_output +690,266508,"TERMINAL",0,0,"⠦",,terminal_output +691,266589,"TERMINAL",0,0,"⠧",,terminal_output +692,266670,"TERMINAL",0,0,"⠇",,terminal_output +693,266751,"TERMINAL",0,0,"⠏",,terminal_output +694,266833,"TERMINAL",0,0,"⠋",,terminal_output +695,266915,"TERMINAL",0,0,"⠙",,terminal_output +696,266995,"TERMINAL",0,0,"⠹",,terminal_output +697,267076,"TERMINAL",0,0,"⠸",,terminal_output +698,267154,"TERMINAL",0,0,"⠼",,terminal_output +699,267238,"TERMINAL",0,0,"⠴",,terminal_output +700,267319,"TERMINAL",0,0,"⠦",,terminal_output +701,267400,"TERMINAL",0,0,"⠧",,terminal_output +702,267480,"TERMINAL",0,0,"⠇",,terminal_output +703,267559,"TERMINAL",0,0,"⠏",,terminal_output +704,267774,"TERMINAL",0,0,"⠋",,terminal_output +705,267845,"TERMINAL",0,0,"\r\n> claude-dev@3.38.3 prepare\r\n> husky\r\n\r\n⠋",,terminal_output +706,268035,"TERMINAL",0,0,"⠙\r\nadded 1505 packages, and audited 1506 packages in 59s\r\n⠙\r\n⠙277 packages are looking for funding\r\n⠙ run `npm fund` for details\r\n⠙\r\nfound 0 vulnerabilities\r\n⠙% \r \r",,terminal_output +707,270587,"TERMINAL",0,0,"zsh",,terminal_focus +708,275890,"TERMINAL",0,0,"cd webview-ui",,terminal_command +709,275896,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +710,278185,"TERMINAL",0,0,"npm run storybook",,terminal_command +711,278237,"TERMINAL",0,0,"]633;C",,terminal_output +712,278317,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\nsh: storybook: command not found\r\n⠙% \r \r",,terminal_output +713,282871,"TERMINAL",0,0,"npm install",,terminal_command +714,282923,"TERMINAL",0,0,"]633;C",,terminal_output +715,283584,"TERMINAL",0,0,"⠙",,terminal_output +716,283850,"TERMINAL",0,0,"⠹",,terminal_output +717,283971,"TERMINAL",0,0,"⠸",,terminal_output +718,284051,"TERMINAL",0,0,"⠼",,terminal_output +719,284140,"TERMINAL",0,0,"⠴",,terminal_output +720,284240,"TERMINAL",0,0,"⠦",,terminal_output +721,284501,"TERMINAL",0,0,"⠧",,terminal_output +722,284569,"TERMINAL",0,0,"⠇",,terminal_output +723,284653,"TERMINAL",0,0,"⠏",,terminal_output +724,284737,"TERMINAL",0,0,"⠋",,terminal_output +725,284817,"TERMINAL",0,0,"⠙",,terminal_output +726,284900,"TERMINAL",0,0,"⠹",,terminal_output +727,284984,"TERMINAL",0,0,"⠸",,terminal_output +728,285063,"TERMINAL",0,0,"⠼",,terminal_output +729,285144,"TERMINAL",0,0,"⠴",,terminal_output +730,285227,"TERMINAL",0,0,"⠦",,terminal_output +731,285306,"TERMINAL",0,0,"⠧",,terminal_output +732,285385,"TERMINAL",0,0,"⠇",,terminal_output +733,285465,"TERMINAL",0,0,"⠏",,terminal_output +734,285545,"TERMINAL",0,0,"⠋",,terminal_output +735,285626,"TERMINAL",0,0,"⠙",,terminal_output +736,285705,"TERMINAL",0,0,"⠹",,terminal_output +737,285785,"TERMINAL",0,0,"⠸",,terminal_output +738,285866,"TERMINAL",0,0,"⠼",,terminal_output +739,285946,"TERMINAL",0,0,"⠴",,terminal_output +740,286027,"TERMINAL",0,0,"⠦",,terminal_output +741,286107,"TERMINAL",0,0,"⠧",,terminal_output +742,286187,"TERMINAL",0,0,"⠇",,terminal_output +743,286300,"TERMINAL",0,0,"⠏",,terminal_output +744,286374,"TERMINAL",0,0,"⠋",,terminal_output +745,286452,"TERMINAL",0,0,"⠙",,terminal_output +746,286534,"TERMINAL",0,0,"⠹",,terminal_output +747,286613,"TERMINAL",0,0,"⠸",,terminal_output +748,286694,"TERMINAL",0,0,"⠼",,terminal_output +749,286774,"TERMINAL",0,0,"⠴",,terminal_output +750,286854,"TERMINAL",0,0,"⠦",,terminal_output +751,286934,"TERMINAL",0,0,"⠧",,terminal_output +752,287014,"TERMINAL",0,0,"⠇",,terminal_output +753,287095,"TERMINAL",0,0,"⠏",,terminal_output +754,287175,"TERMINAL",0,0,"⠋",,terminal_output +755,287256,"TERMINAL",0,0,"⠙",,terminal_output +756,287336,"TERMINAL",0,0,"⠹",,terminal_output +757,287415,"TERMINAL",0,0,"⠸",,terminal_output +758,287496,"TERMINAL",0,0,"⠼",,terminal_output +759,287576,"TERMINAL",0,0,"⠴",,terminal_output +760,287656,"TERMINAL",0,0,"⠦",,terminal_output +761,287734,"TERMINAL",0,0,"⠧",,terminal_output +762,287817,"TERMINAL",0,0,"⠇",,terminal_output +763,287928,"TERMINAL",0,0,"⠏npm warn deprecated @vscode/webview-ui-toolkit@1.4.0: This package has been deprecated, https://github.com/microsoft/vscode-webview-ui-toolkit/issues/561\r\n⠏",,terminal_output +764,288009,"TERMINAL",0,0,"⠋",,terminal_output +765,288091,"TERMINAL",0,0,"⠙",,terminal_output +766,288171,"TERMINAL",0,0,"⠹",,terminal_output +767,288249,"TERMINAL",0,0,"⠸",,terminal_output +768,288329,"TERMINAL",0,0,"⠼",,terminal_output +769,288411,"TERMINAL",0,0,"⠴",,terminal_output +770,288491,"TERMINAL",0,0,"⠦",,terminal_output +771,288572,"TERMINAL",0,0,"⠧",,terminal_output +772,288653,"TERMINAL",0,0,"⠇",,terminal_output +773,288734,"TERMINAL",0,0,"⠏",,terminal_output +774,288814,"TERMINAL",0,0,"⠋",,terminal_output +775,288894,"TERMINAL",0,0,"⠙",,terminal_output +776,288976,"TERMINAL",0,0,"⠹",,terminal_output +777,289057,"TERMINAL",0,0,"⠸",,terminal_output +778,289135,"TERMINAL",0,0,"⠼",,terminal_output +779,289218,"TERMINAL",0,0,"⠴",,terminal_output +780,289298,"TERMINAL",0,0,"⠦",,terminal_output +781,289378,"TERMINAL",0,0,"⠧",,terminal_output +782,289456,"TERMINAL",0,0,"⠇",,terminal_output +783,289537,"TERMINAL",0,0,"⠏",,terminal_output +784,289617,"TERMINAL",0,0,"⠋",,terminal_output +785,289697,"TERMINAL",0,0,"⠙",,terminal_output +786,289776,"TERMINAL",0,0,"⠹",,terminal_output +787,289858,"TERMINAL",0,0,"⠸",,terminal_output +788,289937,"TERMINAL",0,0,"⠼",,terminal_output +789,290016,"TERMINAL",0,0,"⠴",,terminal_output +790,290098,"TERMINAL",0,0,"⠦",,terminal_output +791,290176,"TERMINAL",0,0,"⠧",,terminal_output +792,290258,"TERMINAL",0,0,"⠇",,terminal_output +793,290339,"TERMINAL",0,0,"⠏",,terminal_output +794,290418,"TERMINAL",0,0,"⠋",,terminal_output +795,290496,"TERMINAL",0,0,"⠙",,terminal_output +796,290578,"TERMINAL",0,0,"⠹",,terminal_output +797,290659,"TERMINAL",0,0,"⠸",,terminal_output +798,290738,"TERMINAL",0,0,"⠼",,terminal_output +799,290820,"TERMINAL",0,0,"⠴",,terminal_output +800,290900,"TERMINAL",0,0,"⠦",,terminal_output +801,290982,"TERMINAL",0,0,"⠧",,terminal_output +802,291061,"TERMINAL",0,0,"⠇",,terminal_output +803,291142,"TERMINAL",0,0,"⠏",,terminal_output +804,291222,"TERMINAL",0,0,"⠋",,terminal_output +805,291303,"TERMINAL",0,0,"⠙",,terminal_output +806,291381,"TERMINAL",0,0,"⠹",,terminal_output +807,291463,"TERMINAL",0,0,"⠸",,terminal_output +808,291542,"TERMINAL",0,0,"⠼",,terminal_output +809,291623,"TERMINAL",0,0,"⠴",,terminal_output +810,291702,"TERMINAL",0,0,"⠦",,terminal_output +811,291781,"TERMINAL",0,0,"⠧",,terminal_output +812,291865,"TERMINAL",0,0,"⠇",,terminal_output +813,291944,"TERMINAL",0,0,"⠏",,terminal_output +814,292025,"TERMINAL",0,0,"⠋",,terminal_output +815,292106,"TERMINAL",0,0,"⠙",,terminal_output +816,292186,"TERMINAL",0,0,"⠹",,terminal_output +817,292266,"TERMINAL",0,0,"⠸",,terminal_output +818,292349,"TERMINAL",0,0,"⠼",,terminal_output +819,292426,"TERMINAL",0,0,"⠴",,terminal_output +820,292506,"TERMINAL",0,0,"⠦",,terminal_output +821,292585,"TERMINAL",0,0,"⠧",,terminal_output +822,292667,"TERMINAL",0,0,"⠇",,terminal_output +823,292747,"TERMINAL",0,0,"⠏",,terminal_output +824,292828,"TERMINAL",0,0,"⠋",,terminal_output +825,292908,"TERMINAL",0,0,"⠙",,terminal_output +826,292999,"TERMINAL",0,0,"⠹",,terminal_output +827,293078,"TERMINAL",0,0,"⠸",,terminal_output +828,293160,"TERMINAL",0,0,"⠼",,terminal_output +829,293239,"TERMINAL",0,0,"⠴",,terminal_output +830,293320,"TERMINAL",0,0,"⠦",,terminal_output +831,293399,"TERMINAL",0,0,"⠧",,terminal_output +832,293481,"TERMINAL",0,0,"⠇",,terminal_output +833,293558,"TERMINAL",0,0,"⠏",,terminal_output +834,293639,"TERMINAL",0,0,"⠋",,terminal_output +835,293720,"TERMINAL",0,0,"⠙",,terminal_output +836,293802,"TERMINAL",0,0,"⠹",,terminal_output +837,293880,"TERMINAL",0,0,"⠸",,terminal_output +838,293960,"TERMINAL",0,0,"⠼",,terminal_output +839,294040,"TERMINAL",0,0,"⠴",,terminal_output +840,294121,"TERMINAL",0,0,"⠦",,terminal_output +841,294202,"TERMINAL",0,0,"⠧",,terminal_output +842,294283,"TERMINAL",0,0,"⠇",,terminal_output +843,294363,"TERMINAL",0,0,"⠏",,terminal_output +844,294441,"TERMINAL",0,0,"⠋",,terminal_output +845,294523,"TERMINAL",0,0,"⠙",,terminal_output +846,294621,"TERMINAL",0,0,"⠹",,terminal_output +847,294701,"TERMINAL",0,0,"⠸",,terminal_output +848,294780,"TERMINAL",0,0,"⠼",,terminal_output +849,294861,"TERMINAL",0,0,"⠴",,terminal_output +850,294943,"TERMINAL",0,0,"⠦",,terminal_output +851,295021,"TERMINAL",0,0,"⠧",,terminal_output +852,295102,"TERMINAL",0,0,"⠇",,terminal_output +853,295186,"TERMINAL",0,0,"⠏",,terminal_output +854,295286,"TERMINAL",0,0,"⠋",,terminal_output +855,295359,"TERMINAL",0,0,"⠙",,terminal_output +856,295440,"TERMINAL",0,0,"⠹",,terminal_output +857,295521,"TERMINAL",0,0,"⠸",,terminal_output +858,295601,"TERMINAL",0,0,"⠼",,terminal_output +859,295682,"TERMINAL",0,0,"⠴",,terminal_output +860,295763,"TERMINAL",0,0,"⠦",,terminal_output +861,295844,"TERMINAL",0,0,"⠧",,terminal_output +862,295926,"TERMINAL",0,0,"⠇",,terminal_output +863,296013,"TERMINAL",0,0,"⠏",,terminal_output +864,296086,"TERMINAL",0,0,"⠋",,terminal_output +865,296170,"TERMINAL",0,0,"⠙",,terminal_output +866,296251,"TERMINAL",0,0,"⠹",,terminal_output +867,296330,"TERMINAL",0,0,"⠸",,terminal_output +868,296411,"TERMINAL",0,0,"⠼",,terminal_output +869,296492,"TERMINAL",0,0,"⠴",,terminal_output +870,296572,"TERMINAL",0,0,"⠦",,terminal_output +871,296651,"TERMINAL",0,0,"⠧",,terminal_output +872,296732,"TERMINAL",0,0,"⠇",,terminal_output +873,296815,"TERMINAL",0,0,"⠏",,terminal_output +874,296891,"TERMINAL",0,0,"⠋",,terminal_output +875,296970,"TERMINAL",0,0,"⠙",,terminal_output +876,297050,"TERMINAL",0,0,"⠹",,terminal_output +877,297130,"TERMINAL",0,0,"⠸",,terminal_output +878,297211,"TERMINAL",0,0,"⠼",,terminal_output +879,297291,"TERMINAL",0,0,"⠴",,terminal_output +880,297374,"TERMINAL",0,0,"⠦",,terminal_output +881,297453,"TERMINAL",0,0,"⠧",,terminal_output +882,297533,"TERMINAL",0,0,"⠇",,terminal_output +883,297613,"TERMINAL",0,0,"⠏",,terminal_output +884,297694,"TERMINAL",0,0,"⠋",,terminal_output +885,297775,"TERMINAL",0,0,"⠙",,terminal_output +886,297853,"TERMINAL",0,0,"⠹",,terminal_output +887,297934,"TERMINAL",0,0,"⠸",,terminal_output +888,298014,"TERMINAL",0,0,"⠼",,terminal_output +889,298094,"TERMINAL",0,0,"⠴",,terminal_output +890,298173,"TERMINAL",0,0,"⠦",,terminal_output +891,298254,"TERMINAL",0,0,"⠧",,terminal_output +892,298334,"TERMINAL",0,0,"⠇",,terminal_output +893,298415,"TERMINAL",0,0,"⠏",,terminal_output +894,298494,"TERMINAL",0,0,"⠋",,terminal_output +895,298573,"TERMINAL",0,0,"⠙",,terminal_output +896,298656,"TERMINAL",0,0,"⠹",,terminal_output +897,298737,"TERMINAL",0,0,"⠸",,terminal_output +898,298817,"TERMINAL",0,0,"⠼",,terminal_output +899,298899,"TERMINAL",0,0,"⠴",,terminal_output +900,298979,"TERMINAL",0,0,"⠦",,terminal_output +901,299058,"TERMINAL",0,0,"⠧",,terminal_output +902,299139,"TERMINAL",0,0,"⠇",,terminal_output +903,299220,"TERMINAL",0,0,"⠏",,terminal_output +904,299313,"TERMINAL",0,0,"⠋",,terminal_output +905,299394,"TERMINAL",0,0,"⠙",,terminal_output +906,299475,"TERMINAL",0,0,"⠹",,terminal_output +907,299555,"TERMINAL",0,0,"⠸",,terminal_output +908,299638,"TERMINAL",0,0,"⠼",,terminal_output +909,299717,"TERMINAL",0,0,"⠴",,terminal_output +910,299800,"TERMINAL",0,0,"⠦",,terminal_output +911,299879,"TERMINAL",0,0,"⠧",,terminal_output +912,299961,"TERMINAL",0,0,"⠇",,terminal_output +913,300041,"TERMINAL",0,0,"⠏",,terminal_output +914,300122,"TERMINAL",0,0,"⠋",,terminal_output +915,300204,"TERMINAL",0,0,"⠙",,terminal_output +916,300284,"TERMINAL",0,0,"⠹",,terminal_output +917,300366,"TERMINAL",0,0,"⠸",,terminal_output +918,300447,"TERMINAL",0,0,"⠼",,terminal_output +919,300528,"TERMINAL",0,0,"⠴",,terminal_output +920,300610,"TERMINAL",0,0,"⠦",,terminal_output +921,300691,"TERMINAL",0,0,"⠧",,terminal_output +922,300773,"TERMINAL",0,0,"⠇",,terminal_output +923,300989,"TERMINAL",0,0,"⠏",,terminal_output +924,301052,"TERMINAL",0,0,"⠋\r\nadded 978 packages, and audited 984 packages in 18s\r\n⠋\r\n⠋187 packages are looking for funding\r\n⠋ run `npm fund` for details\r\n⠋\r\nfound 0 vulnerabilities\r\n⠋% \r \r",,terminal_output +925,374417,"TERMINAL",0,0,"npm run storybook",,terminal_command +926,374468,"TERMINAL",0,0,"]633;C",,terminal_output +927,374749,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\n",,terminal_output +928,375925,"TERMINAL",0,0,"storybook v9.1.7\r\n\r\n",,terminal_output +929,376672,"TERMINAL",0,0,"(node:36013) ExperimentalWarning: Type Stripping is an experimental feature and might change at any time\r\n(Use `node --trace-warnings ...` to show where the warning was created)\r\n",,terminal_output +930,376730,"TERMINAL",0,0,"Attention: Storybook now collects completely anonymous telemetry regarding usage. This information is used to shape Storybook's roadmap and prioritize features.\r\nYou can learn more, including how to opt-out if you'd not like to participate in this anonymous program, by visiting the following URL:\r\nhttps://storybook.js.org/telemetry\r\n\r\n",,terminal_output +931,377210,"TERMINAL",0,0,"info => Starting manager..\r\n",,terminal_output +932,377331,"TERMINAL",0,0,"No story files found for the specified pattern: src/**/*.mdx\r\n",,terminal_output +933,377521,"TERMINAL",0,0,"info => Starting preview..\r\n",,terminal_output +934,379138,"TERMINAL",0,0,"Building webview for vscode\r\n",,terminal_output +935,380938,"TERMINAL",0,0,"info Using tsconfig paths for react-docgen\r\nThe `define` option contains an object with ""PATH"" for ""process.env"" key. It looks like you may have passed the entire `process.env` object to `define`, which can unintentionally expose all environment variables. This poses a security risk and is discouraged.\r\n",,terminal_output +936,387648,"TERMINAL",0,0,"╭──────────────────────────────────────────────────────────────────────────────────────────╮\r\n│ │\r\n│ Storybook 9.1.7 for react-vite started │\r\n│ 312 ms for manager and 10 s for preview │\r\n│ │\r\n│ Local: http://localhost:6006/ │\r\n│ On your network: http://192.168.178.68:6006/ │\r\n│ │\r\n│ A new version (10.1.0) is available! │\r\n│ │\r\n│ Upgrade now: npx storybook@latest upgrade │\r\n│ │\r\n│ Read full changelog: https://github.com/storybookjs/storybook/blob/main/CHANGELOG.md │\r\n│ │\r\n╰──────────────────────────────────────────────────────────────────────────────────────────╯\r\n",,terminal_output +937,390868,"TERMINAL",0,0,"[baseline-browser-mapping] The data in this module is over two months old. To ensure accurate Baseline data, please update: `npm i baseline-browser-mapping@latest -D`\r\n",,terminal_output +938,395179,"TERMINAL",0,0,"12:54:54 PM [vite] (client) Pre-transform error: Failed to resolve import ""@shared/proto/cline/common"" from ""src/context/ClineAuthContext.tsx"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/webview-ui/src/context/ClineAuthContext.tsx:2:29\r\n 20 | import { jsxDEV as _jsxDEV } from ""react/jsx-dev-runtime"";\r\n 21 | var _s = $RefreshSig$(), _s1 = $RefreshSig$();\r\n 22 | import { EmptyRequest } from ""@shared/proto/cline/common"";\r\n | ^\r\n 23 | import deepEqual from ""fast-deep-equal"";\r\n 24 | import { createContext, useCallback, useContext, useEffect, useMemo, useState } from ""react"";\r\n12:54:54 PM [vite] (client) Pre-transform error: Failed to resolve import ""../services/grpc-client"" from ""src/context/ExtensionStateContext.tsx"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/webview-ui/src/context/ExtensionStateContext.tsx:31:91\r\n 34 | import { Environment } from ""../../../src/config"";\r\n 35 | import { basetenDefaultModelId, basetenModels, groqDefaultModelId, groqModels, openRouterDefaultModelId, openRouterDe...\r\n 36 | import { McpServiceClient, ModelsServiceClient, StateServiceClient, UiServiceClient } from ""../services/grpc-client"";\r\n | ^\r\n 37 | export const ExtensionStateContext = /*#__PURE__*/ createContext(undefined);\r\n 38 | export const ExtensionStateContextProvider = ({ children })=>{\r\n",,terminal_output +939,395343,"TERMINAL",0,0,"12:54:54 PM [vite] (client) Pre-transform error: Failed to resolve import ""@shared/proto/cline/ui"" from ""../src/shared/proto-conversions/cline-message.ts"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/src/shared/proto-conversions/cline-message.ts:3:88\r\n 1 | import { ClineAsk, ClineMessageType, ClineSay } from ""@shared/proto/cline/ui"";\r\n | ^\r\n 2 | // Helper function to convert ClineAsk string to enum\r\n 3 | function convertClineAskToProtoEnum(ask) {\r\n12:54:54 PM [vite] (client) Pre-transform error: Failed to resolve import ""@shared/proto/cline/mcp"" from ""../src/shared/proto-conversions/mcp/mcp-server-conversion.ts"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/src/shared/proto-conversions/mcp/mcp-server-conversion.ts:7:7\r\n 1 | import { McpServerStatus } from ""@shared/proto/cline/mcp"";\r\n | ^\r\n 2 | // Helper to convert TS status to Proto enum\r\n 3 | function convertMcpStatusToProto(status) {\r\n12:54:54 PM [vite] (client) Pre-transform error: Failed to resolve import ""@shared/proto/cline/models"" from ""../src/shared/proto-conversions/models/typeConversion.ts"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/src/shared/proto-conversions/models/typeConversion.ts:8:7\r\n 1 | import { OpenRouterModelInfo, ThinkingConfig } from ""@shared/proto/cline/models"";\r\n | ^\r\n 2 | /**\r\n 3 | * Convert protobuf ThinkingConfig to application ThinkingConfig\r\n",,terminal_output +940,398882,"TERMINAL",0,0,"\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n12:54:58 PM [vite] Internal server error: Failed to resolve import ""@shared/proto/cline/common"" from ""src/context/ClineAuthContext.tsx"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/webview-ui/src/context/ClineAuthContext.tsx:2:29\r\n 20 | import { jsxDEV as _jsxDEV } from ""react/jsx-dev-runtime"";\r\n 21 | var _s = $RefreshSig$(), _s1 = $RefreshSig$();\r\n 22 | import { EmptyRequest } from ""@shared/proto/cline/common"";\r\n | ^\r\n 23 | import deepEqual from ""fast-deep-equal"";\r\n 24 | import { createContext, useCallback, useContext, useEffect, useMemo, useState } from ""react"";\r\n at TransformPluginContext._formatLog (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:29618:43)\r\n at TransformPluginContext.error (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:29615:14)\r\n at normalizeUrl (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:27738:18)\r\n at async file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:27796:32\r\n at async Promise.all (index 3)\r\n at async TransformPluginContext.transform (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:27764:4)\r\n at async EnvironmentPluginContainer.transform (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:29416:14)\r\n at async loadAndTransform (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:23287:26)\r\n at async viteTransformMiddleware (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:25159:20)\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n12:54:58 PM [vite] Internal server error: Failed to resolve import ""../services/grpc-client"" from ""src/context/ExtensionStateContext.tsx"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/webview-ui/src/context/ExtensionStateContext.tsx:31:91\r\n 34 | import { Environment } from ""../../../src/config"";\r\n 35 | import { basetenDefaultModelId, basetenModels, groqDefaultModelId, groqModels, openRouterDefaultModelId, openRouterDe...\r\n 36 | import { McpServiceClient, ModelsServiceClient, StateServiceClient, UiServiceClient } from ""../services/grpc-client"";\r\n | ^\r\n 37 | export const ExtensionStateContext = /*#__PURE__*/ createContext(undefined);\r\n 38 | export const ExtensionStateContextProvider = ({ children })=>{\r\n at TransformPluginContext._formatLog (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:29618:43)\r\n at TransformPluginContext.error (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:29615:14)\r\n at normalizeUrl (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:27738:18)\r\n at async file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:27796:32\r\n at async Promise.all (index 17)\r\n at async TransformPluginContext.transform (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:27764:4)\r\n at async EnvironmentPluginContainer.transform (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:29416:14)\r\n at async loadAndTransform (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:23287:26)\r\n at async viteTransformMiddleware (file:///Users/franzsrambical/Documents/pdoom/cline/webview-ui/node_modules/vite/dist/node/chunks/config.js:25159:20)\r\n12:54:58 PM [vite] (client) Pre-transform error: Failed to resolve import ""@shared/proto/cline/mcp"" from ""../src/shared/proto-conversions/mcp/mcp-server-conversion.ts"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/src/shared/proto-conversions/mcp/mcp-server-conversion.ts:7:7\r\n 1 | import { McpServerStatus } from ""@shared/proto/cline/mcp"";\r\n | ^\r\n 2 | // Helper to convert TS status to Proto enum\r\n 3 | function convertMcpStatusToProto(status) {\r\n12:54:58 PM [vite] (client) Pre-transform error: Failed to resolve import ""@shared/proto/cline/models"" from ""../src/shared/proto-conversions/models/typeConversion.ts"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/src/shared/proto-conversions/models/typeConversion.ts:8:7\r\n 1 | import { OpenRouterModelInfo, ThinkingConfig } from ""@shared/proto/cline/models"";\r\n | ^\r\n 2 | /**\r\n 3 | * Convert protobuf ThinkingConfig to application ThinkingConfig\r\n12:54:58 PM [vite] (client) Pre-transform error: Failed to resolve import ""@shared/proto/cline/ui"" from ""../src/shared/proto-conversions/cline-message.ts"". Does the file exist?\r\n Plugin: vite:import-analysis\r\n File: /Users/franzsrambical/Documents/pdoom/cline/src/shared/proto-conversions/cline-message.ts:3:88\r\n 1 | import { ClineAsk, ClineMessageType, ClineSay } from ""@shared/proto/cline/ui"";\r\n | ^\r\n 2 | // Helper function to convert ClineAsk string to enum\r\n 3 | function convertClineAskToProtoEnum(ask) {\r\n",,terminal_output +941,2981449,"TERMINAL",0,0,"\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n1:38:01 PM [vite] (client) page reload src/services/grpc-client.ts\r\n",,terminal_output +942,2987755,"TERMINAL",0,0,"1:38:07 PM [vite] (client) ✨ new dependencies optimized: @bufbuild/protobuf/wire, uuid\r\n1:38:07 PM [vite] (client) ✨ optimized dependencies changed. reloading\r\n",,terminal_output +943,2992777,"TERMINAL",0,0,"\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n1:38:12 PM [vite] (client) hmr update /src/index.css\r\n",,terminal_output +944,3008542,"TERMINAL",0,0,"\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n1:38:28 PM [vite] (client) hmr update /src/index.css (x2)\r\n",,terminal_output +945,3026313,"TERMINAL",0,0,"\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n1:38:45 PM [vite] (client) hmr update /src/index.css (x3)\r\n",,terminal_output +946,3034948,"TERMINAL",0,0,"^C",,terminal_output +947,3035530,"TERMINAL",0,0,"⠙% \r \r",,terminal_output +948,3036908,"TERMINAL",0,0,"npm run storybook",,terminal_command +949,3036960,"TERMINAL",0,0,"]633;C",,terminal_output +950,3037124,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\n",,terminal_output +951,3038143,"TERMINAL",0,0,"storybook v9.1.7\r\n\r\n",,terminal_output +952,3038311,"TERMINAL",0,0,"(node:43008) ExperimentalWarning: Type Stripping is an experimental feature and might change at any time\r\n(Use `node --trace-warnings ...` to show where the warning was created)\r\n",,terminal_output +953,3038770,"TERMINAL",0,0,"info => Starting manager..\r\n",,terminal_output +954,3038852,"TERMINAL",0,0,"No story files found for the specified pattern: src/**/*.mdx\r\n",,terminal_output +955,3039062,"TERMINAL",0,0,"info => Starting preview..\r\n",,terminal_output +956,3040101,"TERMINAL",0,0,"Building webview for vscode\r\n",,terminal_output +957,3042195,"TERMINAL",0,0,"info Using tsconfig paths for react-docgen\r\n=> Failed to build the preview\r\nReferenceError: __dirname is not defined\r\n at Object.viteFinal (./.storybook/main.ts:13:17)\r\n at ./node_modules/storybook/dist/common/index.cjs:23650:18\r\n at async createViteServer (./node_modules/@storybook/builder-vite/dist/index.js:92:25)\r\n at async Module.start (./node_modules/@storybook/builder-vite/dist/index.js:92:658)\r\n at async storybookDevServer (./node_modules/storybook/dist/core-server/index.cjs:42917:79)\r\n at async buildOrThrow (./node_modules/storybook/dist/core-server/index.cjs:39285:12)\r\n at async buildDevStandalone (./node_modules/storybook/dist/core-server/index.cjs:44170:78)\r\n at async withTelemetry (./node_modules/storybook/dist/core-server/index.cjs:42279:12)\r\n at async dev (./node_modules/storybook/dist/cli/bin/index.cjs:5905:3)\r\n at async r. (./node_modules/storybook/dist/cli/bin/index.cjs:6015:74)\r\n\r\nBroken build, fix the error above.\r\nYou may need to refresh the browser.\r\n\r\n[?25l? Would you like to help improve Storybook by sending anonymous crash reports? › (Y/n)",,terminal_output +958,3056937,"TERMINAL",0,0,"✔ Would you like to help improve Storybook by sending anonymous crash reports? … no\r\n[?25h",,terminal_output +959,3057351,"TERMINAL",0,0,"⠙% \r \r",,terminal_output +960,3058995,"TERMINAL",0,0,"npm run storybook",,terminal_command +961,3059046,"TERMINAL",0,0,"]633;C",,terminal_output +962,3059206,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\n",,terminal_output +963,3059534,"TERMINAL",0,0,"storybook v9.1.7\r\n\r\n",,terminal_output +964,3059688,"TERMINAL",0,0,"(node:43148) ExperimentalWarning: Type Stripping is an experimental feature and might change at any time\r\n(Use `node --trace-warnings ...` to show where the warning was created)\r\n",,terminal_output +965,3060001,"TERMINAL",0,0,"info => Starting manager..\r\nNo story files found for the specified pattern: src/**/*.mdx\r\n",,terminal_output +966,3060120,"TERMINAL",0,0,"info => Starting preview..\r\n",,terminal_output +967,3060243,"TERMINAL",0,0,"Building webview for vscode\r\n",,terminal_output +968,3060516,"TERMINAL",0,0,"info Using tsconfig paths for react-docgen\r\n=> Failed to build the preview\r\nReferenceError: __dirname is not defined\r\n at Object.viteFinal (./.storybook/main.ts:13:17)\r\n at ./node_modules/storybook/dist/common/index.cjs:23650:18\r\n at async createViteServer (./node_modules/@storybook/builder-vite/dist/index.js:92:25)\r\n at async Module.start (./node_modules/@storybook/builder-vite/dist/index.js:92:658)\r\n at async storybookDevServer (./node_modules/storybook/dist/core-server/index.cjs:42917:79)\r\n at async buildOrThrow (./node_modules/storybook/dist/core-server/index.cjs:39285:12)\r\n at async buildDevStandalone (./node_modules/storybook/dist/core-server/index.cjs:44170:78)\r\n at async withTelemetry (./node_modules/storybook/dist/core-server/index.cjs:42279:12)\r\n at async dev (./node_modules/storybook/dist/cli/bin/index.cjs:5905:3)\r\n at async r. (./node_modules/storybook/dist/cli/bin/index.cjs:6015:74)\r\n\r\nBroken build, fix the error above.\r\nYou may need to refresh the browser.\r\n\r\n",,terminal_output +969,3060956,"TERMINAL",0,0,"⠙% \r \r",,terminal_output +970,3078567,"TERMINAL",0,0,"npm run storybook",,terminal_command +971,3078619,"TERMINAL",0,0,"]633;C",,terminal_output +972,3078759,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\n",,terminal_output +973,3079085,"TERMINAL",0,0,"storybook v9.1.7\r\n\r\n",,terminal_output +974,3079242,"TERMINAL",0,0,"(node:43265) ExperimentalWarning: Type Stripping is an experimental feature and might change at any time\r\n(Use `node --trace-warnings ...` to show where the warning was created)\r\n",,terminal_output +975,3079508,"TERMINAL",0,0,"info => Starting manager..\r\nNo story files found for the specified pattern: src/**/*.mdx\r\n",,terminal_output +976,3079609,"TERMINAL",0,0,"info => Starting preview..\r\n",,terminal_output +977,3079857,"TERMINAL",0,0,"Building webview for vscode\r\n",,terminal_output +978,3080137,"TERMINAL",0,0,"info Using tsconfig paths for react-docgen\r\n=> Failed to build the preview\r\nReferenceError: __dirname is not defined\r\n at Object.viteFinal (./.storybook/main.ts:13:17)\r\n at ./node_modules/storybook/dist/common/index.cjs:23650:18\r\n at async createViteServer (./node_modules/@storybook/builder-vite/dist/index.js:92:25)\r\n at async Module.start (./node_modules/@storybook/builder-vite/dist/index.js:92:658)\r\n at async storybookDevServer (./node_modules/storybook/dist/core-server/index.cjs:42917:79)\r\n at async buildOrThrow (./node_modules/storybook/dist/core-server/index.cjs:39285:12)\r\n at async buildDevStandalone (./node_modules/storybook/dist/core-server/index.cjs:44170:78)\r\n at async withTelemetry (./node_modules/storybook/dist/core-server/index.cjs:42279:12)\r\n at async dev (./node_modules/storybook/dist/cli/bin/index.cjs:5905:3)\r\n at async r. (./node_modules/storybook/dist/cli/bin/index.cjs:6015:74)\r\n\r\nBroken build, fix the error above.\r\nYou may need to refresh the browser.\r\n\r\n",,terminal_output +979,3080559,"TERMINAL",0,0,"⠙% \r \r",,terminal_output +980,3082050,"TERMINAL",0,0,"cd ..",,terminal_command +981,3082051,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +982,3087158,"TERMINAL",0,0,"npm run storybook",,terminal_command +983,3087209,"TERMINAL",0,0,"]633;C",,terminal_output +984,3087345,"TERMINAL",0,0,"\r\n> claude-dev@3.38.3 storybook\r\n> cd webview-ui && npm run storybook\r\n\r\n",,terminal_output +985,3087427,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\n",,terminal_output +986,3087717,"TERMINAL",0,0,"storybook v9.1.7\r\n\r\n",,terminal_output +987,3087864,"TERMINAL",0,0,"(node:43453) ExperimentalWarning: Type Stripping is an experimental feature and might change at any time\r\n(Use `node --trace-warnings ...` to show where the warning was created)\r\n",,terminal_output +988,3088122,"TERMINAL",0,0,"info => Starting manager..\r\nNo story files found for the specified pattern: src/**/*.mdx\r\n",,terminal_output +989,3088223,"TERMINAL",0,0,"info => Starting preview..\r\n",,terminal_output +990,3088362,"TERMINAL",0,0,"Building webview for vscode\r\n",,terminal_output +991,3088601,"TERMINAL",0,0,"info Using tsconfig paths for react-docgen\r\n=> Failed to build the preview\r\nReferenceError: __dirname is not defined\r\n at Object.viteFinal (./.storybook/main.ts:13:17)\r\n at ./node_modules/storybook/dist/common/index.cjs:23650:18\r\n at async createViteServer (./node_modules/@storybook/builder-vite/dist/index.js:92:25)\r\n at async Module.start (./node_modules/@storybook/builder-vite/dist/index.js:92:658)\r\n at async storybookDevServer (./node_modules/storybook/dist/core-server/index.cjs:42917:79)\r\n at async buildOrThrow (./node_modules/storybook/dist/core-server/index.cjs:39285:12)\r\n at async buildDevStandalone (./node_modules/storybook/dist/core-server/index.cjs:44170:78)\r\n at async withTelemetry (./node_modules/storybook/dist/core-server/index.cjs:42279:12)\r\n at async dev (./node_modules/storybook/dist/cli/bin/index.cjs:5905:3)\r\n at async r. (./node_modules/storybook/dist/cli/bin/index.cjs:6015:74)\r\n\r\nBroken build, fix the error above.\r\nYou may need to refresh the browser.\r\n\r\n",,terminal_output +992,3088992,"TERMINAL",0,0,"⠙⠙% \r \r",,terminal_output +993,3161357,"TERMINAL",0,0,"cd webview-ui",,terminal_command +994,3161357,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +995,3162625,"TERMINAL",0,0,"npm run storybook",,terminal_command +996,3162677,"TERMINAL",0,0,"]633;C",,terminal_output +997,3162941,"TERMINAL",0,0,"\r\n> webview-ui@0.3.0 storybook\r\n> storybook dev -p 6006\r\n\r\n",,terminal_output +998,3163284,"TERMINAL",0,0,"storybook v9.1.7\r\n\r\n",,terminal_output +999,3163477,"TERMINAL",0,0,"(node:43708) ExperimentalWarning: Type Stripping is an experimental feature and might change at any time\r\n(Use `node --trace-warnings ...` to show where the warning was created)\r\n",,terminal_output +1000,3163764,"TERMINAL",0,0,"info => Starting manager..\r\nNo story files found for the specified pattern: src/**/*.mdx\r\n",,terminal_output +1001,3163883,"TERMINAL",0,0,"info => Starting preview..\r\n",,terminal_output +1002,3164047,"TERMINAL",0,0,"Building webview for vscode\r\n",,terminal_output +1003,3164319,"TERMINAL",0,0,"info Using tsconfig paths for react-docgen\r\nThe `define` option contains an object with ""PATH"" for ""process.env"" key. It looks like you may have passed the entire `process.env` object to `define`, which can unintentionally expose all environment variables. This poses a security risk and is discouraged.\r\n",,terminal_output +1004,3169884,"TERMINAL",0,0,"╭──────────────────────────────────────────────────────────────────────────────────────────╮\r\n│ │\r\n│ Storybook 9.1.7 for react-vite started │\r\n│ 157 ms for manager and 5.95 s for preview │\r\n│ │\r\n│ Local: http://localhost:6006/ │\r\n│ On your network: http://192.168.178.68:6006/ │\r\n│ │\r\n│ A new version (10.1.0) is available! │\r\n│ │\r\n│ Upgrade now: npx storybook@latest upgrade │\r\n│ │\r\n│ Read full changelog: https://github.com/storybookjs/storybook/blob/main/CHANGELOG.md │\r\n│ │\r\n╰──────────────────────────────────────────────────────────────────────────────────────────╯\r\n",,terminal_output +1005,3170751,"TERMINAL",0,0,"[baseline-browser-mapping] The data in this module is over two months old. To ensure accurate Baseline data, please update: `npm i baseline-browser-mapping@latest -D`\r\n",,terminal_output +1006,3210633,"TERMINAL",0,0,"The request id ""/Users/franzsrambical/Documents/pdoom/cline/node_modules/@vscode/codicons/dist/codicon.ttf"" is outside of Vite serving allow list.\r\n\r\n- /Users/franzsrambical/Documents/pdoom/cline/webview-ui\r\n\r\nRefer to docs https://vite.dev/config/server-options.html#server-fs-allow for configurations and more details.\r\n\r\n",,terminal_output diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-8eaf046e-99c4-4091-a85d-91e359564aa51756825908437-2025_09_02-17.11.50.753/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-8eaf046e-99c4-4091-a85d-91e359564aa51756825908437-2025_09_02-17.11.50.753/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..3156c3d473a0193a2600ed282f62390997a77c30 --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-8eaf046e-99c4-4091-a85d-91e359564aa51756825908437-2025_09_02-17.11.50.753/source.csv @@ -0,0 +1,2870 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,2,"examples/bibliography.bib",0,0,"@article{radford2018improving,\n title = {Improving language understanding by generative pre-training},\n author = {Radford, Alec and Narasimhan, Karthik and Salimans, Tim and\n Sutskever, Ilya and others},\n}\n\n@article{radford2019language,\n title = {Language models are unsupervised multitask learners},\n author = {Radford, Alec and Wu, Jeffrey and Child, Rewon and Luan, David and\n Amodei, Dario and Sutskever, Ilya and others},\n journal = {OpenAI blog},\n volume = {1},\n number = {8},\n pages = {9},\n year = {2019},\n}\n\n@article{brown2020language,\n title = {Language models are few-shot learners},\n author = {Brown, Tom and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie\n and Kaplan, Jared D and Dhariwal, Prafulla and Neelakantan, Arvind\n and Shyam, Pranav and Sastry, Girish and Askell, Amanda and others},\n journal = {Advances in neural information processing systems},\n volume = {33},\n pages = {1877--1901},\n year = {2020},\n}\n\n@article{raffel2020exploring,\n title = {Exploring the limits of transfer learning with a unified text-to-text\n transformer},\n author = {Raffel, Colin and Shazeer, Noam and Roberts, Adam and Lee, Katherine\n and Narang, Sharan and Matena, Michael and Zhou, Yanqi and Li, Wei\n and Liu, Peter J},\n journal = {Journal of machine learning research},\n volume = {21},\n number = {140},\n pages = {1--67},\n year = {2020},\n}\n\n@article{touvron2023llama,\n title = {Llama 2: Open foundation and fine-tuned chat models},\n author = {Touvron, Hugo and Martin, Louis and Stone, Kevin and Albert, Peter\n and Almahairi, Amjad and Babaei, Yasmine and Bashlykov, Nikolay and\n Batra, Soumya and Bhargava, Prajjwal and Bhosale, Shruti and others},\n journal = {arXiv preprint arXiv:2307.09288},\n year = {2023},\n}\n\n@article{bai2023qwen,\n title = {Qwen technical report},\n author = {Bai, Jinze and Bai, Shuai and Chu, Yunfei and Cui, Zeyu and Dang,\n Kai and Deng, Xiaodong and Fan, Yang and Ge, Wenbin and Han, Yu and\n Huang, Fei and others},\n journal = {arXiv preprint arXiv:2309.16609},\n year = {2023},\n}\n\n@article{young2024yi,\n title = {Yi: Open foundation models by 01. ai},\n author = {Young, Alex and Chen, Bei and Li, Chao and Huang, Chengen and Zhang,\n Ge and Zhang, Guanwei and Li, Heng and Zhu, Jiangcheng and Chen,\n Jianqun and Chang, Jing and others},\n journal = {arXiv preprint arXiv:2403.04652},\n year = {2024},\n}\n\n@article{vaswani2017attention,\n title = {Attention is all you need},\n author = {Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit,\n Jakob and Jones, Llion and Gomez, Aidan N and Kaiser, {\L}ukasz and\n Polosukhin, Illia},\n journal = {Advances in neural information processing systems},\n volume = {30},\n year = {2017},\n}\n\n@article{raffel2020exploring,\n title = {Exploring the limits of transfer learning with a unified text-to-text\n transformer},\n author = {Raffel, Colin and Shazeer, Noam and Roberts, Adam and Lee, Katherine\n and Narang, Sharan and Matena, Michael and Zhou, Yanqi and Li, Wei\n and Liu, Peter J},\n journal = {Journal of machine learning research},\n volume = {21},\n number = {140},\n pages = {1--67},\n year = {2020},\n}\n\n@inproceedings{zhou2024what,\n title = {What Algorithms can Transformers Learn? A Study in Length\n Generalization},\n author = {Hattie Zhou and Arwen Bradley and Etai Littwin and Noam Razin and\n Omid Saremi and Joshua M. Susskind and Samy Bengio and Preetum\n Nakkiran},\n booktitle = {The Twelfth International Conference on Learning Representations},\n year = {2024},\n url = {https://openreview.net/forum?id=AssIuHnmHX},\n}\n\n@inproceedings{ding2024causallm,\n title = {Causal{LM} is not optimal for in-context learning},\n author = {Nan Ding and Tomer Levinboim and Jialin Wu and Sebastian Goodman and\n Radu Soricut},\n booktitle = {The Twelfth International Conference on Learning Representations},\n year = {2024},\n url = {https://openreview.net/forum?id=guRNebwZBb},\n}\n\n@article{williams1989learning,\n title = {A learning algorithm for continually running fully recurrent neural\n networks},\n author = {Williams, Ronald J and Zipser, David},\n journal = {Neural computation},\n volume = {1},\n number = {2},\n pages = {270--280},\n year = {1989},\n publisher = {MIT Press One Rogers Street, Cambridge, MA 02142-1209, USA\n journals-info~…},\n}\n\n@article{tay2022ul2,\n title = {Ul2: Unifying language learning paradigms},\n author = {Tay, Yi and Dehghani, Mostafa and Tran, Vinh Q and Garcia, Xavier\n and Wei, Jason and Wang, Xuezhi and Chung, Hyung Won and Shakeri,\n Siamak and Bahri, Dara and Schuster, Tal and others},\n journal = {arXiv preprint arXiv:2205.05131},\n year = {2022},\n}\n\n@misc{pfau2023last,\n title = {Last I checked, it was still not possible for a neural network alone\n (i.e. no MCTS) to beat the world's best Go players...},\n author = {Pfau, David},\n year = {2023},\n url = {https://twitter.com/pfau/status/1732785418565796167},\n note = {Accessed: 2023-12-07},\n}\n\n@article{deepmind2023alphacode,\n title = {AlphaCode 2 Technical Report},\n author = {Team, AlphaCode and Deepmind, Google},\n year = {2023},\n journal = {Google Deepmind},\n url = {\n https://storage.googleapis.com/deepmind-media/AlphaCode2/AlphaCode2_Tech_Report.pdf\n },\n}\n\n@article{reuters2023sam,\n author = {Tong, Anna and Dastin, Jeffrey and Hu, Krystal},\n title = {Sam Altman's ouster from OpenAI was precipitated by letter to board\n about AI breakthrough},\n journal = {Reuters},\n year = {2023},\n url = {\n https://www.reuters.com/technology/sam-altmans-ouster-openai-was-precipitated-by-letter-board-about-ai-breakthrough-2023-11-22/\n },\n note = {Accessed: 2023-12-07},\n}\n\n@misc{imbue2023podcast,\n title = {Noam Brown, FAIR: On achieving human-level performance in poker and\n Diplomacy, and the power of spending compute at inference time},\n author = {Noam Brown},\n howpublished = {\n https://imbue.com/podcast/2023-02-09-podcast-episode-27-noam-brown/\n },\n year = {2023},\n note = {Podcast episode 27, February 9, 2023},\n}\n\n@misc{karpathy2023youtube,\n author = {Karpathy, Andrej},\n title = {[1hr Talk] Intro to Large Language Models},\n howpublished = {YouTube},\n year = {2023},\n note = {Accessed: 2023-12-07},\n url = {https://www.youtube.com/watch?v=zjkBMFhNj_g&t=2100s},\n}\n\n@article{brown2019superhuman,\n title = {Superhuman AI for multiplayer poker},\n author = {Brown, Noam and Sandholm, Tuomas},\n journal = {Science},\n volume = {365},\n number = {6456},\n pages = {885--890},\n year = {2019},\n publisher = {American Association for the Advancement of Science},\n}\n\n@article{silver2016mastering,\n title = {Mastering the game of Go with deep neural networks and tree search},\n author = {Silver, David and Huang, Aja and Maddison, Chris J and Guez, Arthur\n and Sifre, Laurent and Van Den Driessche, George and Schrittwieser,\n Julian and Antonoglou, Ioannis and Panneershelvam, Veda and Lanctot,\n Marc and others},\n journal = {nature},\n volume = {529},\n number = {7587},\n pages = {484--489},\n year = {2016},\n publisher = {Nature Publishing Group},\n}\n\n@article{schrittwieser2020mastering,\n title = {Mastering atari, go, chess and shogi by planning with a learned model\n },\n author = {Schrittwieser, Julian and Antonoglou, Ioannis and Hubert, Thomas and\n Simonyan, Karen and Sifre, Laurent and Schmitt, Simon and Guez,\n Arthur and Lockhart, Edward and Hassabis, Demis and Graepel, Thore\n and others},\n journal = {Nature},\n volume = {588},\n number = {7839},\n pages = {604--609},\n year = {2020},\n publisher = {Nature Publishing Group UK London},\n}\n\n@article{wei2022chain,\n title = {Chain-of-thought prompting elicits reasoning in large language models\n },\n author = {Wei, Jason and Wang, Xuezhi and Schuurmans, Dale and Bosma, Maarten\n and Xia, Fei and Chi, Ed and Le, Quoc V and Zhou, Denny and others},\n journal = {Advances in neural information processing systems},\n volume = {35},\n pages = {24824--24837},\n year = {2022},\n}\n\n@article{yao2024tree,\n title = {Tree of thoughts: Deliberate problem solving with large language\n models},\n author = {Yao, Shunyu and Yu, Dian and Zhao, Jeffrey and Shafran, Izhak and\n Griffiths, Tom and Cao, Yuan and Narasimhan, Karthik},\n journal = {Advances in Neural Information Processing Systems},\n volume = {36},\n year = {2024},\n}\n\n@article{lecun2022path,\n title = {A path towards autonomous machine intelligence version 0.9. 2,\n 2022-06-27},\n author = {LeCun, Yann},\n journal = {Open Review},\n volume = {62},\n number = {1},\n year = {2022},\n}\n\n@article{hoffmann2022training,\n title = {Training compute-optimal large language models},\n author = {Hoffmann, Jordan and Borgeaud, Sebastian and Mensch, Arthur and\n Buchatskaya, Elena and Cai, Trevor and Rutherford, Eliza and Casas,\n Diego de Las and Hendricks, Lisa Anne and Welbl, Johannes and Clark,\n Aidan and others},\n journal = {arXiv preprint arXiv:2203.15556},\n year = {2022},\n}\n\n@article{meta2024introducing,\n title = {Introducing meta llama 3: The most capable openly available llm to\n date},\n author = {Meta, AI},\n journal = {Meta AI.},\n year = {2024},\n}\n\n@misc{riley2024it,\n title = {It's just not a very useful scaling law.},\n author = {@riley_stews},\n year = {2024},\n url = {https://x.com/riley_stews/status/1781019732122198288},\n note = {Accessed: 2023-04-20},\n}\n\n@article{shazeer2017outrageously,\n title = {Outrageously large neural networks: The sparsely-gated\n mixture-of-experts layer},\n author = {Shazeer, Noam and Mirhoseini, Azalia and Maziarz, Krzysztof and\n Davis, Andy and Le, Quoc and Hinton, Geoffrey and Dean, Jeff},\n journal = {arXiv preprint arXiv:1701.06538},\n year = {2017},\n}\n\n@article{fedus2022switch,\n title = {Switch transformers: Scaling to trillion parameter models with simple\n and efficient sparsity},\n author = {Fedus, William and Zoph, Barret and Shazeer, Noam},\n journal = {Journal of Machine Learning Research},\n volume = {23},\n number = {120},\n pages = {1--39},\n year = {2022},\n}\n\n@article{schulman2015high,\n title = {High-dimensional continuous control using generalized advantage\n estimation},\n author = {Schulman, John and Moritz, Philipp and Levine, Sergey and Jordan,\n Michael and Abbeel, Pieter},\n journal = {arXiv preprint arXiv:1506.02438},\n year = {2015},\n}\n\n@article{srambical2025ppo,\n author = {Srambical, Franz},\n title = {PPO Is Secretly Using Monte Carlo Advantage Estimation In LLM\n Post-Training},\n journal = {p(doom) blog},\n year = {2025},\n note = {https://pdoom.org/blog.html},\n}\n\n@article{williams1992simple,\n title = {Simple statistical gradient-following algorithms for connectionist\n reinforcement learning},\n author = {Williams, Ronald J},\n journal = {Machine learning},\n volume = {8},\n pages = {229--256},\n year = {1992},\n publisher = {Springer},\n}\n\n@software{deepmind2020jax,\n title = {The {D}eep{M}ind {JAX} {E}cosystem},\n author = {DeepMind and Babuschkin, Igor and Baumli, Kate and Bell, Alison and\n Bhupatiraju, Surya and Bruce, Jake and Buchlovsky, Peter and Budden,\n David and Cai, Trevor and Clark, Aidan and Danihelka, Ivo and Dedieu,\n Antoine and Fantacci, Claudio and Godwin, Jonathan and Jones, Chris\n and Hemsley, Ross and Hennigan, Tom and Hessel, Matteo and Hou,\n Shaobo and Kapturowski, Steven and Keck, Thomas and Kemaev, Iurii and\n King, Michael and Kunesch, Markus and Martens, Lena and Merzic, Hamza\n and Mikulik, Vladimir and Norman, Tamara and Papamakarios, George and\n Quan, John and Ring, Roman and Ruiz, Francisco and Sanchez, Alvaro\n and Sartran, Laurent and Schneider, Rosalia and Sezener, Eren and\n Spencer, Stephen and Srinivasan, Srivatsan and Stanojevi\'{c}, Milo\v\n {s} and Stokowiec, Wojciech and Wang, Luyu and Zhou, Guangyao and\n Viola, Fabio},\n url = {http://github.com/deepmind},\n year = {2020},\n}\n\n@misc{jax2025jit,\n title = {JAX: Just-in-time compilation},\n author = {James Bradbury and Roy Frostig and Peter Hawkins and Matthew James\n Johnson and Chris Leary and Dougal Maclaurin and George Necula and\n Adam Paszke and Jake Vander{P}las and Skye Wanderman-{M}ilne and Qiao\n Zhang},\n year = {2025},\n url = {https://docs.jax.dev/en/latest/jit-compilation.html},\n note = {Accessed: 2025-03-26},\n}\n\n@misc{jax2025callbacks,\n title = {JAX: External callbacks},\n author = {James Bradbury and Roy Frostig and Peter Hawkins and Matthew James\n Johnson and Chris Leary and Dougal Maclaurin and George Necula and\n Adam Paszke and Jake Vander{P}las and Skye Wanderman-{M}ilne and Qiao\n Zhang},\n year = {2025},\n url = {https://docs.jax.dev/en/latest/external-callbacks.html},\n note = {Accessed: 2025-03-26},\n}\n\n@misc{jax2025checkify,\n title = {JAX: The `checkify` transformation},\n author = {James Bradbury and Roy Frostig and Peter Hawkins and Matthew James\n Johnson and Chris Leary and Dougal Maclaurin and George Necula and\n Adam Paszke and Jake Vander{P}las and Skye Wanderman-{M}ilne and Qiao\n Zhang},\n year = {2025},\n url = {https://docs.jax.dev/en/latest/debugging/checkify_guide.html},\n note = {Accessed: 2025-03-26},\n}\n\n@misc{jax2025key,\n title = {JAX: Key concepts},\n author = {James Bradbury and Roy Frostig and Peter Hawkins and Matthew James\n Johnson and Chris Leary and Dougal Maclaurin and George Necula and\n Adam Paszke and Jake Vander{P}las and Skye Wanderman-{M}ilne and Qiao\n Zhang},\n year = {2025},\n url = {https://docs.jax.dev/en/latest/key-concepts.html},\n note = {Accessed: 2025-03-26},\n}\n\n@software{deepmind2020chex,\n title = {Chex},\n author = {James Bradbury and Roy Frostig and Peter Hawkins and Matthew James\n Johnson and Chris Leary and Dougal Maclaurin and George Necula and\n Adam Paszke and Jake Vander{P}las and Skye Wanderman-{M}ilne and Qiao\n Zhang},\n url = {http://github.com/google-deepmind/chex},\n year = {2020},\n}\n\n@misc{jax2025control,\n title = {JAX: Control flow and logical operators with JIT},\n author = {James Bradbury and Roy Frostig and Peter Hawkins and Matthew James\n Johnson and Chris Leary and Dougal Maclaurin and George Necula and\n Adam Paszke and Jake Vander{P}las and Skye Wanderman-{M}ilne and Qiao\n Zhang},\n year = {2025},\n url = {https://docs.jax.dev/en/latest/control-flow.html},\n note = {Accessed: 2025-03-26},\n}\n\n@misc{xla2025conditional,\n title = {XLA:Operation Semantics:Conditional},\n author = {James Bradbury and Roy Frostig and Peter Hawkins and Matthew James\n Johnson and Chris Leary and Dougal Maclaurin and George Necula and\n Adam Paszke and Jake Vander{P}las and Skye Wanderman-{M}ilne and Qiao\n Zhang},\n year = {2025},\n url = {https://openxla.org/xla/operation_semantics#conditional},\n note = {Accessed: 2025-03-26},\n}\n\n@misc{ayaka76822025error,\n author = {ayaka7682},\n title = {Message on public Discord server: Try this:\n \n import jax from jax._src.error_check import set_error_if, raise_if_error\n \n \n import jax.numpy as jnp\n \n \n @jax.jit\n \n \n def f(x, y):\n \n \n set_error_if(x != 0, 'x must be 0')\n \n \n return jnp.multiply(x, y)\n \n \n f(0, 0)\n \n \n raise_if_error()\n },\n year = {2025},\n url = {\n https://discord.com/channels/1107832795377713302/1107832795688083561/1354171414596419854\n },\n note = {Accessed: 2025-03-26},\n}\n\n@book{sutton1998reinforcement,\n title={Reinforcement learning: An introduction},\n author={Sutton, Richard S and Barto, Andrew G and others},\n volume={1},\n number={1},\n year={1998},\n publisher={MIT press Cambridge}\n}\n\n@article{sutton1999policy,\n title={Policy gradient methods for reinforcement learning with function approximation},\n author={Sutton, Richard S and McAllester, David and Singh, Satinder and Mansour, Yishay},\n journal={Advances in neural information processing systems},\n volume={12},\n year={1999}\n}\n\n@article{degris2012off,\n title={Off-policy actor-critic},\n author={Degris, Thomas and White, Martha and Sutton, Richard S},\n journal={arXiv preprint arXiv:1205.4839},\n year={2012}\n}\n\n@article{schulman2017proximal,\n title={Proximal policy optimization algorithms},\n author={Schulman, John and Wolski, Filip and Dhariwal, Prafulla and Radford, Alec and Klimov, Oleg},\n journal={arXiv preprint arXiv:1707.06347},\n year={2017}\n}\n\n@article{ouyang2022training,\n title={Training language models to follow instructions with human feedback},\n author={Ouyang, Long and Wu, Jeffrey and Jiang, Xu and Almeida, Diogo and Wainwright, Carroll and Mishkin, Pamela and Zhang, Chong and Agarwal, Sandhini and Slama, Katarina and Ray, Alex and others},\n journal={Advances in neural information processing systems},\n volume={35},\n pages={27730--27744},\n year={2022}\n}\n\n\n@misc{openai2025imo,\n title = {We achieved gold medal-level performance 🥇on the 2025 International Mathematical Olympiad with a general-purpose reasoning LLM!},\n author = {OpenAI},\n year = {2025},\n url = {https://x.com/OpenAI/status/1946594928945148246},\n note = {Accessed: 2025-08-05},\n}\n\n@misc{deepmind2025imo,\n title = {Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad},\n author = {Luong, Thang and Lockhart, Edward},\n year = {2025},\n url = {https://deepmind.google/discover/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/},\n note = {DeepMind Blog, July 21, 2025},\n}\n\n@misc{deepseekai2025r1,\n title={DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning}, \n author={DeepSeek-AI},\n year={2025},\n eprint={2501.12948},\n archivePrefix={arXiv},\n primaryClass={cs.CL},\n url={https://arxiv.org/abs/2501.12948}, \n}\n\n@misc{cursor2025tab,\n title = {A New Tab Model},\n author = {Cursor},\n year = {2025},\n url = {https://cursor.com/blog/tab-update},\n note = {Accessed: 2025-08-05},\n}\n\n@inproceedings{bruce2024genie,\n title={Genie: Generative Interactive Environments},\n author={Jake Bruce and Michael D Dennis and Ashley Edwards and Jack Parker-Holder and Yuge Shi and Edward Hughes and Matthew Lai and Aditi Mavalankar and Richie Steigerwald and Chris Apps and Yusuf Aytar and Sarah Maria Elisabeth Bechtle and Feryal Behbahani and Stephanie C.Y. Chan and Nicolas Heess and Lucy Gonzalez and Simon Osindero and Sherjil Ozair and Scott Reed and Jingwei Zhang and Konrad Zolna and Jeff Clune and Nando de Freitas and Satinder Singh and Tim Rockt{\""a}schel},\n booktitle={Forty-first International Conference on Machine Learning},\n year={2024},\n url={https://openreview.net/forum?id=bJbSbJskOS}\n}\n\n@article{parkerholder2024genie2,\n title = {Genie 2: A Large-Scale Foundation World Model},\n author = {Jack Parker-Holder and Philip Ball and Jake Bruce and Vibhavari Dasagi and Kristian Holsheimer and Christos Kaplanis and Alexandre Moufarek and Guy Scully and Jeremy Shar and Jimmy Shi and Stephen Spencer and Jessica Yung and Michael Dennis and Sultan Kenjeyev and Shangbang Long and Vlad Mnih and Harris Chan and Maxime Gazeau and Bonnie Li and Fabio Pardo and Luyu Wang and Lei Zhang and Frederic Besse and Tim Harley and Anna Mitenkova and Jane Wang and Jeff Clune and Demis Hassabis and Raia Hadsell and Adrian Bolton and Satinder Singh and Tim Rockt{\""a}schel},\n year = {2024},\n url = {https://deepmind.google/discover/blog/genie-2-a-large-scale-foundation-world-model/}\n}\n\n@article{deepmind2025genie3,\n title = {Genie 3: A New Frontier for World Models},\n author = {Philip J. Ball and Jakob Bauer and Frank Belletti and Bethanie Brownfield and Ariel Ephrat and Shlomi Fruchter and Agrim Gupta and Kristian Holsheimer and Aleksander Holynski and Jiri Hron and Christos Kaplanis and Marjorie Limont and Matt McGill and Yanko Oliveira and Jack Parker-Holder and Frank Perbet and Guy Scully and Jeremy Shar and Stephen Spencer and Omer Tov and Ruben Villegas and Emma Wang and Jessica Yung and Cip Baetu and Jordi Berbel and David Bridson and Jake Bruce and Gavin Buttimore and Sarah Chakera and Bilva Chandra and Paul Collins and Alex Cullum and Bogdan Damoc and Vibha Dasagi and Maxime Gazeau and Charles Gbadamosi and Woohyun Han and Ed Hirst and Ashyana Kachra and Lucie Kerley and Kristian Kjems and Eva Knoepfel and Vika Koriakin and Jessica Lo and Cong Lu and Zeb Mehring and Alex Moufarek and Henna Nandwani and Valeria Oliveira and Fabio Pardo and Jane Park and Andrew Pierson and Ben Poole and Helen Ran and Tim Salimans and Manuel Sanchez and Igor Saprykin and Amy Shen and Sailesh Sidhwani and Duncan Smith and Joe Stanton and Hamish Tomlinson and Dimple Vijaykumar and Luyu Wang and Piers Wingfield and Nat Wong and Keyang Xu and Christopher Yew and Nick Young and Vadim Zubov and Douglas Eck and Dumitru Erhan and Koray Kavukcuoglu and Demis Hassabis and Zoubin Gharamani and Raia Hadsell and A{\""a}ron van den Oord and Inbar Mosseri and Adrian Bolton and Satinder Singh and Tim Rockt{\""a}schel},\n year = {2025},\n url = {}\n}\n\n@InProceedings{parkerholder2022evolving,\n title = \t {Evolving Curricula with Regret-Based Environment Design},\n author = {Parker-Holder, Jack and Jiang, Minqi and Dennis, Michael and Samvelyan, Mikayel and Foerster, Jakob and Grefenstette, Edward and Rockt{\""a}schel, Tim},\n booktitle = \t {Proceedings of the 39th International Conference on Machine Learning},\n pages = \t {17473--17498},\n year = \t {2022},\n editor = \t {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan},\n volume = \t {162},\n series = \t {Proceedings of Machine Learning Research},\n month = \t {17--23 Jul},\n publisher = {PMLR},\n pdf = \t {https://proceedings.mlr.press/v162/parker-holder22a/parker-holder22a.pdf},\n url = \t {https://proceedings.mlr.press/v162/parker-holder22a.html},\n abstract = \t {Training generally-capable agents with reinforcement learning (RL) remains a significant challenge. A promising avenue for improving the robustness of RL agents is through the use of curricula. One such class of methods frames environment design as a game between a student and a teacher, using regret-based objectives to produce environment instantiations (or levels) at the frontier of the student agent’s capabilities. These methods benefit from theoretical robustness guarantees at equilibrium, yet they often struggle to find effective levels in challenging design spaces in practice. By contrast, evolutionary approaches incrementally alter environment complexity, resulting in potentially open-ended learning, but often rely on domain-specific heuristics and vast amounts of computational resources. This work proposes harnessing the power of evolution in a principled, regret-based curriculum. Our approach, which we call Adversarially Compounding Complexity by Editing Levels (ACCEL), seeks to constantly produce levels at the frontier of an agent’s capabilities, resulting in curricula that start simple but become increasingly complex. ACCEL maintains the theoretical benefits of prior regret-based methods, while providing significant empirical gains in a diverse set of environments. An interactive version of this paper is available at https://accelagent.github.io.}\n}\n\n@article{agarwal2025cosmos,\n title = {Cosmos World Foundation Model Platform for Physical AI},\n author = {Agarwal, Niket and others},\n journal = {arXiv preprint arXiv:2501.03575},\n year = {2025}\n}\n\n@article{bellemare2013arcade,\n title = {The arcade learning environment: An evaluation platform for general agents},\n author = {Bellemare, Marc G and others},\n journal = {Journal of artificial intelligence research},\n volume = {47},\n pages = {253--279},\n year = {2013}\n}\n\n@article{nichol2018retro,\n title={Gotta Learn Fast: A New Benchmark for Generalization in RL},\n author={Nichol, Alex and Pfau, Vicki and Hesse, Christopher and Klimov, Oleg and Schulman, John},\n journal={arXiv preprint arXiv:1804.03720},\n year={2018}\n}\n\n@inproceedings{matthews2024craftax,\n author={Michael Matthews and Michael Beukman and Benjamin Ellis and Mikayel Samvelyan and Matthew Jackson and Samuel Coward and Jakob Foerster},\n title = {Craftax: A Lightning-Fast Benchmark for Open-Ended Reinforcement Learning},\n booktitle = {International Conference on Machine Learning ({ICML})},\n year = {2024}\n}\n\n@inproceedings{NEURIPS2022_9c7008af,\n author = {Baker, Bowen and Akkaya, Ilge and Zhokov, Peter and Huizinga, Joost and Tang, Jie and Ecoffet, Adrien and Houghton, Brandon and Sampedro, Raul and Clune, Jeff},\n booktitle = {Advances in Neural Information Processing Systems},\n editor = {S. Koyejo and S. Mohamed and A. Agarwal and D. Belgrave and K. Cho and A. Oh},\n pages = {24639--24654},\n publisher = {Curran Associates, Inc.},\n title = {Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos},\n url = {https://proceedings.neurips.cc/paper_files/paper/2022/file/9c7008aff45b5d8f0973b23e1a22ada0-Paper-Conference.pdf},\n volume = {35},\n year = {2022}\n}\n\n@inproceedings{osband2020bsuite,\n title={Behaviour Suite for Reinforcement Learning},\n author={Osband, Ian and\n Doron, Yotam and\n Hessel, Matteo and\n Aslanides, John and\n Sezener, Eren and\n Saraiva, Andre and\n McKinney, Katrina and\n Lattimore, Tor and\n {Sz}epesv{\'a}ri, Csaba and\n Singh, Satinder and\n Van Roy, Benjamin and\n Sutton, Richard and\n Silver, David and\n van Hasselt, Hado},\n booktitle={International Conference on Learning Representations},\n year={2020},\n url={https://openreview.net/forum?id=rygf-kSYwH}\n}\n\n@article{nguyen2025crowd-sourcing,\n author = {Nguyen, Alfred and Mahajan, Mihir and Srambical, Franz},\n title = {Crowd-Sourcing A Dataset To Make Agents Code Like Humans},\n journal = {p(doom) blog},\n year = {2025},\n note = {https://pdoom.org/blog.html}\n}",bibtex,tab +2,87,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"5:11:50 PM [info] Activating crowd-code\n5:11:50 PM [info] Recording started\n5:11:50 PM [info] Initializing git provider using file system watchers...\n5:11:50 PM [info] Git repository found\n5:11:50 PM [info] Git provider initialized successfully\n5:11:50 PM [info] Initial git state: [object Object]\n",Log,tab +3,819,"examples/bibliography.bib",0,0,"",bibtex,tab +4,2748,"examples/bibliography.bib",0,0,"",bibtex,selection_command +5,8203,"dist/blog.html",0,0,"\n\n\n\n \n \n \n \n \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n
\n \n \n \n \n \n \n \n \n
\n \n \n\n",html,tab +6,9361,"examples/bibliography.bib",0,0,"",bibtex,tab +7,24646,"examples/act.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

Large language models exhibit remarkable reasoning capabilities with scale. However, a fundamental flaw of current-generation transformer-based language models is their uniform allocation of compute per token.\n

\n \n \n \n 1\n

\n A vanilla transformer (and all currently used variations thereof) generates each token after one (and only one) forward-pass through the network. Intuitively this means that at inference time, the network thinks for the same amount of time before each token generation. While this is not an inherent limitation on the reasoning capabilities of transformer models, it does mean that we would need obscenely large transformers for them to exhibit longer-term planning capabilities. Since we posit that (standalone) models with more than 1 trillion parameters are already unfeasible to serve at scale, we need to overcome this limitation of the transformer architecture.\n

\n

\n

\n While we have been aware of this limitation and of its severity for a long time (much longer than p(doom) even existed for), today it is an open secret in the ML research community. Everyone knows that all big AGI labs (DeepMind, OpenAI, imbue, to name a few) are rushing towards overcoming this limitation , yet nobody has found a solution yet. The market value of an LLM/LMM with agentic reasoning capabilities is difficult to imagine.\n

\n

\n Most public discourse around scaling up reasoning capabilities involves developing methods to shift computational resources from training to inference time: While only a tiny fraction of computational resources of LLMs are used for inference (>99% of compute goes to pretraining), systems like Pluribus , AlphaGo or MuZero leverage vast amounts of compute at inference time, primarily via Monte Carlo Tree Search. A few techniques for shifting compute to inference time already exist, albeit rather simple ones: chain-of-thought , tree-of-thought , sampling+scoring . All of these techniques try to shift computation post-hoc. While only helps if the correct solution is actually in the sampled space, and seem unelegant in execution, somewhat arbitrary as solutions and above all, have not been able to yield agentic systems.\n

\n

\n We posit that one elegant way of addressing the transformer's limitation is via adaptive computation. We want to find a way of letting the model think for as long as it deems necessary before each token generation. That way, we do not hard-code a 'reasoning architecture' into the model and we do not materialize the search. The model's thought process stays in the latent space. We posit that such a model trained at GPT-4 scale will exhibit human-level reasoning.\n

\n \n \n\n \n\n

Contributions

\n

MM worked on research and analysis, FS wrote the manuscript.

\n \n \n \n \n\n \n\n",html,tab +8,29835,"examples/act.html",1301,0,"",html,selection_command +9,30926,"examples/act.html",1318,0,"",html,selection_command +10,31058,"examples/act.html",1326,0,"",html,selection_command +11,31195,"examples/act.html",1360,0,"",html,selection_command +12,31298,"examples/act.html",1363,0,"",html,selection_command +13,31464,"examples/act.html",1372,0,"",html,selection_command +14,31630,"examples/act.html",1338,0,"",html,selection_command +15,31762,"examples/act.html",1343,0,"",html,selection_command +16,31917,"examples/act.html",1351,0,"",html,selection_command +17,33442,"examples/act.html",1352,0,"",html,selection_command +18,33954,"examples/act.html",1352,0,"*",html,content +19,33961,"examples/act.html",1353,0,"",html,selection_keyboard +20,34313,"examples/act.html",1352,0,"",html,selection_command +21,34536,"examples/act.html",1387,0,"",html,selection_command +22,34695,"examples/act.html",1443,0,"",html,selection_command +23,34847,"examples/act.html",1518,0,"",html,selection_command +24,34985,"examples/act.html",1536,0,"",html,selection_command +25,35139,"examples/act.html",1544,0,"",html,selection_command +26,35272,"examples/act.html",1577,0,"",html,selection_command +27,35502,"examples/act.html",1613,0,"",html,selection_command +28,35696,"examples/act.html",1577,0,"",html,selection_command +29,35844,"examples/act.html",1578,0,"",html,selection_command +30,35935,"examples/act.html",1579,0,"",html,selection_command +31,36260,"examples/act.html",1579,0,"*",html,content +32,36263,"examples/act.html",1580,0,"",html,selection_keyboard +33,36537,"examples/act.html",1579,0,"",html,selection_command +34,37868,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +35,39675,"TERMINAL",0,0,"",,terminal_focus +36,39679,"examples/act.html",0,0,"",html,tab +37,41767,"TERMINAL",0,0,"zsh",,terminal_focus +38,46572,"TERMINAL",0,0,"npm run dev",,terminal_command +39,46623,"TERMINAL",0,0,"]633;C",,terminal_output +40,46953,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +41,47135,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +42,47518,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 374ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +43,47919,"TERMINAL",0,0,"created dist/transforms.v2.js in 409ms\r\n\r\n[2025-09-02 17:12:38] waiting for changes...\r\n",,terminal_output +44,49247,"TERMINAL",0,0,"^C⠙npm notice\r\nnpm notice New major version of npm available! 10.9.2 -> 11.5.2\r\nnpm notice Changelog: https://github.com/npm/cli/releases/tag/v11.5.2\r\nnpm notice To update run: npm install -g npm@11.5.2\r\nnpm notice\r\n⠙",,terminal_output +45,49266,"TERMINAL",0,0,"% \r \r",,terminal_output +46,51543,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +47,51566,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +48,56868,"TERMINAL",0,0,"npm run dev",,terminal_command +49,56919,"TERMINAL",0,0,"]633;C",,terminal_output +50,57045,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +51,57206,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +52,57588,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 376ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +53,58022,"TERMINAL",0,0,"created dist/transforms.v2.js in 438ms\r\n\r\n[2025-09-02 17:12:48] waiting for changes...\r\n",,terminal_output +54,99977,"examples/act.html",1616,0,"",html,selection_command +55,100222,"examples/act.html",1661,0,"",html,selection_command +56,100256,"examples/act.html",1709,0,"",html,selection_command +57,100288,"examples/act.html",1716,0,"",html,selection_command +58,100320,"examples/act.html",1731,0,"",html,selection_command +59,100354,"examples/act.html",1753,0,"",html,selection_command +60,100389,"examples/act.html",1788,0,"",html,selection_command +61,100531,"examples/act.html",1817,0,"",html,selection_command +62,100730,"examples/act.html",1823,0,"",html,selection_command +63,101084,"examples/act.html",1836,0,"",html,selection_command +64,167294,"examples/act.html",1825,12," }",html,selection_command +65,167588,"examples/act.html",1819,18," }\n }",html,selection_command +66,167837,"examples/act.html",1811,26," ]\n }\n }",html,selection_command +67,167863,"examples/act.html",1755,82," {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +68,167897,"examples/act.html",1733,104," ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +69,167928,"examples/act.html",1718,119," ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +70,167962,"examples/act.html",1711,126," ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +71,167996,"examples/act.html",1703,134," }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +72,168030,"examples/act.html",1628,209," ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +73,168064,"examples/act.html",1583,254," ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +74,168097,"examples/act.html",1546,291," ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +75,168131,"examples/act.html",1538,299," {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +76,168165,"examples/act.html",1529,308," },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +77,168198,"examples/act.html",1487,350," {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +78,168232,"examples/act.html",1412,425," ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +79,168268,"examples/act.html",1356,481," ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +80,168303,"examples/act.html",1321,516," ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +81,168336,"examples/act.html",1313,524," {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +82,168369,"examples/act.html",1296,541," ""authors"": [\n {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +83,168402,"examples/act.html",1255,582," ""url"": ""https://pdoom.org/act.html"",\n ""authors"": [\n {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +84,168556,"examples/act.html",1217,620," ""published"": ""December 07, 2023"",\n ""url"": ""https://pdoom.org/act.html"",\n ""authors"": [\n {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +85,168811,"examples/act.html",985,852," ""description"": ""Large language models exhibit remarkable reasoning capabilities with scale. However, a fundamental flaw of current-generation transformer-based language models is their uniform allocation of compute per token."",\n ""published"": ""December 07, 2023"",\n ""url"": ""https://pdoom.org/act.html"",\n ""authors"": [\n {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +86,168843,"examples/act.html",935,902," ""title"": ""ACT: Adaptive Compute Transformer"",\n ""description"": ""Large language models exhibit remarkable reasoning capabilities with scale. However, a fundamental flaw of current-generation transformer-based language models is their uniform allocation of compute per token."",\n ""published"": ""December 07, 2023"",\n ""url"": ""https://pdoom.org/act.html"",\n ""authors"": [\n {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },\n {\n ""author"":""Franz Srambical*"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""}]\n }\n ],\n ""katex"": {\n ""delimiters"": [\n {""left"": ""$$"", ""right"": ""$$"", ""display"": false}\n ]\n }\n }",html,selection_command +87,168870,"examples/act.html",878,959," ",html,selection_command +88,169008,"examples/act.html",859,978," \n ",html,selection_command +89,169142,"examples/act.html",853,984," -->\n \n ",html,selection_command +90,169416,"examples/act.html",859,978," \n ",html,selection_command +91,170499,"examples/act.html",853,984," -->\n \n ",html,selection_command +92,171026,"examples/act.html",859,978," \n ",html,selection_command +93,224463,"examples/act.html",1312,0,"",html,selection_mouse +94,224476,"examples/act.html",1311,0,"",html,selection_command +95,257491,"examples/blog.html",0,0,"\n\n\n\n \n \n \n \n \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n
\n \n \n \n \n \n \n \n \n
\n \n \n\n",html,tab +96,258080,"examples/blog.html",31210,0,"",html,selection_command +97,258585,"examples/blog.html",31202,0,"",html,selection_command +98,258842,"examples/blog.html",31164,0,"",html,selection_command +99,258869,"examples/blog.html",31147,0,"",html,selection_command +100,258901,"examples/blog.html",31134,0,"",html,selection_command +101,258935,"examples/blog.html",31117,0,"",html,selection_command +102,258969,"examples/blog.html",31098,0,"",html,selection_command +103,259111,"examples/blog.html",31073,0,"",html,selection_command +104,259274,"examples/blog.html",30817,0,"",html,selection_command +105,259475,"examples/blog.html",30731,0,"",html,selection_command +106,259693,"examples/blog.html",30753,0,"",html,selection_command +107,259943,"examples/blog.html",30754,0,"",html,selection_command +108,259976,"examples/blog.html",30760,0,"",html,selection_command +109,260007,"examples/blog.html",30762,0,"",html,selection_command +110,260041,"examples/blog.html",30769,0,"",html,selection_command +111,260074,"examples/blog.html",30771,0,"",html,selection_command +112,260107,"examples/blog.html",30776,0,"",html,selection_command +113,260262,"examples/blog.html",30777,0,"",html,selection_command +114,260421,"examples/blog.html",30781,0,"",html,selection_command +115,260575,"examples/blog.html",30782,0,"",html,selection_command +116,260726,"examples/blog.html",30789,0,"",html,selection_command +117,261030,"examples/blog.html",30790,0,"",html,selection_command +118,261329,"examples/blog.html",30790,0,"*",html,content +119,261334,"examples/blog.html",30791,0,"",html,selection_keyboard +120,261574,"examples/blog.html",30790,0,"",html,selection_command +121,261848,"examples/blog.html",30791,0,"",html,selection_command +122,262021,"examples/blog.html",30797,0,"",html,selection_command +123,262179,"examples/blog.html",30798,0,"",html,selection_command +124,262345,"examples/blog.html",30802,0,"",html,selection_command +125,262511,"examples/blog.html",30803,0,"",html,selection_command +126,262781,"examples/blog.html",30812,0,"",html,selection_command +127,263019,"examples/blog.html",30814,0,"",html,selection_command +128,263446,"examples/blog.html",30813,0,"",html,selection_command +129,264046,"examples/blog.html",30813,0,"8",html,content +130,264050,"examples/blog.html",30814,0,"",html,selection_keyboard +131,264458,"examples/blog.html",30813,1,"",html,content +132,264644,"examples/blog.html",30813,0,"*",html,content +133,264646,"examples/blog.html",30814,0,"",html,selection_keyboard +134,264837,"examples/blog.html",30813,0,"",html,selection_command +135,267556,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +136,267714,"TERMINAL",0,0,"npm run dev",,terminal_output +137,267869,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +138,268110,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +139,268110,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +140,268128,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +141,268710,"TERMINAL",0,0,"npm run dev",,terminal_command +142,268761,"TERMINAL",0,0,"]633;C",,terminal_output +143,268892,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +144,269078,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +145,269497,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 421ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +146,269967,"TERMINAL",0,0,"created dist/transforms.v2.js in 468ms\r\n\r\n[2025-09-02 17:16:20] waiting for changes...\r\n",,terminal_output +147,275810,"examples/act.html",0,0,"",html,tab +148,276477,"examples/blog.html",0,0,"",html,tab +149,280866,"examples/causal_mask.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

Although ubiquitously used in large-scale language modeling , the necessity of the causal mask is seldom questioned in the literature.\n ""The causal mask is needed to prevent information leakage from future tokens"" is a commonly encountered, almost dogmatically repeated phrase.\n However, among researchers and practitioners alike, there exists a certain confusion around what the causal mask is and why we actually need it.

\n \n \n \n 1\n

A primer on the causal mask

\n

The confusion about the causal mask already starts with its name. The original Transformer paper \n does not mention the term causal mask at all. While we cannot definitively pin-point the origin of the term, its\n first well-known appearance is in the T5 paper , where it is used to describe the\n triangular mask that is applied to the attention weights in the self-attention mechanism (Figure 1, centre). The mask being triangular\n has the effect of only allowing information from previous tokens to be used in the computation of the current token, which already\n leads us to the first common misconception: For causal LMs at inference time, even if $$n$$ tokens have already been generated, token $$k$$ with $$k < n$$\n cannot attend to token $$j$$ with $$k < j < n$$, even though token $$j$$ is already known. From an information-theoretic perspective,\n it is clear that this is suboptimal, and indeed recent work has investigated the algorithmic deficiency of causal language models .\n

\n \n
\n
Figure 1: A schematic of full attention, causal masking, and prefix masking. Figure from . Used under CC-BY 4.0 license.
\n

\n When solely looking at inference time, what we ideally want is full attention, i.e. every (generated) token can attend to every (generated) token (Figure 1, left).\n This is the no-mask regime, where the attention weights are not modified at all. Yet, all LLMs also use\n the causal mask at inference time since omitting the mask would lead to a distribution shift between training and inference, thus impairing performance.\n Since the causal mask is needed at inference time because it was used during training, a natural question to ask is: Why do we need the causal mask during training?\n\n

\n 2\n

On the necessity of the causal mask

\n

\n For brevity's sake, we will focus on the GPT-style pre-training regime , where the model is trained to predict the next token given the previous tokens.\n One of the key advantages of the Transformer architecture over classical RNNs is that we can predict all tokens of a sequence in parallel.\n Specifically, this is achieved using a technique called teacher-forcing , where instead of using the tokens generated by the model, we use the ground-truth previous tokens to predict each 'next token'.\n Clearly, during the model prediction of token $$k$$, its computation should not use information from token $$k$$ and beyond, which brings us to the second common misconception: The causal mask is not needed to prevent information leakage from future tokens during training.\n

\n

\n To illustrate my point, suppose we have a sequence of tokens $$x_1, x_2, x_3, x_4$$ and we want the model to use tokens $$x_1, x_2, x_3$$ to predict token $$x_4$$.\n We need to prohibit tokens from attending to token $$x_4$$ when predicting token $$x_4$$, but we do not need to prohibit token $$x_1$$ from attending to tokens $$\{x_2, x_3\}$$, or token $$x_2$$ from attending to token $$x_3$$, which is exactly what the causal mask does.\n Instead of a triangular mask, using a block-sparse mask would suffice to prevent information leakage from future tokens, while allowing for all tokens in the context to attend to each other.\n

\n

\n The illustration above only depicts the case of a single token prediction during training, raising the question of whether the point holds for parallel training as well.\n In fact, the causal mask is neither needed for parallel training, nor for teacher-forcing, and a block-sparse mask suffices for both.\n

\n 3\n

What about PrefixLMs?

\n

\n We can motivate the use of block-sparse masks by looking at PrefixLMs , which are language models originally designed to solve sequence-to-sequence tasks.\n In PrefixLMs, the model is trained to predict the next token given the previous tokens and a so-called prefix which is input to the model.\n During supervised training of PrefixLMs, the prefix could be a prompt, a question, or any other kind of context that is given to the model.\n The model is then trained to predict the answer or continuation of the prompt. Since the prefix is fixed and not predicted by the model, the tokens in the prefix are not masked at all, while the rest of the sequence is causally masked (Figure 1, right).\n

\n

\n This masking procedure is specifically tailored towards the regime of sequence-to-sequence tasks under explicit supervision.\n However, the great abundance of textual data available on the Internet is unlabeled and largely unstructured, in which case PrefixLM training involves\n randomly splitting text into prefixes and targets. In this case, causal language modeling leads to much denser supervision than prefix language modeling,\n since the prefix does not provide a direct supervisory signal to the model.\n

\n

\n This motivates a procedure that benefits from both dense supervision as well as full attention: Taking each token as the sole target\n and using all previous tokens as the prefix. This is exactly the block-sparse mask and nothing inherently prohibits parallelization in this regime.\n However, the block-sparse mask depends on the position of the token in the sequence, which means that a naïve implementation would require $$n$$ times as much memory\n and $$n$$ times as much computation, since everything after the first attention map computation is token-position dependent (where $$n$$ is the sequence length).\n Thus, the main challenge in moving beyond the causal mask is mitigating the memory and compute overhead that comes with that.\n

\n \n\n \n\n

Contributions

\n

FS worked on all aspects of this post, including research, analysis and writing. This blog post has benefited from various discussions with senior colleagues, among others Preetum Nakkiran and Thomas Scialom.

\n \n \n \n \n\n \n\n",html,tab +150,282560,"examples/causal_mask.html",0,0,"",html,selection_command +151,289059,"examples/thesis.html",0,0,"\n\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

\n Neural networks are mean-seeking. They work well when you run inference on data points that lie around the mean of their training data. They embarrassingly fail otherwise.\n

\n \n \n \n 1\n

\n Currently, we exploit the paradigm of data-driven self-supervision by using neural networks to approximate the underlying data-generating process of a training distribution, until we run out of data. When we have exhausted the data readily accessible to us, a system trained using naïve self-supervision on such a large-scale dataset will have a reasonably good internal model of neighborhoods in data-space that have a high presence in the training data. The further we go along the long tail of the data distribution, the worse a neural network gets at modeling such data points in its representation-space.\n

\n

\n Language models are simulators. They will simulate anything, so long as you let the language model ingest enough imitation data at training time. While general, conversational data may be highly present in a simple data dump of the internet, data of highly skilled behavior is scarce. Upsampling desired behavioral data works until such data is overly scarce. At said skill level, behavior cloning is not feasible anymore, at which point we use reward signals to approximate how we should update the model’s parameters instead. Crucially, the entire progression from naïve modeling of data dumps of the internet towards attaining higher skill levels is handcrafted: dataset curation is handcrafted, upsampling is handcrafted, reward signals are handcrafted.\n

\n

\n

\n They should not be. In systems prospectively as capable as us, they cannot be. At inference time, when given a long-horizon goal such as getting an IMO gold medal, an intelligent system should itself identify whether its internal model and — by extension — the current distribution of training data it has seen is sufficient to solve the task at hand, and if not, what data to gather next. Beyond gathering new data points, such an endeavor necessitates some verification signal, which can be external or intrinsic, in order to gauge whether progress towards the long-horizon goal is being made.\n

\n

\n There are four, largely orthogonal, but additive directions towards AGI: acting, reasoning, continual learning, and the ability to gather the next most useful set of data points. While the first three are active research areas both in the academic literature and in closed AGI labs, the latter is still largely neglected. We motivate the latter by the inherent inability of neural networks to generalize beyond their training distribution. Ultimately, such a capability would solve both curriculum learning and the ‘long-tail problem’ from first principles, allowing us to move beyond power-law scaling and into a new regime. A regime that leads to AGI in a world full of constraints.\n

\n \n\n \n\n

Contributions

\n

MM and FS worked on research and analysis, FS wrote the manuscript.

\n \n \n \n \n\n \n\n\n",html,tab +152,296376,"examples/thesis.html",1315,0,"",html,selection_command +153,296570,"examples/thesis.html",1318,0,"",html,selection_command +154,296741,"examples/thesis.html",1323,0,"",html,selection_command +155,296977,"examples/thesis.html",1333,0,"",html,selection_command +156,306563,"examples/thesis.html",1369,0,"",html,selection_command +157,306760,"examples/thesis.html",1333,0,"",html,selection_command +158,340170,"examples/jax_assert.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

\n Traditional value assertions in jitted JAX lead to performance degredation. A new (not yet public) JAX API fixes this.\n

\n \n \n \n 1\n

\n Jitted JAX does not support traditional python asserts that access JAX arrays .\n Chex and jax.experimental.checkify.check provide ways of wrapping a jitted function\n with decorators to enable value assertions, but they lead to performance-degradation, making them unusable in practical settings.\n

\n

\n For a performance-degradation free way of using value assertions in jitted JAX, we can use a new (as of today still private) JAX API: error_check :\n

\n \n import jax\n from jax._src.error_check import set_error_if, raise_if_error\n import jax.numpy as jnp\n\n @jax.jit\n def f(x, y):\n set_error_if(x != 0, 'x must be 0')\n return jnp.multiply(x, y)\n\n f(1, 0)\n\n raise_if_error()\n \n\n \n Traceback (most recent call last):\n File ""/home/ubuntu/code/temp.py"", line 12, in \n raise_if_error()\n File ""/home/ubuntu/code/.venv/lib/python3.10/site-packages/jax/_src/error_check.py"", line 93, in raise_if_error\n raise exc.with_traceback(filtered_traceback)\n File ""/home/ubuntu/code/temp.py"", line 10, in \n f(1, 0)\n File ""/home/ubuntu/code/temp.py"", line 7, in f\n set_error_if(x != 0, 'x must be 0')\n jax._src.error_check.JaxValueError: x must be 0\n \n

\n This pattern exploits that it suffices to raise an assertion error post-hoc, in this case after the computation of the jitted function.\n Thus, the implementation merely conditionally stores the error in JAX-managed context. While purely functional conditional computation is fully supported by JAX and XLA , and thus fully compatible with XLA graph compilation,\n the error is only raised outside of the jitted function, avoiding the typical performance overhead of value assertions.\n

\n \n \n\n \n\n

Contributions

\n

MM and FS worked on research and analysis, FS wrote the manuscript.

\n \n \n \n \n\n \n\n\n",html,tab +159,340875,"examples/jax_assert.html",0,0,"",html,selection_command +160,341325,"examples/jax_assert.html",794,0,"",html,selection_keyboard +161,341642,"examples/jax_assert.html",802,0,"",html,selection_command +162,341892,"examples/jax_assert.html",803,0,"",html,selection_command +163,341922,"examples/jax_assert.html",810,0,"",html,selection_command +164,341954,"examples/jax_assert.html",817,0,"",html,selection_command +165,341986,"examples/jax_assert.html",853,0,"",html,selection_command +166,342019,"examples/jax_assert.html",859,0,"",html,selection_command +167,342054,"examples/jax_assert.html",878,0,"",html,selection_command +168,342090,"examples/jax_assert.html",935,0,"",html,selection_command +169,342121,"examples/jax_assert.html",1004,0,"",html,selection_command +170,342156,"examples/jax_assert.html",1145,0,"",html,selection_command +171,342190,"examples/jax_assert.html",1180,0,"",html,selection_command +172,342224,"examples/jax_assert.html",1228,0,"",html,selection_command +173,342361,"examples/jax_assert.html",1245,0,"",html,selection_command +174,342542,"examples/jax_assert.html",1253,0,"",html,selection_command +175,345242,"examples/jax_assert.html",1261,0,"",html,selection_command +176,345413,"examples/jax_assert.html",1262,0,"",html,selection_command +177,345579,"examples/jax_assert.html",1268,0,"",html,selection_command +178,345726,"examples/jax_assert.html",1271,0,"",html,selection_command +179,345892,"examples/jax_assert.html",1277,0,"",html,selection_command +180,346159,"examples/jax_assert.html",1284,0,"",html,selection_command +181,346996,"examples/jax_assert.html",1285,0,"",html,selection_command +182,347578,"examples/jax_assert.html",1284,0,"",html,selection_command +183,347945,"examples/jax_assert.html",1284,0,"*",html,content +184,347951,"examples/jax_assert.html",1285,0,"",html,selection_keyboard +185,348291,"examples/jax_assert.html",1284,0,"",html,selection_command +186,349729,"examples/jax_assert.html",1319,0,"",html,selection_command +187,349977,"examples/jax_assert.html",1375,0,"",html,selection_command +188,350010,"examples/jax_assert.html",1450,0,"",html,selection_command +189,350041,"examples/jax_assert.html",1468,0,"",html,selection_command +190,350075,"examples/jax_assert.html",1476,0,"",html,selection_command +191,350237,"examples/jax_assert.html",1509,0,"",html,selection_command +192,350727,"examples/jax_assert.html",1510,0,"",html,selection_command +193,351248,"examples/jax_assert.html",1511,0,"",html,selection_command +194,351526,"examples/jax_assert.html",1511,0,"*",html,content +195,351529,"examples/jax_assert.html",1512,0,"",html,selection_keyboard +196,351793,"examples/jax_assert.html",1511,0,"",html,selection_command +197,358667,"examples/blog.html",0,0,"",html,tab +198,359510,"examples/blog.html",30153,0,"",html,selection_command +199,360576,"examples/blog.html",29148,0,"",html,selection_command +200,361958,"examples/blog.html",29127,0,"",html,selection_command +201,362206,"examples/blog.html",29057,0,"",html,selection_command +202,362240,"examples/blog.html",29020,0,"",html,selection_command +203,362275,"examples/blog.html",28983,0,"",html,selection_command +204,362308,"examples/blog.html",28966,0,"",html,selection_command +205,362341,"examples/blog.html",28947,0,"",html,selection_command +206,362374,"examples/blog.html",28922,0,"",html,selection_command +207,362408,"examples/blog.html",28738,0,"",html,selection_command +208,362441,"examples/blog.html",28672,0,"",html,selection_command +209,362474,"examples/blog.html",28551,0,"",html,selection_command +210,362508,"examples/blog.html",28507,0,"",html,selection_command +211,362541,"examples/blog.html",28473,0,"",html,selection_command +212,362575,"examples/blog.html",28452,0,"",html,selection_command +213,362609,"examples/blog.html",28383,0,"",html,selection_command +214,362643,"examples/blog.html",28346,0,"",html,selection_command +215,363246,"examples/blog.html",28383,0,"",html,selection_command +216,363386,"examples/blog.html",28452,0,"",html,selection_command +217,368098,"examples/blog.html",28383,0,"",html,selection_command +218,368348,"examples/blog.html",28346,0,"",html,selection_command +219,368380,"examples/blog.html",28309,0,"",html,selection_command +220,368412,"examples/blog.html",28292,0,"",html,selection_command +221,368446,"examples/blog.html",28273,0,"",html,selection_command +222,368483,"examples/blog.html",28248,0,"",html,selection_command +223,368517,"examples/blog.html",28083,0,"",html,selection_command +224,368550,"examples/blog.html",27997,0,"",html,selection_command +225,368584,"examples/blog.html",27899,0,"",html,selection_command +226,368617,"examples/blog.html",27855,0,"",html,selection_command +227,368959,"examples/blog.html",27814,0,"",html,selection_command +228,369350,"examples/blog.html",27855,0,"",html,selection_command +229,369535,"examples/blog.html",27899,0,"",html,selection_command +230,370646,"examples/blog.html",27997,0,"",html,selection_command +231,372071,"examples/blog.html",28064,0,"*",html,content +232,372071,"examples/blog.html",28042,0,"*",html,content +233,372077,"examples/blog.html",28066,0,"",html,selection_command +234,376010,"examples/blog.html",27983,0,"",html,selection_command +235,376254,"examples/blog.html",27885,0,"",html,selection_command +236,376508,"examples/blog.html",27841,0,"",html,selection_command +237,376535,"examples/blog.html",27800,0,"",html,selection_command +238,376567,"examples/blog.html",27779,0,"",html,selection_command +239,376600,"examples/blog.html",27713,0,"",html,selection_command +240,376634,"examples/blog.html",27676,0,"",html,selection_command +241,376666,"examples/blog.html",27639,0,"",html,selection_command +242,376699,"examples/blog.html",27622,0,"",html,selection_command +243,376733,"examples/blog.html",27603,0,"",html,selection_command +244,376767,"examples/blog.html",27578,0,"",html,selection_command +245,376802,"examples/blog.html",27350,0,"",html,selection_command +246,376833,"examples/blog.html",27264,0,"",html,selection_command +247,376866,"examples/blog.html",27188,0,"",html,selection_command +248,376900,"examples/blog.html",27144,0,"",html,selection_command +249,376933,"examples/blog.html",27099,0,"",html,selection_command +250,376966,"examples/blog.html",27078,0,"",html,selection_command +251,376999,"examples/blog.html",27012,0,"",html,selection_command +252,377033,"examples/blog.html",26975,0,"",html,selection_command +253,377066,"examples/blog.html",26938,0,"",html,selection_command +254,377099,"examples/blog.html",26921,0,"",html,selection_command +255,377132,"examples/blog.html",26902,0,"",html,selection_command +256,377167,"examples/blog.html",26877,0,"",html,selection_command +257,377201,"examples/blog.html",26627,0,"",html,selection_command +258,377261,"examples/blog.html",26521,0,"",html,selection_command +259,377413,"examples/blog.html",26627,0,"",html,selection_command +260,377668,"examples/blog.html",26877,0,"",html,selection_command +261,377700,"examples/blog.html",26902,0,"",html,selection_command +262,377733,"examples/blog.html",26921,0,"",html,selection_command +263,377766,"examples/blog.html",26938,0,"",html,selection_command +264,377804,"examples/blog.html",26975,0,"",html,selection_command +265,377834,"examples/blog.html",27012,0,"",html,selection_command +266,377866,"examples/blog.html",27078,0,"",html,selection_command +267,377900,"examples/blog.html",27099,0,"",html,selection_command +268,377933,"examples/blog.html",27144,0,"",html,selection_command +269,378107,"examples/blog.html",27188,0,"",html,selection_command +270,383594,"examples/ppo_off_policy.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

\n PPO is commonly referred to as an on-policy algorithm. We argue that this naming scheme is confusing, and show that true on-policy PPO reduces to vanilla policy gradient REINFORCE with baseline.\n

\n \n \n \n 1\n

PPO is off-policy RL

\n

\n Reinforcement learning is used in LLM post-training because we cannot backpropagate through generation .\n The original policy gradient $$\hat{g}=\hat{\mathbb{E}}_t [\nabla_\theta \log \pi_\theta (a_t|s_t)\hat{A}_t]$$ intuitively increases the probability of actions that led to high returns, and decreases that of actions that led to low returns.\n PPO is a two-fold modification of the vanilla policy gradient.\n The first modification is the vanilla policy gradient's extension to the off-policy regime: The policy gradient theorem assumes the behaviour policy to be the target policy due to its expectation.\n When behaviour and target policy are different, a further approximated policy gradient with a corrective term can be used instead: $$\hat{g}=\hat{\mathbb{E}}_t [\frac{\pi_\theta (a_t|s_t)}{\pi_{\theta_\text{old}} (a_t|s_t)} \nabla_\theta \log \pi_\theta (a_t|s_t)\hat{A}_t]$$.\n Instead of nudging the logprobs, the importance sampling ratio $$\frac{\pi_\theta (a_t|s_t)}{\pi_{\theta_\text{old}} (a_t|s_t)}$$ is nudged instead, leading to the objective $$L^{CPI}=\hat{\mathbb{E}}_t [\frac{\pi_\theta (a_t|s_t)}{\pi_{\theta_\text{old}} (a_t|s_t)} \hat{A}_t ]$$.\n A common source of confusion for newcomers to policy gradient methods is the meaning of $$\theta_{\text{old}}$$, which refers to the policy that was used to gather the experience (commonly called behaviour policy).\n PPO's primary contribution over the vanilla policy gradient is increased sample-efficiency by reusing collected trajectories.\n More specifically, trajectories are implicitly reused by performing multiple gradient updates from the same experience. Note that this is different to classical off-policy methods, which usually sample from a replay buffer.\n Therefore, a better (but slightly less scientifically sounding) way to characterize PPO is to call it a on-policy-ish algorithm (where the -ish refers to the fact that the behaviour and target policies in PPO are fairly similar, unlike for classical off-policy methods).\n

\n

\n However, even with the importance sampling ratio, applying multiple gradient steps towards $$L^{CPI}$$ empirically leads to destructively large updates , which leads us to PPO's second contribution:\n Instead of directly optimizing $$L^{CPI}$$, PPO optimizes $$L^{CLIP}=\hat{\mathbb{E}}_t [ \min(r_t(\theta)) \hat{A}_t, \text{clip}(r_t(\theta),1-\epsilon,1+\epsilon)\hat{A}_t ]$$, where the clipping disincentivizes large changes to the policy within a single PPO step.\n

\n 2\n

PPO reduces to REINFORCE with baseline on the first optimizer step

\n

\n In PPO, the behaviour and target policies are identical for the first (of potentially many) optimizer step of each PPO step, but the gradient at said step is still non-zero due to an implicit stop_gradient in the denominator of the importance sampling ratio due to $$\pi_{\theta_{\text{old}}}$$ being collected during the rollout in PPO implementations.\n Specifically, the gradient at the first optimizer step reduces to the gradient of REINFORCE with baseline (and GAE).\n At this initial step, the target parameters $$\theta$$ are identical to the behavior parameters $$\theta_{\text{old}}$$, meaning the ratio $$r_t(\theta) = \frac{\pi_\theta(a_t|s_t)}{\pi_{\theta_{\text{old}}}(a_t|s_t)}$$ numerically evaluates to 1.\n Since $$1-\epsilon < 1 < 1+\epsilon$$, the clipping mechanism is inactive, and the objective matches $$L^{CPI}$$ with the gradient $$ \nabla_\theta L^{CPI} = \hat{\mathbb{E}}_t [ \hat{A}_t \nabla_\theta r_t(\theta) ] $$.\n Using the identity $$ \nabla_\theta r_t(\theta) = r_t(\theta) \nabla_\theta \log \pi_\theta(a_t|s_t) $$, the gradient becomes $$ \nabla_\theta L^{CPI} = \hat{\mathbb{E}}_t [ \hat{A}_t r_t(\theta) \nabla_\theta \log \pi_\theta(a_t|s_t) ] $$. Evaluating this expression when $$r_t(\theta)=1$$ yields:\n $$ \nabla_\theta L^{CLIP}(\theta)|_{\theta=\theta_{\text{old}}} = \hat{\mathbb{E}}_t [ \hat{A}_t \nabla_\theta \log \pi_\theta(a_t|s_t) ] $$\n This is exactly the gradient of REINFORCE using the generalized advantage estimate $$\hat{A}_t$$. PPO's clipping only modifies gradients in subsequent steps as $$\theta$$ diverges from $$\theta_{\text{old}}$$.\n When used with $$\gamma=1$$, $$\lambda=1$$, the first optimizer step of PPO fully reduces to that of REINFORCE with baseline.\n

\n

\n \n\n \n\n

Notes

\n

\n To derive the identity $$ \nabla_\theta r_t(\theta) = r_t(\theta) \nabla_\theta \log \pi_\theta(a_t|s_t) $$, first recognize the identity $$ \nabla_\theta r_t(\theta) = \nabla_\theta [ \frac{\pi_\theta (a_t|s_t)}{\pi_{\theta_{\text{old}}}(a_t|s_t)} ] = \frac{1}{\pi_{\theta_{\text{old}}}(a_t|s_t)} * \nabla_\theta [ \pi_\theta (a_t|s_t) ] $$ since $$ \theta_{\text{old}} $$ is treated as a constant.\n Now we apply the log-derivative trick $$ \nabla_\theta f(\theta) = f(\theta) \nabla_\theta \log f(\theta) $$ to $$ \pi_\theta (a_t|s_t) $$ and get $$ \nabla_\theta \pi_\theta(a_t|s_t) = \pi_\theta(a_t|s_t) \nabla_\theta \log \pi_\theta(a_t|s_t) $$.\n Substituting this back we get $$ \nabla_\theta r_t(\theta) = \frac{1}{\pi_{\theta_{\text{old}}}(a_t|s_t)} * [ \pi_\theta(a_t|s_t) \nabla_\theta \log \pi_\theta (a_t|s_t) ] = \frac{\pi_\theta(a_t|s_t)}{\pi_{\theta_{\text{old}}}(a_t|s_t) } * \nabla_\theta \log \pi_\theta (a_t|s_t) = r_t(\theta) \nabla_\theta \log \pi_\theta(a_t|s_t) $$.\n\n

\n

Contributions

\n

MM and FS worked on research and analysis, FS wrote the manuscript. We thank Gemini 2.5 Pro for deriving the identity.

\n \n \n \n \n\n \n\n\n",html,tab +271,400549,"examples/ppo_off_policy.html",31,0,"",html,selection_command +272,400797,"examples/ppo_off_policy.html",97,0,"",html,selection_command +273,400830,"examples/ppo_off_policy.html",164,0,"",html,selection_command +274,400862,"examples/ppo_off_policy.html",206,0,"",html,selection_command +275,400895,"examples/ppo_off_policy.html",207,0,"",html,selection_command +276,400928,"examples/ppo_off_policy.html",257,0,"",html,selection_command +277,400963,"examples/ppo_off_policy.html",258,0,"",html,selection_command +278,400995,"examples/ppo_off_policy.html",328,0,"",html,selection_command +279,401030,"examples/ppo_off_policy.html",396,0,"",html,selection_command +280,401063,"examples/ppo_off_policy.html",471,0,"",html,selection_command +281,401096,"examples/ppo_off_policy.html",541,0,"",html,selection_command +282,401129,"examples/ppo_off_policy.html",574,0,"",html,selection_command +283,401162,"examples/ppo_off_policy.html",578,0,"",html,selection_command +284,401197,"examples/ppo_off_policy.html",594,0,"",html,selection_command +285,401230,"examples/ppo_off_policy.html",595,0,"",html,selection_command +286,401265,"examples/ppo_off_policy.html",602,0,"",html,selection_command +287,401298,"examples/ppo_off_policy.html",643,0,"",html,selection_command +288,401330,"examples/ppo_off_policy.html",714,0,"",html,selection_command +289,401362,"examples/ppo_off_policy.html",738,0,"",html,selection_command +290,401397,"examples/ppo_off_policy.html",794,0,"",html,selection_command +291,401430,"examples/ppo_off_policy.html",802,0,"",html,selection_command +292,401563,"examples/ppo_off_policy.html",794,0,"",html,selection_command +293,401812,"examples/ppo_off_policy.html",802,0,"",html,selection_command +294,401850,"examples/ppo_off_policy.html",794,0,"",html,selection_command +295,402094,"examples/ppo_off_policy.html",738,0,"",html,selection_command +296,402126,"examples/ppo_off_policy.html",714,0,"",html,selection_command +297,402158,"examples/ppo_off_policy.html",643,0,"",html,selection_command +298,402190,"examples/ppo_off_policy.html",602,0,"",html,selection_command +299,402225,"examples/ppo_off_policy.html",595,0,"",html,selection_command +300,402550,"examples/ppo_off_policy.html",594,0,"",html,selection_command +301,402831,"examples/ppo_off_policy.html",595,0,"",html,selection_command +302,403080,"examples/ppo_off_policy.html",602,0,"",html,selection_command +303,403109,"examples/ppo_off_policy.html",643,0,"",html,selection_command +304,403141,"examples/ppo_off_policy.html",714,0,"",html,selection_command +305,403175,"examples/ppo_off_policy.html",738,0,"",html,selection_command +306,403206,"examples/ppo_off_policy.html",794,0,"",html,selection_command +307,403241,"examples/ppo_off_policy.html",802,0,"",html,selection_command +308,403274,"examples/ppo_off_policy.html",803,0,"",html,selection_command +309,403307,"examples/ppo_off_policy.html",810,0,"",html,selection_command +310,403341,"examples/ppo_off_policy.html",817,0,"",html,selection_command +311,403374,"examples/ppo_off_policy.html",853,0,"",html,selection_command +312,403408,"examples/ppo_off_policy.html",859,0,"",html,selection_command +313,403442,"examples/ppo_off_policy.html",878,0,"",html,selection_command +314,403476,"examples/ppo_off_policy.html",935,0,"",html,selection_command +315,403510,"examples/ppo_off_policy.html",982,0,"",html,selection_command +316,403547,"examples/ppo_off_policy.html",1186,0,"",html,selection_command +317,403581,"examples/ppo_off_policy.html",1221,0,"",html,selection_command +318,403614,"examples/ppo_off_policy.html",1273,0,"",html,selection_command +319,403775,"examples/ppo_off_policy.html",1290,0,"",html,selection_command +320,403944,"examples/ppo_off_policy.html",1298,0,"",html,selection_command +321,404941,"examples/ppo_off_policy.html",1306,0,"",html,selection_command +322,405117,"examples/ppo_off_policy.html",1312,0,"",html,selection_command +323,405332,"examples/ppo_off_policy.html",1315,0,"",html,selection_command +324,405798,"examples/ppo_off_policy.html",1331,0,"*",html,content +325,405803,"examples/ppo_off_policy.html",1332,0,"",html,selection_command +326,407020,"examples/ppo_off_policy.html",1503,0,"*",html,content +327,407023,"examples/ppo_off_policy.html",1504,0,"",html,selection_command +328,407919,"examples/ppo_off_policy.html",1472,0,"",html,selection_command +329,426665,"examples/crowd_code.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

\n We introduce crowd-code, a VS Code/Cursor extension that allows anyone to participate in crowd-sourcing a software engineering dataset to eventually finetune models on.\n Install once, and forget about it.\n

\n \n \n \n 1\n

Data Is The New Oil

\n

\n Neural networks are simulators. Anything you want your model to do, you have to teach it. Base models represent the mean of the data distribution of the internet.\n Post-training permits shifting that distribution towards desired behaviours. The higher the required skill, the scarcer the corresponding data on the internet.\n

\n

\n The most straightforward way to teach a model to do something is to give it a dataset to behaviour-clone off of. No software engineer in the world writes code linearly,\n in a single branch, in a single commit, in a single PR, without jumping around the codebase, without making mistakes, without debugging. crowd-code is our first attempt\n at capturing a broad spectrum of human software engineering, including character-level text insertions, deletions, undo, redo, cursor movement, file switches, jumping\n to function defintions, git checkouts, terminal command execution, autocompletes, LLM changes, and more.\n

\n 2\n

Going Beyond Open-Source Towards Open-Engineering

\n

\n Every day, millions of developers are writing open-source code. We propose going beyond open-source and towards open-engineering, a paradigm where the value of the\n developer's time is not just captured by the code they produce, but also by the mere act of engineering.\n

\n

\n We introduce crowd-code, a VS Code/Cursor extension that allows anyone to participate in crowd-sourcing a software engineering dataset to eventually finetune models on.\n We want to make it as easy as possible for developers to participate in crowd-sourcing. All you need to do is \n install the crowd-code extension, and forget about it.\n

\n
\n
Figure 1: A preview of the crowd-code extension in action. Figure from Mattia Consiglio.
\n

\n The extension periodically uploads the user's captured IDE actions to a server, where they are cleaned, filtered, thoroughly anonymized, and periodically released to the\n public under the most permissive Creative Commons license (CC0).\n An ongoing recording is transparently indicated in the IDE's status bar, and can be stopped at any time. If the user has inserted sensitive data, they can simply\n press the 'panic button' in the status bar to remove the last actions from the recording before they even leave the user's machine. Additionally, the user is asked for\n consent to participate in crowd-sourcing upon extension installation, and can opt-out at any time. We take user privacy very seriously and welcome any feedback on how to\n make data collection and release more transparent.\n

\n

\n Beyond behaviour-cloning on a dataset crowd-sourced using crowd-code, we eventually want to use crowd-code to annotate the entirety of IDE screencasts on the internet \n using an inverse dynamics model trained on screen recordings paired with crowd-code's IDE action annotations. This would unlock an entirely new trove of training data\n for software engineering agents.\n

\n

\n We are excited to see what the community builds with this dataset. We want to democratize AI research. We are greater than the sum of our parts. Together.\n

\n \n\n \n\n

Contributions

\n

AN, MM and FS worked on ideation and implementation. FS wrote the manuscript. We thank Gemini Code Assist and Cursor for their help in writing the extension.

\n \n \n \n \n\n \n\n\n",html,tab +330,428142,"examples/crowd_code.html",0,0,"",html,selection_command +331,429111,"examples/crowd_code.html",794,0,"",html,selection_keyboard +332,429579,"examples/crowd_code.html",802,0,"",html,selection_command +333,429826,"examples/crowd_code.html",803,0,"",html,selection_command +334,429856,"examples/crowd_code.html",810,0,"",html,selection_command +335,429890,"examples/crowd_code.html",817,0,"",html,selection_command +336,429922,"examples/crowd_code.html",853,0,"",html,selection_command +337,429957,"examples/crowd_code.html",859,0,"",html,selection_command +338,429989,"examples/crowd_code.html",878,0,"",html,selection_command +339,430023,"examples/crowd_code.html",935,0,"",html,selection_command +340,430057,"examples/crowd_code.html",1008,0,"",html,selection_command +341,430094,"examples/crowd_code.html",1234,0,"",html,selection_command +342,430125,"examples/crowd_code.html",1268,0,"",html,selection_command +343,430159,"examples/crowd_code.html",1316,0,"",html,selection_command +344,430192,"examples/crowd_code.html",1333,0,"",html,selection_command +345,430234,"examples/crowd_code.html",1341,0,"",html,selection_command +346,430263,"examples/crowd_code.html",1375,0,"",html,selection_command +347,430298,"examples/crowd_code.html",1433,0,"",html,selection_command +348,430331,"examples/crowd_code.html",1508,0,"",html,selection_command +349,430536,"examples/crowd_code.html",1550,0,"",html,selection_command +350,430743,"examples/crowd_code.html",1559,0,"",html,selection_command +351,430957,"examples/crowd_code.html",1567,0,"",html,selection_command +352,431142,"examples/crowd_code.html",1601,0,"",html,selection_command +353,489988,"examples/crowd_code.html",1567,0,"",html,selection_command +354,490239,"examples/crowd_code.html",1559,0,"",html,selection_command +355,490265,"examples/crowd_code.html",1550,0,"",html,selection_command +356,490299,"examples/crowd_code.html",1508,0,"",html,selection_command +357,490366,"examples/crowd_code.html",1433,0,"",html,selection_command +358,490625,"examples/crowd_code.html",1508,0,"",html,selection_command +359,490872,"examples/crowd_code.html",1550,0,"",html,selection_command +360,490899,"examples/crowd_code.html",1559,0,"",html,selection_command +361,490932,"examples/crowd_code.html",1567,0,"",html,selection_command +362,490965,"examples/crowd_code.html",1601,0,"",html,selection_command +363,491000,"examples/crowd_code.html",1657,0,"",html,selection_command +364,491032,"examples/crowd_code.html",1732,0,"",html,selection_command +365,491226,"examples/crowd_code.html",1774,0,"",html,selection_command +366,491384,"examples/crowd_code.html",1783,0,"",html,selection_command +367,491521,"examples/crowd_code.html",1791,0,"",html,selection_command +368,491842,"examples/crowd_code.html",1783,0,"",html,selection_command +369,492096,"examples/crowd_code.html",1783,7," {",html,selection_command +370,492275,"examples/crowd_code.html",1783,43," {\n ""author"":""Franz Srambical"",",html,selection_command +371,492525,"examples/crowd_code.html",1783,88," {\n ""author"":""Franz Srambical"",\n ""authorURL"":""https://srambical.fr/"",",html,selection_command +372,492555,"examples/crowd_code.html",1783,163," {\n ""author"":""Franz Srambical"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},",html,selection_command +373,492668,"examples/crowd_code.html",1783,205," {\n ""author"":""Franz Srambical"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]",html,selection_command +374,492826,"examples/crowd_code.html",1783,213," {\n ""author"":""Franz Srambical"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n }",html,selection_command +375,493912,"examples/crowd_code.html",1783,214,"",html,content +376,493937,"examples/crowd_code.html",1787,0,"",html,selection_command +377,494127,"examples/crowd_code.html",1778,0,"",html,selection_command +378,494375,"examples/crowd_code.html",1736,0,"",html,selection_command +379,494407,"examples/crowd_code.html",1661,0,"",html,selection_command +380,494442,"examples/crowd_code.html",1605,0,"",html,selection_command +381,494472,"examples/crowd_code.html",1571,0,"",html,selection_command +382,494508,"examples/crowd_code.html",1563,0,"",html,selection_command +383,494540,"examples/crowd_code.html",1554,0,"",html,selection_command +384,494575,"examples/crowd_code.html",1512,0,"",html,selection_command +385,494608,"examples/crowd_code.html",1437,0,"",html,selection_command +386,494642,"examples/crowd_code.html",1379,0,"",html,selection_command +387,494771,"examples/crowd_code.html",1345,0,"",html,selection_command +388,494939,"examples/crowd_code.html",1337,0,"",html,selection_command +389,495106,"examples/crowd_code.html",1320,0,"",html,selection_command +390,495298,"examples/crowd_code.html",1332,0,"\n {\n ""author"":""Franz Srambical"",\n ""authorURL"":""https://srambical.fr/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n }",html,content +391,495301,"examples/crowd_code.html",1339,0,"",html,selection_command +392,495693,"examples/crowd_code.html",1347,0,"",html,selection_command +393,495855,"examples/crowd_code.html",1383,0,"",html,selection_command +394,496006,"examples/crowd_code.html",1428,0,"",html,selection_command +395,496135,"examples/crowd_code.html",1503,0,"",html,selection_command +396,496296,"examples/crowd_code.html",1545,0,"",html,selection_command +397,496593,"examples/crowd_code.html",1546,0,"",html,selection_command +398,496770,"examples/crowd_code.html",1546,0,",",html,content +399,496772,"examples/crowd_code.html",1547,0,"",html,selection_keyboard +400,496928,"examples/crowd_code.html",1546,0,"",html,selection_command +401,497245,"examples/crowd_code.html",1554,0,"",html,selection_command +402,497493,"examples/crowd_code.html",1563,0,"",html,selection_command +403,497523,"examples/crowd_code.html",1597,0,"",html,selection_command +404,497556,"examples/crowd_code.html",1655,0,"",html,selection_command +405,497587,"examples/crowd_code.html",1730,0,"",html,selection_command +406,497621,"examples/crowd_code.html",1772,0,"",html,selection_command +407,497654,"examples/crowd_code.html",1780,0,"",html,selection_command +408,497835,"examples/crowd_code.html",1789,0,"",html,selection_command +409,498093,"examples/crowd_code.html",1823,0,"",html,selection_command +410,498121,"examples/crowd_code.html",1879,0,"",html,selection_command +411,498148,"examples/crowd_code.html",1954,0,"",html,selection_command +412,498184,"examples/crowd_code.html",1996,0,"",html,selection_command +413,498307,"examples/crowd_code.html",2003,0,"",html,selection_command +414,498572,"examples/crowd_code.html",1996,0,"",html,selection_command +415,499381,"examples/crowd_code.html",1996,1,"",html,content +416,499395,"examples/crowd_code.html",1995,0,"",html,selection_command +417,503689,"examples/crowd_code.html",1953,0,"",html,selection_command +418,503938,"examples/crowd_code.html",1878,0,"",html,selection_command +419,503966,"examples/crowd_code.html",1822,0,"",html,selection_command +420,504000,"examples/crowd_code.html",1788,0,"",html,selection_command +421,504035,"examples/crowd_code.html",1780,0,"",html,selection_command +422,504065,"examples/crowd_code.html",1771,0,"",html,selection_command +423,504101,"examples/crowd_code.html",1729,0,"",html,selection_command +424,504133,"examples/crowd_code.html",1654,0,"",html,selection_command +425,504169,"examples/crowd_code.html",1596,0,"",html,selection_command +426,504203,"examples/crowd_code.html",1562,0,"",html,selection_command +427,504237,"examples/crowd_code.html",1554,0,"",html,selection_command +428,504271,"examples/crowd_code.html",1545,0,"",html,selection_command +429,504304,"examples/crowd_code.html",1503,0,"",html,selection_command +430,504337,"examples/crowd_code.html",1428,0,"",html,selection_command +431,504371,"examples/crowd_code.html",1383,0,"",html,selection_command +432,504495,"examples/crowd_code.html",1347,0,"",html,selection_command +433,508044,"examples/crowd_code.html",1383,0,"",html,selection_command +434,508287,"examples/crowd_code.html",1428,0,"",html,selection_command +435,508317,"examples/crowd_code.html",1503,0,"",html,selection_command +436,508348,"examples/crowd_code.html",1545,0,"",html,selection_command +437,508383,"examples/crowd_code.html",1554,0,"",html,selection_command +438,508416,"examples/crowd_code.html",1562,0,"",html,selection_command +439,508449,"examples/crowd_code.html",1596,0,"",html,selection_command +440,508485,"examples/crowd_code.html",1654,0,"",html,selection_command +441,508518,"examples/crowd_code.html",1729,0,"",html,selection_command +442,508554,"examples/crowd_code.html",1771,0,"",html,selection_command +443,508791,"examples/crowd_code.html",1780,0,"",html,selection_command +444,509042,"examples/crowd_code.html",1774,7," {",html,selection_command +445,509155,"examples/crowd_code.html",1774,41," {\n ""author"":""Mihir Mahajan"",",html,selection_command +446,509307,"examples/crowd_code.html",1774,97," {\n ""author"":""Mihir Mahajan"",\n ""authorURL"":""https://maharajamihir.github.io/"",",html,selection_command +447,509475,"examples/crowd_code.html",1774,172," {\n ""author"":""Mihir Mahajan"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},",html,selection_command +448,509596,"examples/crowd_code.html",1774,214," {\n ""author"":""Mihir Mahajan"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]",html,selection_command +449,509756,"examples/crowd_code.html",1774,222," {\n ""author"":""Mihir Mahajan"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n }",html,selection_command +450,510516,"examples/crowd_code.html",1774,223,"",html,content +451,510536,"examples/crowd_code.html",1778,0,"",html,selection_command +452,510686,"examples/crowd_code.html",1769,0,"",html,selection_command +453,510937,"examples/crowd_code.html",1727,0,"",html,selection_command +454,510970,"examples/crowd_code.html",1652,0,"",html,selection_command +455,511004,"examples/crowd_code.html",1594,0,"",html,selection_command +456,511037,"examples/crowd_code.html",1560,0,"",html,selection_command +457,511139,"examples/crowd_code.html",1552,0,"",html,selection_command +458,511303,"examples/crowd_code.html",1543,0,"",html,selection_command +459,511674,"examples/crowd_code.html",1547,0,"\n {\n ""author"":""Mihir Mahajan"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n }",html,content +460,511680,"examples/crowd_code.html",1554,0,"",html,selection_command +461,511955,"examples/crowd_code.html",1562,0,"",html,selection_command +462,512122,"examples/crowd_code.html",1596,0,"",html,selection_command +463,512255,"examples/crowd_code.html",1652,0,"",html,selection_command +464,512393,"examples/crowd_code.html",1727,0,"",html,selection_command +465,512523,"examples/crowd_code.html",1769,0,"",html,selection_command +466,512760,"examples/crowd_code.html",1770,0,"",html,selection_command +467,512889,"examples/crowd_code.html",1770,0,",",html,content +468,512890,"examples/crowd_code.html",1771,0,"",html,selection_keyboard +469,513071,"examples/crowd_code.html",1770,0,"",html,selection_command +470,513160,"examples/crowd_code.html",1778,0,"",html,selection_command +471,513421,"examples/crowd_code.html",1787,0,"",html,selection_command +472,513449,"examples/crowd_code.html",1821,0,"",html,selection_command +473,513492,"examples/crowd_code.html",1879,0,"",html,selection_command +474,513514,"examples/crowd_code.html",1954,0,"",html,selection_command +475,513548,"examples/crowd_code.html",1996,0,"",html,selection_command +476,513581,"examples/crowd_code.html",2003,0,"",html,selection_command +477,513906,"examples/crowd_code.html",1996,0,"",html,selection_command +478,514055,"examples/crowd_code.html",1996,1,"",html,content +479,514059,"examples/crowd_code.html",1995,0,"",html,selection_command +480,515803,"examples/crowd_code.html",1989,0,"",html,selection_command +481,523142,"examples/blog.html",0,0,"",html,tab +482,524537,"examples/blog.html",27144,0,"",html,selection_command +483,524777,"examples/blog.html",27099,0,"",html,selection_command +484,525059,"examples/blog.html",27144,0,"",html,selection_command +485,525292,"examples/blog.html",27188,0,"",html,selection_command +486,526162,"examples/blog.html",27264,0,"",html,selection_command +487,528050,"examples/blog.html",27345,0,"*",html,content +488,528051,"examples/blog.html",27325,0,"*",html,content +489,528058,"examples/blog.html",27347,0,"",html,selection_command +490,530383,"examples/blog.html",27264,0,"",html,selection_command +491,540268,"examples/ppo_off_policy.html",0,0,"",html,tab +492,543279,"examples/blog.html",0,0,"",html,tab +493,544491,"examples/blog.html",27188,0,"",html,selection_command +494,544742,"examples/blog.html",27144,0,"",html,selection_command +495,544771,"examples/blog.html",27099,0,"",html,selection_command +496,544804,"examples/blog.html",27078,0,"",html,selection_command +497,544837,"examples/blog.html",27012,0,"",html,selection_command +498,544870,"examples/blog.html",26975,0,"",html,selection_command +499,544904,"examples/blog.html",26938,0,"",html,selection_command +500,544937,"examples/blog.html",26921,0,"",html,selection_command +501,544969,"examples/blog.html",26902,0,"",html,selection_command +502,545004,"examples/blog.html",26877,0,"",html,selection_command +503,545037,"examples/blog.html",26627,0,"",html,selection_command +504,545070,"examples/blog.html",26521,0,"",html,selection_command +505,545104,"examples/blog.html",26419,0,"",html,selection_command +506,545137,"examples/blog.html",26375,0,"",html,selection_command +507,545171,"examples/blog.html",26265,0,"",html,selection_command +508,545204,"examples/blog.html",26224,0,"",html,selection_command +509,545237,"examples/blog.html",26203,0,"",html,selection_command +510,545271,"examples/blog.html",26138,0,"",html,selection_command +511,545303,"examples/blog.html",26101,0,"",html,selection_command +512,545336,"examples/blog.html",26064,0,"",html,selection_command +513,545370,"examples/blog.html",26047,0,"",html,selection_command +514,545403,"examples/blog.html",26028,0,"",html,selection_command +515,545436,"examples/blog.html",26003,0,"",html,selection_command +516,545470,"examples/blog.html",25796,0,"",html,selection_command +517,545504,"examples/blog.html",25671,0,"",html,selection_command +518,545537,"examples/blog.html",25543,0,"",html,selection_command +519,545570,"examples/blog.html",25499,0,"",html,selection_command +520,545613,"examples/blog.html",25543,0,"",html,selection_command +521,545871,"examples/blog.html",25671,0,"",html,selection_command +522,545899,"examples/blog.html",25796,0,"",html,selection_command +523,545932,"examples/blog.html",26003,0,"",html,selection_command +524,545965,"examples/blog.html",26028,0,"",html,selection_command +525,546000,"examples/blog.html",26047,0,"",html,selection_command +526,546033,"examples/blog.html",26064,0,"",html,selection_command +527,546067,"examples/blog.html",26101,0,"",html,selection_command +528,546099,"examples/blog.html",26138,0,"",html,selection_command +529,546362,"examples/blog.html",26203,0,"",html,selection_command +530,546557,"examples/blog.html",26224,0,"",html,selection_command +531,559846,"examples/blog.html",26265,0,"",html,selection_command +532,560004,"examples/blog.html",26375,0,"",html,selection_command +533,560150,"examples/blog.html",26419,0,"",html,selection_command +534,560327,"examples/blog.html",26521,0,"",html,selection_command +535,561090,"examples/blog.html",26543,0,"",html,selection_command +536,561337,"examples/blog.html",26544,0,"",html,selection_command +537,561365,"examples/blog.html",26546,0,"",html,selection_command +538,561398,"examples/blog.html",26551,0,"",html,selection_command +539,561633,"examples/blog.html",26553,0,"",html,selection_command +540,561800,"examples/blog.html",26560,0,"",html,selection_command +541,561974,"examples/blog.html",26562,0,"",html,selection_command +542,562951,"examples/blog.html",26568,0,"",html,selection_command +543,563197,"examples/blog.html",26569,0,"",html,selection_command +544,563226,"examples/blog.html",26573,0,"",html,selection_command +545,563259,"examples/blog.html",26574,0,"",html,selection_command +546,563292,"examples/blog.html",26580,0,"",html,selection_command +547,563324,"examples/blog.html",26582,0,"",html,selection_command +548,563359,"examples/blog.html",26587,0,"",html,selection_command +549,563504,"examples/blog.html",26588,0,"",html,selection_command +550,563692,"examples/blog.html",26592,0,"",html,selection_command +551,563823,"examples/blog.html",26593,0,"",html,selection_command +552,564176,"examples/blog.html",26600,0,"",html,selection_command +553,564390,"examples/blog.html",26602,0,"",html,selection_command +554,565227,"examples/blog.html",26602,5,"",html,content +555,565385,"examples/blog.html",26602,0,"e",html,content +556,565386,"examples/blog.html",26603,0,"",html,selection_keyboard +557,565956,"examples/blog.html",26602,0,"",html,selection_command +558,566111,"examples/blog.html",26602,1,"Franz",html,content +559,566781,"examples/blog.html",26602,1,"F",html,selection_command +560,566857,"examples/blog.html",26602,5,"Franz",html,selection_command +561,567031,"examples/blog.html",26602,6,"Franz&",html,selection_command +562,567201,"examples/blog.html",26602,10,"Franz ",html,selection_command +563,567357,"examples/blog.html",26602,11,"Franz ",html,selection_command +564,567676,"examples/blog.html",26602,20,"Franz Srambical",html,selection_command +565,568312,"examples/blog.html",26602,20,"",html,content +566,568624,"examples/blog.html",26600,0,"",html,selection_command +567,568873,"examples/blog.html",26593,0,"",html,selection_command +568,568906,"examples/blog.html",26592,0,"",html,selection_command +569,568938,"examples/blog.html",26588,0,"",html,selection_command +570,568972,"examples/blog.html",26587,0,"",html,selection_command +571,569005,"examples/blog.html",26582,0,"",html,selection_command +572,569039,"examples/blog.html",26580,0,"",html,selection_command +573,569171,"examples/blog.html",26574,0,"",html,selection_command +574,569326,"examples/blog.html",26573,0,"",html,selection_command +575,569463,"examples/blog.html",26569,0,"",html,selection_command +576,569595,"examples/blog.html",26568,0,"",html,selection_command +577,569803,"examples/blog.html",26562,0,"",html,selection_command +578,570194,"examples/blog.html",26561,0,"",html,selection_command +579,570358,"examples/blog.html",26562,0,"Franz Srambical",html,content +580,570372,"examples/blog.html",26581,0,"",html,selection_command +581,573057,"examples/blog.html",26582,0,"",html,selection_command +582,573221,"examples/blog.html",26582,0,",",html,content +583,573224,"examples/blog.html",26583,0,"",html,selection_keyboard +584,573332,"examples/blog.html",26583,0," ",html,content +585,573335,"examples/blog.html",26584,0,"",html,selection_keyboard +586,573548,"examples/blog.html",26583,0,"",html,selection_command +587,573766,"examples/blog.html",26584,0,"",html,selection_command +588,574064,"examples/blog.html",26590,0,"",html,selection_command +589,574241,"examples/blog.html",26591,0,"",html,selection_command +590,574409,"examples/blog.html",26595,0,"",html,selection_command +591,574527,"examples/blog.html",26596,0,"",html,selection_command +592,574684,"examples/blog.html",26602,0,"",html,selection_command +593,574873,"examples/blog.html",26604,0,"",html,selection_command +594,575177,"examples/blog.html",26604,1,"M",html,selection_command +595,575259,"examples/blog.html",26604,5,"Mihir",html,selection_command +596,575444,"examples/blog.html",26604,6,"Mihir&",html,selection_command +597,575609,"examples/blog.html",26604,10,"Mihir ",html,selection_command +598,575793,"examples/blog.html",26604,11,"Mihir ",html,selection_command +599,575957,"examples/blog.html",26604,18,"Mihir Mahajan",html,selection_command +600,576125,"examples/blog.html",26604,19,"Mihir Mahajan,",html,selection_command +601,576452,"examples/blog.html",26604,20,"Mihir Mahajan, ",html,selection_command +602,576991,"examples/blog.html",26604,20,"",html,content +603,577236,"examples/blog.html",26602,0,"",html,selection_command +604,577377,"examples/blog.html",26596,0,"",html,selection_command +605,577512,"examples/blog.html",26595,0,"",html,selection_command +606,577644,"examples/blog.html",26591,0,"",html,selection_command +607,577780,"examples/blog.html",26590,0,"",html,selection_command +608,577923,"examples/blog.html",26584,0,"",html,selection_command +609,578998,"examples/blog.html",26583,0,"",html,selection_command +610,579226,"examples/blog.html",26584,0,"Mihir Mahajan, ",html,content +611,579237,"examples/blog.html",26603,0,"",html,selection_command +612,580975,"examples/blog.html",26609,0,"",html,selection_command +613,581199,"examples/blog.html",26610,0,"",html,selection_command +614,581525,"examples/blog.html",26622,0,"",html,selection_command +615,581995,"examples/blog.html",26622,1,"",html,content +616,582128,"examples/blog.html",26622,1,"",html,content +617,582456,"examples/blog.html",26521,0,"",html,selection_command +618,589276,"examples/crowd_code.html",0,0,"",html,tab +619,591116,"examples/blog.html",0,0,"",html,tab +620,592762,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +621,592940,"TERMINAL",0,0,"npm run dev",,terminal_output +622,593065,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +623,593554,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +624,593554,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +625,593572,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +626,594220,"TERMINAL",0,0,"npm run dev",,terminal_command +627,594271,"TERMINAL",0,0,"]633;C",,terminal_output +628,594400,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +629,594586,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +630,594995,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 410ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +631,595462,"TERMINAL",0,0,"created dist/transforms.v2.js in 469ms\r\n\r\n[2025-09-02 17:21:46] waiting for changes...\r\n",,terminal_output +632,686321,"examples/jasmine.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

\n We introduce Jasmine, a production-ready JAX-based codebase for world modeling from unlabeled videos.\n Scale from single hosts to hundreds of xPUs thanks to XLA.\n

\n \n \n \n 1\n
\n
Figure 1: Jasmine in action.
\n

Introduction

\n

\n We are at the cusp of an intelligence revolution. Neural networks are able to clone the behaviour of peak human intellectual performance \n given enough compute, data, and the right algorithms . While an increasing amount of capital expenditure is allocated to compute clusters, and a well-working\n recipe of equipping models with the required priors and capacity to reason is publicly available, the path to human-level intelligence with the ability to automate\n large fractions of the economy will increasingly be shaped by paradigms that are able to find and efficiently use untouched data troves.\n

\n

\n While product-feedback-loops constitute an adaptive data trove, many domains like robotics are not mature enough to yield a product with wide enough\n adoption to create a feedback-loop of sufficient magnitude, prompting the search for alternatives.\n One paradigm proposed by the research community to overcome the data scarcity in those domains is that of world models. While world models can help frontier model\n development in numerous ways, an ambitious goal of the community is to train a world model to act as a simulation of the world , in order to\n train an agent in that simulation, via an adaptive curriculum or otherwise.\n

\n

Deriving Empirical Environment Complexity Scaling Trends

\n

\n While numerous previous works have investigated large-scale world modeling and its application to robotics , world modeling for agent training calls for a vastly different treatment.\n Such regime requires the compounding error of world models to be orders of magnitude smaller than when solely used for short-term look-ahead. The feasibility of such a world model in its truest sense is entirely\n understudied, and Jasmine, a world modeling codebase, is our first milestone towards studying the setting using rigorous evaluations. Specifically, we want to develop Empirical Environment Complexity Scaling Trends, where we train world models to full convergence\n in environments of increasing complexity (Atari , RetroGym , Craftax , Minecraft )\n and under the synthetic infinite-data regime. Subsequently, we want to evaluate those models two-fold: i) via a taxonomy of granular benchmarks probing\n specific world modeling capabilities (reconstruction quality, environment dynamics at the body/tail of the data distribution, long-horizon consistency) , and ii) by training reinforcement learning (RL) agents in both\n the world model and the corresponding ground-truth environment, and measuring the performance difference between those agents.\n

\n

\n Ultimately, such treatment permits us to derive empirical estimates of compute and data requirements to model environments of increasing complexity sufficiently well (as determined by our evaluation procedure). Only given such estimates can we try to draw conclusions\n about the feasibility of world modeling of environments as complex as the real world for agent training. If our empirical estimates show resource requirement trends that are feasible under the assumption of the continuation of Moore's Law and increased capital\n expenditure, that would manifest world modeling as a paradigm with high likelihood of success in overcoming the data-scarcity in domains as general as (humanoid) robotics. Otherwise, the world modeling research community must realign its direction with downstream goals\n that are feasible.\n

\n

A batteries-included foundation for world modeling research

\n

\n Jasmine, our first milestone towards deriving Empirical Environment Complexity Scaling Trends, is the result of weeks of infrastructure work to make large-scale world modeling research more accessible. What started off as a fork of\n Jafar grew into a full-fledged world\n modeling codebase amenable to large-scale training, implementing multiple dynamics model baselines, asynchronous checkpointing, process-parallel dataloading, checkpointing of model weights, optimizer and dataloader states, checkpointing policies, full reproducibility with identical\n training curves, mixed precision training, optimized FlashAttention (via cuDNN SDPA), activation checkpointing, DDP\n (with FSDP/HSDP requiring changing a singe LoC), WSD schedule, index-shuffling during dataloading, and native Treescope support. Jasmine implements the new\n flax.nnx API and strictly adheres to Noam Shazeer's shape suffix convention, thereby providing\n a didactic implementation of world modeling architectures. Jasmine solely depends\n on battle-tested libraries from the Google ecosystem (Flax, Optax, Orbax, Grain,\n PIX, ArrayRecord).\n

\n

Releasing a dataset of fine-grained research engineering

\n

\n We captured every step of the research engineering process behind Jasmine using crowd-code ,\n a VS Code/ Cursor extension that captures fine-grained IDE interactions (character-level edits, navigation, debugging patterns, terminal usage) and allows researchers to contribute their \n engineering process to a crowd-sourced dataset. Today, we release crowd-code-0.1, our first dataset of dense IDE interactions, which encompasses the entire development of Jasmine.\n crowd-code-0.1 is unfiltered, uncleaned, and uncurated, but only contains IDE interactions of the Jasmine authors. We are actively working on cleaning and curating the full dataset,\n which will be released in the future.\n

\n \n\n \n\n

Contributions

\n

MM, AN and FS worked on research, ideation and implementation. FS wrote the manuscript. SB provided feedback and guidance.

\n \n \n \n \n\n \n\n\n",html,tab +633,686952,"examples/jasmine.html",0,0,"",html,selection_command +634,687723,"examples/jasmine.html",5,0,"",html,selection_command +635,687977,"examples/jasmine.html",30,0,"",html,selection_command +636,688005,"examples/jasmine.html",31,0,"",html,selection_command +637,688039,"examples/jasmine.html",97,0,"",html,selection_command +638,688071,"examples/jasmine.html",164,0,"",html,selection_command +639,688105,"examples/jasmine.html",206,0,"",html,selection_command +640,688138,"examples/jasmine.html",207,0,"",html,selection_command +641,688172,"examples/jasmine.html",257,0,"",html,selection_command +642,688205,"examples/jasmine.html",258,0,"",html,selection_command +643,688240,"examples/jasmine.html",328,0,"",html,selection_command +644,688273,"examples/jasmine.html",396,0,"",html,selection_command +645,688307,"examples/jasmine.html",471,0,"",html,selection_command +646,688340,"examples/jasmine.html",541,0,"",html,selection_command +647,688375,"examples/jasmine.html",574,0,"",html,selection_command +648,688407,"examples/jasmine.html",578,0,"",html,selection_command +649,688442,"examples/jasmine.html",594,0,"",html,selection_command +650,688477,"examples/jasmine.html",595,0,"",html,selection_command +651,688512,"examples/jasmine.html",602,0,"",html,selection_command +652,688545,"examples/jasmine.html",643,0,"",html,selection_command +653,688579,"examples/jasmine.html",714,0,"",html,selection_command +654,688612,"examples/jasmine.html",738,0,"",html,selection_command +655,688646,"examples/jasmine.html",794,0,"",html,selection_command +656,688678,"examples/jasmine.html",802,0,"",html,selection_command +657,688714,"examples/jasmine.html",803,0,"",html,selection_command +658,688746,"examples/jasmine.html",810,0,"",html,selection_command +659,688779,"examples/jasmine.html",817,0,"",html,selection_command +660,688812,"examples/jasmine.html",853,0,"",html,selection_command +661,688846,"examples/jasmine.html",859,0,"",html,selection_command +662,688878,"examples/jasmine.html",878,0,"",html,selection_command +663,688912,"examples/jasmine.html",935,0,"",html,selection_command +664,688946,"examples/jasmine.html",1034,0,"",html,selection_command +665,688979,"examples/jasmine.html",1217,0,"",html,selection_command +666,689013,"examples/jasmine.html",1252,0,"",html,selection_command +667,689046,"examples/jasmine.html",1297,0,"",html,selection_command +668,689079,"examples/jasmine.html",1314,0,"",html,selection_command +669,689113,"examples/jasmine.html",1322,0,"",html,selection_command +670,689146,"examples/jasmine.html",1356,0,"",html,selection_command +671,689180,"examples/jasmine.html",1412,0,"",html,selection_command +672,690574,"examples/jasmine.html",1420,0,"",html,selection_command +673,696243,"examples/jasmine.html",1364,0,"",html,selection_command +674,696422,"examples/jasmine.html",1330,0,"",html,selection_command +675,697087,"examples/jasmine.html",1322,0,"",html,selection_command +676,701191,"examples/jasmine.html",1330,0,"",html,selection_command +677,701350,"examples/jasmine.html",1336,0,"",html,selection_command +678,701503,"examples/jasmine.html",1339,0,"",html,selection_command +679,701648,"examples/jasmine.html",1344,0,"",html,selection_command +680,701817,"examples/jasmine.html",1352,0,"",html,selection_command +681,702186,"examples/jasmine.html",1353,0,"",html,selection_command +682,702492,"examples/jasmine.html",1353,0,"*",html,content +683,702496,"examples/jasmine.html",1354,0,"",html,selection_keyboard +684,702850,"examples/jasmine.html",1353,0,"",html,selection_command +685,703071,"examples/jasmine.html",1388,0,"",html,selection_command +686,703204,"examples/jasmine.html",1444,0,"",html,selection_command +687,703339,"examples/jasmine.html",1519,0,"",html,selection_command +688,703472,"examples/jasmine.html",1537,0,"",html,selection_command +689,703617,"examples/jasmine.html",1545,0,"",html,selection_command +690,703757,"examples/jasmine.html",1578,0,"",html,selection_command +691,704566,"examples/jasmine.html",1578,0,"*",html,content +692,704568,"examples/jasmine.html",1579,0,"",html,selection_keyboard +693,704888,"examples/jasmine.html",1578,0,"",html,selection_command +694,705059,"examples/jasmine.html",1613,0,"",html,selection_command +695,705170,"examples/jasmine.html",1666,0,"",html,selection_command +696,705422,"examples/jasmine.html",1741,0,"",html,selection_command +697,705536,"examples/jasmine.html",1759,0,"",html,selection_command +698,705705,"examples/jasmine.html",1767,0,"",html,selection_command +699,705849,"examples/jasmine.html",1800,0,"",html,selection_command +700,706036,"examples/jasmine.html",1836,0,"",html,selection_command +701,706368,"examples/jasmine.html",1800,0,"",html,selection_command +702,706494,"examples/jasmine.html",1801,0,"",html,selection_command +703,706582,"examples/jasmine.html",1802,0,"",html,selection_command +704,707017,"examples/jasmine.html",1802,0,"*",html,content +705,707019,"examples/jasmine.html",1803,0,"",html,selection_keyboard +706,707273,"examples/jasmine.html",1802,0,"",html,selection_command +707,711409,"examples/blog.html",0,0,"",html,tab +708,711974,"examples/blog.html",26419,0,"",html,selection_command +709,712224,"examples/blog.html",26375,0,"",html,selection_command +710,712257,"examples/blog.html",26265,0,"",html,selection_command +711,712291,"examples/blog.html",26224,0,"",html,selection_command +712,712324,"examples/blog.html",26203,0,"",html,selection_command +713,712358,"examples/blog.html",26138,0,"",html,selection_command +714,712393,"examples/blog.html",26101,0,"",html,selection_command +715,712428,"examples/blog.html",26064,0,"",html,selection_command +716,712468,"examples/blog.html",26047,0,"",html,selection_command +717,712496,"examples/blog.html",26028,0,"",html,selection_command +718,712531,"examples/blog.html",26003,0,"",html,selection_command +719,712562,"examples/blog.html",25796,0,"",html,selection_command +720,712853,"examples/blog.html",25671,0,"",html,selection_command +721,713009,"examples/blog.html",25543,0,"",html,selection_command +722,713544,"examples/blog.html",25671,0,"",html,selection_command +723,716403,"examples/blog.html",25772,0,"*",html,content +724,716403,"examples/blog.html",25750,0,"*",html,content +725,716403,"examples/blog.html",25730,0,"*",html,content +726,716407,"examples/blog.html",25775,0,"",html,selection_command +727,719662,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +728,719743,"TERMINAL",0,0,"npm run dev",,terminal_output +729,719896,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +730,720129,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +731,720129,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +732,720148,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +733,720707,"TERMINAL",0,0,"npm run dev",,terminal_command +734,720758,"TERMINAL",0,0,"]633;C",,terminal_output +735,720886,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +736,721058,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +737,721436,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 377ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +738,721866,"TERMINAL",0,0,"created dist/transforms.v2.js in 429ms\r\n\r\n[2025-09-02 17:23:52] waiting for changes...\r\n",,terminal_output +739,922411,"examples/jax_assert.html",0,0,"",html,tab +740,923322,"examples/jax_assert.html",1476,0,"",html,selection_command +741,923567,"examples/jax_assert.html",1468,0,"",html,selection_command +742,923598,"examples/jax_assert.html",1452,0,"",html,selection_command +743,923628,"examples/jax_assert.html",1377,0,"",html,selection_command +744,923662,"examples/jax_assert.html",1321,0,"",html,selection_command +745,923696,"examples/jax_assert.html",1286,0,"",html,selection_command +746,923818,"examples/jax_assert.html",1321,0,"",html,selection_command +747,924069,"examples/jax_assert.html",1377,0,"",html,selection_command +748,924394,"examples/jax_assert.html",1321,0,"",html,selection_command +749,924703,"examples/jax_assert.html",1286,0,"",html,selection_command +750,937541,"examples/jax_assert.html",1251,0,"",html,selection_command +751,937890,"examples/jax_assert.html",1245,7," {",html,selection_command +752,938076,"examples/jax_assert.html",1245,42," {\n ""author"":""Mihir Mahajan*"",",html,selection_command +753,938329,"examples/jax_assert.html",1245,98," {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",",html,selection_command +754,938358,"examples/jax_assert.html",1245,173," {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},",html,selection_command +755,938391,"examples/jax_assert.html",1245,215," {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]",html,selection_command +756,938499,"examples/jax_assert.html",1245,224," {\n ""author"":""Mihir Mahajan*"",\n ""authorURL"":""https://maharajamihir.github.io/"",\n ""affiliations"": [{""name"": ""p(doom)"", ""url"": ""https://pdoom.org/""},\n {""name"": ""TUM""}]\n },",html,selection_command +757,938832,"examples/jax_assert.html",1245,225,"",html,content +758,938850,"examples/jax_assert.html",1251,0,"",html,selection_command +759,939657,"examples/jax_assert.html",1259,0,"",html,selection_command +760,939804,"examples/jax_assert.html",1296,0,"",html,selection_command +761,939970,"examples/jax_assert.html",1341,0,"",html,selection_command +762,940157,"examples/jax_assert.html",1416,0,"",html,selection_command +763,940346,"examples/jax_assert.html",1341,0,"",html,selection_command +764,940669,"examples/jax_assert.html",1259,0,"",html,selection_command +765,941256,"examples/jax_assert.html",1261,0,"",html,selection_command +766,941538,"examples/jax_assert.html",1267,0,"",html,selection_command +767,941684,"examples/jax_assert.html",1270,0,"",html,selection_command +768,941854,"examples/jax_assert.html",1275,0,"",html,selection_command +769,942019,"examples/jax_assert.html",1285,0,"",html,selection_command +770,942201,"examples/jax_assert.html",1288,0,"",html,selection_command +771,942573,"examples/jax_assert.html",1287,0,"",html,selection_command +772,942707,"examples/jax_assert.html",1286,0,"",html,selection_command +773,942839,"examples/jax_assert.html",1286,1,"",html,content +774,944636,"examples/blog.html",0,0,"",html,tab +775,947240,"examples/blog.html",27930,0,"",html,selection_command +776,947635,"examples/blog.html",28028,0,"",html,selection_command +777,948127,"examples/blog.html",28029,0,"",html,selection_command +778,948610,"examples/blog.html",28029,21,"",html,content +779,950081,"examples/blog.html",28033,0,"",html,selection_command +780,950221,"examples/blog.html",28034,0,"",html,selection_command +781,950380,"examples/blog.html",28038,0,"",html,selection_command +782,950552,"examples/blog.html",28039,0,"",html,selection_command +783,950733,"examples/blog.html",28048,0,"",html,selection_command +784,950940,"examples/blog.html",28051,0,"",html,selection_command +785,951239,"examples/blog.html",28050,0,"",html,selection_command +786,951401,"examples/blog.html",28049,0,"",html,selection_command +787,951568,"examples/blog.html",28049,1,"",html,content +788,952768,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +789,952924,"TERMINAL",0,0,"npm run dev",,terminal_output +790,953061,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +791,953289,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +792,953289,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +793,953310,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +794,953870,"TERMINAL",0,0,"npm run dev",,terminal_command +795,953921,"TERMINAL",0,0,"]633;C",,terminal_output +796,954056,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +797,954395,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +798,954673,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 420ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +799,955021,"TERMINAL",0,0,"created dist/transforms.v2.js in 348ms\r\n\r\n[2025-09-02 17:27:45] waiting for changes...\r\n",,terminal_output +800,83976931,"examples/crowd_code.html",0,0,"",html,tab +801,83980532,"examples/crowd_code.html",3575,0,"",html,selection_command +802,83981120,"examples/crowd_code.html",3576,0,"",html,selection_command +803,83981225,"examples/crowd_code.html",3577,0,"",html,selection_command +804,83981368,"examples/crowd_code.html",3578,0,"",html,selection_command +805,83981520,"examples/crowd_code.html",3579,0,"",html,selection_command +806,83981879,"examples/crowd_code.html",3580,0,"",html,selection_command +807,83981897,"examples/crowd_code.html",3580,0,"i",html,content +808,83981902,"examples/crowd_code.html",3581,0,"",html,selection_keyboard +809,83982124,"examples/crowd_code.html",3580,0,"",html,selection_command +810,84025213,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +811,84025567,"TERMINAL",0,0,"npm run dev",,terminal_output +812,84053794,"TERMINAL",0,0,"npm run dev",,terminal_command +813,84053795,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;npm run dev;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +814,84053844,"TERMINAL",0,0,"]633;C",,terminal_output +815,84054062,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +816,84054252,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +817,84054782,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 532ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +818,84055164,"TERMINAL",0,0,"created dist/transforms.v2.js in 379ms\r\n\r\n[2025-09-03 16:32:45] waiting for changes...\r\n",,terminal_output +819,84749782,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +820,84751925,"examples/crowd_code.html",903,0,"",html,selection_keyboard +821,84752091,"examples/crowd_code.html",0,0,"",html,selection_keyboard +822,84764503,"examples/act.html",0,0,"",html,tab +823,84775570,"TERMINAL",0,0,"\r\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B",,terminal_output +824,84776206,"TERMINAL",0,0,"npm run dev",,terminal_output +825,84776406,"TERMINAL",0,0,"npm run dev",,terminal_command +826,84776406,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;npm run dev;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +827,84776456,"TERMINAL",0,0,"]633;C",,terminal_output +828,84776663,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +829,84776848,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +830,84777223,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 376ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +831,84777626,"TERMINAL",0,0,"created dist/transforms.v2.js in 404ms\r\n\r\n[2025-09-03 16:44:48] waiting for changes...\r\n",,terminal_output +832,84827337,"examples/crowd_code.html",0,0,"",html,tab +833,84827711,"examples/crowd_code.html",6817,0,"",html,selection_command +834,84828878,"examples/act.html",0,0,"",html,tab +835,84830988,"examples/crowd_code.html",0,0,"",html,tab +836,84848190,"examples/crowd_code.html",0,0,"",html,selection_command +837,84850048,"examples/crowd_code.html",3575,0,"",html,selection_command +838,84853460,"TERMINAL",0,0,"^P",,terminal_output +839,84854117,"TERMINAL",0,0,",c",,terminal_output +840,84854546,"TERMINAL",0,0,"^C⠙% \r \r",,terminal_output +841,84854605,"TERMINAL",0,0,"",,terminal_command +842,84854605,"TERMINAL",0,0,"]633;C",,terminal_output +843,84855640,"examples/act.html",0,0,"",html,tab +844,84856703,"examples/crowd_code.html",0,0,"",html,tab +845,84862103,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +846,84862121,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +847,84862913,"TERMINAL",0,0,"npm run dev",,terminal_command +848,84862965,"TERMINAL",0,0,"]633;C",,terminal_output +849,84863094,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +850,84863273,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +851,84863700,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 429ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +852,84864087,"TERMINAL",0,0,"created dist/transforms.v2.js in 386ms\r\n\r\n[2025-09-03 16:46:14] waiting for changes...\r\n",,terminal_output +853,84871572,"examples/crowd_code.html",6817,0,"",html,selection_command +854,84872268,"examples/crowd_code.html",6809,0,"",html,selection_command +855,84872522,"examples/crowd_code.html",6808,0,"",html,selection_command +856,84872554,"examples/crowd_code.html",6772,0,"",html,selection_command +857,84872587,"examples/crowd_code.html",6771,0,"",html,selection_command +858,84872618,"examples/crowd_code.html",6755,0,"",html,selection_command +859,84872653,"examples/crowd_code.html",6731,0,"",html,selection_command +860,84872682,"examples/crowd_code.html",6708,0,"",html,selection_command +861,84872714,"examples/crowd_code.html",6647,0,"",html,selection_command +862,84872858,"examples/crowd_code.html",6478,0,"",html,selection_command +863,84873307,"examples/crowd_code.html",6482,0,"",html,selection_command +864,84873757,"examples/crowd_code.html",6483,0,"",html,selection_command +865,84874123,"examples/crowd_code.html",6484,0,"",html,selection_command +866,84874528,"examples/crowd_code.html",6485,0,"",html,selection_command +867,84878185,"examples/crowd_code.html",6487,0,"",html,selection_command +868,84878436,"examples/crowd_code.html",6489,0,"",html,selection_command +869,84878468,"examples/crowd_code.html",6492,0,"",html,selection_command +870,84878500,"examples/crowd_code.html",6496,0,"",html,selection_command +871,84878534,"examples/crowd_code.html",6499,0,"",html,selection_command +872,84878563,"examples/crowd_code.html",6506,0,"",html,selection_command +873,84878721,"examples/crowd_code.html",6509,0,"",html,selection_command +874,84878921,"examples/crowd_code.html",6518,0,"",html,selection_command +875,84879088,"examples/crowd_code.html",6522,0,"",html,selection_command +876,84879338,"examples/crowd_code.html",6536,0,"",html,selection_command +877,84879371,"examples/crowd_code.html",6538,0,"",html,selection_command +878,84879404,"examples/crowd_code.html",6541,0,"",html,selection_command +879,84879638,"examples/crowd_code.html",6547,0,"",html,selection_command +880,84880738,"examples/crowd_code.html",6541,0,"",html,selection_command +881,84880991,"examples/crowd_code.html",6538,0,"",html,selection_command +882,84881020,"examples/crowd_code.html",6536,0,"",html,selection_command +883,84881052,"examples/crowd_code.html",6522,0,"",html,selection_command +884,84881082,"examples/crowd_code.html",6518,0,"",html,selection_command +885,84881115,"examples/crowd_code.html",6509,0,"",html,selection_command +886,84881150,"examples/crowd_code.html",6506,0,"",html,selection_command +887,84883173,"examples/crowd_code.html",6506,1,"o",html,selection_command +888,84883282,"examples/crowd_code.html",6506,2,"on",html,selection_command +889,84883512,"examples/crowd_code.html",6506,11,"on ideation",html,selection_command +890,84883695,"examples/crowd_code.html",6506,15,"on ideation and",html,selection_command +891,84884127,"examples/crowd_code.html",6506,16,"on ideation and ",html,selection_command +892,84884239,"examples/crowd_code.html",6506,16,"",html,content +893,84886772,"examples/crowd_code.html",6519,0,"",html,selection_command +894,84888322,"examples/crowd_code.html",6520,0,"",html,selection_command +895,84889622,"examples/crowd_code.html",6521,0,"",html,selection_command +896,84889926,"examples/crowd_code.html",6521,0," ",html,content +897,84889931,"examples/crowd_code.html",6522,0,"",html,selection_keyboard +898,84893455,"examples/crowd_code.html",6522,0,"M",html,content +899,84893458,"examples/crowd_code.html",6523,0,"",html,selection_keyboard +900,84894026,"examples/crowd_code.html",6523,0,"S",html,content +901,84894030,"examples/crowd_code.html",6524,0,"",html,selection_keyboard +902,84894955,"examples/crowd_code.html",6523,1,"",html,content +903,84895358,"examples/crowd_code.html",6523,0,"M",html,content +904,84895360,"examples/crowd_code.html",6524,0,"",html,selection_keyboard +905,84895754,"examples/crowd_code.html",6524,0," ",html,content +906,84895758,"examples/crowd_code.html",6525,0,"",html,selection_keyboard +907,84897221,"examples/crowd_code.html",6525,0,"w",html,content +908,84897224,"examples/crowd_code.html",6526,0,"",html,selection_keyboard +909,84897300,"examples/crowd_code.html",6526,0,"o",html,content +910,84897302,"examples/crowd_code.html",6527,0,"",html,selection_keyboard +911,84897410,"examples/crowd_code.html",6527,0,"r",html,content +912,84897412,"examples/crowd_code.html",6528,0,"",html,selection_keyboard +913,84897574,"examples/crowd_code.html",6528,0,"k",html,content +914,84897577,"examples/crowd_code.html",6529,0,"",html,selection_keyboard +915,84897595,"examples/crowd_code.html",6529,0,"e",html,content +916,84897597,"examples/crowd_code.html",6530,0,"",html,selection_keyboard +917,84897825,"examples/crowd_code.html",6530,0,"d",html,content +918,84897829,"examples/crowd_code.html",6531,0,"",html,selection_keyboard +919,84897936,"examples/crowd_code.html",6531,0," ",html,content +920,84897939,"examples/crowd_code.html",6532,0,"",html,selection_keyboard +921,84897956,"examples/crowd_code.html",6532,0,"o",html,content +922,84897957,"examples/crowd_code.html",6533,0,"",html,selection_keyboard +923,84898026,"examples/crowd_code.html",6533,0,"n",html,content +924,84898027,"examples/crowd_code.html",6534,0,"",html,selection_keyboard +925,84898203,"examples/crowd_code.html",6534,0," ",html,content +926,84898204,"examples/crowd_code.html",6535,0,"",html,selection_keyboard +927,84898206,"examples/crowd_code.html",6535,0,"t",html,content +928,84898207,"examples/crowd_code.html",6536,0,"",html,selection_keyboard +929,84898315,"examples/crowd_code.html",6536,0,"h",html,content +930,84898317,"examples/crowd_code.html",6537,0,"",html,selection_keyboard +931,84898423,"examples/crowd_code.html",6537,0,"e",html,content +932,84898424,"examples/crowd_code.html",6538,0,"",html,selection_keyboard +933,84898586,"examples/crowd_code.html",6538,0," ",html,content +934,84898589,"examples/crowd_code.html",6539,0,"",html,selection_keyboard +935,84898824,"examples/crowd_code.html",6539,0,"b",html,content +936,84898826,"examples/crowd_code.html",6540,0,"",html,selection_keyboard +937,84899037,"examples/crowd_code.html",6540,0,"a",html,content +938,84899038,"examples/crowd_code.html",6541,0,"",html,selection_keyboard +939,84899095,"examples/crowd_code.html",6541,0,"c",html,content +940,84899097,"examples/crowd_code.html",6542,0,"",html,selection_keyboard +941,84899278,"examples/crowd_code.html",6542,0,"k",html,content +942,84899280,"examples/crowd_code.html",6543,0,"",html,selection_keyboard +943,84899305,"examples/crowd_code.html",6543,0,"e",html,content +944,84899307,"examples/crowd_code.html",6544,0,"",html,selection_keyboard +945,84899422,"examples/crowd_code.html",6544,0,"n",html,content +946,84899424,"examples/crowd_code.html",6545,0,"",html,selection_keyboard +947,84899573,"examples/crowd_code.html",6545,0,"d",html,content +948,84899576,"examples/crowd_code.html",6546,0,"",html,selection_keyboard +949,84899687,"examples/crowd_code.html",6546,0,",",html,content +950,84899688,"examples/crowd_code.html",6547,0,"",html,selection_keyboard +951,84899838,"examples/crowd_code.html",6547,0," ",html,content +952,84899839,"examples/crowd_code.html",6548,0,"",html,selection_keyboard +953,84899883,"examples/crowd_code.html",6548,0,"w",html,content +954,84899885,"examples/crowd_code.html",6549,0,"",html,selection_keyboard +955,84900423,"examples/crowd_code.html",6549,0,"h",html,content +956,84900429,"examples/crowd_code.html",6550,0,"",html,selection_keyboard +957,84900525,"examples/crowd_code.html",6550,0,"i",html,content +958,84900527,"examples/crowd_code.html",6551,0,"",html,selection_keyboard +959,84900655,"examples/crowd_code.html",6551,0,"l",html,content +960,84900657,"examples/crowd_code.html",6552,0,"",html,selection_keyboard +961,84900661,"examples/crowd_code.html",6552,0,"e",html,content +962,84900662,"examples/crowd_code.html",6553,0,"",html,selection_keyboard +963,84900807,"examples/crowd_code.html",6553,0," ",html,content +964,84900809,"examples/crowd_code.html",6554,0,"",html,selection_keyboard +965,84902671,"examples/crowd_code.html",6548,6,"",html,content +966,84903067,"examples/crowd_code.html",6547,1,"",html,content +967,84903236,"examples/crowd_code.html",6546,1,"",html,content +968,84903679,"examples/crowd_code.html",6546,0,".",html,content +969,84903681,"examples/crowd_code.html",6547,0,"",html,selection_keyboard +970,84904258,"examples/crowd_code.html",6547,0," ",html,content +971,84904260,"examples/crowd_code.html",6548,0,"",html,selection_keyboard +972,84904605,"examples/crowd_code.html",6548,0,"F",html,content +973,84904607,"examples/crowd_code.html",6549,0,"",html,selection_keyboard +974,84905387,"examples/crowd_code.html",6548,1,"",html,content +975,84905560,"examples/crowd_code.html",6547,1,"",html,content +976,84905954,"examples/crowd_code.html",6546,0,"",html,selection_command +977,84907032,"examples/crowd_code.html",6549,0,"",html,selection_command +978,84907223,"examples/crowd_code.html",6555,0,"",html,selection_command +979,84907772,"examples/crowd_code.html",6551,0,"",html,selection_command +980,84907963,"examples/crowd_code.html",6548,0,"",html,selection_command +981,84908138,"examples/crowd_code.html",6546,0,"",html,selection_command +982,84908789,"examples/crowd_code.html",6549,0,"",html,selection_command +983,84908972,"examples/crowd_code.html",6555,0,"",html,selection_command +984,84909906,"examples/crowd_code.html",6551,0,"",html,selection_command +985,84910321,"examples/crowd_code.html",6551,0,"w",html,content +986,84910323,"examples/crowd_code.html",6552,0,"",html,selection_keyboard +987,84910390,"examples/crowd_code.html",6552,0,"o",html,content +988,84910392,"examples/crowd_code.html",6553,0,"",html,selection_keyboard +989,84910486,"examples/crowd_code.html",6553,0,"r",html,content +990,84910488,"examples/crowd_code.html",6554,0,"",html,selection_keyboard +991,84910670,"examples/crowd_code.html",6554,0,"k",html,content +992,84910672,"examples/crowd_code.html",6555,0,"",html,selection_keyboard +993,84910677,"examples/crowd_code.html",6555,0,"e",html,content +994,84910678,"examples/crowd_code.html",6556,0,"",html,selection_keyboard +995,84910937,"examples/crowd_code.html",6556,0,"d",html,content +996,84910941,"examples/crowd_code.html",6557,0,"",html,selection_keyboard +997,84911000,"examples/crowd_code.html",6557,0," ",html,content +998,84911003,"examples/crowd_code.html",6558,0,"",html,selection_keyboard +999,84911031,"examples/crowd_code.html",6558,0,"o",html,content +1000,84911034,"examples/crowd_code.html",6559,0,"",html,selection_keyboard +1001,84911126,"examples/crowd_code.html",6559,0,"n",html,content +1002,84911128,"examples/crowd_code.html",6560,0,"",html,selection_keyboard +1003,84911275,"examples/crowd_code.html",6560,0," ",html,content +1004,84911279,"examples/crowd_code.html",6561,0,"",html,selection_keyboard +1005,84911410,"examples/crowd_code.html",6561,0,"i",html,content +1006,84911412,"examples/crowd_code.html",6562,0,"",html,selection_keyboard +1007,84911777,"examples/crowd_code.html",6562,0,"d",html,content +1008,84911779,"examples/crowd_code.html",6563,0,"",html,selection_keyboard +1009,84911865,"examples/crowd_code.html",6563,0,"e",html,content +1010,84911869,"examples/crowd_code.html",6564,0,"",html,selection_keyboard +1011,84912136,"examples/crowd_code.html",6564,0,"a",html,content +1012,84912139,"examples/crowd_code.html",6565,0,"",html,selection_keyboard +1013,84912143,"examples/crowd_code.html",6565,0,"t",html,content +1014,84912144,"examples/crowd_code.html",6566,0,"",html,selection_keyboard +1015,84912157,"examples/crowd_code.html",6566,0,"i",html,content +1016,84912159,"examples/crowd_code.html",6567,0,"",html,selection_keyboard +1017,84912182,"examples/crowd_code.html",6567,0,"o",html,content +1018,84912185,"examples/crowd_code.html",6568,0,"",html,selection_keyboard +1019,84912257,"examples/crowd_code.html",6568,0,"n",html,content +1020,84912259,"examples/crowd_code.html",6569,0,"",html,selection_keyboard +1021,84912527,"examples/crowd_code.html",6569,0,",",html,content +1022,84912530,"examples/crowd_code.html",6570,0,"",html,selection_keyboard +1023,84912605,"examples/crowd_code.html",6570,0," ",html,content +1024,84912607,"examples/crowd_code.html",6571,0,"",html,selection_keyboard +1025,84927904,"examples/crowd_code.html",6570,1,"",html,content +1026,84928103,"examples/crowd_code.html",6569,1,"",html,content +1027,84929244,"examples/crowd_code.html",6569,0,",",html,content +1028,84929249,"examples/crowd_code.html",6570,0,"",html,selection_keyboard +1029,84929322,"examples/crowd_code.html",6570,0," ",html,content +1030,84929323,"examples/crowd_code.html",6571,0,"",html,selection_keyboard +1031,84929704,"examples/crowd_code.html",6571,0,"t",html,content +1032,84929705,"examples/crowd_code.html",6572,0,"",html,selection_keyboard +1033,84929783,"examples/crowd_code.html",6572,0,"h",html,content +1034,84929784,"examples/crowd_code.html",6573,0,"",html,selection_keyboard +1035,84929866,"examples/crowd_code.html",6573,0,"e",html,content +1036,84929867,"examples/crowd_code.html",6574,0,"",html,selection_keyboard +1037,84930104,"examples/crowd_code.html",6574,0," ",html,content +1038,84930106,"examples/crowd_code.html",6575,0,"",html,selection_keyboard +1039,84930109,"examples/crowd_code.html",6575,0,"e",html,content +1040,84930110,"examples/crowd_code.html",6576,0,"",html,selection_keyboard +1041,84930306,"examples/crowd_code.html",6576,0,"x",html,content +1042,84930310,"examples/crowd_code.html",6577,0,"",html,selection_keyboard +1043,84930408,"examples/crowd_code.html",6577,0,"t",html,content +1044,84930412,"examples/crowd_code.html",6578,0,"",html,selection_keyboard +1045,84930444,"examples/crowd_code.html",6578,0,"e",html,content +1046,84930447,"examples/crowd_code.html",6579,0,"",html,selection_keyboard +1047,84930549,"examples/crowd_code.html",6579,0,"n",html,content +1048,84930553,"examples/crowd_code.html",6580,0,"",html,selection_keyboard +1049,84930705,"examples/crowd_code.html",6580,0,"s",html,content +1050,84930709,"examples/crowd_code.html",6581,0,"",html,selection_keyboard +1051,84930767,"examples/crowd_code.html",6581,0,"i",html,content +1052,84930770,"examples/crowd_code.html",6582,0,"",html,selection_keyboard +1053,84931372,"examples/crowd_code.html",6582,0,"o",html,content +1054,84931376,"examples/crowd_code.html",6583,0,"",html,selection_keyboard +1055,84931456,"examples/crowd_code.html",6583,0,"n",html,content +1056,84931459,"examples/crowd_code.html",6584,0,"",html,selection_keyboard +1057,84931722,"examples/crowd_code.html",6584,0,",",html,content +1058,84931726,"examples/crowd_code.html",6585,0,"",html,selection_keyboard +1059,84931827,"examples/crowd_code.html",6585,0," ",html,content +1060,84931830,"examples/crowd_code.html",6586,0,"",html,selection_keyboard +1061,84931960,"examples/crowd_code.html",6586,0,"a",html,content +1062,84931965,"examples/crowd_code.html",6587,0,"",html,selection_keyboard +1063,84932005,"examples/crowd_code.html",6587,0,"n",html,content +1064,84932007,"examples/crowd_code.html",6588,0,"",html,selection_keyboard +1065,84932126,"examples/crowd_code.html",6588,0,"d",html,content +1066,84932128,"examples/crowd_code.html",6589,0,"",html,selection_keyboard +1067,84932327,"examples/crowd_code.html",6588,0,"",html,selection_command +1068,84933061,"examples/crowd_code.html",6589,0,"",html,selection_command +1069,84933091,"examples/crowd_code.html",6589,0," ",html,content +1070,84933093,"examples/crowd_code.html",6590,0,"",html,selection_keyboard +1071,84933341,"examples/crowd_code.html",6589,0,"",html,selection_command +1072,84941007,"examples/crowd_code.html",6586,0,"",html,selection_command +1073,84941143,"examples/crowd_code.html",6584,0,"",html,selection_command +1074,84941323,"examples/crowd_code.html",6575,0,"",html,selection_command +1075,84941506,"examples/crowd_code.html",6571,0,"",html,selection_command +1076,84942106,"examples/crowd_code.html",6570,0,"",html,selection_command +1077,84942250,"examples/crowd_code.html",6569,0,"",html,selection_command +1078,84942509,"examples/crowd_code.html",6569,1,"",html,content +1079,84943121,"examples/crowd_code.html",6569,0," ",html,content +1080,84943124,"examples/crowd_code.html",6570,0,"",html,selection_keyboard +1081,84943220,"examples/crowd_code.html",6570,0,"a",html,content +1082,84943222,"examples/crowd_code.html",6571,0,"",html,selection_keyboard +1083,84943241,"examples/crowd_code.html",6571,0,"n",html,content +1084,84943243,"examples/crowd_code.html",6572,0,"",html,selection_keyboard +1085,84943407,"examples/crowd_code.html",6572,0,"d",html,content +1086,84943411,"examples/crowd_code.html",6573,0,"",html,selection_keyboard +1087,84943601,"examples/crowd_code.html",6572,0,"",html,selection_command +1088,84943836,"examples/crowd_code.html",6576,0,"",html,selection_command +1089,84944001,"examples/crowd_code.html",6586,0,"",html,selection_command +1090,84944212,"examples/crowd_code.html",6587,0,"",html,selection_command +1091,84944825,"examples/crowd_code.html",6587,1,".",html,content +1092,84945529,"examples/crowd_code.html",6588,0,"",html,selection_command +1093,84946025,"examples/crowd_code.html",6589,0,"",html,selection_command +1094,84946358,"examples/crowd_code.html",6589,0,"F",html,content +1095,84946360,"examples/crowd_code.html",6590,0,"",html,selection_keyboard +1096,84946416,"examples/crowd_code.html",6590,0,"S",html,content +1097,84946419,"examples/crowd_code.html",6591,0,"",html,selection_keyboard +1098,84946841,"examples/crowd_code.html",6591,0," ",html,content +1099,84946844,"examples/crowd_code.html",6592,0,"",html,selection_keyboard +1100,84947083,"examples/crowd_code.html",6591,0,"",html,selection_command +1101,84947625,"examples/crowd_code.html",6592,0,"",html,selection_command +1102,84947861,"examples/crowd_code.html",6592,4,"",html,content +1103,84962675,"examples/crowd_code.html",6589,0,"",html,selection_command +1104,84962809,"examples/crowd_code.html",6587,0,"",html,selection_command +1105,84962973,"examples/crowd_code.html",6578,0,"",html,selection_command +1106,84963112,"examples/crowd_code.html",6574,0,"",html,selection_command +1107,84964506,"examples/crowd_code.html",6578,0,"",html,selection_command +1108,84966022,"examples/crowd_code.html",6574,0,"",html,selection_command +1109,84966175,"examples/crowd_code.html",6570,0,"",html,selection_command +1110,84967337,"examples/crowd_code.html",6570,1,"a",html,selection_command +1111,84967374,"examples/crowd_code.html",6570,3,"and",html,selection_command +1112,84967569,"examples/crowd_code.html",6570,7,"and the",html,selection_command +1113,84968589,"examples/crowd_code.html",6576,0,"",html,selection_command +1114,84968864,"examples/crowd_code.html",6574,0,"",html,selection_command +1115,84969012,"examples/crowd_code.html",6570,0,"",html,selection_command +1116,84969172,"examples/crowd_code.html",6561,0,"",html,selection_command +1117,84969326,"examples/crowd_code.html",6558,0,"",html,selection_command +1118,84969587,"examples/crowd_code.html",6561,0,"",html,selection_command +1119,84969788,"examples/crowd_code.html",6570,0,"",html,selection_command +1120,84969955,"examples/crowd_code.html",6574,0,"",html,selection_command +1121,84970109,"examples/crowd_code.html",6578,0,"",html,selection_command +1122,84970263,"examples/crowd_code.html",6587,0,"",html,selection_command +1123,84970464,"examples/crowd_code.html",6589,0,"",html,selection_command +1124,84971968,"TERMINAL",0,0,"^P",,terminal_output +1125,84972661,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +1126,84973162,"TERMINAL",0,0,"npm run dev",,terminal_output +1127,84973247,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +1128,84973500,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +1129,84973501,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +1130,84973519,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +1131,84974074,"TERMINAL",0,0,"npm run dev",,terminal_command +1132,84974126,"TERMINAL",0,0,"]633;C",,terminal_output +1133,84974254,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +1134,84974457,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +1135,84974929,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 473ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +1136,84975288,"TERMINAL",0,0,"created dist/transforms.v2.js in 359ms\r\n\r\n[2025-09-03 16:48:05] waiting for changes...\r\n",,terminal_output +1137,84982121,"examples/crowd_code.html",6587,0,"",html,selection_command +1138,84982224,"examples/crowd_code.html",6578,0,"",html,selection_command +1139,84982475,"examples/crowd_code.html",6574,0,"",html,selection_command +1140,84982508,"examples/crowd_code.html",6570,0,"",html,selection_command +1141,84982541,"examples/crowd_code.html",6561,0,"",html,selection_command +1142,84982575,"examples/crowd_code.html",6558,0,"",html,selection_command +1143,84982609,"examples/crowd_code.html",6551,0,"",html,selection_command +1144,84982648,"examples/crowd_code.html",6548,0,"",html,selection_command +1145,84982793,"examples/crowd_code.html",6546,0,"",html,selection_command +1146,84982938,"examples/crowd_code.html",6539,0,"",html,selection_command +1147,84983144,"examples/crowd_code.html",6535,0,"",html,selection_command +1148,84983351,"examples/crowd_code.html",6532,0,"",html,selection_command +1149,84983527,"examples/crowd_code.html",6525,0,"",html,selection_command +1150,84983676,"examples/crowd_code.html",6522,0,"",html,selection_command +1151,84983899,"examples/crowd_code.html",6520,0,"",html,selection_command +1152,84984210,"examples/crowd_code.html",6506,0,"",html,selection_command +1153,84984729,"examples/crowd_code.html",6506,0,"o",html,content +1154,84984734,"examples/crowd_code.html",6507,0,"",html,selection_keyboard +1155,84984807,"examples/crowd_code.html",6507,0,"n",html,content +1156,84984808,"examples/crowd_code.html",6508,0,"",html,selection_keyboard +1157,84984985,"examples/crowd_code.html",6508,0," ",html,content +1158,84984986,"examples/crowd_code.html",6509,0,"",html,selection_keyboard +1159,84985100,"examples/crowd_code.html",6509,0,"t",html,content +1160,84985102,"examples/crowd_code.html",6510,0,"",html,selection_keyboard +1161,84985189,"examples/crowd_code.html",6510,0,"h",html,content +1162,84985190,"examples/crowd_code.html",6511,0,"",html,selection_keyboard +1163,84985283,"examples/crowd_code.html",6511,0,"e",html,content +1164,84985285,"examples/crowd_code.html",6512,0,"",html,selection_keyboard +1165,84985413,"examples/crowd_code.html",6512,0," ",html,content +1166,84985414,"examples/crowd_code.html",6513,0,"",html,selection_keyboard +1167,84985622,"examples/crowd_code.html",6512,0,"",html,selection_command +1168,84987146,"TERMINAL",0,0,"^P",,terminal_output +1169,84987726,"TERMINAL",0,0,"",,terminal_command +1170,84987726,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h[?2004l\r\r\n% \r \r]633;E;;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +1171,84987726,"TERMINAL",0,0,"]633;C",,terminal_output +1172,84988411,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +1173,84988431,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +1174,84988932,"TERMINAL",0,0,"npm run dev",,terminal_command +1175,84988984,"TERMINAL",0,0,"]633;C",,terminal_output +1176,84989100,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +1177,84989251,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +1178,84989672,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 424ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +1179,84990078,"TERMINAL",0,0,"created dist/transforms.v2.js in 404ms\r\n\r\n[2025-09-03 16:48:20] waiting for changes...\r\n",,terminal_output +1180,84996294,"examples/crowd_code.html",6509,0,"",html,selection_command +1181,84996443,"examples/crowd_code.html",6506,0,"",html,selection_command +1182,84996841,"examples/crowd_code.html",6509,0,"",html,selection_command +1183,84997408,"examples/crowd_code.html",6509,4,"",html,content +1184,84999406,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0^C\rfranzsrambical@MBF6N9WFVKFV pdoom.org % [?2004h",,terminal_output +1185,84999876,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +1186,85000272,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1]633;C% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +1187,85000828,"TERMINAL",0,0,"cp examples/* distnpm run dev ",,terminal_output +1188,85001039,"TERMINAL",0,0,"npm run dev",,terminal_command +1189,85001039,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;npm run dev;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +1190,85001090,"TERMINAL",0,0,"]633;C",,terminal_output +1191,85001218,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +1192,85001373,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +1193,85001765,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 394ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +1194,85002287,"TERMINAL",0,0,"created dist/transforms.v2.js in 520ms\r\n\r\n[2025-09-03 16:48:32] waiting for changes...\r\n",,terminal_output +1195,85017006,"examples/crowd_code.html",6522,0,"",html,selection_command +1196,85017255,"examples/crowd_code.html",6523,0,"",html,selection_command +1197,85017282,"examples/crowd_code.html",6526,0,"",html,selection_command +1198,85017314,"examples/crowd_code.html",6533,0,"",html,selection_command +1199,85017348,"examples/crowd_code.html",6536,0,"",html,selection_command +1200,85017384,"examples/crowd_code.html",6540,0,"",html,selection_command +1201,85017416,"examples/crowd_code.html",6548,0,"",html,selection_command +1202,85017449,"examples/crowd_code.html",6549,0,"",html,selection_command +1203,85017484,"examples/crowd_code.html",6552,0,"",html,selection_command +1204,85017519,"examples/crowd_code.html",6559,0,"",html,selection_command +1205,85017771,"examples/crowd_code.html",6562,0,"",html,selection_command +1206,85017955,"examples/crowd_code.html",6571,0,"",html,selection_command +1207,85018105,"examples/crowd_code.html",6575,0,"",html,selection_command +1208,85018253,"examples/crowd_code.html",6579,0,"",html,selection_command +1209,85018671,"examples/crowd_code.html",6580,0,"",html,selection_command +1210,85018990,"examples/crowd_code.html",6580,0," ",html,content +1211,85018995,"examples/crowd_code.html",6581,0,"",html,selection_keyboard +1212,85019695,"examples/crowd_code.html",6581,0,"I",html,content +1213,85019697,"examples/crowd_code.html",6582,0,"",html,selection_keyboard +1214,85020271,"examples/crowd_code.html",6582,0,"D",html,content +1215,85020276,"examples/crowd_code.html",6583,0,"",html,selection_keyboard +1216,85020357,"examples/crowd_code.html",6583,0,"E",html,content +1217,85020360,"examples/crowd_code.html",6584,0,"",html,selection_keyboard +1218,85020572,"examples/crowd_code.html",6583,0,"",html,selection_command +1219,85022535,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +1220,85022861,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +1221,85023132,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +1222,85023132,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +1223,85023148,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +1224,85024058,"TERMINAL",0,0,"npm run dev",,terminal_command +1225,85024109,"TERMINAL",0,0,"]633;C",,terminal_output +1226,85024229,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +1227,85024379,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +1228,85024748,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 369ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +1229,85025219,"TERMINAL",0,0,"created dist/transforms.v2.js in 472ms\r\n\r\n[2025-09-03 16:48:55] waiting for changes...\r\n",,terminal_output +1230,85070742,"examples/crowd_code.html",6581,0,"",html,selection_command +1231,85070991,"examples/crowd_code.html",6577,0,"",html,selection_command +1232,85071020,"examples/crowd_code.html",6573,0,"",html,selection_command +1233,85071054,"examples/crowd_code.html",6564,0,"",html,selection_command +1234,85071086,"examples/crowd_code.html",6561,0,"",html,selection_command +1235,85071120,"examples/crowd_code.html",6554,0,"",html,selection_command +1236,85071272,"examples/crowd_code.html",6551,0,"",html,selection_command +1237,85071441,"examples/crowd_code.html",6549,0,"",html,selection_command +1238,85071590,"examples/crowd_code.html",6542,0,"",html,selection_command +1239,85071792,"examples/crowd_code.html",6538,0,"",html,selection_command +1240,85072044,"examples/crowd_code.html",6535,0,"",html,selection_command +1241,85072209,"examples/crowd_code.html",6528,0,"",html,selection_command +1242,85072367,"examples/crowd_code.html",6525,0,"",html,selection_command +1243,85072527,"examples/crowd_code.html",6523,0,"",html,selection_command +1244,85072650,"examples/crowd_code.html",6509,0,"",html,selection_command +1245,85075292,"examples/crowd_code.html",6506,0,"",html,selection_command +1246,85075462,"examples/crowd_code.html",6499,0,"",html,selection_command +1247,85076908,"examples/crowd_code.html",6499,1,"w",html,selection_command +1248,85076969,"examples/crowd_code.html",6499,6,"worked",html,selection_command +1249,85077175,"examples/crowd_code.html",6499,9,"worked on",html,selection_command +1250,85077367,"examples/crowd_code.html",6499,24,"worked on implementation",html,selection_command +1251,85079272,"examples/crowd_code.html",6522,0,"",html,selection_command +1252,85079573,"examples/crowd_code.html",6509,0,"",html,selection_command +1253,85086641,"examples/crowd_code.html",6506,0,"",html,selection_command +1254,85086796,"examples/crowd_code.html",6499,0,"",html,selection_command +1255,85087629,"examples/crowd_code.html",6506,0,"",html,selection_command +1256,85087821,"examples/crowd_code.html",6509,0,"",html,selection_command +1257,85088155,"examples/crowd_code.html",6506,0,"",html,selection_command +1258,85088380,"examples/crowd_code.html",6499,0,"",html,selection_command +1259,85088671,"examples/crowd_code.html",6499,1,"w",html,selection_command +1260,85088774,"examples/crowd_code.html",6499,6,"worked",html,selection_command +1261,85088987,"examples/crowd_code.html",6499,9,"worked on",html,selection_command +1262,85089325,"examples/crowd_code.html",6499,24,"worked on implementation",html,selection_command +1263,85090037,"examples/crowd_code.html",6522,0,"",html,selection_command +1264,85090179,"examples/crowd_code.html",6509,0,"",html,selection_command +1265,85094087,"examples/crowd_code.html",6506,0,"",html,selection_command +1266,85094292,"examples/crowd_code.html",6499,0,"",html,selection_command +1267,85095010,"examples/crowd_code.html",6506,0,"",html,selection_command +1268,85095591,"examples/crowd_code.html",6506,1,"o",html,selection_command +1269,85095638,"examples/crowd_code.html",6506,2,"on",html,selection_command +1270,85095838,"examples/crowd_code.html",6506,17,"on implementation",html,selection_command +1271,85097022,"examples/crowd_code.html",6506,17,"",html,content +1272,85097956,"examples/crowd_code.html",6499,7,"",html,content +1273,85100258,"examples/crowd_code.html",6498,0,"",html,selection_command +1274,85103471,"examples/crowd_code.html",6499,0,"",html,selection_command +1275,85104171,"examples/crowd_code.html",6496,3,"",html,content +1276,85104358,"examples/crowd_code.html",6492,4,"",html,content +1277,85104542,"examples/crowd_code.html",6489,3,"",html,content +1278,85104738,"examples/crowd_code.html",6487,2,"",html,content +1279,85105490,"examples/crowd_code.html",6488,0,"",html,selection_command +1280,85105651,"examples/crowd_code.html",6489,0,"",html,selection_command +1281,85106075,"examples/crowd_code.html",6488,1,"",html,content +1282,85106238,"examples/crowd_code.html",6487,1,"",html,content +1283,85106386,"examples/crowd_code.html",6486,1,"",html,content +1284,85106536,"examples/crowd_code.html",6485,1,"",html,content +1285,85107025,"examples/crowd_code.html",6485,0," ",html,content +1286,85107027,"examples/crowd_code.html",6486,0,"",html,selection_keyboard +1287,85107754,"examples/crowd_code.html",6485,1,"",html,content +1288,85107907,"examples/crowd_code.html",6484,0,"",html,selection_command +1289,85109241,"examples/crowd_code.html",6486,0,"",html,selection_command +1290,85109491,"examples/crowd_code.html",6493,0,"",html,selection_command +1291,85109524,"examples/crowd_code.html",6496,0,"",html,selection_command +1292,85109558,"examples/crowd_code.html",6500,0,"",html,selection_command +1293,85109591,"examples/crowd_code.html",6508,0,"",html,selection_command +1294,85109624,"examples/crowd_code.html",6509,0,"",html,selection_command +1295,85109822,"examples/crowd_code.html",6512,0,"",html,selection_command +1296,85110069,"examples/crowd_code.html",6519,0,"",html,selection_command +1297,85110098,"examples/crowd_code.html",6522,0,"",html,selection_command +1298,85110131,"examples/crowd_code.html",6531,0,"",html,selection_command +1299,85110165,"examples/crowd_code.html",6535,0,"",html,selection_command +1300,85110199,"examples/crowd_code.html",6539,0,"",html,selection_command +1301,85110354,"examples/crowd_code.html",6543,0,"",html,selection_command +1302,85110534,"examples/crowd_code.html",6553,0,"",html,selection_command +1303,85110721,"examples/crowd_code.html",6554,0,"",html,selection_command +1304,85110895,"examples/crowd_code.html",6557,0,"",html,selection_command +1305,85111075,"examples/crowd_code.html",6563,0,"",html,selection_command +1306,85111255,"examples/crowd_code.html",6567,0,"",html,selection_command +1307,85111472,"examples/crowd_code.html",6578,0,"",html,selection_command +1308,85111645,"examples/crowd_code.html",6579,0,"",html,selection_command +1309,85112184,"examples/crowd_code.html",6580,0,"",html,selection_command +1310,85112402,"examples/crowd_code.html",6580,0," ",html,content +1311,85112405,"examples/crowd_code.html",6581,0,"",html,selection_keyboard +1312,85112670,"examples/crowd_code.html",6581,0,"A",html,content +1313,85112673,"examples/crowd_code.html",6582,0,"",html,selection_keyboard +1314,85112906,"examples/crowd_code.html",6582,0,"l",html,content +1315,85112908,"examples/crowd_code.html",6583,0,"",html,selection_keyboard +1316,85113093,"examples/crowd_code.html",6583,0,"l",html,content +1317,85113095,"examples/crowd_code.html",6584,0,"",html,selection_keyboard +1318,85113132,"examples/crowd_code.html",6584,0," ",html,content +1319,85113135,"examples/crowd_code.html",6585,0,"",html,selection_keyboard +1320,85113299,"examples/crowd_code.html",6585,0,"a",html,content +1321,85113301,"examples/crowd_code.html",6586,0,"",html,selection_keyboard +1322,85113322,"examples/crowd_code.html",6586,0,"u",html,content +1323,85113324,"examples/crowd_code.html",6587,0,"",html,selection_keyboard +1324,85113419,"examples/crowd_code.html",6587,0,"t",html,content +1325,85113420,"examples/crowd_code.html",6588,0,"",html,selection_keyboard +1326,85113517,"examples/crowd_code.html",6588,0,"h",html,content +1327,85113518,"examples/crowd_code.html",6589,0,"",html,selection_keyboard +1328,85113601,"examples/crowd_code.html",6589,0,"o",html,content +1329,85113605,"examples/crowd_code.html",6590,0,"",html,selection_keyboard +1330,85113647,"examples/crowd_code.html",6590,0,"r",html,content +1331,85113651,"examples/crowd_code.html",6591,0,"",html,selection_keyboard +1332,85113841,"examples/crowd_code.html",6591,0,"s",html,content +1333,85113846,"examples/crowd_code.html",6592,0,"",html,selection_keyboard +1334,85120818,"examples/crowd_code.html",6585,7,"",html,content +1335,85121051,"examples/crowd_code.html",6581,4,"",html,content +1336,85121681,"examples/crowd_code.html",6580,1,"",html,content +1337,85121887,"examples/crowd_code.html",6579,0,"",html,selection_command +1338,85125060,"examples/crowd_code.html",6580,0,"",html,selection_command +1339,85125230,"examples/crowd_code.html",6580,0," ",html,content +1340,85125234,"examples/crowd_code.html",6581,0,"",html,selection_keyboard +1341,85125509,"examples/crowd_code.html",6581,0,"A",html,content +1342,85125510,"examples/crowd_code.html",6582,0,"",html,selection_keyboard +1343,85125744,"examples/crowd_code.html",6582,0,"l",html,content +1344,85125746,"examples/crowd_code.html",6583,0,"",html,selection_keyboard +1345,85125911,"examples/crowd_code.html",6583,0,"l",html,content +1346,85125913,"examples/crowd_code.html",6584,0,"",html,selection_keyboard +1347,85126002,"examples/crowd_code.html",6584,0," ",html,content +1348,85126003,"examples/crowd_code.html",6585,0,"",html,selection_keyboard +1349,85127124,"examples/crowd_code.html",6585,0,"a",html,content +1350,85127129,"examples/crowd_code.html",6586,0,"",html,selection_keyboard +1351,85127188,"examples/crowd_code.html",6586,0,"t",html,content +1352,85127191,"examples/crowd_code.html",6587,0,"",html,selection_keyboard +1353,85127195,"examples/crowd_code.html",6587,0,"u",html,content +1354,85127196,"examples/crowd_code.html",6588,0,"",html,selection_keyboard +1355,85127721,"examples/crowd_code.html",6587,1,"",html,content +1356,85127858,"examples/crowd_code.html",6586,1,"",html,content +1357,85127919,"examples/crowd_code.html",6586,0,"u",html,content +1358,85127920,"examples/crowd_code.html",6587,0,"",html,selection_keyboard +1359,85128187,"examples/crowd_code.html",6587,0,"t",html,content +1360,85128189,"examples/crowd_code.html",6588,0,"",html,selection_keyboard +1361,85128489,"examples/crowd_code.html",6587,1,"",html,content +1362,85128822,"examples/crowd_code.html",6587,0,"h",html,content +1363,85128823,"examples/crowd_code.html",6588,0,"",html,selection_keyboard +1364,85129123,"examples/crowd_code.html",6587,1,"",html,content +1365,85129155,"examples/crowd_code.html",6587,0,"t",html,content +1366,85129156,"examples/crowd_code.html",6588,0,"",html,selection_keyboard +1367,85129223,"examples/crowd_code.html",6588,0,"h",html,content +1368,85129226,"examples/crowd_code.html",6589,0,"",html,selection_keyboard +1369,85129321,"examples/crowd_code.html",6589,0,"o",html,content +1370,85129323,"examples/crowd_code.html",6590,0,"",html,selection_keyboard +1371,85129382,"examples/crowd_code.html",6590,0,"r",html,content +1372,85129385,"examples/crowd_code.html",6591,0,"",html,selection_keyboard +1373,85129556,"examples/crowd_code.html",6591,0,"s",html,content +1374,85129558,"examples/crowd_code.html",6592,0,"",html,selection_keyboard +1375,85129652,"examples/crowd_code.html",6592,0," ",html,content +1376,85129654,"examples/crowd_code.html",6593,0,"",html,selection_keyboard +1377,85129888,"examples/crowd_code.html",6593,0,"c",html,content +1378,85129889,"examples/crowd_code.html",6594,0,"",html,selection_keyboard +1379,85129969,"examples/crowd_code.html",6594,0,"o",html,content +1380,85129970,"examples/crowd_code.html",6595,0,"",html,selection_keyboard +1381,85130076,"examples/crowd_code.html",6595,0,"n",html,content +1382,85130077,"examples/crowd_code.html",6596,0,"",html,selection_keyboard +1383,85130165,"examples/crowd_code.html",6596,0,"t",html,content +1384,85130169,"examples/crowd_code.html",6597,0,"",html,selection_keyboard +1385,85130608,"examples/crowd_code.html",6597,0,"r",html,content +1386,85130615,"examples/crowd_code.html",6598,0,"",html,selection_keyboard +1387,85130763,"examples/crowd_code.html",6598,0,"i",html,content +1388,85130769,"examples/crowd_code.html",6599,0,"",html,selection_keyboard +1389,85130848,"examples/crowd_code.html",6599,0,"b",html,content +1390,85130853,"examples/crowd_code.html",6600,0,"",html,selection_keyboard +1391,85131005,"examples/crowd_code.html",6600,0,"u",html,content +1392,85131011,"examples/crowd_code.html",6601,0,"",html,selection_keyboard +1393,85131056,"examples/crowd_code.html",6601,0,"t",html,content +1394,85131060,"examples/crowd_code.html",6602,0,"",html,selection_keyboard +1395,85131194,"examples/crowd_code.html",6602,0,"e",html,content +1396,85131199,"examples/crowd_code.html",6603,0,"",html,selection_keyboard +1397,85131437,"examples/crowd_code.html",6603,0,"d",html,content +1398,85131443,"examples/crowd_code.html",6604,0,"",html,selection_keyboard +1399,85131534,"examples/crowd_code.html",6604,0," ",html,content +1400,85131537,"examples/crowd_code.html",6605,0,"",html,selection_keyboard +1401,85131872,"examples/crowd_code.html",6605,0,"t",html,content +1402,85131873,"examples/crowd_code.html",6606,0,"",html,selection_keyboard +1403,85131942,"examples/crowd_code.html",6606,0,"e",html,content +1404,85131943,"examples/crowd_code.html",6607,0,"",html,selection_keyboard +1405,85132469,"examples/crowd_code.html",6607,0,"c",html,content +1406,85132471,"examples/crowd_code.html",6608,0,"",html,selection_keyboard +1407,85132606,"examples/crowd_code.html",6608,0,"h",html,content +1408,85132608,"examples/crowd_code.html",6609,0,"",html,selection_keyboard +1409,85134310,"examples/crowd_code.html",6605,4,"",html,content +1410,85134448,"examples/crowd_code.html",6593,12,"",html,content +1411,85134606,"examples/crowd_code.html",6585,8,"",html,content +1412,85134772,"examples/crowd_code.html",6581,4,"",html,content +1413,85135255,"examples/crowd_code.html",6580,1,"",html,content +1414,85135422,"examples/crowd_code.html",6579,0,"",html,selection_command +1415,85141809,"examples/crowd_code.html",6569,0,"",html,selection_command +1416,85142058,"examples/crowd_code.html",6565,0,"",html,selection_command +1417,85142088,"examples/crowd_code.html",6559,0,"",html,selection_command +1418,85142121,"examples/crowd_code.html",6556,0,"",html,selection_command +1419,85142154,"examples/crowd_code.html",6554,0,"",html,selection_command +1420,85142186,"examples/crowd_code.html",6545,0,"",html,selection_command +1421,85142220,"examples/crowd_code.html",6541,0,"",html,selection_command +1422,85142253,"examples/crowd_code.html",6537,0,"",html,selection_command +1423,85142286,"examples/crowd_code.html",6533,0,"",html,selection_command +1424,85142319,"examples/crowd_code.html",6524,0,"",html,selection_command +1425,85142353,"examples/crowd_code.html",6521,0,"",html,selection_command +1426,85142386,"examples/crowd_code.html",6514,0,"",html,selection_command +1427,85142419,"examples/crowd_code.html",6511,0,"",html,selection_command +1428,85142453,"examples/crowd_code.html",6509,0,"",html,selection_command +1429,85142486,"examples/crowd_code.html",6502,0,"",html,selection_command +1430,85142520,"examples/crowd_code.html",6498,0,"",html,selection_command +1431,85142553,"examples/crowd_code.html",6495,0,"",html,selection_command +1432,85142586,"examples/crowd_code.html",6488,0,"",html,selection_command +1433,85142619,"examples/crowd_code.html",6485,0,"",html,selection_command +1434,85142653,"examples/crowd_code.html",6484,0,"",html,selection_command +1435,85143164,"examples/crowd_code.html",6483,0,"",html,selection_command +1436,85143414,"examples/crowd_code.html",6482,0,"",html,selection_command +1437,85143444,"examples/crowd_code.html",6476,0,"",html,selection_command +1438,85143477,"examples/crowd_code.html",6474,0,"",html,selection_command +1439,85143510,"examples/crowd_code.html",6472,0,"",html,selection_command +1440,85143725,"examples/crowd_code.html",6474,0,"",html,selection_command +1441,85143977,"examples/crowd_code.html",6476,0,"",html,selection_command +1442,85144010,"examples/crowd_code.html",6482,0,"",html,selection_command +1443,85144048,"examples/crowd_code.html",6483,0,"",html,selection_command +1444,85144300,"examples/crowd_code.html",6484,0,"",html,selection_command +1445,85144556,"examples/crowd_code.html",6485,0,"",html,selection_command +1446,85144587,"examples/crowd_code.html",6488,0,"",html,selection_command +1447,85144620,"examples/crowd_code.html",6495,0,"",html,selection_command +1448,85144653,"examples/crowd_code.html",6498,0,"",html,selection_command +1449,85144928,"examples/crowd_code.html",6502,0,"",html,selection_command +1450,85145106,"examples/crowd_code.html",6509,0,"",html,selection_command +1451,85145274,"examples/crowd_code.html",6511,0,"",html,selection_command +1452,85145555,"examples/crowd_code.html",6509,0,"",html,selection_command +1453,85145805,"examples/crowd_code.html",6502,0,"",html,selection_command +1454,85146005,"examples/crowd_code.html",6498,0,"",html,selection_command +1455,85146186,"examples/crowd_code.html",6495,0,"",html,selection_command +1456,85146341,"examples/crowd_code.html",6488,0,"",html,selection_command +1457,85146524,"examples/crowd_code.html",6485,0,"",html,selection_command +1458,85146818,"examples/crowd_code.html",6488,0,"",html,selection_command +1459,85147013,"examples/crowd_code.html",6495,0,"",html,selection_command +1460,85147189,"examples/crowd_code.html",6498,0,"",html,selection_command +1461,85147355,"examples/crowd_code.html",6502,0,"",html,selection_command +1462,85147991,"examples/crowd_code.html",6498,0,"",html,selection_command +1463,85148140,"examples/crowd_code.html",6495,0,"",html,selection_command +1464,85148542,"examples/crowd_code.html",6498,0,"",html,selection_command +1465,85153355,"examples/crowd_code.html",6502,0,"",html,selection_command +1466,85154777,"examples/crowd_code.html",6502,7,"",html,content +1467,85155091,"examples/crowd_code.html",6502,0,"s",html,content +1468,85155093,"examples/crowd_code.html",6503,0,"",html,selection_keyboard +1469,85155096,"examples/crowd_code.html",6503,0,"e",html,content +1470,85155098,"examples/crowd_code.html",6504,0,"",html,selection_keyboard +1471,85155106,"examples/crowd_code.html",6504,0,"r",html,content +1472,85155107,"examples/crowd_code.html",6505,0,"",html,selection_keyboard +1473,85155419,"examples/crowd_code.html",6505,0,"v",html,content +1474,85155421,"examples/crowd_code.html",6506,0,"",html,selection_keyboard +1475,85155567,"examples/crowd_code.html",6506,0,"e",html,content +1476,85155569,"examples/crowd_code.html",6507,0,"",html,selection_keyboard +1477,85155638,"examples/crowd_code.html",6507,0,"r",html,content +1478,85155640,"examples/crowd_code.html",6508,0,"",html,selection_keyboard +1479,85156364,"examples/crowd_code.html",6508,0,"-",html,content +1480,85156370,"examples/crowd_code.html",6509,0,"",html,selection_keyboard +1481,85156888,"examples/crowd_code.html",6509,0,"s",html,content +1482,85156893,"examples/crowd_code.html",6510,0,"",html,selection_keyboard +1483,85156913,"examples/crowd_code.html",6510,0,"i",html,content +1484,85156915,"examples/crowd_code.html",6511,0,"",html,selection_keyboard +1485,85157062,"examples/crowd_code.html",6511,0,"d",html,content +1486,85157066,"examples/crowd_code.html",6512,0,"",html,selection_keyboard +1487,85157151,"examples/crowd_code.html",6512,0,"e",html,content +1488,85157153,"examples/crowd_code.html",6513,0,"",html,selection_keyboard +1489,85157613,"examples/crowd_code.html",6513,0," ",html,content +1490,85157616,"examples/crowd_code.html",6514,0,"",html,selection_keyboard +1491,85158382,"examples/crowd_code.html",6514,0,"c",html,content +1492,85158383,"examples/crowd_code.html",6515,0,"",html,selection_keyboard +1493,85158875,"examples/crowd_code.html",6514,1,"",html,content +1494,85159301,"examples/crowd_code.html",6513,1,"",html,content +1495,85159609,"examples/crowd_code.html",6512,0,"",html,selection_command +1496,85160256,"examples/crowd_code.html",6509,0,"",html,selection_command +1497,85160426,"examples/crowd_code.html",6508,0,"",html,selection_command +1498,85160606,"examples/crowd_code.html",6502,0,"",html,selection_command +1499,85160790,"examples/crowd_code.html",6498,0,"",html,selection_command +1500,85160952,"examples/crowd_code.html",6495,0,"",html,selection_command +1501,85161125,"examples/crowd_code.html",6488,0,"",html,selection_command +1502,85167807,"examples/crowd_code.html",6488,1,"w",html,selection_command +1503,85167879,"examples/crowd_code.html",6488,6,"worked",html,selection_command +1504,85168059,"examples/crowd_code.html",6488,9,"worked on",html,selection_command +1505,85168224,"examples/crowd_code.html",6488,13,"worked on the",html,selection_command +1506,85168374,"examples/crowd_code.html",6488,20,"worked on the server",html,selection_command +1507,85168539,"examples/crowd_code.html",6488,21,"worked on the server-",html,selection_command +1508,85168787,"examples/crowd_code.html",6488,21,"",html,content +1509,85169412,"examples/crowd_code.html",6487,0,"",html,selection_command +1510,85169727,"examples/crowd_code.html",6488,0,"",html,selection_command +1511,85170248,"examples/crowd_code.html",6488,4,"",html,content +1512,85170426,"examples/crowd_code.html",6488,0,"i",html,content +1513,85170429,"examples/crowd_code.html",6489,0,"",html,selection_keyboard +1514,85170505,"examples/crowd_code.html",6489,0,"m",html,content +1515,85170506,"examples/crowd_code.html",6490,0,"",html,selection_keyboard +1516,85170590,"examples/crowd_code.html",6490,0,"p",html,content +1517,85170591,"examples/crowd_code.html",6491,0,"",html,selection_keyboard +1518,85170903,"examples/crowd_code.html",6491,0,"l",html,content +1519,85170906,"examples/crowd_code.html",6492,0,"",html,selection_keyboard +1520,85171271,"examples/crowd_code.html",6492,0,"e",html,content +1521,85171277,"examples/crowd_code.html",6493,0,"",html,selection_keyboard +1522,85171390,"examples/crowd_code.html",6493,0,"m",html,content +1523,85171396,"examples/crowd_code.html",6494,0,"",html,selection_keyboard +1524,85171503,"examples/crowd_code.html",6494,0,"e",html,content +1525,85171506,"examples/crowd_code.html",6495,0,"",html,selection_keyboard +1526,85171570,"examples/crowd_code.html",6495,0,"n",html,content +1527,85171573,"examples/crowd_code.html",6496,0,"",html,selection_keyboard +1528,85171653,"examples/crowd_code.html",6496,0,"t",html,content +1529,85171656,"examples/crowd_code.html",6497,0,"",html,selection_keyboard +1530,85171744,"examples/crowd_code.html",6497,0,"e",html,content +1531,85171748,"examples/crowd_code.html",6498,0,"",html,selection_keyboard +1532,85172237,"examples/crowd_code.html",6498,0,"d",html,content +1533,85172241,"examples/crowd_code.html",6499,0,"",html,selection_keyboard +1534,85172321,"examples/crowd_code.html",6499,0," ",html,content +1535,85172324,"examples/crowd_code.html",6500,0,"",html,selection_keyboard +1536,85172432,"examples/crowd_code.html",6500,0,"t",html,content +1537,85172435,"examples/crowd_code.html",6501,0,"",html,selection_keyboard +1538,85172621,"examples/crowd_code.html",6501,0,"h",html,content +1539,85172625,"examples/crowd_code.html",6502,0,"",html,selection_keyboard +1540,85172697,"examples/crowd_code.html",6502,0,"e",html,content +1541,85172701,"examples/crowd_code.html",6503,0,"",html,selection_keyboard +1542,85172856,"examples/crowd_code.html",6503,0," ",html,content +1543,85172861,"examples/crowd_code.html",6504,0,"",html,selection_keyboard +1544,85173021,"examples/crowd_code.html",6504,0,"s",html,content +1545,85173025,"examples/crowd_code.html",6505,0,"",html,selection_keyboard +1546,85173030,"examples/crowd_code.html",6505,0,"e",html,content +1547,85173034,"examples/crowd_code.html",6506,0,"",html,selection_keyboard +1548,85173042,"examples/crowd_code.html",6506,0,"r",html,content +1549,85173044,"examples/crowd_code.html",6507,0,"",html,selection_keyboard +1550,85173199,"examples/crowd_code.html",6507,0,"v",html,content +1551,85173202,"examples/crowd_code.html",6508,0,"",html,selection_keyboard +1552,85173341,"examples/crowd_code.html",6508,0,"e",html,content +1553,85173343,"examples/crowd_code.html",6509,0,"",html,selection_keyboard +1554,85173409,"examples/crowd_code.html",6509,0,"r",html,content +1555,85173414,"examples/crowd_code.html",6510,0,"",html,selection_keyboard +1556,85174106,"examples/crowd_code.html",6510,0,"-",html,content +1557,85174113,"examples/crowd_code.html",6511,0,"",html,selection_keyboard +1558,85174487,"examples/crowd_code.html",6511,0,"s",html,content +1559,85174491,"examples/crowd_code.html",6512,0,"",html,selection_keyboard +1560,85174518,"examples/crowd_code.html",6512,0,"i",html,content +1561,85174521,"examples/crowd_code.html",6513,0,"",html,selection_keyboard +1562,85174639,"examples/crowd_code.html",6513,0,"d",html,content +1563,85174642,"examples/crowd_code.html",6514,0,"",html,selection_keyboard +1564,85174739,"examples/crowd_code.html",6514,0,"e",html,content +1565,85174741,"examples/crowd_code.html",6515,0,"",html,selection_keyboard +1566,85174889,"examples/crowd_code.html",6515,0," ",html,content +1567,85174892,"examples/crowd_code.html",6516,0,"",html,selection_keyboard +1568,85175104,"examples/crowd_code.html",6516,0,"c",html,content +1569,85175106,"examples/crowd_code.html",6517,0,"",html,selection_keyboard +1570,85175216,"examples/crowd_code.html",6517,0,"o",html,content +1571,85175218,"examples/crowd_code.html",6518,0,"",html,selection_keyboard +1572,85175488,"examples/crowd_code.html",6517,1,"",html,content +1573,85175637,"examples/crowd_code.html",6516,1,"",html,content +1574,85182393,"examples/crowd_code.html",6516,0,"d",html,content +1575,85182396,"examples/crowd_code.html",6517,0,"",html,selection_keyboard +1576,85183788,"examples/crowd_code.html",6517,0,"a",html,content +1577,85183791,"examples/crowd_code.html",6518,0,"",html,selection_keyboard +1578,85183904,"examples/crowd_code.html",6518,0,"t",html,content +1579,85183905,"examples/crowd_code.html",6519,0,"",html,selection_keyboard +1580,85184170,"examples/crowd_code.html",6519,0,"a",html,content +1581,85184171,"examples/crowd_code.html",6520,0,"",html,selection_keyboard +1582,85184239,"examples/crowd_code.html",6520,0," ",html,content +1583,85184241,"examples/crowd_code.html",6521,0,"",html,selection_keyboard +1584,85184448,"examples/crowd_code.html",6521,0,"c",html,content +1585,85184449,"examples/crowd_code.html",6522,0,"",html,selection_keyboard +1586,85184526,"examples/crowd_code.html",6522,0,"o",html,content +1587,85184527,"examples/crowd_code.html",6523,0,"",html,selection_keyboard +1588,85184743,"examples/crowd_code.html",6523,0,"l",html,content +1589,85184744,"examples/crowd_code.html",6524,0,"",html,selection_keyboard +1590,85184917,"examples/crowd_code.html",6524,0,"l",html,content +1591,85184920,"examples/crowd_code.html",6525,0,"",html,selection_keyboard +1592,85184925,"examples/crowd_code.html",6525,0,"e",html,content +1593,85184926,"examples/crowd_code.html",6526,0,"",html,selection_keyboard +1594,85185042,"examples/crowd_code.html",6526,0,"c",html,content +1595,85185043,"examples/crowd_code.html",6527,0,"",html,selection_keyboard +1596,85185239,"examples/crowd_code.html",6527,0,"t",html,content +1597,85185242,"examples/crowd_code.html",6528,0,"",html,selection_keyboard +1598,85185331,"examples/crowd_code.html",6528,0,"i",html,content +1599,85185335,"examples/crowd_code.html",6529,0,"",html,selection_keyboard +1600,85185366,"examples/crowd_code.html",6529,0,"o",html,content +1601,85185370,"examples/crowd_code.html",6530,0,"",html,selection_keyboard +1602,85185472,"examples/crowd_code.html",6530,0,"n",html,content +1603,85185473,"examples/crowd_code.html",6531,0,"",html,selection_keyboard +1604,85186172,"examples/crowd_code.html",6531,0," ",html,content +1605,85186176,"examples/crowd_code.html",6532,0,"",html,selection_keyboard +1606,85186972,"examples/crowd_code.html",6532,0,"p",html,content +1607,85186977,"examples/crowd_code.html",6533,0,"",html,selection_keyboard +1608,85187031,"examples/crowd_code.html",6533,0,"i",html,content +1609,85187035,"examples/crowd_code.html",6534,0,"",html,selection_keyboard +1610,85187186,"examples/crowd_code.html",6534,0,"p",html,content +1611,85187190,"examples/crowd_code.html",6535,0,"",html,selection_keyboard +1612,85187553,"examples/crowd_code.html",6535,0,"e",html,content +1613,85187558,"examples/crowd_code.html",6536,0,"",html,selection_keyboard +1614,85187973,"examples/crowd_code.html",6536,0,"e",html,content +1615,85187977,"examples/crowd_code.html",6537,0,"",html,selection_keyboard +1616,85188630,"examples/crowd_code.html",6536,0,"",html,selection_command +1617,85193112,"examples/crowd_code.html",6537,0,"",html,selection_command +1618,85193458,"examples/crowd_code.html",6536,1,"",html,content +1619,85193824,"examples/crowd_code.html",6536,0,"l",html,content +1620,85193826,"examples/crowd_code.html",6537,0,"",html,selection_keyboard +1621,85193830,"examples/crowd_code.html",6537,0,"i",html,content +1622,85193834,"examples/crowd_code.html",6538,0,"",html,selection_keyboard +1623,85193939,"examples/crowd_code.html",6538,0,"n",html,content +1624,85193942,"examples/crowd_code.html",6539,0,"",html,selection_keyboard +1625,85194058,"examples/crowd_code.html",6539,0,"e",html,content +1626,85194060,"examples/crowd_code.html",6540,0,"",html,selection_keyboard +1627,85194262,"examples/crowd_code.html",6539,0,"",html,selection_command +1628,85199746,"examples/crowd_code.html",6540,0,"",html,selection_command +1629,85199892,"examples/crowd_code.html",6542,0,"",html,selection_command +1630,85200068,"examples/crowd_code.html",6545,0,"",html,selection_command +1631,85200227,"examples/crowd_code.html",6552,0,"",html,selection_command +1632,85200374,"examples/crowd_code.html",6555,0,"",html,selection_command +1633,85200547,"examples/crowd_code.html",6564,0,"",html,selection_command +1634,85200704,"examples/crowd_code.html",6568,0,"",html,selection_command +1635,85201273,"examples/crowd_code.html",6568,3,"",html,content +1636,85201505,"examples/crowd_code.html",6568,0,"i",html,content +1637,85201506,"examples/crowd_code.html",6569,0,"",html,selection_keyboard +1638,85201621,"examples/crowd_code.html",6569,0,"p",html,content +1639,85201623,"examples/crowd_code.html",6570,0,"",html,selection_keyboard +1640,85201640,"examples/crowd_code.html",6570,0,"m",html,content +1641,85201641,"examples/crowd_code.html",6571,0,"",html,selection_keyboard +1642,85202196,"examples/crowd_code.html",6570,1,"",html,content +1643,85202333,"examples/crowd_code.html",6569,1,"",html,content +1644,85202402,"examples/crowd_code.html",6569,0,"m",html,content +1645,85202404,"examples/crowd_code.html",6570,0,"",html,selection_keyboard +1646,85202483,"examples/crowd_code.html",6570,0,"p",html,content +1647,85202485,"examples/crowd_code.html",6571,0,"",html,selection_keyboard +1648,85202739,"examples/crowd_code.html",6571,0,"l",html,content +1649,85202743,"examples/crowd_code.html",6572,0,"",html,selection_keyboard +1650,85202758,"examples/crowd_code.html",6572,0,"e",html,content +1651,85202759,"examples/crowd_code.html",6573,0,"",html,selection_keyboard +1652,85202892,"examples/crowd_code.html",6573,0,"m",html,content +1653,85202895,"examples/crowd_code.html",6574,0,"",html,selection_keyboard +1654,85202965,"examples/crowd_code.html",6574,0,"e",html,content +1655,85202968,"examples/crowd_code.html",6575,0,"",html,selection_keyboard +1656,85203070,"examples/crowd_code.html",6575,0,"n",html,content +1657,85203074,"examples/crowd_code.html",6576,0,"",html,selection_keyboard +1658,85203149,"examples/crowd_code.html",6576,0,"t",html,content +1659,85203153,"examples/crowd_code.html",6577,0,"",html,selection_keyboard +1660,85205816,"examples/crowd_code.html",6577,0,"ation of the",html,content +1661,85206135,"examples/crowd_code.html",6588,0,"",html,selection_command +1662,85207056,"examples/crowd_code.html",6586,0,"",html,selection_command +1663,85207303,"examples/crowd_code.html",6583,0,"",html,selection_command +1664,85207337,"examples/crowd_code.html",6568,0,"",html,selection_command +1665,85207369,"examples/crowd_code.html",6564,0,"",html,selection_command +1666,85207403,"examples/crowd_code.html",6555,0,"",html,selection_command +1667,85207435,"examples/crowd_code.html",6552,0,"",html,selection_command +1668,85207469,"examples/crowd_code.html",6545,0,"",html,selection_command +1669,85207503,"examples/crowd_code.html",6542,0,"",html,selection_command +1670,85207536,"examples/crowd_code.html",6540,0,"",html,selection_command +1671,85209404,"examples/crowd_code.html",6532,0,"",html,selection_command +1672,85209655,"examples/crowd_code.html",6521,0,"",html,selection_command +1673,85209684,"examples/crowd_code.html",6516,0,"",html,selection_command +1674,85209720,"examples/crowd_code.html",6511,0,"",html,selection_command +1675,85209748,"examples/crowd_code.html",6510,0,"",html,selection_command +1676,85209781,"examples/crowd_code.html",6504,0,"",html,selection_command +1677,85209816,"examples/crowd_code.html",6500,0,"",html,selection_command +1678,85209849,"examples/crowd_code.html",6488,0,"",html,selection_command +1679,85209883,"examples/crowd_code.html",6485,0,"",html,selection_command +1680,85210188,"examples/crowd_code.html",6484,0,"",html,selection_command +1681,85210414,"examples/crowd_code.html",6483,0,"",html,selection_command +1682,85210705,"examples/crowd_code.html",6484,0,"",html,selection_command +1683,85210906,"examples/crowd_code.html",6485,0,"",html,selection_command +1684,85211157,"examples/crowd_code.html",6488,0,"",html,selection_command +1685,85211188,"examples/crowd_code.html",6500,0,"",html,selection_command +1686,85211222,"examples/crowd_code.html",6504,0,"",html,selection_command +1687,85211254,"examples/crowd_code.html",6510,0,"",html,selection_command +1688,85211286,"examples/crowd_code.html",6511,0,"",html,selection_command +1689,85211320,"examples/crowd_code.html",6516,0,"",html,selection_command +1690,85211353,"examples/crowd_code.html",6521,0,"",html,selection_command +1691,85211386,"examples/crowd_code.html",6532,0,"",html,selection_command +1692,85211606,"examples/crowd_code.html",6540,0,"",html,selection_command +1693,85211776,"examples/crowd_code.html",6542,0,"",html,selection_command +1694,85212039,"examples/crowd_code.html",6542,1,"F",html,selection_command +1695,85212122,"examples/crowd_code.html",6542,2,"FS",html,selection_command +1696,85212301,"examples/crowd_code.html",6542,9,"FS worked",html,selection_command +1697,85212461,"examples/crowd_code.html",6542,12,"FS worked on",html,selection_command +1698,85212630,"examples/crowd_code.html",6542,21,"FS worked on ideation",html,selection_command +1699,85212797,"examples/crowd_code.html",6542,25,"FS worked on ideation and",html,selection_command +1700,85212973,"examples/crowd_code.html",6542,40,"FS worked on ideation and implementation",html,selection_command +1701,85213129,"examples/crowd_code.html",6542,43,"FS worked on ideation and implementation of",html,selection_command +1702,85213297,"examples/crowd_code.html",6542,47,"FS worked on ideation and implementation of the",html,selection_command +1703,85213442,"examples/crowd_code.html",6542,51,"FS worked on ideation and implementation of the IDE",html,selection_command +1704,85213638,"examples/crowd_code.html",6542,61,"FS worked on ideation and implementation of the IDE extension",html,selection_command +1705,85213784,"examples/crowd_code.html",6542,62,"FS worked on ideation and implementation of the IDE extension.",html,selection_command +1706,85214227,"examples/crowd_code.html",6542,63,"FS worked on ideation and implementation of the IDE extension. ",html,selection_command +1707,85214353,"examples/crowd_code.html",6542,63,"",html,content +1708,85214690,"examples/crowd_code.html",6540,0,"",html,selection_command +1709,85214891,"examples/crowd_code.html",6532,0,"",html,selection_command +1710,85215028,"examples/crowd_code.html",6521,0,"",html,selection_command +1711,85215203,"examples/crowd_code.html",6516,0,"",html,selection_command +1712,85215377,"examples/crowd_code.html",6511,0,"",html,selection_command +1713,85215554,"examples/crowd_code.html",6510,0,"",html,selection_command +1714,85215705,"examples/crowd_code.html",6504,0,"",html,selection_command +1715,85215857,"examples/crowd_code.html",6500,0,"",html,selection_command +1716,85216022,"examples/crowd_code.html",6488,0,"",html,selection_command +1717,85216148,"examples/crowd_code.html",6485,0,"",html,selection_command +1718,85216321,"examples/crowd_code.html",6484,0,"",html,selection_command +1719,85216789,"examples/crowd_code.html",6483,0,"",html,selection_command +1720,85216810,"examples/crowd_code.html",6484,0,"FS worked on ideation and implementation of the IDE extension. ",html,content +1721,85216813,"examples/crowd_code.html",6546,0,"",html,selection_command +1722,85217637,"examples/crowd_code.html",6484,63,"",html,content +1723,85217645,"examples/crowd_code.html",6483,0,"",html,selection_command +1724,85218556,"examples/crowd_code.html",6484,0,"",html,selection_command +1725,85218653,"examples/crowd_code.html",6485,0,"FS worked on ideation and implementation of the IDE extension. ",html,content +1726,85218655,"examples/crowd_code.html",6547,0,"",html,selection_command +1727,85233938,"examples/crowd_code.html",6548,0,"",html,selection_command +1728,85234188,"examples/crowd_code.html",6551,0,"",html,selection_command +1729,85234221,"examples/crowd_code.html",6563,0,"",html,selection_command +1730,85234253,"examples/crowd_code.html",6567,0,"",html,selection_command +1731,85234286,"examples/crowd_code.html",6573,0,"",html,selection_command +1732,85234319,"examples/crowd_code.html",6574,0,"",html,selection_command +1733,85234357,"examples/crowd_code.html",6579,0,"",html,selection_command +1734,85234388,"examples/crowd_code.html",6584,0,"",html,selection_command +1735,85234419,"examples/crowd_code.html",6595,0,"",html,selection_command +1736,85235120,"examples/crowd_code.html",6584,0,"",html,selection_command +1737,85235336,"examples/crowd_code.html",6579,0,"",html,selection_command +1738,85235495,"examples/crowd_code.html",6574,0,"",html,selection_command +1739,85236054,"examples/crowd_code.html",6573,0,"",html,selection_command +1740,85236305,"examples/crowd_code.html",6567,0,"",html,selection_command +1741,85236337,"examples/crowd_code.html",6563,0,"",html,selection_command +1742,85236369,"examples/crowd_code.html",6551,0,"",html,selection_command +1743,85236399,"examples/crowd_code.html",6548,0,"",html,selection_command +1744,85237288,"examples/crowd_code.html",6547,0,"",html,selection_command +1745,85243342,"examples/crowd_code.html",6546,0,"",html,selection_command +1746,85243592,"examples/crowd_code.html",6537,0,"",html,selection_command +1747,85243621,"examples/crowd_code.html",6533,0,"",html,selection_command +1748,85243655,"examples/crowd_code.html",6529,0,"",html,selection_command +1749,85243686,"examples/crowd_code.html",6526,0,"",html,selection_command +1750,85243720,"examples/crowd_code.html",6511,0,"",html,selection_command +1751,85243753,"examples/crowd_code.html",6507,0,"",html,selection_command +1752,85243786,"examples/crowd_code.html",6498,0,"",html,selection_command +1753,85243820,"examples/crowd_code.html",6495,0,"",html,selection_command +1754,85243853,"examples/crowd_code.html",6488,0,"",html,selection_command +1755,85244499,"examples/crowd_code.html",6485,0,"",html,selection_command +1756,85251860,"examples/crowd_code.html",6488,0,"",html,selection_command +1757,85252109,"examples/crowd_code.html",6495,0,"",html,selection_command +1758,85252140,"examples/crowd_code.html",6498,0,"",html,selection_command +1759,85252173,"examples/crowd_code.html",6507,0,"",html,selection_command +1760,85252206,"examples/crowd_code.html",6511,0,"",html,selection_command +1761,85252240,"examples/crowd_code.html",6526,0,"",html,selection_command +1762,85252274,"examples/crowd_code.html",6529,0,"",html,selection_command +1763,85252702,"examples/crowd_code.html",6533,0,"",html,selection_command +1764,85262977,"examples/crowd_code.html",6529,0,"",html,selection_command +1765,85263171,"examples/crowd_code.html",6526,0,"",html,selection_command +1766,85263390,"examples/crowd_code.html",6511,0,"",html,selection_command +1767,85263581,"examples/crowd_code.html",6507,0,"",html,selection_command +1768,85263692,"examples/crowd_code.html",6498,0,"",html,selection_command +1769,85264048,"examples/crowd_code.html",6505,0,"",html,selection_command +1770,85264328,"examples/crowd_code.html",6506,0,"",html,selection_command +1771,85264454,"examples/crowd_code.html",6506,0," ",html,content +1772,85264461,"examples/crowd_code.html",6507,0,"",html,selection_keyboard +1773,85264571,"examples/crowd_code.html",6507,0,"o",html,content +1774,85264573,"examples/crowd_code.html",6508,0,"",html,selection_keyboard +1775,85264742,"examples/crowd_code.html",6508,0,"f",html,content +1776,85264743,"examples/crowd_code.html",6509,0,"",html,selection_keyboard +1777,85264867,"examples/crowd_code.html",6509,0," ",html,content +1778,85264868,"examples/crowd_code.html",6510,0,"",html,selection_keyboard +1779,85264874,"examples/crowd_code.html",6510,0,"t",html,content +1780,85264875,"examples/crowd_code.html",6511,0,"",html,selection_keyboard +1781,85265019,"examples/crowd_code.html",6511,0,"h",html,content +1782,85265020,"examples/crowd_code.html",6512,0,"",html,selection_keyboard +1783,85265099,"examples/crowd_code.html",6512,0,"e",html,content +1784,85265101,"examples/crowd_code.html",6513,0,"",html,selection_keyboard +1785,85265260,"examples/crowd_code.html",6513,0," ",html,content +1786,85265261,"examples/crowd_code.html",6514,0,"",html,selection_keyboard +1787,85265302,"examples/crowd_code.html",6514,0,"p",html,content +1788,85265303,"examples/crowd_code.html",6515,0,"",html,selection_keyboard +1789,85265373,"examples/crowd_code.html",6515,0,"r",html,content +1790,85265374,"examples/crowd_code.html",6516,0,"",html,selection_keyboard +1791,85265483,"examples/crowd_code.html",6516,0,"o",html,content +1792,85265485,"examples/crowd_code.html",6517,0,"",html,selection_keyboard +1793,85265650,"examples/crowd_code.html",6517,0,"j",html,content +1794,85265652,"examples/crowd_code.html",6518,0,"",html,selection_keyboard +1795,85265656,"examples/crowd_code.html",6518,0,"e",html,content +1796,85265657,"examples/crowd_code.html",6519,0,"",html,selection_keyboard +1797,85265767,"examples/crowd_code.html",6519,0,"c",html,content +1798,85265770,"examples/crowd_code.html",6520,0,"",html,selection_keyboard +1799,85265937,"examples/crowd_code.html",6520,0,"t",html,content +1800,85265938,"examples/crowd_code.html",6521,0,"",html,selection_keyboard +1801,85266086,"examples/crowd_code.html",6521,0," ",html,content +1802,85266088,"examples/crowd_code.html",6522,0,"",html,selection_keyboard +1803,85266221,"examples/crowd_code.html",6522,0,"a",html,content +1804,85266223,"examples/crowd_code.html",6523,0,"",html,selection_keyboard +1805,85266241,"examples/crowd_code.html",6523,0,"n",html,content +1806,85266242,"examples/crowd_code.html",6524,0,"",html,selection_keyboard +1807,85266387,"examples/crowd_code.html",6524,0,"d",html,content +1808,85266388,"examples/crowd_code.html",6525,0,"",html,selection_keyboard +1809,85266688,"examples/crowd_code.html",6524,1,"",html,content +1810,85266830,"examples/crowd_code.html",6523,1,"",html,content +1811,85266973,"examples/crowd_code.html",6522,1,"",html,content +1812,85267125,"examples/crowd_code.html",6521,1,"",html,content +1813,85267286,"examples/crowd_code.html",6520,0,"",html,selection_command +1814,85270852,"examples/crowd_code.html",6522,0,"",html,selection_command +1815,85271104,"examples/crowd_code.html",6526,0,"",html,selection_command +1816,85271135,"examples/crowd_code.html",6541,0,"",html,selection_command +1817,85271165,"examples/crowd_code.html",6544,0,"",html,selection_command +1818,85271371,"examples/crowd_code.html",6548,0,"",html,selection_command +1819,85272172,"examples/crowd_code.html",6544,0,"",html,selection_command +1820,85272314,"examples/crowd_code.html",6541,0,"",html,selection_command +1821,85272500,"examples/crowd_code.html",6526,0,"",html,selection_command +1822,85272858,"examples/crowd_code.html",6526,1,"i",html,selection_command +1823,85272920,"examples/crowd_code.html",6526,16,"implementation o",html,selection_command +1824,85274304,"examples/crowd_code.html",6526,15,"implementation ",html,selection_command +1825,85274504,"examples/crowd_code.html",6526,15,"",html,content +1826,85274902,"examples/crowd_code.html",6526,0,"i",html,content +1827,85274905,"examples/crowd_code.html",6527,0,"",html,selection_keyboard +1828,85275002,"examples/crowd_code.html",6527,0,"m",html,content +1829,85275003,"examples/crowd_code.html",6528,0,"",html,selection_keyboard +1830,85275036,"examples/crowd_code.html",6528,0,"p",html,content +1831,85275037,"examples/crowd_code.html",6529,0,"",html,selection_keyboard +1832,85275737,"examples/crowd_code.html",6529,0,"l",html,content +1833,85275739,"examples/crowd_code.html",6530,0,"",html,selection_keyboard +1834,85276293,"examples/crowd_code.html",6530,0,"e",html,content +1835,85276296,"examples/crowd_code.html",6531,0,"",html,selection_keyboard +1836,85276385,"examples/crowd_code.html",6531,0,"m",html,content +1837,85276387,"examples/crowd_code.html",6532,0,"",html,selection_keyboard +1838,85276495,"examples/crowd_code.html",6532,0,"e",html,content +1839,85276496,"examples/crowd_code.html",6533,0,"",html,selection_keyboard +1840,85276558,"examples/crowd_code.html",6533,0,"n",html,content +1841,85276560,"examples/crowd_code.html",6534,0,"",html,selection_keyboard +1842,85276648,"examples/crowd_code.html",6534,0,"t",html,content +1843,85276650,"examples/crowd_code.html",6535,0,"",html,selection_keyboard +1844,85276722,"examples/crowd_code.html",6535,0,"e",html,content +1845,85276723,"examples/crowd_code.html",6536,0,"",html,selection_keyboard +1846,85276974,"examples/crowd_code.html",6536,0,"d",html,content +1847,85276976,"examples/crowd_code.html",6537,0,"",html,selection_keyboard +1848,85277036,"examples/crowd_code.html",6537,0," ",html,content +1849,85277037,"examples/crowd_code.html",6538,0,"",html,selection_keyboard +1850,85277913,"examples/crowd_code.html",6537,0,"",html,selection_command +1851,85278142,"examples/crowd_code.html",6538,0,"",html,selection_command +1852,85278709,"examples/crowd_code.html",6538,3,"",html,content +1853,85280726,"examples/crowd_code.html",6526,0,"",html,selection_command +1854,85280972,"examples/crowd_code.html",6522,0,"",html,selection_command +1855,85281004,"examples/crowd_code.html",6514,0,"",html,selection_command +1856,85281037,"examples/crowd_code.html",6510,0,"",html,selection_command +1857,85281069,"examples/crowd_code.html",6507,0,"",html,selection_command +1858,85281104,"examples/crowd_code.html",6498,0,"",html,selection_command +1859,85281136,"examples/crowd_code.html",6495,0,"",html,selection_command +1860,85281170,"examples/crowd_code.html",6488,0,"",html,selection_command +1861,85281391,"examples/crowd_code.html",6485,0,"",html,selection_command +1862,85282987,"examples/crowd_code.html",6488,0,"",html,selection_command +1863,85283238,"examples/crowd_code.html",6495,0,"",html,selection_command +1864,85283271,"examples/crowd_code.html",6498,0,"",html,selection_command +1865,85283303,"examples/crowd_code.html",6507,0,"",html,selection_command +1866,85283331,"examples/crowd_code.html",6510,0,"",html,selection_command +1867,85283364,"examples/crowd_code.html",6514,0,"",html,selection_command +1868,85283398,"examples/crowd_code.html",6522,0,"",html,selection_command +1869,85283431,"examples/crowd_code.html",6526,0,"",html,selection_command +1870,85283464,"examples/crowd_code.html",6538,0,"",html,selection_command +1871,85283496,"examples/crowd_code.html",6542,0,"",html,selection_command +1872,85283530,"examples/crowd_code.html",6546,0,"",html,selection_command +1873,85283922,"examples/crowd_code.html",6555,0,"",html,selection_command +1874,85284081,"examples/crowd_code.html",6557,0,"",html,selection_command +1875,85284271,"examples/crowd_code.html",6560,0,"",html,selection_command +1876,85284750,"examples/crowd_code.html",6572,0,"",html,selection_command +1877,85285156,"examples/crowd_code.html",6576,0,"",html,selection_command +1878,85285396,"examples/crowd_code.html",6582,0,"",html,selection_command +1879,85285432,"examples/crowd_code.html",6583,0,"",html,selection_command +1880,85285465,"examples/crowd_code.html",6588,0,"",html,selection_command +1881,85285499,"examples/crowd_code.html",6593,0,"",html,selection_command +1882,85285740,"examples/crowd_code.html",6604,0,"",html,selection_command +1883,85285929,"examples/crowd_code.html",6612,0,"",html,selection_command +1884,85286112,"examples/crowd_code.html",6614,0,"",html,selection_command +1885,85286365,"examples/crowd_code.html",6617,0,"",html,selection_command +1886,85286392,"examples/crowd_code.html",6623,0,"",html,selection_command +1887,85286424,"examples/crowd_code.html",6627,0,"",html,selection_command +1888,85286470,"examples/crowd_code.html",6637,0,"",html,selection_command +1889,85288257,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +1890,85288662,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +1891,85288912,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +1892,85288913,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +1893,85288937,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +1894,85290078,"TERMINAL",0,0,"npm run dev",,terminal_command +1895,85290130,"TERMINAL",0,0,"]633;C",,terminal_output +1896,85290264,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +1897,85290434,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +1898,85290816,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 383ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +1899,85291320,"TERMINAL",0,0,"created dist/transforms.v2.js in 502ms\r\n\r\n[2025-09-03 16:53:22] waiting for changes...\r\n",,terminal_output +1900,85312356,"examples/crowd_code.html",6638,0,"",html,selection_command +1901,85313635,"examples/crowd_code.html",6637,0,"",html,selection_command +1902,85313835,"examples/crowd_code.html",6627,0,"",html,selection_command +1903,85314083,"examples/crowd_code.html",6623,0,"",html,selection_command +1904,85314117,"examples/crowd_code.html",6617,0,"",html,selection_command +1905,85314385,"examples/crowd_code.html",6614,0,"",html,selection_command +1906,85315936,"examples/crowd_code.html",6614,0,"A",html,content +1907,85315940,"examples/crowd_code.html",6615,0,"",html,selection_keyboard +1908,85316119,"examples/crowd_code.html",6615,0,"l",html,content +1909,85316121,"examples/crowd_code.html",6616,0,"",html,selection_keyboard +1910,85316278,"examples/crowd_code.html",6616,0,"l",html,content +1911,85316281,"examples/crowd_code.html",6617,0,"",html,selection_keyboard +1912,85316393,"examples/crowd_code.html",6617,0," ",html,content +1913,85316394,"examples/crowd_code.html",6618,0,"",html,selection_keyboard +1914,85316550,"examples/crowd_code.html",6618,0,"a",html,content +1915,85316552,"examples/crowd_code.html",6619,0,"",html,selection_keyboard +1916,85316568,"examples/crowd_code.html",6619,0,"u",html,content +1917,85316569,"examples/crowd_code.html",6620,0,"",html,selection_keyboard +1918,85316650,"examples/crowd_code.html",6620,0,"t",html,content +1919,85316651,"examples/crowd_code.html",6621,0,"",html,selection_keyboard +1920,85316733,"examples/crowd_code.html",6621,0,"h",html,content +1921,85316736,"examples/crowd_code.html",6622,0,"",html,selection_keyboard +1922,85316818,"examples/crowd_code.html",6622,0,"o",html,content +1923,85316822,"examples/crowd_code.html",6623,0,"",html,selection_keyboard +1924,85316919,"examples/crowd_code.html",6623,0,"r",html,content +1925,85316921,"examples/crowd_code.html",6624,0,"",html,selection_keyboard +1926,85317069,"examples/crowd_code.html",6624,0,"s",html,content +1927,85317071,"examples/crowd_code.html",6625,0,"",html,selection_keyboard +1928,85317170,"examples/crowd_code.html",6625,0," ",html,content +1929,85317171,"examples/crowd_code.html",6626,0,"",html,selection_keyboard +1930,85317501,"examples/crowd_code.html",6626,0,"c",html,content +1931,85317503,"examples/crowd_code.html",6627,0,"",html,selection_keyboard +1932,85317625,"examples/crowd_code.html",6627,0,"o",html,content +1933,85317627,"examples/crowd_code.html",6628,0,"",html,selection_keyboard +1934,85317675,"examples/crowd_code.html",6628,0,"n",html,content +1935,85317678,"examples/crowd_code.html",6629,0,"",html,selection_keyboard +1936,85318388,"examples/crowd_code.html",6629,0,"t",html,content +1937,85318391,"examples/crowd_code.html",6630,0,"",html,selection_keyboard +1938,85318552,"examples/crowd_code.html",6630,0,"r",html,content +1939,85318554,"examples/crowd_code.html",6631,0,"",html,selection_keyboard +1940,85318935,"examples/crowd_code.html",6631,0,"i",html,content +1941,85318938,"examples/crowd_code.html",6632,0,"",html,selection_keyboard +1942,85319001,"examples/crowd_code.html",6632,0,"b",html,content +1943,85319003,"examples/crowd_code.html",6633,0,"",html,selection_keyboard +1944,85319202,"examples/crowd_code.html",6633,0,"t",html,content +1945,85319204,"examples/crowd_code.html",6634,0,"",html,selection_keyboard +1946,85319218,"examples/crowd_code.html",6634,0,"u",html,content +1947,85319220,"examples/crowd_code.html",6635,0,"",html,selection_keyboard +1948,85319420,"examples/crowd_code.html",6635,0,"e",html,content +1949,85319421,"examples/crowd_code.html",6636,0,"",html,selection_keyboard +1950,85319706,"examples/crowd_code.html",6636,0,"d",html,content +1951,85319707,"examples/crowd_code.html",6637,0,"",html,selection_keyboard +1952,85319750,"examples/crowd_code.html",6637,0," ",html,content +1953,85319752,"examples/crowd_code.html",6638,0,"",html,selection_keyboard +1954,85319971,"examples/crowd_code.html",6637,1,"",html,content +1955,85320120,"examples/crowd_code.html",6636,1,"",html,content +1956,85320254,"examples/crowd_code.html",6635,1,"",html,content +1957,85320375,"examples/crowd_code.html",6634,1,"",html,content +1958,85320520,"examples/crowd_code.html",6633,1,"",html,content +1959,85320602,"examples/crowd_code.html",6633,0,"u",html,content +1960,85320604,"examples/crowd_code.html",6634,0,"",html,selection_keyboard +1961,85320667,"examples/crowd_code.html",6634,0,"t",html,content +1962,85320668,"examples/crowd_code.html",6635,0,"",html,selection_keyboard +1963,85320736,"examples/crowd_code.html",6635,0,"e",html,content +1964,85320737,"examples/crowd_code.html",6636,0,"",html,selection_keyboard +1965,85320993,"examples/crowd_code.html",6636,0,"d",html,content +1966,85320995,"examples/crowd_code.html",6637,0,"",html,selection_keyboard +1967,85321093,"examples/crowd_code.html",6637,0," ",html,content +1968,85321094,"examples/crowd_code.html",6638,0,"",html,selection_keyboard +1969,85321534,"examples/crowd_code.html",6638,0,"t",html,content +1970,85321535,"examples/crowd_code.html",6639,0,"",html,selection_keyboard +1971,85321606,"examples/crowd_code.html",6639,0,"e",html,content +1972,85321608,"examples/crowd_code.html",6640,0,"",html,selection_keyboard +1973,85321756,"examples/crowd_code.html",6640,0,"c",html,content +1974,85321758,"examples/crowd_code.html",6641,0,"",html,selection_keyboard +1975,85321832,"examples/crowd_code.html",6641,0,"h",html,content +1976,85321837,"examples/crowd_code.html",6642,0,"",html,selection_keyboard +1977,85322004,"examples/crowd_code.html",6642,0,"n",html,content +1978,85322005,"examples/crowd_code.html",6643,0,"",html,selection_keyboard +1979,85322173,"examples/crowd_code.html",6643,0,"i",html,content +1980,85322175,"examples/crowd_code.html",6644,0,"",html,selection_keyboard +1981,85322287,"examples/crowd_code.html",6644,0,"c",html,content +1982,85322290,"examples/crowd_code.html",6645,0,"",html,selection_keyboard +1983,85322376,"examples/crowd_code.html",6645,0,"q",html,content +1984,85322377,"examples/crowd_code.html",6646,0,"",html,selection_keyboard +1985,85322438,"examples/crowd_code.html",6646,0,"a",html,content +1986,85322441,"examples/crowd_code.html",6647,0,"",html,selection_keyboard +1987,85322507,"examples/crowd_code.html",6647,0,"l",html,content +1988,85322509,"examples/crowd_code.html",6648,0,"",html,selection_keyboard +1989,85322916,"examples/crowd_code.html",6647,1,"",html,content +1990,85323083,"examples/crowd_code.html",6646,1,"",html,content +1991,85323216,"examples/crowd_code.html",6645,1,"",html,content +1992,85323453,"examples/crowd_code.html",6645,0,"a",html,content +1993,85323454,"examples/crowd_code.html",6646,0,"",html,selection_keyboard +1994,85323517,"examples/crowd_code.html",6646,0,"l",html,content +1995,85323521,"examples/crowd_code.html",6647,0,"",html,selection_keyboard +1996,85323700,"examples/crowd_code.html",6647,0,"l",html,content +1997,85323702,"examples/crowd_code.html",6648,0,"",html,selection_keyboard +1998,85323715,"examples/crowd_code.html",6648,0,"y",html,content +1999,85323716,"examples/crowd_code.html",6649,0,"",html,selection_keyboard +2000,85323970,"examples/crowd_code.html",6649,0,".",html,content +2001,85323973,"examples/crowd_code.html",6650,0,"",html,selection_keyboard +2002,85324074,"examples/crowd_code.html",6650,0," ",html,content +2003,85324075,"examples/crowd_code.html",6651,0,"",html,selection_keyboard +2004,85324357,"examples/crowd_code.html",6650,0,"",html,selection_command +2005,85326291,"examples/crowd_code.html",6649,0,"",html,selection_command +2006,85326484,"examples/crowd_code.html",6638,0,"",html,selection_command +2007,85346305,"examples/crowd_code.html",6626,0,"",html,selection_command +2008,85346554,"examples/crowd_code.html",6618,0,"",html,selection_command +2009,85346586,"examples/crowd_code.html",6614,0,"",html,selection_command +2010,85346619,"examples/crowd_code.html",6612,0,"",html,selection_command +2011,85346654,"examples/crowd_code.html",6604,0,"",html,selection_command +2012,85346698,"examples/crowd_code.html",6593,0,"",html,selection_command +2013,85346719,"examples/crowd_code.html",6588,0,"",html,selection_command +2014,85346753,"examples/crowd_code.html",6583,0,"",html,selection_command +2015,85346786,"examples/crowd_code.html",6582,0,"",html,selection_command +2016,85346819,"examples/crowd_code.html",6576,0,"",html,selection_command +2017,85346853,"examples/crowd_code.html",6572,0,"",html,selection_command +2018,85346886,"examples/crowd_code.html",6560,0,"",html,selection_command +2019,85346920,"examples/crowd_code.html",6557,0,"",html,selection_command +2020,85346956,"examples/crowd_code.html",6555,0,"",html,selection_command +2021,85346986,"examples/crowd_code.html",6546,0,"",html,selection_command +2022,85347020,"examples/crowd_code.html",6542,0,"",html,selection_command +2023,85347053,"examples/crowd_code.html",6538,0,"",html,selection_command +2024,85347086,"examples/crowd_code.html",6526,0,"",html,selection_command +2025,85347120,"examples/crowd_code.html",6522,0,"",html,selection_command +2026,85347261,"examples/crowd_code.html",6514,0,"",html,selection_command +2027,85347438,"examples/crowd_code.html",6510,0,"",html,selection_command +2028,85347547,"examples/crowd_code.html",6507,0,"",html,selection_command +2029,85347780,"examples/crowd_code.html",6498,0,"",html,selection_command +2030,85348380,"examples/crowd_code.html",6498,8,"",html,content +2031,85348653,"examples/crowd_code.html",6498,0,"c",html,content +2032,85348655,"examples/crowd_code.html",6499,0,"",html,selection_keyboard +2033,85348806,"examples/crowd_code.html",6499,0,"o",html,content +2034,85348807,"examples/crowd_code.html",6500,0,"",html,selection_keyboard +2035,85348900,"examples/crowd_code.html",6500,0,"n",html,content +2036,85348902,"examples/crowd_code.html",6501,0,"",html,selection_keyboard +2037,85349049,"examples/crowd_code.html",6501,0,"c",html,content +2038,85349051,"examples/crowd_code.html",6502,0,"",html,selection_keyboard +2039,85349251,"examples/crowd_code.html",6502,0,"e",html,content +2040,85349253,"examples/crowd_code.html",6503,0,"",html,selection_keyboard +2041,85349343,"examples/crowd_code.html",6503,0,"p",html,content +2042,85349346,"examples/crowd_code.html",6504,0,"",html,selection_keyboard +2043,85349439,"examples/crowd_code.html",6504,0,"t",html,content +2044,85349442,"examples/crowd_code.html",6505,0,"",html,selection_keyboard +2045,85349549,"examples/crowd_code.html",6505,0,"u",html,content +2046,85349550,"examples/crowd_code.html",6506,0,"",html,selection_keyboard +2047,85349721,"examples/crowd_code.html",6506,0,"a",html,content +2048,85349723,"examples/crowd_code.html",6507,0,"",html,selection_keyboard +2049,85349821,"examples/crowd_code.html",6507,0,"l",html,content +2050,85349822,"examples/crowd_code.html",6508,0,"",html,selection_keyboard +2051,85350074,"examples/crowd_code.html",6508,0,"i",html,content +2052,85350076,"examples/crowd_code.html",6509,0,"",html,selection_keyboard +2053,85350396,"examples/crowd_code.html",6509,0,"z",html,content +2054,85350401,"examples/crowd_code.html",6510,0,"",html,selection_keyboard +2055,85350707,"examples/crowd_code.html",6510,0,"a",html,content +2056,85350711,"examples/crowd_code.html",6511,0,"",html,selection_keyboard +2057,85350718,"examples/crowd_code.html",6511,0,"t",html,content +2058,85350722,"examples/crowd_code.html",6512,0,"",html,selection_keyboard +2059,85350814,"examples/crowd_code.html",6512,0,"i",html,content +2060,85350816,"examples/crowd_code.html",6513,0,"",html,selection_keyboard +2061,85350847,"examples/crowd_code.html",6513,0,"o",html,content +2062,85350849,"examples/crowd_code.html",6514,0,"",html,selection_keyboard +2063,85350905,"examples/crowd_code.html",6514,0,"n",html,content +2064,85350906,"examples/crowd_code.html",6515,0,"",html,selection_keyboard +2065,85351140,"examples/crowd_code.html",6514,0,"",html,selection_command +2066,85390556,"examples/crowd_code.html",6478,290,"

FS worked on conceptualization of the project and implemented the IDE extension. MM implemented the server-side data collection pipeline. All authors contributed technically. FS wrote the manuscript. We thank Gemini Code Assist and Cursor for their help in writing the extension.

",html,selection_command +2067,85390788,"examples/crowd_code.html",6478,351,"

FS worked on conceptualization of the project and implemented the IDE extension. MM implemented the server-side data collection pipeline. All authors contributed technically. FS wrote the manuscript. We thank Gemini Code Assist and Cursor for their help in writing the extension.

\n ",html,selection_command +2068,85391189,"examples/crowd_code.html",6805,0,"",html,selection_command +2069,85392661,"examples/crowd_code.html",6514,0,"",html,selection_command +2070,85393509,"examples/crowd_code.html",6478,290,"

FS worked on conceptualization of the project and implemented the IDE extension. MM implemented the server-side data collection pipeline. All authors contributed technically. FS wrote the manuscript. We thank Gemini Code Assist and Cursor for their help in writing the extension.

",html,selection_command +2071,85394087,"examples/crowd_code.html",6514,0,"",html,selection_command +2072,85488535,"examples/act.html",0,0,"",html,tab +2073,85488882,"examples/act.html",6720,0,"",html,selection_command +2074,85489455,"examples/act.html",6719,0,"",html,selection_command +2075,85489702,"examples/act.html",6683,0,"",html,selection_command +2076,85489736,"examples/act.html",6682,0,"",html,selection_command +2077,85489769,"examples/act.html",6666,0,"",html,selection_command +2078,85489802,"examples/act.html",6642,0,"",html,selection_command +2079,85489836,"examples/act.html",6619,0,"",html,selection_command +2080,85489869,"examples/act.html",6558,0,"",html,selection_command +2081,85490186,"examples/act.html",6486,0,"",html,selection_command +2082,85490624,"examples/act.html",6490,0,"",html,selection_command +2083,85490774,"examples/act.html",6491,0,"",html,selection_command +2084,85490990,"examples/act.html",6492,0,"",html,selection_command +2085,85491355,"examples/act.html",6494,0,"",html,selection_command +2086,85491679,"examples/act.html",6495,0,"",html,selection_command +2087,85491928,"examples/act.html",6495,0," ",html,content +2088,85491932,"examples/act.html",6496,0,"",html,selection_keyboard +2089,85492052,"examples/act.html",6496,0,"a",html,content +2090,85492054,"examples/act.html",6497,0,"",html,selection_keyboard +2091,85492120,"examples/act.html",6497,0,"n",html,content +2092,85492122,"examples/act.html",6498,0,"",html,selection_keyboard +2093,85492276,"examples/act.html",6498,0,"d",html,content +2094,85492277,"examples/act.html",6499,0,"",html,selection_keyboard +2095,85492346,"examples/act.html",6499,0," ",html,content +2096,85492348,"examples/act.html",6500,0,"",html,selection_keyboard +2097,85492580,"examples/act.html",6500,0,"F",html,content +2098,85492582,"examples/act.html",6501,0,"",html,selection_keyboard +2099,85492616,"examples/act.html",6501,0,"S",html,content +2100,85492618,"examples/act.html",6502,0,"",html,selection_keyboard +2101,85492958,"examples/act.html",6501,0,"",html,selection_command +2102,85494882,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2103,85495348,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +2104,85495604,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2105,85495605,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2106,85495621,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2107,85496159,"TERMINAL",0,0,"npm run dev",,terminal_command +2108,85496211,"TERMINAL",0,0,"]633;C",,terminal_output +2109,85496340,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2110,85496518,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2111,85496920,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 402ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2112,85497383,"TERMINAL",0,0,"created dist/transforms.v2.js in 462ms\r\n\r\n[2025-09-03 16:56:48] waiting for changes...\r\n",,terminal_output +2113,85601568,"examples/jax_assert.html",0,0,"",html,tab +2114,85602874,"examples/act.html",0,0,"",html,tab +2115,85610643,"examples/ppo_off_policy.html",0,0,"",html,tab +2116,85611024,"examples/ppo_off_policy.html",8554,0,"",html,selection_command +2117,85611536,"examples/ppo_off_policy.html",8546,0,"",html,selection_command +2118,85611782,"examples/ppo_off_policy.html",8545,0,"",html,selection_command +2119,85611820,"examples/ppo_off_policy.html",8509,0,"",html,selection_command +2120,85611851,"examples/ppo_off_policy.html",8508,0,"",html,selection_command +2121,85611882,"examples/ppo_off_policy.html",8492,0,"",html,selection_command +2122,85611915,"examples/ppo_off_policy.html",8468,0,"",html,selection_command +2123,85611948,"examples/ppo_off_policy.html",8445,0,"",html,selection_command +2124,85611982,"examples/ppo_off_policy.html",8384,0,"",html,selection_command +2125,85612017,"examples/ppo_off_policy.html",8254,0,"",html,selection_command +2126,85616748,"examples/ppo.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

\n When using PPO in LLM post-training, hyperparameter settings turn Generalized Advantage Estimation into Monte Carlo Advantage Estimation.\n

\n \n \n \n 1\n

\n Although PPO originally uses Generalized Advantage Estimation ,\n modern LLM post-training usually employs $$\lambda=1$$, $$\gamma=1$$, which means that we do not introduce bias to reduce variance\n and do not discount future rewards. Disabling discounting is natural in the post-training regime, since the trajectories\n are not very long.\n In post-training, we want the policy updates to be as stable as possible. However, we also want them to be as\n correct as possible. Since there are other ways of reducing variance (e.g. increasing batch size), we hold off\n from increasing bias.\n

\n\n

\n The original authors define generalized advantage as\n $$\hat{A}_t^{GAE(\gamma,\lambda)} := \sum_{l=0}^{\infty}(\gamma \lambda)^l \delta^V_{t+l}$$.\n\n They already shows that in the setting $$\lambda=1$$,\n the advantage estimation reduces to $$\hat{A}_t := \sum_{l=0}^{\infty}\gamma^l \delta_{t+l} = \sum_{l=0}^{\infty}\gamma^l r_{t+l} - V(s_t)$$.\n\n Additionally setting $$\gamma=1$$ reduces this to $$\hat{A}_t := \sum_{l=0}^{\infty}r_{t+l} - V(s_t)$$, the empirical returns minus a value function baseline.\n\n Since in post-training we only get sparse rewards at the end of the trajectory, the advantage becomes $$\hat{A}_t := r_{T} - V(s_t)$$.\n

\n

\n This is simply Monte Carlo Advantage Estimation. Thus in the LLM post-training setting, the difference between REINFORCE with baseline\n and PPO boils down to PPO's clipping and the use of a likelihood ratio:\n

\n
    \n
  • \n $$\mathcal{J}^{\text{CLIP}}=\mathbb{E}[\text{min}(\text{ratio}_t(\theta)(r_T-V(s_t)), \text{clip}(\text{ratio}_t(\theta), 1-\epsilon, \epsilon)(r_T-V(s_t)))]$$\n , where $$\text{ratio}_t=\frac{\pi_\theta(a_t|s_t)}{\pi_{\theta_{\text{old}}}(a_t|s_t)}$$.\n
    • \n
  • \n $$\mathcal{J}^{\text{REINFORCE}}=\mathbb{E}[\log \pi_\theta(a_t|s_t)(r_T-V(s_t))]$$\n
    • \n
\n \n\n \n\n

Contributions

\n

FS worked on all aspects of this post, including research, analysis and writing.

\n \n \n \n \n\n \n\n\n",html,tab +2127,85617204,"examples/ppo.html",4290,0,"",html,selection_command +2128,85617801,"examples/ppo.html",4282,0,"",html,selection_command +2129,85618056,"examples/ppo.html",4281,0,"",html,selection_command +2130,85618086,"examples/ppo.html",4245,0,"",html,selection_command +2131,85618119,"examples/ppo.html",4244,0,"",html,selection_command +2132,85618153,"examples/ppo.html",4228,0,"",html,selection_command +2133,85618186,"examples/ppo.html",4204,0,"",html,selection_command +2134,85618220,"examples/ppo.html",4181,0,"",html,selection_command +2135,85618369,"examples/ppo.html",4120,0,"",html,selection_command +2136,85618704,"examples/ppo.html",4028,0,"",html,selection_command +2137,85618959,"examples/ppo.html",4032,0,"",html,selection_command +2138,85619122,"examples/ppo.html",4033,0,"",html,selection_command +2139,85619292,"examples/ppo.html",4034,0,"",html,selection_command +2140,85619541,"examples/ppo.html",4034,1,">",html,selection_command +2141,85620241,"examples/ppo.html",4034,0,"",html,selection_command +2142,85620608,"examples/ppo.html",4035,0,"",html,selection_command +2143,85620706,"examples/ppo.html",4035,1,"F",html,selection_command +2144,85620811,"examples/ppo.html",4035,2,"FS",html,selection_command +2145,85620989,"examples/ppo.html",4035,9,"FS worked",html,selection_command +2146,85621239,"examples/ppo.html",4035,12,"FS worked on",html,selection_command +2147,85621287,"examples/ppo.html",4035,16,"FS worked on all",html,selection_command +2148,85621306,"examples/ppo.html",4035,24,"FS worked on all aspects",html,selection_command +2149,85621331,"examples/ppo.html",4035,27,"FS worked on all aspects of",html,selection_command +2150,85621366,"examples/ppo.html",4035,32,"FS worked on all aspects of this",html,selection_command +2151,85621400,"examples/ppo.html",4035,37,"FS worked on all aspects of this post",html,selection_command +2152,85621434,"examples/ppo.html",4035,38,"FS worked on all aspects of this post,",html,selection_command +2153,85621469,"examples/ppo.html",4035,48,"FS worked on all aspects of this post, including",html,selection_command +2154,85621502,"examples/ppo.html",4035,57,"FS worked on all aspects of this post, including research",html,selection_command +2155,85621536,"examples/ppo.html",4035,58,"FS worked on all aspects of this post, including research,",html,selection_command +2156,85621674,"examples/ppo.html",4035,67,"FS worked on all aspects of this post, including research, analysis",html,selection_command +2157,85621855,"examples/ppo.html",4035,71,"FS worked on all aspects of this post, including research, analysis and",html,selection_command +2158,85622008,"examples/ppo.html",4035,79,"FS worked on all aspects of this post, including research, analysis and writing",html,selection_command +2159,85622159,"examples/ppo.html",4035,82,"FS worked on all aspects of this post, including research, analysis and writing. distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2206,85636386,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2207,85636756,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 371ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2208,85637249,"TERMINAL",0,0,"created dist/transforms.v2.js in 492ms\r\n\r\n[2025-09-03 16:59:07] waiting for changes...\r\n",,terminal_output +2209,85697086,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2210,85704677,"examples/jax_assert.html",0,0,"",html,selection_command +2211,85706535,"dist/act.html",0,0,"\n\n\n\n \n \n \n \n\n\n\n \n \n \n \n \n

Large language models exhibit remarkable reasoning capabilities with scale. However, a fundamental flaw of current-generation transformer-based language models is their uniform allocation of compute per token.\n

\n \n \n \n 1\n

\n A vanilla transformer (and all currently used variations thereof) generates each token after one (and only one) forward-pass through the network. Intuitively this means that at inference time, the network thinks for the same amount of time before each token generation. While this is not an inherent limitation on the reasoning capabilities of transformer models, it does mean that we would need obscenely large transformers for them to exhibit longer-term planning capabilities. Since we posit that (standalone) models with more than 1 trillion parameters are already unfeasible to serve at scale, we need to overcome this limitation of the transformer architecture.\n

\n

\n

\n While we have been aware of this limitation and of its severity for a long time (much longer than p(doom) even existed for), today it is an open secret in the ML research community. Everyone knows that all big AGI labs (DeepMind, OpenAI, imbue, to name a few) are rushing towards overcoming this limitation , yet nobody has found a solution yet. The market value of an LLM/LMM with agentic reasoning capabilities is difficult to imagine.\n

\n

\n Most public discourse around scaling up reasoning capabilities involves developing methods to shift computational resources from training to inference time: While only a tiny fraction of computational resources of LLMs are used for inference (>99% of compute goes to pretraining), systems like Pluribus , AlphaGo or MuZero leverage vast amounts of compute at inference time, primarily via Monte Carlo Tree Search. A few techniques for shifting compute to inference time already exist, albeit rather simple ones: chain-of-thought , tree-of-thought , sampling+scoring . All of these techniques try to shift computation post-hoc. While only helps if the correct solution is actually in the sampled space, and seem unelegant in execution, somewhat arbitrary as solutions and above all, have not been able to yield agentic systems.\n

\n

\n We posit that one elegant way of addressing the transformer's limitation is via adaptive computation. We want to find a way of letting the model think for as long as it deems necessary before each token generation. That way, we do not hard-code a 'reasoning architecture' into the model and we do not materialize the search. The model's thought process stays in the latent space. We posit that such a model trained at GPT-4 scale will exhibit human-level reasoning.\n

\n \n \n\n \n\n

Contributions

\n

MM and FS worked on research and analysis, FS wrote the manuscript.

\n \n \n \n \n\n \n\n",html,tab +2212,85707416,"examples/jax_assert.html",0,0,"",html,tab +2213,85709057,"examples/act.html",0,0,"",html,tab +2214,85709872,"examples/act.html",0,0,"",html,selection_command +2215,85743133,"examples/act.html",594,0,"",html,selection_mouse +2216,85744149,"TERMINAL",0,0,"\r\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B",,terminal_output +2217,85744852,"TERMINAL",0,0,"\r\r\nbck-i-search: _",,terminal_output +2218,85745160,"TERMINAL",0,0,"cp examples/* dists_",,terminal_output +2219,85745217,"TERMINAL",0,0,"ssh franz.srambical@hpc-submit02.scidom.deu_",,terminal_output +2220,85745292,"TERMINAL",0,0,"ubb_",,terminal_output +2221,85745516,"TERMINAL",0,0,"git subtree push --prefix dist origin gh-pagest_",,terminal_output +2222,85745713,"TERMINAL",0,0,"trr_",,terminal_output +2223,85745796,"TERMINAL",0,0,"ree_",,terminal_output +2224,85745995,"TERMINAL",0,0,"eee_",,terminal_output +2225,85747521,"TERMINAL",0,0,"git subtree push --prefix dist origin gh-pages",,terminal_command +2226,85747521,"TERMINAL",0,0,"subtree\r[?2004l\r]633;E;git subtree push --prefix dist origin gh-pages;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2227,85747571,"TERMINAL",0,0,"]633;C",,terminal_output +2228,85752519,"TERMINAL",0,0,"git push using: origin gh-pages\r\n1/ 441 (0) [0]\r2/ 441 (1) [0]\r3/ 441 (2) [0]\r4/ 441 (3) [0]\r5/ 441 (4) [0]\r6/ 441 (5) [0]\r7/ 441 (6) [0]\r8/ 441 (7) [0]\r9/ 441 (8) [0]\r10/ 441 (9) [0]\r11/ 441 (10) [0]\r12/ 441 (11) [0]\r13/ 441 (12) [0]\r14/ 441 (13) [0]\r15/ 441 (14) [0]\r16/ 441 (15) [0]\r17/ 441 (16) [0]\r18/ 441 (17) [0]\r19/ 441 (18) [0]\r20/ 441 (19) [0]\r21/ 441 (20) [0]\r22/ 441 (21) [0]\r23/ 441 (22) [0]\r24/ 441 (23) [0]\r25/ 441 (24) [0]\r26/ 441 (25) [0]\r27/ 441 (26) [0]\r28/ 441 (27) [0]\r29/ 441 (28) [0]\r30/ 441 (29) [0]\r31/ 441 (30) [0]\r32/ 441 (31) [0]\r33/ 441 (32) [0]\r34/ 441 (33) [0]\r35/ 441 (34) [0]\r36/ 441 (35) [0]\r37/ 441 (36) [0]\r38/ 441 (37) [0]\r39/ 441 (38) [0]\r40/ 441 (39) [0]\r41/ 441 (40) [0]\r42/ 441 (41) [0]\r43/ 441 (42) [0]\r44/ 441 (43) [0]\r45/ 441 (44) [0]\r46/ 441 (45) [0]\r47/ 441 (46) [0]\r48/ 441 (47) [0]\r49/ 441 (48) [0]\r50/ 441 (49) [0]\r51/ 441 (50) [0]\r52/ 441 (51) [0]\r53/ 441 (52) [0]\r54/ 441 (53) [0]\r55/ 441 (54) [0]\r56/ 441 (55) [0]\r57/ 441 (56) [0]\r58/ 441 (57) [0]\r59/ 441 (58) [0]\r60/ 441 (59) [0]\r61/ 441 (60) [0]\r62/ 441 (61) [0]\r63/ 441 (62) [0]\r64/ 441 (63) [0]\r65/ 441 (64) [0]\r66/ 441 (65) [0]\r67/ 441 (66) [0]\r68/ 441 (67) [0]\r69/ 441 (68) [0]\r70/ 441 (69) [0]\r71/ 441 (70) [0]\r72/ 441 (71) [0]\r73/ 441 (72) [0]\r74/ 441 (73) [0]\r75/ 441 (74) [0]\r76/ 441 (75) [0]\r77/ 441 (76) [0]\r78/ 441 (77) [0]\r79/ 441 (78) [0]\r80/ 441 (79) [0]\r81/ 441 (80) [0]\r82/ 441 (81) [0]\r83/ 441 (82) [0]\r84/ 441 (83) [0]\r85/ 441 (84) [0]\r86/ 441 (85) [0]\r87/ 441 (86) [0]\r88/ 441 (87) [0]\r89/ 441 (88) [0]\r90/ 441 (89) [0]\r91/ 441 (90) [0]\r92/ 441 (91) [0]\r93/ 441 (92) [0]\r94/ 441 (93) [0]\r95/ 441 (94) [0]\r96/ 441 (95) [0]\r97/ 441 (96) [0]\r98/ 441 (97) [0]\r99/ 441 (98) [0]\r100/ 441 (99) [0]\r101/ 441 (100) [0]\r102/ 441 (101) [0]\r103/ 441 (102) [0]\r104/ 441 (103) [0]\r105/ 441 (104) [0]\r106/ 441 (105) [0]\r107/ 441 (106) [0]\r108/ 441 (107) [0]\r109/ 441 (108) [0]\r110/ 441 (109) [0]\r111/ 441 (110) [0]\r112/ 441 (111) [0]\r113/ 441 (112) [0]\r114/ 441 (113) [0]\r115/ 441 (114) [0]\r116/ 441 (115) [0]\r117/ 441 (116) [0]\r118/ 441 (117) [0]\r119/ 441 (118) [0]\r120/ 441 (119) [0]\r121/ 441 (120) [0]\r122/ 441 (121) [0]\r123/ 441 (122) [0]\r124/ 441 (123) [0]\r125/ 441 (124) [0]\r126/ 441 (125) [0]\r127/ 441 (126) [0]\r128/ 441 (127) [0]\r129/ 441 (128) [0]\r130/ 441 (129) [0]\r131/ 441 (130) [0]\r132/ 441 (131) [0]\r133/ 441 (132) [0]\r134/ 441 (133) [0]\r135/ 441 (134) [0]\r136/ 441 (135) [0]\r137/ 441 (136) [0]\r138/ 441 (137) [0]\r139/ 441 (138) [0]\r140/ 441 (139) [0]\r141/ 441 (140) [0]\r142/ 441 (141) [0]\r143/ 441 (142) [0]\r144/ 441 (143) [0]\r145/ 441 (144) [0]\r146/ 441 (145) [0]\r147/ 441 (146) [0]\r148/ 441 (147) [0]\r149/ 441 (148) [0]\r150/ 441 (149) [0]\r151/ 441 (150) [0]\r152/ 441 (151) [0]\r153/ 441 (152) [0]\r154/ 441 (153) [0]\r155/ 441 (154) [0]\r156/ 441 (155) [0]\r157/ 441 (156) [0]\r158/ 441 (157) [0]\r159/ 441 (158) [0]\r160/ 441 (159) [0]\r161/ 441 (160) [0]\r162/ 441 (161) [0]\r163/ 441 (162) [0]\r164/ 441 (163) [0]\r165/ 441 (164) [0]\r166/ 441 (165) [0]\r167/ 441 (166) [0]\r168/ 441 (167) [0]\r169/ 441 (168) [0]\r170/ 441 (169) [0]\r171/ 441 (170) [0]\r172/ 441 (171) [0]\r173/ 441 (172) [0]\r174/ 441 (173) [0]\r175/ 441 (174) [0]\r176/ 441 (175) [0]\r177/ 441 (176) [0]\r178/ 441 (177) [0]\r179/ 441 (178) [0]\r180/ 441 (179) [0]\r181/ 441 (180) [0]\r182/ 441 (181) [0]\r183/ 441 (182) [0]\r184/ 441 (183) [0]\r185/ 441 (184) [0]\r186/ 441 (185) [0]\r187/ 441 (186) [0]\r188/ 441 (187) [0]\r189/ 441 (188) [0]\r190/ 441 (189) [0]\r191/ 441 (190) [0]\r192/ 441 (191) [0]\r193/ 441 (192) [0]\r194/ 441 (193) [0]\r195/ 441 (194) [0]\r196/ 441 (195) [0]\r197/ 441 (196) [0]\r198/ 441 (197) [0]\r199/ 441 (198) [0]\r200/ 441 (199) [0]\r201/ 441 (200) [0]\r202/ 441 (201) [0]\r203/ 441 (202) [0]\r204/ 441 (203) [0]\r205/ 441 (204) [0]\r206/ 441 (205) [0]\r207/ 441 (206) [0]\r208/ 441 (207) [0]\r209/ 441 (208) [0]\r210/ 441 (209) [0]\r211/ 441 (210) [0]\r212/ 441 (211) [0]\r213/ 441 (212) [0]\r214/ 441 (213) [0]\r215/ 441 (214) [0]\r216/ 441 (215) [0]\r217/ 441 (216) [0]\r218/ 441 (217) [0]\r219/ 441 (218) [0]\r220/ 441 (219) [0]\r221/ 441 (220) [0]\r222/ 441 (221) [0]\r223/ 441 (222) [0]\r224/ 441 (223) [0]\r225/ 441 (224) [0]\r226/ 441 (225) [0]\r227/ 441 (226) [0]\r228/ 441 (227) [0]\r229/ 441 (228) [0]\r230/ 441 (229) [0]\r231/ 441 (230) [0]\r232/ 441 (231) [0]\r233/ 441 (232) [0]\r234/ 441 (233) [0]\r235/ 441 (234) [0]\r236/ 441 (235) [0]\r237/ 441 (236) [0]\r238/ 441 (237) [0]\r239/ 441 (238) [0]\r240/ 441 (239) [0]\r241/ 441 (240) [0]\r242/ 441 (241) [0]\r243/ 441 (242) [0]\r244/ 441 (243) [0]\r245/ 441 (244) [0]\r246/ 441 (245) [0]\r247/ 441 (246) [0]\r248/ 441 (247) [0]\r249/ 441 (248) [0]\r250/ 441 (249) [0]\r251/ 441 (250) [0]\r252/ 441 (251) [0]\r253/ 441 (252) [0]\r254/ 441 (253) [0]\r255/ 441 (254) [0]\r256/ 441 (255) [0]\r257/ 441 (256) [0]\r258/ 441 (257) [0]\r259/ 441 (258) [0]\r260/ 441 (259) [0]\r261/ 441 (260) [0]\r262/ 441 (261) [0]\r263/ 441 (262) [0]\r264/ 441 (263) [0]\r265/ 441 (264) [0]\r266/ 441 (265) [0]\r267/ 441 (266) [0]\r268/ 441 (267) [0]\r269/ 441 (268) [0]\r270/ 441 (269) [0]\r271/ 441 (270) [0]\r272/ 441 (271) [0]\r273/ 441 (272) [0]\r274/ 441 (273) [0]\r275/ 441 (274) [0]\r276/ 441 (275) [0]\r277/ 441 (276) [0]\r278/ 441 (277) [0]\r279/ 441 (278) [0]\r280/ 441 (279) [0]\r281/ 441 (280) [0]\r282/ 441 (281) [0]\r283/ 441 (282) [0]\r284/ 441 (283) [0]\r285/ 441 (284) [0]\r286/ 441 (285) [0]\r287/ 441 (286) [0]\r288/ 441 (287) [0]\r289/ 441 (288) [0]\r290/ 441 (289) [0]\r291/ 441 (290) [0]\r292/ 441 (291) [0]\r293/ 441 (292) [0]\r294/ 441 (293) [0]\r295/ 441 (294) [0]\r296/ 441 (295) [0]\r297/ 441 (296) [0]\r298/ 441 (297) [0]\r299/ 441 (298) [0]\r300/ 441 (299) [0]\r301/ 441 (300) [0]\r302/ 441 (301) [0]\r303/ 441 (302) [0]\r304/ 441 (303) [0]\r305/ 441 (304) [0]\r306/ 441 (305) [0]\r307/ 441 (306) [0]\r308/ 441 (307) [0]\r309/ 441 (308) [0]\r310/ 441 (309) [0]\r311/ 441 (310) [0]\r312/ 441 (311) [0]\r313/ 441 (312) [0]\r314/ 441 (313) [0]\r315/ 441 (314) [0]\r316/ 441 (315) [0]\r317/ 441 (316) [0]\r318/ 441 (317) [0]\r319/ 441 (318) [0]\r320/ 441 (319) [0]\r321/ 441 (320) [0]\r322/ 441 (321) [0]\r323/ 441 (322) [0]\r324/ 441 (323) [0]\r325/ 441 (324) [0]\r326/ 441 (325) [0]\r327/ 441 (326) [0]\r328/ 441 (327) [0]\r329/ 441 (328) [0]\r330/ 441 (329) [0]\r331/ 441 (330) [0]\r332/ 441 (331) [0]\r333/ 441 (332) [0]\r334/ 441 (333) [0]\r335/ 441 (334) [0]\r336/ 441 (335) [0]\r337/ 441 (336) [0]\r338/ 441 (337) [0]\r339/ 441 (338) [0]\r340/ 441 (339) [0]\r341/ 441 (340) [0]\r342/ 441 (341) [0]\r343/ 441 (342) [0]\r344/ 441 (343) [0]\r345/ 441 (344) [0]\r346/ 441 (345) [0]\r347/ 441 (346) [0]\r348/ 441 (347) [0]\r349/ 441 (348) [0]\r350/ 441 (349) [0]\r351/ 441 (350) [0]\r352/ 441 (351) [0]\r353/ 441 (352) [0]\r354/ 441 (353) [0]\r355/ 441 (354) [0]\r356/ 441 (355) [0]\r357/ 441 (356) [0]\r358/ 441 (357) [0]\r359/ 441 (358) [0]\r360/ 441 (359) [0]\r361/ 441 (360) [0]\r362/ 441 (361) [0]\r363/ 441 (362) [0]\r364/ 441 (363) [0]\r365/ 441 (364) [0]\r366/ 441 (365) [0]\r367/ 441 (366) [0]\r368/ 441 (367) [0]\r369/ 441 (368) [0]\r370/ 441 (369) [0]\r371/ 441 (370) [0]\r372/ 441 (371) [0]\r373/ 441 (372) [0]\r374/ 441 (373) [0]\r375/ 441 (374) [0]\r376/ 441 (375) [0]\r377/ 441 (376) [0]\r378/ 441 (377) [0]\r379/ 441 (378) [0]\r380/ 441 (379) [0]\r381/ 441 (380) [0]\r382/ 441 (381) [0]\r383/ 441 (382) [0]\r384/ 441 (383) [0]\r385/ 441 (384) [0]\r386/ 441 (385) [0]\r387/ 441 (386) [0]\r388/ 441 (387) [0]\r389/ 441 (388) [0]\r390/ 441 (389) [0]\r391/ 441 (390) [0]\r392/ 441 (391) [0]\r393/ 441 (392) [0]\r394/ 441 (393) [0]\r395/ 441 (394) [0]\r396/ 441 (395) [0]\r397/ 441 (396) [0]\r398/ 441 (397) [0]\r399/ 441 (398) [0]\r400/ 441 (399) [0]\r401/ 441 (400) [0]\r402/ 441 (401) [0]\r403/ 441 (402) [0]\r404/ 441 (403) [0]\r405/ 441 (404) [0]\r406/ 441 (405) [0]\r407/ 441 (406) [0]\r408/ 441 (407) [0]\r409/ 441 (408) [0]\r410/ 441 (409) [0]\r411/ 441 (410) [0]\r412/ 441 (411) [0]\r413/ 441 (412) [0]\r414/ 441 (413) [0]\r415/ 441 (414) [0]\r416/ 441 (415) [0]\r417/ 441 (416) [0]\r418/ 441 (417) [0]\r419/ 441 (418) [0]\r420/ 441 (419) [0]\r421/ 441 (420) [0]\r422/ 441 (421) [0]\r423/ 441 (422) [0]\r424/ 441 (423) [0]\r425/ 441 (424) [0]\r426/ 441 (425) [0]\r427/ 441 (426) [0]\r428/ 441 (427) [0]\r429/ 441 (428) [0]\r430/ 441 (429) [0]\r431/ 441 (430) [0]\r432/ 441 (431) [0]\r433/ 441 (432) [0]\r434/ 441 (433) [0]\r435/ 441 (434) [0]\r436/ 441 (435) [0]\r437/ 441 (436) [0]\r438/ 441 (437) [0]\r439/ 441 (438) [0]\r440/ 441 (439) [0]\r441/ 441 (440) [0]\r",,terminal_output +2229,85752997,"TERMINAL",0,0,"Enumerating objects: 15, done.\r\nCounting objects: 6% (1/15)\rCounting objects: 13% (2/15)\rCounting objects: 20% (3/15)\rCounting objects: 26% (4/15)\rCounting objects: 33% (5/15)\rCounting objects: 40% (6/15)\rCounting objects: 46% (7/15)\rCounting objects: 53% (8/15)\rCounting objects: 60% (9/15)\rCounting objects: 66% (10/15)\rCounting objects: 73% (11/15)\rCounting objects: 80% (12/15)\rCounting objects: 86% (13/15)\rCounting objects: 93% (14/15)\rCounting objects: 100% (15/15)\rCounting objects: 100% (15/15), done.\r\nDelta compression using up to 8 threads\r\nCompressing objects: 12% (1/8)\rCompressing objects: 25% (2/8)\rCompressing objects: 37% (3/8)\rCompressing objects: 50% (4/8)\rCompressing objects: 62% (5/8)\rCompressing objects: 75% (6/8)\rCompressing objects: 87% (7/8)\rCompressing objects: 100% (8/8)\rCompressing objects: 100% (8/8), done.\r\nWriting objects: 12% (1/8)\rWriting objects: 25% (2/8)\rWriting objects: 37% (3/8)\rWriting objects: 50% (4/8)\rWriting objects: 62% (5/8)\rWriting objects: 75% (6/8)\rWriting objects: 87% (7/8)\rWriting objects: 100% (8/8)\rWriting objects: 100% (8/8), 1.02 KiB | 1.02 MiB/s, done.\r\nTotal 8 (delta 7), reused 0 (delta 0), pack-reused 0\r\n",,terminal_output +2230,85753529,"TERMINAL",0,0,"remote: Resolving deltas: 0% (0/7)\rremote: Resolving deltas: 14% (1/7)\rremote: Resolving deltas: 28% (2/7)\rremote: Resolving deltas: 42% (3/7)\rremote: Resolving deltas: 57% (4/7)\rremote: Resolving deltas: 71% (5/7)\rremote: Resolving deltas: 85% (6/7)\rremote: Resolving deltas: 100% (7/7)\rremote: Resolving deltas: 100% (7/7), completed with 7 local objects.\r\nTo https://github.com/emergenz/pdoom.org\r\n 7371cf5..2f3effc 2f3effc49a7b72da00fb42dcfdf27abfca5171ec -> gh-pages\r\n% \r \r",,terminal_output +2231,85761558,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2232,85761578,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2233,85763442,"TERMINAL",0,0,"npm run dev",,terminal_command +2234,85763492,"TERMINAL",0,0,"]633;C",,terminal_output +2235,85763688,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2236,85763862,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2237,85764238,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 378ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2238,85764737,"TERMINAL",0,0,"created dist/transforms.v2.js in 497ms\r\n\r\n[2025-09-03 17:01:15] waiting for changes...\r\n",,terminal_output +2239,85782555,"examples/act.html",595,0,"",html,selection_command +2240,85782803,"examples/act.html",602,0,"",html,selection_command +2241,85782833,"examples/act.html",643,0,"",html,selection_command +2242,85782864,"examples/act.html",714,0,"",html,selection_command +2243,85782899,"examples/act.html",738,0,"",html,selection_command +2244,85782933,"examples/act.html",794,0,"",html,selection_command +2245,85782967,"examples/act.html",802,0,"",html,selection_command +2246,85783002,"examples/act.html",803,0,"",html,selection_command +2247,85783037,"examples/act.html",810,0,"",html,selection_command +2248,85783069,"examples/act.html",817,0,"",html,selection_command +2249,85783103,"examples/act.html",853,0,"",html,selection_command +2250,85783137,"examples/act.html",859,0,"",html,selection_command +2251,85783169,"examples/act.html",878,0,"",html,selection_command +2252,85783294,"examples/act.html",935,0,"",html,selection_command +2253,85783454,"examples/act.html",985,0,"",html,selection_command +2254,85800726,"examples/act.html",1217,0,"",html,selection_command +2255,85800982,"examples/act.html",1255,0,"",html,selection_command +2256,85801017,"examples/act.html",1296,0,"",html,selection_command +2257,85801047,"examples/act.html",1313,0,"",html,selection_command +2258,85801079,"examples/act.html",1321,0,"",html,selection_command +2259,85801114,"examples/act.html",1356,0,"",html,selection_command +2260,85801149,"examples/act.html",1412,0,"",html,selection_command +2261,85801278,"examples/act.html",1356,0,"",html,selection_command +2262,85801526,"examples/act.html",1321,0,"",html,selection_command +2263,85801564,"examples/act.html",1313,0,"",html,selection_command +2264,85801597,"examples/act.html",1296,0,"",html,selection_command +2265,85801632,"examples/act.html",1255,0,"",html,selection_command +2266,85801667,"examples/act.html",1217,0,"",html,selection_command +2267,85801698,"examples/act.html",985,0,"",html,selection_command +2268,85801731,"examples/act.html",935,0,"",html,selection_command +2269,85801855,"examples/act.html",878,0,"",html,selection_command +2270,85802290,"examples/act.html",935,0,"",html,selection_command +2271,85802541,"examples/act.html",878,0,"",html,selection_command +2272,85803037,"examples/act.html",882,0,"",html,selection_command +2273,85803205,"examples/act.html",883,0,"",html,selection_command +2274,85803389,"examples/act.html",890,0,"",html,selection_command +2275,85803533,"examples/act.html",892,0,"",html,selection_command +2276,85803704,"examples/act.html",894,0,"",html,selection_command +2277,85804090,"examples/act.html",892,0,"",html,selection_command +2278,85804235,"examples/act.html",890,0,"",html,selection_command +2279,85804405,"examples/act.html",883,0,"",html,selection_command +2280,85804574,"examples/act.html",882,0,"",html,selection_command +2281,85825164,"examples/act.html",1725,0,"",html,selection_mouse +2282,85836303,"examples/act.html",1824,0,"",html,selection_mouse +2283,85836307,"examples/act.html",1823,0,"",html,selection_command +2284,85836735,"examples/act.html",1834,0,"",html,selection_mouse +2285,85837120,"examples/act.html",1836,0,"",html,selection_mouse +2286,85843771,"examples/act.html",1849,0,"",html,selection_command +2287,85847875,"examples/act.html",1857,0,"\n ",html,content +2288,85849922,"examples/act.html",1860,0,"a",html,content +2289,85849971,"examples/act.html",1861,0,"s",html,content +2290,85849973,"examples/act.html",1862,0,"",html,selection_keyboard +2291,85850004,"examples/act.html",1862,0,"i",html,content +2292,85850005,"examples/act.html",1863,0,"",html,selection_keyboard +2293,85850321,"examples/act.html",1863,0,"e",html,content +2294,85850323,"examples/act.html",1864,0,"",html,selection_keyboard +2295,85850964,"examples/act.html",1863,1,"",html,content +2296,85851124,"examples/act.html",1863,0,"d",html,content +2297,85851126,"examples/act.html",1864,0,"",html,selection_keyboard +2298,85851219,"examples/act.html",1864,0,"e",html,content +2299,85851221,"examples/act.html",1865,0,"",html,selection_keyboard +2300,85851487,"examples/act.html",1860,5,"",html,content +2301,85852722,"examples/act.html",1867,0,"A",html,content +2302,85852725,"examples/act.html",1868,0,"",html,selection_keyboard +2303,85852926,"examples/act.html",1868,0,"l",html,content +2304,85852927,"examples/act.html",1869,0,"",html,selection_keyboard +2305,85853104,"examples/act.html",1869,0,"l",html,content +2306,85853106,"examples/act.html",1870,0,"",html,selection_keyboard +2307,85855226,"examples/act.html",1870,0," ",html,content +2308,85855232,"examples/act.html",1871,0,"",html,selection_keyboard +2309,85856041,"examples/act.html",1871,0,"a",html,content +2310,85856044,"examples/act.html",1872,0,"",html,selection_keyboard +2311,85856075,"examples/act.html",1872,0,"u",html,content +2312,85856078,"examples/act.html",1873,0,"",html,selection_keyboard +2313,85856259,"examples/act.html",1873,0,"h",html,content +2314,85856263,"examples/act.html",1874,0,"",html,selection_keyboard +2315,85856342,"examples/act.html",1874,0,"o",html,content +2316,85856344,"examples/act.html",1875,0,"",html,selection_keyboard +2317,85856380,"examples/act.html",1875,0,"r",html,content +2318,85856382,"examples/act.html",1876,0,"",html,selection_keyboard +2319,85856638,"examples/act.html",1876,0,"s",html,content +2320,85856641,"examples/act.html",1877,0,"",html,selection_keyboard +2321,85856691,"examples/act.html",1877,0," ",html,content +2322,85856693,"examples/act.html",1878,0,"",html,selection_keyboard +2323,85857085,"examples/act.html",1871,7,"",html,content +2324,85857198,"examples/act.html",1867,4,"",html,content +2325,85857712,"examples/act.html",1867,0,"T",html,content +2326,85857714,"examples/act.html",1868,0,"",html,selection_keyboard +2327,85857837,"examples/act.html",1868,0,"h",html,content +2328,85857838,"examples/act.html",1869,0,"",html,selection_keyboard +2329,85857879,"examples/act.html",1869,0,"e",html,content +2330,85857881,"examples/act.html",1870,0,"",html,selection_keyboard +2331,85857986,"examples/act.html",1870,0,"r",html,content +2332,85857987,"examples/act.html",1871,0,"",html,selection_keyboard +2333,85858065,"examples/act.html",1871,0,"e",html,content +2334,85858069,"examples/act.html",1872,0,"",html,selection_keyboard +2335,85858336,"examples/act.html",1871,1,"",html,content +2336,85858474,"examples/act.html",1870,1,"",html,content +2337,85858624,"examples/act.html",1870,0,"s",html,content +2338,85858625,"examples/act.html",1871,0,"",html,selection_keyboard +2339,85858629,"examples/act.html",1871,0,"e",html,content +2340,85858630,"examples/act.html",1872,0,"",html,selection_keyboard +2341,85858738,"examples/act.html",1872,0," ",html,content +2342,85858740,"examples/act.html",1873,0,"",html,selection_keyboard +2343,85858842,"examples/act.html",1873,0,"a",html,content +2344,85858843,"examples/act.html",1874,0,"",html,selection_keyboard +2345,85858890,"examples/act.html",1874,0,"u",html,content +2346,85858891,"examples/act.html",1875,0,"",html,selection_keyboard +2347,85858998,"examples/act.html",1875,0,"t",html,content +2348,85859000,"examples/act.html",1876,0,"",html,selection_keyboard +2349,85859100,"examples/act.html",1876,0,"h",html,content +2350,85859102,"examples/act.html",1877,0,"",html,selection_keyboard +2351,85859217,"examples/act.html",1877,0,"o",html,content +2352,85859219,"examples/act.html",1878,0,"",html,selection_keyboard +2353,85859302,"examples/act.html",1878,0,"r",html,content +2354,85859305,"examples/act.html",1879,0,"",html,selection_keyboard +2355,85859455,"examples/act.html",1879,0,"s",html,content +2356,85859461,"examples/act.html",1880,0,"",html,selection_keyboard +2357,85859787,"examples/act.html",1873,7,"",html,content +2358,85859864,"examples/act.html",1867,6,"",html,content +2359,85860403,"examples/act.html",1867,0,"E",html,content +2360,85860405,"examples/act.html",1868,0,"",html,selection_keyboard +2361,85860537,"examples/act.html",1868,0,"q",html,content +2362,85860539,"examples/act.html",1869,0,"",html,selection_keyboard +2363,85860649,"examples/act.html",1869,0,"u",html,content +2364,85860650,"examples/act.html",1870,0,"",html,selection_keyboard +2365,85860871,"examples/act.html",1870,0,"a",html,content +2366,85860874,"examples/act.html",1871,0,"",html,selection_keyboard +2367,85860931,"examples/act.html",1871,0,"l",html,content +2368,85860934,"examples/act.html",1872,0,"",html,selection_keyboard +2369,85861067,"examples/act.html",1872,0," ",html,content +2370,85861071,"examples/act.html",1873,0,"",html,selection_keyboard +2371,85861204,"examples/act.html",1873,0,"c",html,content +2372,85861207,"examples/act.html",1874,0,"",html,selection_keyboard +2373,85861303,"examples/act.html",1874,0,"o",html,content +2374,85861305,"examples/act.html",1875,0,"",html,selection_keyboard +2375,85861366,"examples/act.html",1875,0,"n",html,content +2376,85861369,"examples/act.html",1876,0,"",html,selection_keyboard +2377,85861504,"examples/act.html",1876,0,"t",html,content +2378,85861508,"examples/act.html",1877,0,"",html,selection_keyboard +2379,85861705,"examples/act.html",1877,0,"r",html,content +2380,85861710,"examples/act.html",1878,0,"",html,selection_keyboard +2381,85861837,"examples/act.html",1878,0,"i",html,content +2382,85861842,"examples/act.html",1879,0,"",html,selection_keyboard +2383,85861927,"examples/act.html",1879,0,"b",html,content +2384,85861930,"examples/act.html",1880,0,"",html,selection_keyboard +2385,85862060,"examples/act.html",1880,0,"u",html,content +2386,85862064,"examples/act.html",1881,0,"",html,selection_keyboard +2387,85862097,"examples/act.html",1881,0,"t",html,content +2388,85862100,"examples/act.html",1882,0,"",html,selection_keyboard +2389,85862288,"examples/act.html",1882,0,"i",html,content +2390,85862294,"examples/act.html",1883,0,"",html,selection_keyboard +2391,85862338,"examples/act.html",1883,0,"o",html,content +2392,85862341,"examples/act.html",1884,0,"",html,selection_keyboard +2393,85862414,"examples/act.html",1884,0,"n",html,content +2394,85862417,"examples/act.html",1885,0,"",html,selection_keyboard +2395,85862589,"examples/act.html",1884,0,"",html,selection_command +2396,85862737,"examples/act.html",1873,0,"",html,selection_command +2397,85862876,"examples/act.html",1867,0,"",html,selection_command +2398,85863594,"examples/act.html",1867,0,"&",html,content +2399,85863596,"examples/act.html",1868,0,"",html,selection_keyboard +2400,85863975,"examples/act.html",1867,1,"",html,content +2401,85864322,"examples/act.html",1867,0,"*",html,content +2402,85864324,"examples/act.html",1868,0,"",html,selection_keyboard +2403,85864524,"examples/act.html",1868,0," ",html,content +2404,85864526,"examples/act.html",1869,0,"",html,selection_keyboard +2405,85864770,"examples/act.html",1868,0,"",html,selection_command +2406,85865939,"examples/act.html",1869,0,"",html,selection_command +2407,85866174,"examples/act.html",1868,1,"",html,content +2408,85866434,"examples/act.html",1867,0,"",html,selection_command +2409,85868071,"TERMINAL",0,0,"",,terminal_command +2410,85868072,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h[?2004l\r\r\n% \r \r]633;E;;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2411,85868072,"TERMINAL",0,0,"]633;C",,terminal_output +2412,85869672,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2413,85869693,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2414,85870337,"TERMINAL",0,0,"npm run dev",,terminal_command +2415,85870385,"TERMINAL",0,0,"]633;C",,terminal_output +2416,85870506,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2417,85870691,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2418,85871081,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 390ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2419,85871639,"TERMINAL",0,0,"created dist/transforms.v2.js in 559ms\r\n\r\n[2025-09-03 17:03:02] waiting for changes...\r\n",,terminal_output +2420,85878495,"examples/act.html",1858,37,"",html,content +2421,85878513,"examples/act.html",1860,0,"",html,selection_command +2422,85878925,"examples/act.html",1872,0,"",html,selection_command +2423,85879003,"examples/act.html",2089,0,"",html,selection_command +2424,85879138,"examples/act.html",2098,0,"",html,selection_command +2425,85879282,"examples/act.html",2111,0,"",html,selection_command +2426,85879407,"examples/act.html",2135,0,"",html,selection_command +2427,85879776,"examples/act.html",2146,0,"\n ",html,content +2428,85879780,"examples/act.html",2149,0,"",html,selection_command +2429,85884053,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2430,85884400,"TERMINAL",0,0,"npm run dev",,terminal_output +2431,85884488,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +2432,85884753,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2433,85884753,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2434,85884772,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2435,85885572,"TERMINAL",0,0,"npm run dev",,terminal_command +2436,85885623,"TERMINAL",0,0,"]633;C",,terminal_output +2437,85885736,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2438,85885888,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2439,85886263,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 376ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2440,85886759,"TERMINAL",0,0,"created dist/transforms.v2.js in 495ms\r\n\r\n[2025-09-03 17:03:17] waiting for changes...\r\n",,terminal_output +2441,85918454,"examples/ppo.html",0,0,"",html,tab +2442,85921071,"examples/ppo.html",0,0,"",html,selection_command +2443,85925888,"examples/ppo_off_policy.html",0,0,"",html,tab +2444,85926705,"examples/ppo_off_policy.html",0,0,"",html,selection_command +2445,85927138,"examples/ppo_off_policy.html",5,0,"",html,selection_command +2446,85927386,"examples/ppo_off_policy.html",30,0,"",html,selection_command +2447,85927421,"examples/ppo_off_policy.html",31,0,"",html,selection_command +2448,85927453,"examples/ppo_off_policy.html",97,0,"",html,selection_command +2449,85927486,"examples/ppo_off_policy.html",164,0,"",html,selection_command +2450,85927520,"examples/ppo_off_policy.html",206,0,"",html,selection_command +2451,85927553,"examples/ppo_off_policy.html",207,0,"",html,selection_command +2452,85927586,"examples/ppo_off_policy.html",257,0,"",html,selection_command +2453,85927620,"examples/ppo_off_policy.html",258,0,"",html,selection_command +2454,85927653,"examples/ppo_off_policy.html",328,0,"",html,selection_command +2455,85927686,"examples/ppo_off_policy.html",396,0,"",html,selection_command +2456,85927719,"examples/ppo_off_policy.html",471,0,"",html,selection_command +2457,85927752,"examples/ppo_off_policy.html",541,0,"",html,selection_command +2458,85927786,"examples/ppo_off_policy.html",574,0,"",html,selection_command +2459,85927819,"examples/ppo_off_policy.html",578,0,"",html,selection_command +2460,85927852,"examples/ppo_off_policy.html",594,0,"",html,selection_command +2461,85927886,"examples/ppo_off_policy.html",595,0,"",html,selection_command +2462,85927919,"examples/ppo_off_policy.html",602,0,"",html,selection_command +2463,85927952,"examples/ppo_off_policy.html",643,0,"",html,selection_command +2464,85927986,"examples/ppo_off_policy.html",714,0,"",html,selection_command +2465,85928019,"examples/ppo_off_policy.html",738,0,"",html,selection_command +2466,85928052,"examples/ppo_off_policy.html",794,0,"",html,selection_command +2467,85928086,"examples/ppo_off_policy.html",802,0,"",html,selection_command +2468,85928119,"examples/ppo_off_policy.html",803,0,"",html,selection_command +2469,85928152,"examples/ppo_off_policy.html",810,0,"",html,selection_command +2470,85928186,"examples/ppo_off_policy.html",817,0,"",html,selection_command +2471,85928219,"examples/ppo_off_policy.html",853,0,"",html,selection_command +2472,85928253,"examples/ppo_off_policy.html",859,0,"",html,selection_command +2473,85928286,"examples/ppo_off_policy.html",878,0,"",html,selection_command +2474,85928320,"examples/ppo_off_policy.html",935,0,"",html,selection_command +2475,85928353,"examples/ppo_off_policy.html",982,0,"",html,selection_command +2476,85928386,"examples/ppo_off_policy.html",1186,0,"",html,selection_command +2477,85928420,"examples/ppo_off_policy.html",1221,0,"",html,selection_command +2478,85928453,"examples/ppo_off_policy.html",1273,0,"",html,selection_command +2479,85928486,"examples/ppo_off_policy.html",1290,0,"",html,selection_command +2480,85928519,"examples/ppo_off_policy.html",1298,0,"",html,selection_command +2481,85928553,"examples/ppo_off_policy.html",1335,0,"",html,selection_command +2482,85928587,"examples/ppo_off_policy.html",1380,0,"",html,selection_command +2483,85928620,"examples/ppo_off_policy.html",1455,0,"",html,selection_command +2484,85928653,"examples/ppo_off_policy.html",1464,0,"",html,selection_command +2485,85928687,"examples/ppo_off_policy.html",1472,0,"",html,selection_command +2486,85928720,"examples/ppo_off_policy.html",1507,0,"",html,selection_command +2487,85928752,"examples/ppo_off_policy.html",1563,0,"",html,selection_command +2488,85928787,"examples/ppo_off_policy.html",1638,0,"",html,selection_command +2489,85928820,"examples/ppo_off_policy.html",1680,0,"",html,selection_command +2490,85928853,"examples/ppo_off_policy.html",1688,0,"",html,selection_command +2491,85928887,"examples/ppo_off_policy.html",1695,0,"",html,selection_command +2492,85928919,"examples/ppo_off_policy.html",1710,0,"",html,selection_command +2493,85928952,"examples/ppo_off_policy.html",1732,0,"",html,selection_command +2494,85928985,"examples/ppo_off_policy.html",1788,0,"",html,selection_command +2495,85929018,"examples/ppo_off_policy.html",1796,0,"",html,selection_command +2496,85929052,"examples/ppo_off_policy.html",1802,0,"",html,selection_command +2497,85929085,"examples/ppo_off_policy.html",1815,0,"",html,selection_command +2498,85929119,"examples/ppo_off_policy.html",1835,0,"",html,selection_command +2499,85929152,"examples/ppo_off_policy.html",1847,0,"",html,selection_command +2500,85929185,"examples/ppo_off_policy.html",1855,0,"",html,selection_command +2501,85929219,"examples/ppo_off_policy.html",2058,0,"",html,selection_command +2502,85929253,"examples/ppo_off_policy.html",2067,0,"",html,selection_command +2503,85929285,"examples/ppo_off_policy.html",2080,0,"",html,selection_command +2504,85929319,"examples/ppo_off_policy.html",2104,0,"",html,selection_command +2505,85929519,"examples/ppo_off_policy.html",2080,0,"",html,selection_command +2506,85929769,"examples/ppo_off_policy.html",2067,0,"",html,selection_command +2507,85929804,"examples/ppo_off_policy.html",2058,0,"",html,selection_command +2508,85929835,"examples/ppo_off_policy.html",1855,0,"",html,selection_command +2509,85929866,"examples/ppo_off_policy.html",1847,0,"",html,selection_command +2510,85930138,"examples/ppo_off_policy.html",1847,0,"",html,selection_command +2511,85930392,"examples/ppo_off_policy.html",1855,0,"",html,selection_command +2512,85930473,"examples/ppo_off_policy.html",2058,0,"",html,selection_command +2513,85930623,"examples/ppo_off_policy.html",2067,0,"",html,selection_command +2514,85930774,"examples/ppo_off_policy.html",2080,0,"",html,selection_command +2515,85930969,"examples/ppo_off_policy.html",2104,0,"",html,selection_command +2516,85932678,"examples/ppo_off_policy.html",2117,0,"\n ",html,content +2517,85934817,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2518,85935160,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +2519,85935432,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2520,85935433,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2521,85935453,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2522,85936208,"TERMINAL",0,0,"npm run dev",,terminal_command +2523,85936260,"TERMINAL",0,0,"]633;C",,terminal_output +2524,85936387,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2525,85936565,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2526,85936933,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 370ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2527,85937400,"TERMINAL",0,0,"created dist/transforms.v2.js in 467ms\r\n\r\n[2025-09-03 17:04:08] waiting for changes...\r\n",,terminal_output +2528,85969392,"examples/ppo_off_policy.html",2080,0,"",html,selection_command +2529,85969786,"examples/ppo_off_policy.html",2104,0,"",html,selection_command +2530,85969937,"examples/ppo_off_policy.html",2118,0,"",html,selection_command +2531,85970577,"examples/ppo_off_policy.html",2118,39,"",html,content +2532,85970596,"examples/ppo_off_policy.html",2122,0,"",html,selection_command +2533,85970705,"examples/ppo_off_policy.html",2108,0,"",html,selection_command +2534,85970857,"examples/ppo_off_policy.html",2084,0,"",html,selection_command +2535,85971247,"examples/ppo_off_policy.html",2103,0,"\n ",html,content +2536,85971262,"examples/ppo_off_policy.html",2108,0,"",html,selection_command +2537,85972285,"examples/ppo_off_policy.html",2109,0,"",html,selection_command +2538,85973545,"examples/ppo_off_policy.html",2108,0,"",html,selection_command +2539,85975920,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2540,85976314,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +2541,85976515,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2542,85976517,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2543,85976536,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2544,85977220,"TERMINAL",0,0,"npm run dev",,terminal_command +2545,85977271,"TERMINAL",0,0,"]633;C",,terminal_output +2546,85977401,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2547,85977575,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2548,85978020,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 443ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2549,85978387,"TERMINAL",0,0,"created dist/transforms.v2.js in 367ms\r\n\r\n[2025-09-03 17:04:49] waiting for changes...\r\n",,terminal_output +2550,85981805,"examples/ppo_off_policy.html",2106,39,"",html,content +2551,85981824,"examples/ppo_off_policy.html",2084,0,"",html,selection_command +2552,85981938,"examples/ppo_off_policy.html",2124,0,"side>*Equal contribution\n distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2564,85985450,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2565,85985866,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 417ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2566,85986296,"TERMINAL",0,0,"created dist/transforms.v2.js in 428ms\r\n\r\n[2025-09-03 17:04:56] waiting for changes...\r\n",,terminal_output +2567,85992548,"examples/ppo_off_policy.html",2118,39,"",html,content +2568,85992566,"examples/ppo_off_policy.html",2122,0,"",html,selection_command +2569,85992999,"examples/ppo_off_policy.html",2108,0,"",html,selection_command +2570,85993249,"examples/ppo_off_policy.html",2084,0,"",html,selection_command +2571,85993283,"examples/ppo_off_policy.html",2071,0,"",html,selection_command +2572,85993317,"examples/ppo_off_policy.html",2062,0,"",html,selection_command +2573,85993352,"examples/ppo_off_policy.html",1859,0,"",html,selection_command +2574,85993386,"examples/ppo_off_policy.html",1851,0,"",html,selection_command +2575,85993419,"examples/ppo_off_policy.html",1839,0,"",html,selection_command +2576,85993452,"examples/ppo_off_policy.html",1819,0,"",html,selection_command +2577,85993485,"examples/ppo_off_policy.html",1806,0,"",html,selection_command +2578,85993519,"examples/ppo_off_policy.html",1800,0,"",html,selection_command +2579,85993552,"examples/ppo_off_policy.html",1792,0,"",html,selection_command +2580,85993586,"examples/ppo_off_policy.html",1736,0,"",html,selection_command +2581,85993618,"examples/ppo_off_policy.html",1714,0,"",html,selection_command +2582,85993652,"examples/ppo_off_policy.html",1699,0,"",html,selection_command +2583,85993686,"examples/ppo_off_policy.html",1692,0,"",html,selection_command +2584,85993720,"examples/ppo_off_policy.html",1684,0,"",html,selection_command +2585,85993752,"examples/ppo_off_policy.html",1642,0,"",html,selection_command +2586,85993785,"examples/ppo_off_policy.html",1567,0,"",html,selection_command +2587,85993819,"examples/ppo_off_policy.html",1511,0,"",html,selection_command +2588,85993852,"examples/ppo_off_policy.html",1476,0,"",html,selection_command +2589,85993886,"examples/ppo_off_policy.html",1468,0,"",html,selection_command +2590,85993919,"examples/ppo_off_policy.html",1459,0,"",html,selection_command +2591,85993952,"examples/ppo_off_policy.html",1384,0,"",html,selection_command +2592,85993986,"examples/ppo_off_policy.html",1339,0,"",html,selection_command +2593,85994019,"examples/ppo_off_policy.html",1302,0,"",html,selection_command +2594,85994052,"examples/ppo_off_policy.html",1294,0,"",html,selection_command +2595,85994085,"examples/ppo_off_policy.html",1277,0,"",html,selection_command +2596,85994119,"examples/ppo_off_policy.html",1225,0,"",html,selection_command +2597,85994153,"examples/ppo_off_policy.html",1190,0,"",html,selection_command +2598,85994186,"examples/ppo_off_policy.html",986,0,"",html,selection_command +2599,85994225,"examples/ppo_off_policy.html",939,0,"",html,selection_command +2600,85994252,"examples/ppo_off_policy.html",882,0,"",html,selection_command +2601,85994286,"examples/ppo_off_policy.html",863,0,"",html,selection_command +2602,85994318,"examples/ppo_off_policy.html",857,0,"",html,selection_command +2603,85994626,"examples/ppo_off_policy.html",863,0,"",html,selection_command +2604,85995074,"examples/ppo_off_policy.html",857,0,"",html,selection_command +2605,85995707,"examples/ppo_off_policy.html",858,0,"\n ",html,content +2606,85995710,"examples/ppo_off_policy.html",863,0,"",html,selection_command +2607,85997298,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2608,85997674,"TERMINAL",0,0,"npm run dev",,terminal_output +2609,85997845,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +2610,85998106,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2611,85998106,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2612,85998115,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2613,85998674,"TERMINAL",0,0,"npm run dev",,terminal_command +2614,85998726,"TERMINAL",0,0,"]633;C",,terminal_output +2615,85998822,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2616,85998939,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2617,85999305,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 368ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2618,85999818,"TERMINAL",0,0,"created dist/transforms.v2.js in 509ms\r\n\r\n[2025-09-03 17:05:10] waiting for changes...\r\n",,terminal_output +2619,86006104,"examples/ppo_off_policy.html",861,39,"",html,content +2620,86006126,"examples/ppo_off_policy.html",857,0,"",html,selection_command +2621,86006267,"examples/ppo_off_policy.html",2124,0,"side>*Equal contribution\n distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2632,86009629,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2633,86009994,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 366ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2634,86010477,"TERMINAL",0,0,"created dist/transforms.v2.js in 482ms\r\n\r\n[2025-09-03 17:05:21] waiting for changes...\r\n",,terminal_output +2635,86042293,"examples/ppo_off_policy.html",2118,39,"",html,content +2636,86042319,"examples/ppo_off_policy.html",2122,0,"",html,selection_command +2637,86043954,"examples/ppo_off_policy.html",6522,0,"",html,selection_command +2638,86045057,"examples/ppo_off_policy.html",6671,0,"",html,selection_command +2639,86045306,"examples/ppo_off_policy.html",6889,0,"",html,selection_command +2640,86045339,"examples/ppo_off_policy.html",7023,0,"",html,selection_command +2641,86045372,"examples/ppo_off_policy.html",7039,0,"",html,selection_command +2642,86045406,"examples/ppo_off_policy.html",7120,0,"",html,selection_command +2643,86045532,"examples/ppo_off_policy.html",7136,0,"",html,selection_command +2644,86045705,"examples/ppo_off_policy.html",7146,0,"",html,selection_command +2645,86046056,"examples/ppo_off_policy.html",7136,0,"",html,selection_command +2646,86046671,"examples/ppo_off_policy.html",7137,0,"\n ",html,content +2647,86046674,"examples/ppo_off_policy.html",7142,0,"",html,selection_command +2648,86048494,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2649,86048764,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +2650,86049044,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2651,86049044,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2652,86049056,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2653,86049564,"TERMINAL",0,0,"npm run dev",,terminal_command +2654,86049615,"TERMINAL",0,0,"]633;C",,terminal_output +2655,86049714,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2656,86049833,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2657,86050255,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 423ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2658,86050654,"TERMINAL",0,0,"created dist/transforms.v2.js in 398ms\r\n\r\n[2025-09-03 17:06:01] waiting for changes...\r\n",,terminal_output +2659,86061359,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2660,86062318,"examples/ppo_off_policy.html",7143,39,"",html,content +2661,86062323,"examples/ppo_off_policy.html",7136,0,"",html,selection_command +2662,86062466,"examples/ppo_off_policy.html",2124,0,"side>*Equal contribution\n distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2672,86067526,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2673,86067973,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 446ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2674,86068406,"TERMINAL",0,0,"created dist/transforms.v2.js in 432ms\r\n\r\n[2025-09-03 17:06:19] waiting for changes...\r\n",,terminal_output +2675,86109122,"examples/jasmine.html",0,0,"",html,tab +2676,86110673,"examples/jasmine.html",1839,0,"",html,selection_command +2677,86110922,"examples/jasmine.html",1884,0,"",html,selection_command +2678,86110953,"examples/jasmine.html",1959,0,"",html,selection_command +2679,86110986,"examples/jasmine.html",1975,0,"",html,selection_command +2680,86111019,"examples/jasmine.html",1983,0,"",html,selection_command +2681,86111052,"examples/jasmine.html",2016,0,"",html,selection_command +2682,86111086,"examples/jasmine.html",2051,0,"",html,selection_command +2683,86111119,"examples/jasmine.html",2121,0,"",html,selection_command +2684,86111152,"examples/jasmine.html",2136,0,"",html,selection_command +2685,86111185,"examples/jasmine.html",2143,0,"",html,selection_command +2686,86111219,"examples/jasmine.html",2158,0,"",html,selection_command +2687,86111252,"examples/jasmine.html",2180,0,"",html,selection_command +2688,86111286,"examples/jasmine.html",2215,0,"",html,selection_command +2689,86111320,"examples/jasmine.html",2244,0,"",html,selection_command +2690,86111353,"examples/jasmine.html",2250,0,"",html,selection_command +2691,86111386,"examples/jasmine.html",2263,0,"",html,selection_command +2692,86111420,"examples/jasmine.html",2283,0,"",html,selection_command +2693,86111452,"examples/jasmine.html",2295,0,"",html,selection_command +2694,86111486,"examples/jasmine.html",2303,0,"",html,selection_command +2695,86111519,"examples/jasmine.html",2338,0,"",html,selection_command +2696,86111552,"examples/jasmine.html",2496,0,"",html,selection_command +2697,86111585,"examples/jasmine.html",2537,0,"",html,selection_command +2698,86111618,"examples/jasmine.html",2550,0,"",html,selection_command +2699,86111652,"examples/jasmine.html",2574,0,"",html,selection_command +2700,86111685,"examples/jasmine.html",2588,0,"",html,selection_command +2701,86111718,"examples/jasmine.html",2623,0,"",html,selection_command +2702,86111752,"examples/jasmine.html",2697,0,"",html,selection_command +2703,86111786,"examples/jasmine.html",2822,0,"",html,selection_command +2704,86111819,"examples/jasmine.html",2878,0,"",html,selection_command +2705,86111853,"examples/jasmine.html",3040,0,"",html,selection_command +2706,86112056,"examples/jasmine.html",2878,0,"",html,selection_command +2707,86112302,"examples/jasmine.html",2822,0,"",html,selection_command +2708,86112337,"examples/jasmine.html",2697,0,"",html,selection_command +2709,86112367,"examples/jasmine.html",2623,0,"",html,selection_command +2710,86112397,"examples/jasmine.html",2588,0,"",html,selection_command +2711,86112431,"examples/jasmine.html",2574,0,"",html,selection_command +2712,86112462,"examples/jasmine.html",2550,0,"",html,selection_command +2713,86112497,"examples/jasmine.html",2537,0,"",html,selection_command +2714,86112530,"examples/jasmine.html",2496,0,"",html,selection_command +2715,86112564,"examples/jasmine.html",2338,0,"",html,selection_command +2716,86112598,"examples/jasmine.html",2303,0,"",html,selection_command +2717,86113020,"examples/jasmine.html",2295,0,"",html,selection_command +2718,86113340,"examples/jasmine.html",2283,0,"",html,selection_command +2719,86113569,"examples/jasmine.html",2295,0,"",html,selection_command +2720,86113733,"examples/jasmine.html",2303,0,"",html,selection_command +2721,86114370,"examples/jasmine.html",2295,0,"",html,selection_command +2722,86115137,"examples/jasmine.html",2303,0,"",html,selection_command +2723,86115388,"examples/jasmine.html",2338,0,"",html,selection_command +2724,86115420,"examples/jasmine.html",2496,0,"",html,selection_command +2725,86115805,"examples/jasmine.html",2537,0,"",html,selection_command +2726,86116000,"examples/jasmine.html",2550,0,"",html,selection_command +2727,86116272,"examples/jasmine.html",2574,0,"",html,selection_command +2728,86118084,"examples/jasmine.html",2589,0,"\n ",html,content +2729,86118094,"examples/jasmine.html",2594,0,"",html,selection_command +2730,86119792,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2731,86120088,"TERMINAL",0,0,"npm run dev",,terminal_output +2732,86120155,"TERMINAL",0,0,"cp examples/* dist",,terminal_output +2733,86120437,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2734,86120438,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2735,86120458,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2736,86121120,"TERMINAL",0,0,"npm run dev",,terminal_command +2737,86121171,"TERMINAL",0,0,"]633;C",,terminal_output +2738,86121303,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2739,86121475,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2740,86121971,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 496ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2741,86122376,"TERMINAL",0,0,"created dist/transforms.v2.js in 405ms\r\n\r\n[2025-09-03 17:07:13] waiting for changes...\r\n",,terminal_output +2742,86189760,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2743,86190177,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +2744,86190463,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2745,86190464,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;cp examples/* dist;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2746,86190486,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2747,86191061,"TERMINAL",0,0,"npm run dev",,terminal_command +2748,86191113,"TERMINAL",0,0,"]633;C",,terminal_output +2749,86191245,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2750,86191416,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2751,86191785,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 370ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2752,86192296,"TERMINAL",0,0,"created dist/transforms.v2.js in 509ms\r\n\r\n[2025-09-03 17:08:22] waiting for changes...\r\n",,terminal_output +2753,86197229,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2754,86218965,"TERMINAL",0,0,"npm run devcp examples/* dist",,terminal_output +2755,86219427,"TERMINAL",0,0,"npm run dev ",,terminal_output +2756,86220282,"TERMINAL",0,0,"npm run dev",,terminal_command +2757,86220284,"TERMINAL",0,0,"[?2004l\r\r\n]633;E;npm run dev;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2758,86220333,"TERMINAL",0,0,"]633;C",,terminal_output +2759,86220461,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n\r\n",,terminal_output +2760,86220604,"TERMINAL",0,0,"\r\ncrollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2761,86220981,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 373ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2762,86221431,"TERMINAL",0,0,"created dist/transforms.v2.js in 448ms\r\n\r\n[2025-09-03 17:08:52] waiting for changes...\r\n",,terminal_output +2763,86223365,"TERMINAL",0,0,"^C⠙% \r \r",,terminal_output +2764,86230224,"TERMINAL",0,0,"git subtree push --prefix dist origin gh-pages",,terminal_command +2765,86230276,"TERMINAL",0,0,"]633;C",,terminal_output +2766,86235199,"TERMINAL",0,0,"git push using: origin gh-pages\r\n1/ 442 (0) [0]\r2/ 442 (1) [0]\r3/ 442 (2) [0]\r4/ 442 (3) [0]\r5/ 442 (4) [0]\r6/ 442 (5) [0]\r7/ 442 (6) [0]\r8/ 442 (7) [0]\r9/ 442 (8) [0]\r10/ 442 (9) [0]\r11/ 442 (10) [0]\r12/ 442 (11) [0]\r13/ 442 (12) [0]\r14/ 442 (13) [0]\r15/ 442 (14) [0]\r16/ 442 (15) [0]\r17/ 442 (16) [0]\r18/ 442 (17) [0]\r19/ 442 (18) [0]\r20/ 442 (19) [0]\r21/ 442 (20) [0]\r22/ 442 (21) [0]\r23/ 442 (22) [0]\r24/ 442 (23) [0]\r25/ 442 (24) [0]\r26/ 442 (25) [0]\r27/ 442 (26) [0]\r28/ 442 (27) [0]\r29/ 442 (28) [0]\r30/ 442 (29) [0]\r31/ 442 (30) [0]\r32/ 442 (31) [0]\r33/ 442 (32) [0]\r34/ 442 (33) [0]\r35/ 442 (34) [0]\r36/ 442 (35) [0]\r37/ 442 (36) [0]\r38/ 442 (37) [0]\r39/ 442 (38) [0]\r40/ 442 (39) [0]\r41/ 442 (40) [0]\r42/ 442 (41) [0]\r43/ 442 (42) [0]\r44/ 442 (43) [0]\r45/ 442 (44) [0]\r46/ 442 (45) [0]\r47/ 442 (46) [0]\r48/ 442 (47) [0]\r49/ 442 (48) [0]\r50/ 442 (49) [0]\r51/ 442 (50) [0]\r52/ 442 (51) [0]\r53/ 442 (52) [0]\r54/ 442 (53) [0]\r55/ 442 (54) [0]\r56/ 442 (55) [0]\r57/ 442 (56) [0]\r58/ 442 (57) [0]\r59/ 442 (58) [0]\r60/ 442 (59) [0]\r61/ 442 (60) [0]\r62/ 442 (61) [0]\r63/ 442 (62) [0]\r64/ 442 (63) [0]\r65/ 442 (64) [0]\r66/ 442 (65) [0]\r67/ 442 (66) [0]\r68/ 442 (67) [0]\r69/ 442 (68) [0]\r70/ 442 (69) [0]\r71/ 442 (70) [0]\r72/ 442 (71) [0]\r73/ 442 (72) [0]\r74/ 442 (73) [0]\r75/ 442 (74) [0]\r76/ 442 (75) [0]\r77/ 442 (76) [0]\r78/ 442 (77) [0]\r79/ 442 (78) [0]\r80/ 442 (79) [0]\r81/ 442 (80) [0]\r82/ 442 (81) [0]\r83/ 442 (82) [0]\r84/ 442 (83) [0]\r85/ 442 (84) [0]\r86/ 442 (85) [0]\r87/ 442 (86) [0]\r88/ 442 (87) [0]\r89/ 442 (88) [0]\r90/ 442 (89) [0]\r91/ 442 (90) [0]\r92/ 442 (91) [0]\r93/ 442 (92) [0]\r94/ 442 (93) [0]\r95/ 442 (94) [0]\r96/ 442 (95) [0]\r97/ 442 (96) [0]\r98/ 442 (97) [0]\r99/ 442 (98) [0]\r100/ 442 (99) [0]\r101/ 442 (100) [0]\r102/ 442 (101) [0]\r103/ 442 (102) [0]\r104/ 442 (103) [0]\r105/ 442 (104) [0]\r106/ 442 (105) [0]\r107/ 442 (106) [0]\r108/ 442 (107) [0]\r109/ 442 (108) [0]\r110/ 442 (109) [0]\r111/ 442 (110) [0]\r112/ 442 (111) [0]\r113/ 442 (112) [0]\r114/ 442 (113) [0]\r115/ 442 (114) [0]\r116/ 442 (115) [0]\r117/ 442 (116) [0]\r118/ 442 (117) [0]\r119/ 442 (118) [0]\r120/ 442 (119) [0]\r121/ 442 (120) [0]\r122/ 442 (121) [0]\r123/ 442 (122) [0]\r124/ 442 (123) [0]\r125/ 442 (124) [0]\r126/ 442 (125) [0]\r127/ 442 (126) [0]\r128/ 442 (127) [0]\r129/ 442 (128) [0]\r130/ 442 (129) [0]\r131/ 442 (130) [0]\r132/ 442 (131) [0]\r133/ 442 (132) [0]\r134/ 442 (133) [0]\r135/ 442 (134) [0]\r136/ 442 (135) [0]\r137/ 442 (136) [0]\r138/ 442 (137) [0]\r139/ 442 (138) [0]\r140/ 442 (139) [0]\r141/ 442 (140) [0]\r142/ 442 (141) [0]\r143/ 442 (142) [0]\r144/ 442 (143) [0]\r145/ 442 (144) [0]\r146/ 442 (145) [0]\r147/ 442 (146) [0]\r148/ 442 (147) [0]\r149/ 442 (148) [0]\r150/ 442 (149) [0]\r151/ 442 (150) [0]\r152/ 442 (151) [0]\r153/ 442 (152) [0]\r154/ 442 (153) [0]\r155/ 442 (154) [0]\r156/ 442 (155) [0]\r157/ 442 (156) [0]\r158/ 442 (157) [0]\r159/ 442 (158) [0]\r160/ 442 (159) [0]\r161/ 442 (160) [0]\r162/ 442 (161) [0]\r163/ 442 (162) [0]\r164/ 442 (163) [0]\r165/ 442 (164) [0]\r166/ 442 (165) [0]\r167/ 442 (166) [0]\r168/ 442 (167) [0]\r169/ 442 (168) [0]\r170/ 442 (169) [0]\r171/ 442 (170) [0]\r172/ 442 (171) [0]\r173/ 442 (172) [0]\r174/ 442 (173) [0]\r175/ 442 (174) [0]\r176/ 442 (175) [0]\r177/ 442 (176) [0]\r178/ 442 (177) [0]\r179/ 442 (178) [0]\r180/ 442 (179) [0]\r181/ 442 (180) [0]\r182/ 442 (181) [0]\r183/ 442 (182) [0]\r184/ 442 (183) [0]\r185/ 442 (184) [0]\r186/ 442 (185) [0]\r187/ 442 (186) [0]\r188/ 442 (187) [0]\r189/ 442 (188) [0]\r190/ 442 (189) [0]\r191/ 442 (190) [0]\r192/ 442 (191) [0]\r193/ 442 (192) [0]\r194/ 442 (193) [0]\r195/ 442 (194) [0]\r196/ 442 (195) [0]\r197/ 442 (196) [0]\r198/ 442 (197) [0]\r199/ 442 (198) [0]\r200/ 442 (199) [0]\r201/ 442 (200) [0]\r202/ 442 (201) [0]\r203/ 442 (202) [0]\r204/ 442 (203) [0]\r205/ 442 (204) [0]\r206/ 442 (205) [0]\r207/ 442 (206) [0]\r208/ 442 (207) [0]\r209/ 442 (208) [0]\r210/ 442 (209) [0]\r211/ 442 (210) [0]\r212/ 442 (211) [0]\r213/ 442 (212) [0]\r214/ 442 (213) [0]\r215/ 442 (214) [0]\r216/ 442 (215) [0]\r217/ 442 (216) [0]\r218/ 442 (217) [0]\r219/ 442 (218) [0]\r220/ 442 (219) [0]\r221/ 442 (220) [0]\r222/ 442 (221) [0]\r223/ 442 (222) [0]\r224/ 442 (223) [0]\r225/ 442 (224) [0]\r226/ 442 (225) [0]\r227/ 442 (226) [0]\r228/ 442 (227) [0]\r229/ 442 (228) [0]\r230/ 442 (229) [0]\r231/ 442 (230) [0]\r232/ 442 (231) [0]\r233/ 442 (232) [0]\r234/ 442 (233) [0]\r235/ 442 (234) [0]\r236/ 442 (235) [0]\r237/ 442 (236) [0]\r238/ 442 (237) [0]\r239/ 442 (238) [0]\r240/ 442 (239) [0]\r241/ 442 (240) [0]\r242/ 442 (241) [0]\r243/ 442 (242) [0]\r244/ 442 (243) [0]\r245/ 442 (244) [0]\r246/ 442 (245) [0]\r247/ 442 (246) [0]\r248/ 442 (247) [0]\r249/ 442 (248) [0]\r250/ 442 (249) [0]\r251/ 442 (250) [0]\r252/ 442 (251) [0]\r253/ 442 (252) [0]\r254/ 442 (253) [0]\r255/ 442 (254) [0]\r256/ 442 (255) [0]\r257/ 442 (256) [0]\r258/ 442 (257) [0]\r259/ 442 (258) [0]\r260/ 442 (259) [0]\r261/ 442 (260) [0]\r262/ 442 (261) [0]\r263/ 442 (262) [0]\r264/ 442 (263) [0]\r265/ 442 (264) [0]\r266/ 442 (265) [0]\r267/ 442 (266) [0]\r268/ 442 (267) [0]\r269/ 442 (268) [0]\r270/ 442 (269) [0]\r271/ 442 (270) [0]\r272/ 442 (271) [0]\r273/ 442 (272) [0]\r274/ 442 (273) [0]\r275/ 442 (274) [0]\r276/ 442 (275) [0]\r277/ 442 (276) [0]\r278/ 442 (277) [0]\r279/ 442 (278) [0]\r280/ 442 (279) [0]\r281/ 442 (280) [0]\r282/ 442 (281) [0]\r283/ 442 (282) [0]\r284/ 442 (283) [0]\r285/ 442 (284) [0]\r286/ 442 (285) [0]\r287/ 442 (286) [0]\r288/ 442 (287) [0]\r289/ 442 (288) [0]\r290/ 442 (289) [0]\r291/ 442 (290) [0]\r292/ 442 (291) [0]\r293/ 442 (292) [0]\r294/ 442 (293) [0]\r295/ 442 (294) [0]\r296/ 442 (295) [0]\r297/ 442 (296) [0]\r298/ 442 (297) [0]\r299/ 442 (298) [0]\r300/ 442 (299) [0]\r301/ 442 (300) [0]\r302/ 442 (301) [0]\r303/ 442 (302) [0]\r304/ 442 (303) [0]\r305/ 442 (304) [0]\r306/ 442 (305) [0]\r307/ 442 (306) [0]\r308/ 442 (307) [0]\r309/ 442 (308) [0]\r310/ 442 (309) [0]\r311/ 442 (310) [0]\r312/ 442 (311) [0]\r313/ 442 (312) [0]\r314/ 442 (313) [0]\r315/ 442 (314) [0]\r316/ 442 (315) [0]\r317/ 442 (316) [0]\r318/ 442 (317) [0]\r319/ 442 (318) [0]\r320/ 442 (319) [0]\r321/ 442 (320) [0]\r322/ 442 (321) [0]\r323/ 442 (322) [0]\r324/ 442 (323) [0]\r325/ 442 (324) [0]\r326/ 442 (325) [0]\r327/ 442 (326) [0]\r328/ 442 (327) [0]\r329/ 442 (328) [0]\r330/ 442 (329) [0]\r331/ 442 (330) [0]\r332/ 442 (331) [0]\r333/ 442 (332) [0]\r334/ 442 (333) [0]\r335/ 442 (334) [0]\r336/ 442 (335) [0]\r337/ 442 (336) [0]\r338/ 442 (337) [0]\r339/ 442 (338) [0]\r340/ 442 (339) [0]\r341/ 442 (340) [0]\r342/ 442 (341) [0]\r343/ 442 (342) [0]\r344/ 442 (343) [0]\r345/ 442 (344) [0]\r346/ 442 (345) [0]\r347/ 442 (346) [0]\r348/ 442 (347) [0]\r349/ 442 (348) [0]\r350/ 442 (349) [0]\r351/ 442 (350) [0]\r352/ 442 (351) [0]\r353/ 442 (352) [0]\r354/ 442 (353) [0]\r355/ 442 (354) [0]\r356/ 442 (355) [0]\r357/ 442 (356) [0]\r358/ 442 (357) [0]\r359/ 442 (358) [0]\r360/ 442 (359) [0]\r361/ 442 (360) [0]\r362/ 442 (361) [0]\r363/ 442 (362) [0]\r364/ 442 (363) [0]\r365/ 442 (364) [0]\r366/ 442 (365) [0]\r367/ 442 (366) [0]\r368/ 442 (367) [0]\r369/ 442 (368) [0]\r370/ 442 (369) [0]\r371/ 442 (370) [0]\r372/ 442 (371) [0]\r373/ 442 (372) [0]\r374/ 442 (373) [0]\r375/ 442 (374) [0]\r376/ 442 (375) [0]\r377/ 442 (376) [0]\r378/ 442 (377) [0]\r379/ 442 (378) [0]\r380/ 442 (379) [0]\r381/ 442 (380) [0]\r382/ 442 (381) [0]\r383/ 442 (382) [0]\r384/ 442 (383) [0]\r385/ 442 (384) [0]\r386/ 442 (385) [0]\r387/ 442 (386) [0]\r388/ 442 (387) [0]\r389/ 442 (388) [0]\r390/ 442 (389) [0]\r391/ 442 (390) [0]\r392/ 442 (391) [0]\r393/ 442 (392) [0]\r394/ 442 (393) [0]\r395/ 442 (394) [0]\r396/ 442 (395) [0]\r397/ 442 (396) [0]\r398/ 442 (397) [0]\r399/ 442 (398) [0]\r400/ 442 (399) [0]\r401/ 442 (400) [0]\r402/ 442 (401) [0]\r403/ 442 (402) [0]\r404/ 442 (403) [0]\r405/ 442 (404) [0]\r406/ 442 (405) [0]\r407/ 442 (406) [0]\r408/ 442 (407) [0]\r409/ 442 (408) [0]\r410/ 442 (409) [0]\r411/ 442 (410) [0]\r412/ 442 (411) [0]\r413/ 442 (412) [0]\r414/ 442 (413) [0]\r415/ 442 (414) [0]\r416/ 442 (415) [0]\r417/ 442 (416) [0]\r418/ 442 (417) [0]\r419/ 442 (418) [0]\r420/ 442 (419) [0]\r421/ 442 (420) [0]\r422/ 442 (421) [0]\r423/ 442 (422) [0]\r424/ 442 (423) [0]\r425/ 442 (424) [0]\r426/ 442 (425) [0]\r427/ 442 (426) [0]\r428/ 442 (427) [0]\r429/ 442 (428) [0]\r430/ 442 (429) [0]\r431/ 442 (430) [0]\r432/ 442 (431) [0]\r433/ 442 (432) [0]\r434/ 442 (433) [0]\r435/ 442 (434) [0]\r436/ 442 (435) [0]\r437/ 442 (436) [0]\r438/ 442 (437) [0]\r439/ 442 (438) [0]\r440/ 442 (439) [0]\r441/ 442 (440) [0]\r442/ 442 (441) [0]\r",,terminal_output +2767,86235648,"TERMINAL",0,0,"Enumerating objects: 9, done.\r\nCounting objects: 11% (1/9)\rCounting objects: 22% (2/9)\rCounting objects: 33% (3/9)\rCounting objects: 44% (4/9)\rCounting objects: 55% (5/9)\rCounting objects: 66% (6/9)\rCounting objects: 77% (7/9)\rCounting objects: 88% (8/9)\rCounting objects: 100% (9/9)\rCounting objects: 100% (9/9), done.\r\nDelta compression using up to 8 threads\r\nCompressing objects: 20% (1/5)\rCompressing objects: 40% (2/5)\rCompressing objects: 60% (3/5)\rCompressing objects: 80% (4/5)\rCompressing objects: 100% (5/5)\rCompressing objects: 100% (5/5), done.\r\nWriting objects: 20% (1/5)\rWriting objects: 40% (2/5)\rWriting objects: 60% (3/5)\rWriting objects: 80% (4/5)\rWriting objects: 100% (5/5)\rWriting objects: 100% (5/5), 548 bytes | 548.00 KiB/s, done.\r\nTotal 5 (delta 4), reused 0 (delta 0), pack-reused 0\r\n",,terminal_output +2768,86236037,"TERMINAL",0,0,"remote: Resolving deltas: 0% (0/4)\rremote: Resolving deltas: 25% (1/4)\rremote: Resolving deltas: 50% (2/4)\rremote: Resolving deltas: 75% (3/4)\rremote: Resolving deltas: 100% (4/4)\rremote: Resolving deltas: 100% (4/4), completed with 4 local objects.\r\nTo https://github.com/emergenz/pdoom.org\r\n 2f3effc..9ebe017 9ebe0170bae66942f9f499b4d6c50a3dd1749734 -> gh-pages\r\n% \r \r",,terminal_output +2769,86298517,"examples/jasmine.html",2633,0,"",html,selection_command +2770,86299059,"examples/jasmine.html",2629,74,"",html,content +2771,86299087,"examples/jasmine.html",2633,0,"",html,selection_command +2772,86299686,"examples/jasmine.html",2758,0,"",html,selection_command +2773,86299768,"examples/jasmine.html",2814,0,"",html,selection_command +2774,86299931,"examples/jasmine.html",2985,0,"",html,selection_command +2775,86300239,"examples/jasmine.html",2814,0,"",html,selection_command +2776,86300610,"examples/jasmine.html",2980,0,"\n 1",html,content +2777,86300615,"examples/jasmine.html",2985,0,"",html,selection_command +2778,86302213,"TERMINAL",0,0,"",,terminal_command +2779,86302213,"TERMINAL",0,0,"]633;C",,terminal_output +2780,86304302,"TERMINAL",0,0,"cp examples/* dist",,terminal_command +2781,86304320,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +2782,86306602,"TERMINAL",0,0,"npm run dev",,terminal_command +2783,86306654,"TERMINAL",0,0,"]633;C",,terminal_output +2784,86306749,"TERMINAL",0,0,"\r\n> distill-template@2.8.0 dev\r\n> rollup -c rollup.config.dev.js -w\r\n\r\n",,terminal_output +2785,86306866,"TERMINAL",0,0,"crollup v2.7.3\r\nbundles src/components.js → dist/template.v2.js...\r\n",,terminal_output +2786,86307238,"TERMINAL",0,0,"http://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/dist\r\nhttp://localhost:8088 -> /Users/franzsrambical/Documents/pdoom/pdoom.org/examples\r\n(!) Circular dependencies\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/array.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-interpolate/src/value.js -> node_modules/d3-interpolate/src/object.js -> node_modules/d3-interpolate/src/value.js\r\nnode_modules/d3-selection/src/selection/index.js -> node_modules/d3-selection/src/selection/select.js -> node_modules/d3-selection/src/selection/index.js\r\n...and 7 more\r\ncreated dist/template.v2.js in 373ms\r\nbundles src/transforms.js → dist/transforms.v2.js...\r\n",,terminal_output +2787,86307689,"TERMINAL",0,0,"created dist/transforms.v2.js in 449ms\r\n\r\n[2025-09-03 17:10:18] waiting for changes...\r\n",,terminal_output +2788,86338875,"TERMINAL",0,0,"^C⠙% \r \r]633;D;0]633;P;Cwd=/Users/franzsrambical/Documents/pdoom/pdoom.org\r]633;Afranzsrambical@MBF6N9WFVKFV pdoom.org % ]633;B[?2004h",,terminal_output +2789,86339947,"TERMINAL",0,0,"\r\r\nbck-i-search: _",,terminal_output +2790,86340246,"TERMINAL",0,0,"cp examples/* dists_git subtree push --prefix dist origin gh-pagesu_",,terminal_output +2791,86340312,"TERMINAL",0,0,"ubb_",,terminal_output +2792,86340612,"TERMINAL",0,0,"git subtree push --prefix dist origin gh-pages",,terminal_command +2793,86340612,"TERMINAL",0,0,"sub\r[?2004l\r]633;E;git subtree push --prefix dist origin gh-pages;b07e0e29-8b6e-4188-b95b-fd2ce17f4bb1",,terminal_output +2794,86340663,"TERMINAL",0,0,"]633;C",,terminal_output +2795,86342830,"TERMINAL",0,0,"git push using: origin gh-pages\r\n1/ 443 (0) [0]\r2/ 443 (1) [0]\r3/ 443 (2) [0]\r4/ 443 (3) [0]\r5/ 443 (4) [0]\r6/ 443 (5) [0]\r7/ 443 (6) [0]\r8/ 443 (7) [0]\r9/ 443 (8) [0]\r10/ 443 (9) [0]\r11/ 443 (10) [0]\r12/ 443 (11) [0]\r13/ 443 (12) [0]\r14/ 443 (13) [0]\r15/ 443 (14) [0]\r16/ 443 (15) [0]\r17/ 443 (16) [0]\r18/ 443 (17) [0]\r19/ 443 (18) [0]\r20/ 443 (19) [0]\r21/ 443 (20) [0]\r22/ 443 (21) [0]\r23/ 443 (22) [0]\r24/ 443 (23) [0]\r25/ 443 (24) [0]\r26/ 443 (25) [0]\r27/ 443 (26) [0]\r28/ 443 (27) [0]\r29/ 443 (28) [0]\r30/ 443 (29) [0]\r31/ 443 (30) [0]\r32/ 443 (31) [0]\r33/ 443 (32) [0]\r34/ 443 (33) [0]\r35/ 443 (34) [0]\r36/ 443 (35) [0]\r37/ 443 (36) [0]\r38/ 443 (37) [0]\r39/ 443 (38) [0]\r40/ 443 (39) [0]\r41/ 443 (40) [0]\r42/ 443 (41) [0]\r43/ 443 (42) [0]\r44/ 443 (43) [0]\r45/ 443 (44) [0]\r46/ 443 (45) [0]\r47/ 443 (46) [0]\r48/ 443 (47) [0]\r49/ 443 (48) [0]\r50/ 443 (49) [0]\r51/ 443 (50) [0]\r52/ 443 (51) [0]\r53/ 443 (52) [0]\r54/ 443 (53) [0]\r55/ 443 (54) [0]\r56/ 443 (55) [0]\r57/ 443 (56) [0]\r58/ 443 (57) [0]\r59/ 443 (58) [0]\r60/ 443 (59) [0]\r61/ 443 (60) [0]\r62/ 443 (61) [0]\r63/ 443 (62) [0]\r64/ 443 (63) [0]\r65/ 443 (64) [0]\r66/ 443 (65) [0]\r67/ 443 (66) [0]\r68/ 443 (67) [0]\r69/ 443 (68) [0]\r70/ 443 (69) [0]\r71/ 443 (70) [0]\r72/ 443 (71) [0]\r73/ 443 (72) [0]\r74/ 443 (73) [0]\r75/ 443 (74) [0]\r76/ 443 (75) [0]\r77/ 443 (76) [0]\r78/ 443 (77) [0]\r79/ 443 (78) [0]\r80/ 443 (79) [0]\r81/ 443 (80) [0]\r82/ 443 (81) [0]\r83/ 443 (82) [0]\r84/ 443 (83) [0]\r85/ 443 (84) [0]\r86/ 443 (85) [0]\r87/ 443 (86) [0]\r88/ 443 (87) [0]\r89/ 443 (88) [0]\r90/ 443 (89) [0]\r91/ 443 (90) [0]\r92/ 443 (91) [0]\r93/ 443 (92) [0]\r94/ 443 (93) [0]\r95/ 443 (94) [0]\r96/ 443 (95) [0]\r97/ 443 (96) [0]\r98/ 443 (97) [0]\r99/ 443 (98) [0]\r100/ 443 (99) [0]\r101/ 443 (100) [0]\r102/ 443 (101) [0]\r103/ 443 (102) [0]\r104/ 443 (103) [0]\r105/ 443 (104) [0]\r106/ 443 (105) [0]\r107/ 443 (106) [0]\r108/ 443 (107) [0]\r109/ 443 (108) [0]\r110/ 443 (109) [0]\r111/ 443 (110) [0]\r112/ 443 (111) [0]\r113/ 443 (112) [0]\r114/ 443 (113) [0]\r115/ 443 (114) [0]\r116/ 443 (115) [0]\r117/ 443 (116) [0]\r118/ 443 (117) [0]\r119/ 443 (118) [0]\r120/ 443 (119) [0]\r121/ 443 (120) [0]\r122/ 443 (121) [0]\r123/ 443 (122) [0]\r124/ 443 (123) [0]\r125/ 443 (124) [0]\r126/ 443 (125) [0]\r127/ 443 (126) [0]\r128/ 443 (127) [0]\r129/ 443 (128) [0]\r130/ 443 (129) [0]\r131/ 443 (130) [0]\r132/ 443 (131) [0]\r133/ 443 (132) [0]\r134/ 443 (133) [0]\r135/ 443 (134) [0]\r",,terminal_output +2796,86345623,"TERMINAL",0,0,"136/ 443 (135) [0]\r137/ 443 (136) [0]\r138/ 443 (137) [0]\r139/ 443 (138) [0]\r140/ 443 (139) [0]\r141/ 443 (140) [0]\r142/ 443 (141) [0]\r143/ 443 (142) [0]\r144/ 443 (143) [0]\r145/ 443 (144) [0]\r146/ 443 (145) [0]\r147/ 443 (146) [0]\r148/ 443 (147) [0]\r149/ 443 (148) [0]\r150/ 443 (149) [0]\r151/ 443 (150) [0]\r152/ 443 (151) [0]\r153/ 443 (152) [0]\r154/ 443 (153) [0]\r155/ 443 (154) [0]\r156/ 443 (155) [0]\r157/ 443 (156) [0]\r158/ 443 (157) [0]\r159/ 443 (158) [0]\r160/ 443 (159) [0]\r161/ 443 (160) [0]\r162/ 443 (161) [0]\r163/ 443 (162) [0]\r164/ 443 (163) [0]\r165/ 443 (164) [0]\r166/ 443 (165) [0]\r167/ 443 (166) [0]\r168/ 443 (167) [0]\r169/ 443 (168) [0]\r170/ 443 (169) [0]\r171/ 443 (170) [0]\r172/ 443 (171) [0]\r173/ 443 (172) [0]\r174/ 443 (173) [0]\r175/ 443 (174) [0]\r176/ 443 (175) [0]\r177/ 443 (176) [0]\r178/ 443 (177) [0]\r179/ 443 (178) [0]\r180/ 443 (179) [0]\r181/ 443 (180) [0]\r182/ 443 (181) [0]\r183/ 443 (182) [0]\r184/ 443 (183) [0]\r185/ 443 (184) [0]\r186/ 443 (185) [0]\r187/ 443 (186) [0]\r188/ 443 (187) [0]\r189/ 443 (188) [0]\r190/ 443 (189) [0]\r191/ 443 (190) [0]\r192/ 443 (191) [0]\r193/ 443 (192) [0]\r194/ 443 (193) [0]\r195/ 443 (194) [0]\r196/ 443 (195) [0]\r197/ 443 (196) [0]\r198/ 443 (197) [0]\r199/ 443 (198) [0]\r200/ 443 (199) [0]\r201/ 443 (200) [0]\r202/ 443 (201) [0]\r203/ 443 (202) [0]\r204/ 443 (203) [0]\r205/ 443 (204) [0]\r206/ 443 (205) [0]\r207/ 443 (206) [0]\r208/ 443 (207) [0]\r209/ 443 (208) [0]\r210/ 443 (209) [0]\r211/ 443 (210) [0]\r212/ 443 (211) [0]\r213/ 443 (212) [0]\r214/ 443 (213) [0]\r215/ 443 (214) [0]\r216/ 443 (215) [0]\r217/ 443 (216) [0]\r218/ 443 (217) [0]\r219/ 443 (218) [0]\r220/ 443 (219) [0]\r221/ 443 (220) [0]\r222/ 443 (221) [0]\r223/ 443 (222) [0]\r224/ 443 (223) [0]\r225/ 443 (224) [0]\r226/ 443 (225) [0]\r227/ 443 (226) [0]\r228/ 443 (227) [0]\r229/ 443 (228) [0]\r230/ 443 (229) [0]\r231/ 443 (230) [0]\r232/ 443 (231) [0]\r233/ 443 (232) [0]\r234/ 443 (233) [0]\r235/ 443 (234) [0]\r236/ 443 (235) [0]\r237/ 443 (236) [0]\r238/ 443 (237) [0]\r239/ 443 (238) [0]\r240/ 443 (239) [0]\r241/ 443 (240) [0]\r242/ 443 (241) [0]\r243/ 443 (242) [0]\r244/ 443 (243) [0]\r245/ 443 (244) [0]\r246/ 443 (245) [0]\r247/ 443 (246) [0]\r248/ 443 (247) [0]\r249/ 443 (248) [0]\r250/ 443 (249) [0]\r251/ 443 (250) [0]\r252/ 443 (251) [0]\r253/ 443 (252) [0]\r254/ 443 (253) [0]\r255/ 443 (254) [0]\r256/ 443 (255) [0]\r257/ 443 (256) [0]\r258/ 443 (257) [0]\r259/ 443 (258) [0]\r260/ 443 (259) [0]\r261/ 443 (260) [0]\r262/ 443 (261) [0]\r263/ 443 (262) [0]\r264/ 443 (263) [0]\r265/ 443 (264) [0]\r266/ 443 (265) [0]\r267/ 443 (266) [0]\r268/ 443 (267) [0]\r269/ 443 (268) [0]\r270/ 443 (269) [0]\r271/ 443 (270) [0]\r272/ 443 (271) [0]\r273/ 443 (272) [0]\r274/ 443 (273) [0]\r275/ 443 (274) [0]\r276/ 443 (275) [0]\r277/ 443 (276) [0]\r278/ 443 (277) [0]\r279/ 443 (278) [0]\r280/ 443 (279) [0]\r281/ 443 (280) [0]\r282/ 443 (281) [0]\r283/ 443 (282) [0]\r284/ 443 (283) [0]\r285/ 443 (284) [0]\r286/ 443 (285) [0]\r287/ 443 (286) [0]\r288/ 443 (287) [0]\r289/ 443 (288) [0]\r290/ 443 (289) [0]\r291/ 443 (290) [0]\r292/ 443 (291) [0]\r293/ 443 (292) [0]\r294/ 443 (293) [0]\r295/ 443 (294) [0]\r296/ 443 (295) [0]\r297/ 443 (296) [0]\r298/ 443 (297) [0]\r299/ 443 (298) [0]\r300/ 443 (299) [0]\r301/ 443 (300) [0]\r302/ 443 (301) [0]\r303/ 443 (302) [0]\r304/ 443 (303) [0]\r305/ 443 (304) [0]\r306/ 443 (305) [0]\r307/ 443 (306) [0]\r308/ 443 (307) [0]\r309/ 443 (308) [0]\r310/ 443 (309) [0]\r311/ 443 (310) [0]\r312/ 443 (311) [0]\r313/ 443 (312) [0]\r314/ 443 (313) [0]\r315/ 443 (314) [0]\r316/ 443 (315) [0]\r317/ 443 (316) [0]\r318/ 443 (317) [0]\r319/ 443 (318) [0]\r320/ 443 (319) [0]\r321/ 443 (320) [0]\r322/ 443 (321) [0]\r323/ 443 (322) [0]\r324/ 443 (323) [0]\r325/ 443 (324) [0]\r326/ 443 (325) [0]\r327/ 443 (326) [0]\r328/ 443 (327) [0]\r329/ 443 (328) [0]\r330/ 443 (329) [0]\r331/ 443 (330) [0]\r332/ 443 (331) [0]\r333/ 443 (332) [0]\r334/ 443 (333) [0]\r335/ 443 (334) [0]\r336/ 443 (335) [0]\r337/ 443 (336) [0]\r338/ 443 (337) [0]\r339/ 443 (338) [0]\r340/ 443 (339) [0]\r341/ 443 (340) [0]\r342/ 443 (341) [0]\r343/ 443 (342) [0]\r344/ 443 (343) [0]\r345/ 443 (344) [0]\r346/ 443 (345) [0]\r347/ 443 (346) [0]\r348/ 443 (347) [0]\r349/ 443 (348) [0]\r350/ 443 (349) [0]\r351/ 443 (350) [0]\r352/ 443 (351) [0]\r353/ 443 (352) [0]\r354/ 443 (353) [0]\r355/ 443 (354) [0]\r356/ 443 (355) [0]\r357/ 443 (356) [0]\r358/ 443 (357) [0]\r359/ 443 (358) [0]\r360/ 443 (359) [0]\r361/ 443 (360) [0]\r362/ 443 (361) [0]\r363/ 443 (362) [0]\r364/ 443 (363) [0]\r365/ 443 (364) [0]\r366/ 443 (365) [0]\r367/ 443 (366) [0]\r368/ 443 (367) [0]\r369/ 443 (368) [0]\r370/ 443 (369) [0]\r371/ 443 (370) [0]\r372/ 443 (371) [0]\r373/ 443 (372) [0]\r374/ 443 (373) [0]\r375/ 443 (374) [0]\r376/ 443 (375) [0]\r377/ 443 (376) [0]\r378/ 443 (377) [0]\r379/ 443 (378) [0]\r380/ 443 (379) [0]\r381/ 443 (380) [0]\r382/ 443 (381) [0]\r383/ 443 (382) [0]\r384/ 443 (383) [0]\r385/ 443 (384) [0]\r386/ 443 (385) [0]\r387/ 443 (386) [0]\r388/ 443 (387) [0]\r389/ 443 (388) [0]\r390/ 443 (389) [0]\r391/ 443 (390) [0]\r392/ 443 (391) [0]\r393/ 443 (392) [0]\r394/ 443 (393) [0]\r395/ 443 (394) [0]\r396/ 443 (395) [0]\r397/ 443 (396) [0]\r398/ 443 (397) [0]\r399/ 443 (398) [0]\r400/ 443 (399) [0]\r401/ 443 (400) [0]\r402/ 443 (401) [0]\r403/ 443 (402) [0]\r404/ 443 (403) [0]\r405/ 443 (404) [0]\r406/ 443 (405) [0]\r407/ 443 (406) [0]\r408/ 443 (407) [0]\r409/ 443 (408) [0]\r410/ 443 (409) [0]\r411/ 443 (410) [0]\r412/ 443 (411) [0]\r413/ 443 (412) [0]\r414/ 443 (413) [0]\r415/ 443 (414) [0]\r416/ 443 (415) [0]\r417/ 443 (416) [0]\r418/ 443 (417) [0]\r419/ 443 (418) [0]\r420/ 443 (419) [0]\r421/ 443 (420) [0]\r422/ 443 (421) [0]\r423/ 443 (422) [0]\r424/ 443 (423) [0]\r425/ 443 (424) [0]\r426/ 443 (425) [0]\r427/ 443 (426) [0]\r428/ 443 (427) [0]\r429/ 443 (428) [0]\r430/ 443 (429) [0]\r431/ 443 (430) [0]\r432/ 443 (431) [0]\r433/ 443 (432) [0]\r434/ 443 (433) [0]\r435/ 443 (434) [0]\r436/ 443 (435) [0]\r437/ 443 (436) [0]\r438/ 443 (437) [0]\r439/ 443 (438) [0]\r440/ 443 (439) [0]\r441/ 443 (440) [0]\r442/ 443 (441) [0]\r443/ 443 (442) [0]\r",,terminal_output +2797,86346131,"TERMINAL",0,0,"Enumerating objects: 5, done.\r\nCounting objects: 20% (1/5)\rCounting objects: 40% (2/5)\rCounting objects: 60% (3/5)\rCounting objects: 80% (4/5)\rCounting objects: 100% (5/5)\rCounting objects: 100% (5/5), done.\r\nDelta compression using up to 8 threads\r\nCompressing objects: 33% (1/3)\rCompressing objects: 66% (2/3)\rCompressing objects: 100% (3/3)\rCompressing objects: 100% (3/3), done.\r\nWriting objects: 33% (1/3)\rWriting objects: 66% (2/3)\rWriting objects: 100% (3/3)\rWriting objects: 100% (3/3), 323 bytes | 323.00 KiB/s, done.\r\nTotal 3 (delta 2), reused 0 (delta 0), pack-reused 0\r\n",,terminal_output +2798,86346627,"TERMINAL",0,0,"remote: Resolving deltas: 0% (0/2)\rremote: Resolving deltas: 50% (1/2)\rremote: Resolving deltas: 100% (2/2)\rremote: Resolving deltas: 100% (2/2), completed with 2 local objects.\r\n",,terminal_output +2799,86346817,"TERMINAL",0,0,"To https://github.com/emergenz/pdoom.org\r\n 9ebe017..c414c90 c414c900b3fdbbf1c5a5f8144510f5107a20a104 -> gh-pages\r\n% \r \r",,terminal_output +2800,86414408,"examples/jasmine.html",10548,0,"",html,selection_command +2801,86418272,"examples/jasmine.html",10540,0,"",html,selection_command +2802,86418522,"examples/jasmine.html",10539,0,"",html,selection_command +2803,86418554,"examples/jasmine.html",10503,0,"",html,selection_command +2804,86418586,"examples/jasmine.html",10502,0,"",html,selection_command +2805,86418620,"examples/jasmine.html",10486,0,"",html,selection_command +2806,86418653,"examples/jasmine.html",10462,0,"",html,selection_command +2807,86418790,"examples/jasmine.html",10439,0,"",html,selection_command +2808,86418955,"examples/jasmine.html",10378,0,"",html,selection_command +2809,86419204,"examples/jasmine.html",10244,0,"",html,selection_command +2810,86429361,"dist/template.v2.js",0,0,"(function (factory) {\n typeof define === 'function' && define.amd ? define(factory) :\n factory();\n}((function () { 'use strict';\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n const days = ['Sunday', 'Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday'];\n const months = ['Jan.', 'Feb.', 'March', 'April', 'May', 'June', 'July', 'Aug.', 'Sept.', 'Oct.', 'Nov.', 'Dec.'];\n const zeroPad = n => n < 10 ? '0' + n : n;\n\n const RFC = function(date) {\n const day = days[date.getDay()].substring(0, 3);\n const paddedDate = zeroPad(date.getDate());\n const month = months[date.getMonth()].substring(0,3);\n const year = date.getFullYear().toString();\n const hours = date.getUTCHours().toString();\n const minutes = date.getUTCMinutes().toString();\n const seconds = date.getUTCSeconds().toString();\n return `${day}, ${paddedDate} ${month} ${year} ${hours}:${minutes}:${seconds} Z`;\n };\n\n const objectFromMap = function(map) {\n const object = Array.from(map).reduce((object, [key, value]) => (\n Object.assign(object, { [key]: value }) // Be careful! Maps can have non-String keys; object literals can't.\n ), {});\n return object;\n };\n\n const mapFromObject = function(object) {\n const map = new Map();\n for (var property in object) {\n if (object.hasOwnProperty(property)) {\n map.set(property, object[property]);\n }\n }\n return map;\n };\n\n class Author {\n\n // constructor(name='', personalURL='', affiliation='', affiliationURL='') {\n // this.name = name; // 'Chris Olah'\n // this.personalURL = personalURL; // 'https://colah.github.io'\n // this.affiliation = affiliation; // 'Google Brain'\n // this.affiliationURL = affiliationURL; // 'https://g.co/brain'\n // }\n\n constructor(object) {\n this.name = object.author; // 'Chris Olah'\n this.personalURL = object.authorURL; // 'https://colah.github.io'\n this.affiliation = object.affiliation; // 'Google Brain'\n this.affiliationURL = object.affiliationURL; // 'https://g.co/brain'\n this.affiliations = object.affiliations || []; // new-style affiliations\n }\n\n // 'Chris'\n get firstName() {\n const names = this.name.split(' ');\n return names.slice(0, names.length - 1).join(' ');\n }\n\n // 'Olah'\n get lastName() {\n const names = this.name.split(' ');\n return names[names.length -1];\n }\n }\n\n function mergeFromYMLFrontmatter(target, source) {\n target.title = source.title;\n if (source.published) {\n if (source.published instanceof Date) {\n target.publishedDate = source.published;\n } else if (source.published.constructor === String) {\n target.publishedDate = new Date(source.published);\n }\n }\n if (source.publishedDate) {\n if (source.publishedDate instanceof Date) {\n target.publishedDate = source.publishedDate;\n } else if (source.publishedDate.constructor === String) {\n target.publishedDate = new Date(source.publishedDate);\n } else {\n console.error('Don\'t know what to do with published date: ' + source.publishedDate);\n }\n }\n target.description = source.description;\n target.authors = source.authors.map( (authorObject) => new Author(authorObject));\n target.katex = source.katex;\n target.password = source.password;\n if (source.doi) {\n target.doi = source.doi;\n }\n }\n\n class FrontMatter {\n constructor() {\n this.title = 'unnamed article'; // 'Attention and Augmented Recurrent Neural Networks'\n this.description = ''; // 'A visual overview of neural attention...'\n this.authors = []; // Array of Author(s)\n\n this.bibliography = new Map();\n this.bibliographyParsed = false;\n // {\n // 'gregor2015draw': {\n // 'title': 'DRAW: A recurrent neural network for image generation',\n // 'author': 'Gregor, Karol and Danihelka, Ivo and Graves, Alex and Rezende, Danilo Jimenez and Wierstra, Daan',\n // 'journal': 'arXiv preprint arXiv:1502.04623',\n // 'year': '2015',\n // 'url': 'https://arxiv.org/pdf/1502.04623.pdf',\n // 'type': 'article'\n // },\n // }\n\n // Citation keys should be listed in the order that they are appear in the document.\n // Each key refers to a key in the bibliography dictionary.\n this.citations = []; // [ 'gregor2015draw', 'mercier2011humans' ]\n this.citationsCollected = false;\n\n //\n // Assigned from posts.csv\n //\n\n // publishedDate: 2016-09-08T07:00:00.000Z,\n // tags: [ 'rnn' ],\n // distillPath: '2016/augmented-rnns',\n // githubPath: 'distillpub/post--augmented-rnns',\n // doiSuffix: 1,\n\n //\n // Assigned from journal\n //\n this.journal = {};\n // journal: {\n // 'title': 'Distill',\n // 'full_title': 'Distill',\n // 'abbrev_title': 'Distill',\n // 'url': 'http://distill.pub',\n // 'doi': '10.23915/distill',\n // 'publisherName': 'Distill Working Group',\n // 'publisherEmail': 'admin@distill.pub',\n // 'issn': '2476-0757',\n // 'editors': [...],\n // 'committee': [...]\n // }\n // volume: 1,\n // issue: 9,\n\n this.katex = {};\n\n //\n // Assigned from publishing process\n //\n\n // githubCompareUpdatesUrl: 'https://github.com/distillpub/post--augmented-rnns/compare/1596e094d8943d2dc0ea445d92071129c6419c59...3bd9209e0c24d020f87cf6152dcecc6017cbc193',\n // updatedDate: 2017-03-21T07:13:16.000Z,\n // doi: '10.23915/distill.00001',\n this.doi = undefined;\n this.publishedDate = undefined;\n }\n\n // Example:\n // title: Demo Title Attention and Augmented Recurrent Neural Networks\n // published: Jan 10, 2017\n // authors:\n // - Chris Olah:\n // - Shan Carter: http://shancarter.com\n // affiliations:\n // - Google Brain:\n // - Google Brain: http://g.co/brain\n\n //\n // Computed Properties\n //\n\n // 'http://distill.pub/2016/augmented-rnns',\n set url(value) {\n this._url = value;\n }\n get url() {\n if (this._url) {\n return this._url;\n } else if (this.distillPath && this.journal.url) {\n return this.journal.url + '/' + this.distillPath;\n } else if (this.journal.url) {\n return this.journal.url;\n }\n }\n\n // 'https://github.com/distillpub/post--augmented-rnns',\n get githubUrl() {\n if (this.githubPath) {\n return 'https://github.com/' + this.githubPath;\n } else {\n return undefined;\n }\n }\n\n // TODO resolve differences in naming of URL/Url/url.\n // 'http://distill.pub/2016/augmented-rnns/thumbnail.jpg',\n set previewURL(value) {\n this._previewURL = value;\n }\n get previewURL() {\n return this._previewURL ? this._previewURL : this.url + '/thumbnail.jpg';\n }\n\n // 'Thu, 08 Sep 2016 00:00:00 -0700',\n get publishedDateRFC() {\n return RFC(this.publishedDate);\n }\n\n // 'Thu, 08 Sep 2016 00:00:00 -0700',\n get updatedDateRFC() {\n return RFC(this.updatedDate);\n }\n\n // 2016,\n get publishedYear() {\n return this.publishedDate.getFullYear();\n }\n\n // 'Sept',\n get publishedMonth() {\n return months[this.publishedDate.getMonth()];\n }\n\n // 8,\n get publishedDay() {\n return this.publishedDate.getDate();\n }\n\n // '09',\n get publishedMonthPadded() {\n return zeroPad(this.publishedDate.getMonth() + 1);\n }\n\n // '08',\n get publishedDayPadded() {\n return zeroPad(this.publishedDate.getDate());\n }\n\n get publishedISODateOnly() {\n return this.publishedDate.toISOString().split('T')[0];\n }\n\n get volume() {\n const volume = this.publishedYear - 2015;\n if (volume < 1) {\n throw new Error('Invalid publish date detected during computing volume');\n }\n return volume;\n }\n\n get issue() {\n return this.publishedDate.getMonth() + 1;\n }\n\n // 'Olah & Carter',\n get concatenatedAuthors() {\n if (this.authors.length > 2) {\n return this.authors[0].lastName + ', et al.';\n } else if (this.authors.length === 2) {\n return this.authors[0].lastName + ' & ' + this.authors[1].lastName;\n } else if (this.authors.length === 1) {\n return this.authors[0].lastName;\n }\n }\n\n // 'Olah, Chris and Carter, Shan',\n get bibtexAuthors() {\n return this.authors.map(author => {\n return author.lastName + ', ' + author.firstName;\n }).join(' and ');\n }\n\n // 'olah2016attention'\n get slug() {\n let slug = '';\n if (this.authors.length) {\n slug += this.authors[0].lastName.toLowerCase();\n slug += this.publishedYear;\n slug += this.title.split(' ')[0].toLowerCase();\n }\n return slug || 'Untitled';\n }\n\n get bibliographyEntries() {\n return new Map(this.citations.map( citationKey => {\n const entry = this.bibliography.get(citationKey);\n return [citationKey, entry];\n }));\n }\n\n set bibliography(bibliography) {\n if (bibliography instanceof Map) {\n this._bibliography = bibliography;\n } else if (typeof bibliography === 'object') {\n this._bibliography = mapFromObject(bibliography);\n }\n }\n\n get bibliography() {\n return this._bibliography;\n }\n\n static fromObject(source) {\n const frontMatter = new FrontMatter();\n Object.assign(frontMatter, source);\n return frontMatter;\n }\n\n assignToObject(target) {\n Object.assign(target, this);\n target.bibliography = objectFromMap(this.bibliographyEntries);\n target.url = this.url;\n target.doi = this.doi;\n target.githubUrl = this.githubUrl;\n target.previewURL = this.previewURL;\n if (this.publishedDate) {\n target.volume = this.volume;\n target.issue = this.issue;\n target.publishedDateRFC = this.publishedDateRFC;\n target.publishedYear = this.publishedYear;\n target.publishedMonth = this.publishedMonth;\n target.publishedDay = this.publishedDay;\n target.publishedMonthPadded = this.publishedMonthPadded;\n target.publishedDayPadded = this.publishedDayPadded;\n }\n if (this.updatedDate) {\n target.updatedDateRFC = this.updatedDateRFC;\n }\n target.concatenatedAuthors = this.concatenatedAuthors;\n target.bibtexAuthors = this.bibtexAuthors;\n target.slug = this.slug;\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n const Mutating = (superclass) => {\n return class extends superclass {\n\n constructor() {\n super();\n\n // set up mutation observer\n const options = {childList: true, characterData: true, subtree: true};\n const observer = new MutationObserver( () => {\n observer.disconnect();\n this.renderIfPossible();\n observer.observe(this, options);\n });\n\n // ...and listen for changes\n observer.observe(this, options);\n }\n\n connectedCallback() {\n super.connectedCallback();\n\n this.renderIfPossible();\n }\n\n // potential TODO: check if this is enough for all our usecases\n // maybe provide a custom function to tell if we have enough information to render\n renderIfPossible() {\n if (this.textContent && this.root) {\n this.renderContent();\n }\n }\n\n renderContent() {\n console.error(`Your class ${this.constructor.name} must provide a custom renderContent() method!` );\n }\n\n }; // end class\n }; // end mixin function\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n /*global ShadyCSS*/\n\n const Template = (name, templateString, useShadow = true) => {\n\n return (superclass) => {\n\n const template = document.createElement('template');\n template.innerHTML = templateString;\n\n if (useShadow && 'ShadyCSS' in window) {\n ShadyCSS.prepareTemplate(template, name);\n }\n\n return class extends superclass {\n\n static get is() { return name; }\n\n constructor() {\n super();\n\n this.clone = document.importNode(template.content, true);\n if (useShadow) {\n this.attachShadow({mode: 'open'});\n this.shadowRoot.appendChild(this.clone);\n }\n }\n\n connectedCallback() {\n if (this.hasAttribute('distill-prerendered')) {\n return;\n }\n if (useShadow) {\n if ('ShadyCSS' in window) {\n ShadyCSS.styleElement(this);\n }\n } else {\n this.insertBefore(this.clone, this.firstChild);\n }\n }\n\n get root() {\n if (useShadow) {\n return this.shadowRoot;\n } else {\n return this;\n }\n }\n\n /* TODO: Are we using these? Should we even? */\n $(query) {\n return this.root.querySelector(query);\n }\n\n $$(query) {\n return this.root.querySelectorAll(query);\n }\n };\n };\n };\n\n var math = ""/*\n * Copyright 2018 The Distill Template Authors\n *\n * Licensed under the Apache License, Version 2.0 (the \""License\"");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \""AS IS\"" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n */\n\nspan.katex-display {\n text-align: left;\n padding: 8px 0 8px 0;\n margin: 0.5em 0 0.5em 1em;\n}\n\nspan.katex {\n -webkit-font-smoothing: antialiased;\n color: rgba(0, 0, 0, 0.8);\n font-size: 1.18em;\n}\n"";\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n // This is a straight concatenation of code from KaTeX's contrib folder,\n // but we aren't using some of their helpers that don't work well outside a browser environment.\n\n /*global katex */\n\n const findEndOfMath = function(delimiter, text, startIndex) {\n // Adapted from\n // https://github.com/Khan/perseus/blob/master/src/perseus-markdown.jsx\n let index = startIndex;\n let braceLevel = 0;\n\n const delimLength = delimiter.length;\n\n while (index < text.length) {\n const character = text[index];\n\n if (\n braceLevel <= 0 &&\n text.slice(index, index + delimLength) === delimiter\n ) {\n return index;\n } else if (character === ""\\"") {\n index++;\n } else if (character === ""{"") {\n braceLevel++;\n } else if (character === ""}"") {\n braceLevel--;\n }\n\n index++;\n }\n\n return -1;\n };\n\n const splitAtDelimiters = function(startData, leftDelim, rightDelim, display) {\n const finalData = [];\n\n for (let i = 0; i < startData.length; i++) {\n if (startData[i].type === ""text"") {\n const text = startData[i].data;\n\n let lookingForLeft = true;\n let currIndex = 0;\n let nextIndex;\n\n nextIndex = text.indexOf(leftDelim);\n if (nextIndex !== -1) {\n currIndex = nextIndex;\n finalData.push({\n type: ""text"",\n data: text.slice(0, currIndex)\n });\n lookingForLeft = false;\n }\n\n while (true) {\n // eslint-disable-line no-constant-condition\n if (lookingForLeft) {\n nextIndex = text.indexOf(leftDelim, currIndex);\n if (nextIndex === -1) {\n break;\n }\n\n finalData.push({\n type: ""text"",\n data: text.slice(currIndex, nextIndex)\n });\n\n currIndex = nextIndex;\n } else {\n nextIndex = findEndOfMath(\n rightDelim,\n text,\n currIndex + leftDelim.length\n );\n if (nextIndex === -1) {\n break;\n }\n\n finalData.push({\n type: ""math"",\n data: text.slice(currIndex + leftDelim.length, nextIndex),\n rawData: text.slice(currIndex, nextIndex + rightDelim.length),\n display: display\n });\n\n currIndex = nextIndex + rightDelim.length;\n }\n\n lookingForLeft = !lookingForLeft;\n }\n\n finalData.push({\n type: ""text"",\n data: text.slice(currIndex)\n });\n } else {\n finalData.push(startData[i]);\n }\n }\n\n return finalData;\n };\n\n const splitWithDelimiters = function(text, delimiters) {\n let data = [{ type: ""text"", data: text }];\n for (let i = 0; i < delimiters.length; i++) {\n const delimiter = delimiters[i];\n data = splitAtDelimiters(\n data,\n delimiter.left,\n delimiter.right,\n delimiter.display || false\n );\n }\n return data;\n };\n\n /* Note: optionsCopy is mutated by this method. If it is ever exposed in the\n * API, we should copy it before mutating.\n */\n const renderMathInText = function(text, optionsCopy) {\n const data = splitWithDelimiters(text, optionsCopy.delimiters);\n const fragment = document.createDocumentFragment();\n\n for (let i = 0; i < data.length; i++) {\n if (data[i].type === ""text"") {\n fragment.appendChild(document.createTextNode(data[i].data));\n } else {\n const tag = document.createElement(""d-math"");\n const math = data[i].data;\n // Override any display mode defined in the settings with that\n // defined by the text itself\n optionsCopy.displayMode = data[i].display;\n try {\n tag.textContent = math;\n if (optionsCopy.displayMode) {\n tag.setAttribute(""block"", """");\n }\n } catch (e) {\n if (!(e instanceof katex.ParseError)) {\n throw e;\n }\n optionsCopy.errorCallback(\n ""KaTeX auto-render: Failed to parse `"" + data[i].data + ""` with "",\n e\n );\n fragment.appendChild(document.createTextNode(data[i].rawData));\n continue;\n }\n fragment.appendChild(tag);\n }\n }\n\n return fragment;\n };\n\n const renderElem = function(elem, optionsCopy) {\n for (let i = 0; i < elem.childNodes.length; i++) {\n const childNode = elem.childNodes[i];\n if (childNode.nodeType === 3) {\n // Text node\n const text = childNode.textContent;\n if (optionsCopy.mightHaveMath(text)) {\n const frag = renderMathInText(text, optionsCopy);\n i += frag.childNodes.length - 1;\n elem.replaceChild(frag, childNode);\n }\n } else if (childNode.nodeType === 1) {\n // Element node\n const shouldRender =\n optionsCopy.ignoredTags.indexOf(childNode.nodeName.toLowerCase()) ===\n -1;\n\n if (shouldRender) {\n renderElem(childNode, optionsCopy);\n }\n }\n // Otherwise, it's something else, and ignore it.\n }\n };\n\n const defaultAutoRenderOptions = {\n delimiters: [\n { left: ""$$"", right: ""$$"", display: true },\n { left: ""\\["", right: ""\\]"", display: true },\n { left: ""\\("", right: ""\\)"", display: false }\n // LaTeX uses this, but it ruins the display of normal `$` in text:\n // {left: '$', right: '$', display: false},\n ],\n\n ignoredTags: [\n ""script"",\n ""noscript"",\n ""style"",\n ""textarea"",\n ""pre"",\n ""code"",\n ""svg""\n ],\n\n errorCallback: function(msg, err) {\n console.error(msg, err);\n }\n };\n\n const renderMathInElement = function(elem, options) {\n if (!elem) {\n throw new Error(""No element provided to render"");\n }\n\n const optionsCopy = Object.assign({}, defaultAutoRenderOptions, options);\n const delimiterStrings = optionsCopy.delimiters.flatMap(d => [\n d.left,\n d.right\n ]);\n const mightHaveMath = text =>\n delimiterStrings.some(d => text.indexOf(d) !== -1);\n optionsCopy.mightHaveMath = mightHaveMath;\n renderElem(elem, optionsCopy);\n };\n\n // Copyright 2018 The Distill Template Authors\n\n const katexJSURL = 'https://distill.pub/third-party/katex/katex.min.js';\n const katexCSSTag = '';\n\n const T = Template('d-math', `\n${katexCSSTag}\n\n\n`);\n\n // DMath, not Math, because that would conflict with the JS built-in\n class DMath extends Mutating(T(HTMLElement)) {\n\n static set katexOptions(options) {\n DMath._katexOptions = options;\n if (DMath.katexOptions.delimiters) {\n if (!DMath.katexAdded) {\n DMath.addKatex();\n } else {\n DMath.katexLoadedCallback();\n }\n }\n }\n\n static get katexOptions() {\n if (!DMath._katexOptions) {\n DMath._katexOptions = {\n delimiters: [ { 'left':'$$', 'right':'$$', 'display': false } ]\n };\n }\n return DMath._katexOptions;\n }\n\n static katexLoadedCallback() {\n // render all d-math tags\n const mathTags = document.querySelectorAll('d-math');\n for (const mathTag of mathTags) {\n mathTag.renderContent();\n }\n // transform inline delimited math to d-math tags\n if (DMath.katexOptions.delimiters) {\n renderMathInElement(document.body, DMath.katexOptions);\n }\n }\n\n static addKatex() {\n // css tag can use this convenience function\n document.head.insertAdjacentHTML('beforeend', katexCSSTag);\n // script tag has to be created to work properly\n const scriptTag = document.createElement('script');\n scriptTag.src = katexJSURL;\n scriptTag.async = true;\n scriptTag.onload = DMath.katexLoadedCallback;\n scriptTag.crossorigin = 'anonymous';\n document.head.appendChild(scriptTag);\n\n DMath.katexAdded = true;\n }\n\n get options() {\n const localOptions = { displayMode: this.hasAttribute('block') };\n return Object.assign(localOptions, DMath.katexOptions);\n }\n\n connectedCallback() {\n super.connectedCallback();\n if (!DMath.katexAdded) {\n DMath.addKatex();\n }\n }\n\n renderContent() {\n if (typeof katex !== 'undefined') {\n const container = this.root.querySelector('#katex-container');\n katex.render(this.textContent, container, this.options);\n }\n }\n\n }\n\n DMath.katexAdded = false;\n DMath.inlineMathRendered = false;\n window.DMath = DMath; // TODO: check if this can be removed, or if we should expose a distill global\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n function collect_citations(dom = document) {\n const citations = new Set();\n const citeTags = dom.querySelectorAll(""d-cite"");\n for (const tag of citeTags) {\n const keyString = tag.getAttribute(""key"") || tag.getAttribute(""bibtex-key"");\n const keys = keyString.split("","").map(k => k.trim());\n for (const key of keys) {\n citations.add(key);\n }\n }\n return [...citations];\n }\n\n function author_string(ent, template, sep, finalSep) {\n if (ent.author == null) {\n return """";\n }\n var names = ent.author.split("" and "");\n let name_strings = names.map(name => {\n name = name.trim();\n if (name.indexOf("","") != -1) {\n var last = name.split("","")[0].trim();\n var firsts = name.split("","")[1];\n } else if (name.indexOf("" "") != -1) {\n var last = name\n .split("" "")\n .slice(-1)[0]\n .trim();\n var firsts = name\n .split("" "")\n .slice(0, -1)\n .join("" "");\n } else {\n var last = name.trim();\n }\n var initials = """";\n if (firsts != undefined) {\n initials = firsts\n .trim()\n .split("" "")\n .map(s => s.trim()[0]);\n initials = initials.join(""."") + ""."";\n }\n return template\n .replace(""${F}"", firsts)\n .replace(""${L}"", last)\n .replace(""${I}"", initials)\n .trim(); // in case one of first or last was empty\n });\n if (names.length > 1) {\n var str = name_strings.slice(0, names.length - 1).join(sep);\n str += (finalSep || sep) + name_strings[names.length - 1];\n return str;\n } else {\n return name_strings[0];\n }\n }\n\n function venue_string(ent) {\n var cite = ent.journal || ent.booktitle || """";\n if (""volume"" in ent) {\n var issue = ent.issue || ent.number;\n issue = issue != undefined ? ""("" + issue + "")"" : """";\n cite += "", Vol "" + ent.volume + issue;\n }\n if (""pages"" in ent) {\n cite += "", pp. "" + ent.pages;\n }\n if (cite != """") cite += "". "";\n if (""publisher"" in ent) {\n cite += ent.publisher;\n if (cite[cite.length - 1] != ""."") cite += ""."";\n }\n return cite;\n }\n\n function link_string(ent) {\n if (""url"" in ent) {\n var url = ent.url;\n var arxiv_match = /arxiv\.org\/abs\/([0-9\.]*)/.exec(url);\n if (arxiv_match != null) {\n url = `http://arxiv.org/pdf/${arxiv_match[1]}.pdf`;\n }\n\n if (url.slice(-4) == "".pdf"") {\n var label = ""PDF"";\n } else if (url.slice(-5) == "".html"") {\n var label = ""HTML"";\n }\n return `  [${label || ""link""}]`;\n } /* else if (""doi"" in ent){\n return `  [DOI]`;\n }*/ else {\n return """";\n }\n }\n function doi_string(ent, new_line) {\n if (""doi"" in ent) {\n return `${new_line ? ""
"" : """"} DOI: ${ent.doi}`;\n } else {\n return """";\n }\n }\n\n function title_string(ent) {\n return '' + ent.title + "" "";\n }\n\n function bibliography_cite(ent, fancy) {\n if (ent) {\n var cite = title_string(ent);\n cite += link_string(ent) + ""
"";\n if (ent.author) {\n cite += author_string(ent, ""${L}, ${I}"", "", "", "" and "");\n if (ent.year || ent.date) {\n cite += "", "";\n }\n }\n if (ent.year || ent.date) {\n cite += (ent.year || ent.date) + "". "";\n } else {\n cite += "". "";\n }\n cite += venue_string(ent);\n cite += doi_string(ent);\n return cite;\n /*var cite = author_string(ent, ""${L}, ${I}"", "", "", "" and "");\n if (ent.year || ent.date){\n cite += "", "" + (ent.year || ent.date) + "". ""\n } else {\n cite += "". ""\n }\n cite += """" + ent.title + "". "";\n cite += venue_string(ent);\n cite += doi_string(ent);\n cite += link_string(ent);\n return cite*/\n } else {\n return ""?"";\n }\n }\n\n function hover_cite(ent) {\n if (ent) {\n var cite = """";\n cite += """" + ent.title + """";\n cite += link_string(ent);\n cite += ""
"";\n\n var a_str = author_string(ent, ""${I} ${L}"", "", "") + ""."";\n var v_str =\n venue_string(ent).trim() + "" "" + ent.year + "". "" + doi_string(ent, true);\n\n if ((a_str + v_str).length < Math.min(40, ent.title.length)) {\n cite += a_str + "" "" + v_str;\n } else {\n cite += a_str + ""
"" + v_str;\n }\n return cite;\n } else {\n return ""?"";\n }\n }\n\n function domContentLoaded() {\n return ['interactive', 'complete'].indexOf(document.readyState) !== -1;\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n function _moveLegacyAffiliationFormatIntoArray(frontMatter) {\n // authors used to have propoerties ""affiliation"" and ""affiliationURL"".\n // We now encourage using an array for affiliations containing objects with\n // properties ""name"" and ""url"".\n for (let author of frontMatter.authors) {\n const hasOldStyle = Boolean(author.affiliation);\n const hasNewStyle = Boolean(author.affiliations);\n if (!hasOldStyle) continue;\n if (hasNewStyle) {\n console.warn(`Author ${author.author} has both old-style (""affiliation"" & ""affiliationURL"") and new style (""affiliations"") affiliation information!`);\n } else {\n let newAffiliation = {\n ""name"": author.affiliation\n };\n if (author.affiliationURL) newAffiliation.url = author.affiliationURL;\n author.affiliations = [newAffiliation];\n }\n }\n return frontMatter\n }\n\n function parseFrontmatter(element) {\n const scriptTag = element.firstElementChild;\n if (scriptTag) {\n const type = scriptTag.getAttribute('type');\n if (type.split('/')[1] == 'json') {\n const content = scriptTag.textContent;\n const parsed = JSON.parse(content);\n return _moveLegacyAffiliationFormatIntoArray(parsed);\n } else {\n console.error('Distill only supports JSON frontmatter tags anymore; no more YAML.');\n }\n } else {\n console.error('You added a frontmatter tag but did not provide a script tag with front matter data in it. Please take a look at our templates.');\n }\n return {};\n }\n\n class FrontMatter$1 extends HTMLElement {\n\n static get is() { return 'd-front-matter'; }\n\n constructor() {\n super();\n\n const options = {childList: true, characterData: true, subtree: true};\n const observer = new MutationObserver( (entries) => {\n for (const entry of entries) {\n if (entry.target.nodeName === 'SCRIPT' || entry.type === 'characterData') {\n const data = parseFrontmatter(this);\n this.notify(data);\n }\n }\n });\n observer.observe(this, options);\n }\n\n notify(data) {\n const options = { detail: data, bubbles: true };\n const event = new CustomEvent('onFrontMatterChanged', options);\n document.dispatchEvent(event);\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n // no appendix -> add appendix\n // title in front, no h1 -> add it\n // no title in front, h1 -> read and put into frontMatter\n // footnote -> footnote list\n // break up bib\n // if citation, no bib-list -> add citation-list\n\n // if authors, no byline -> add byline\n\n function optionalComponents(dom, data) {\n const body = dom.body;\n const article = body.querySelector('d-article');\n\n // If we don't have an article tag, something weird is going on—giving up.\n if (!article) {\n console.warn('No d-article tag found; skipping adding optional components!');\n return;\n }\n\n let byline = dom.querySelector('d-byline');\n if (!byline) {\n if (data.authors) {\n byline = dom.createElement('d-byline');\n body.insertBefore(byline, article);\n } else {\n console.warn('No authors found in front matter; please add them before submission!');\n }\n }\n\n let title = dom.querySelector('d-title');\n if (!title) {\n title = dom.createElement('d-title');\n body.insertBefore(title, byline);\n }\n\n let h1 = title.querySelector('h1');\n if (!h1) {\n h1 = dom.createElement('h1');\n h1.textContent = data.title;\n title.insertBefore(h1, title.firstChild);\n }\n\n const hasPassword = typeof data.password !== 'undefined';\n let interstitial = body.querySelector('d-interstitial');\n if (hasPassword && !interstitial) {\n const inBrowser = typeof window !== 'undefined';\n const onLocalhost = inBrowser && window.location.hostname.includes('localhost');\n if (!inBrowser || !onLocalhost) {\n interstitial = dom.createElement('d-interstitial');\n interstitial.password = data.password;\n body.insertBefore(interstitial, body.firstChild);\n }\n } else if (!hasPassword && interstitial) {\n interstitial.parentElement.removeChild(this);\n }\n\n let appendix = dom.querySelector('d-appendix');\n if (!appendix) {\n appendix = dom.createElement('d-appendix');\n dom.body.appendChild(appendix);\n }\n\n let footnoteList = dom.querySelector('d-footnote-list');\n if (!footnoteList) {\n footnoteList = dom.createElement('d-footnote-list');\n appendix.appendChild(footnoteList);\n }\n\n let citationList = dom.querySelector('d-citation-list');\n if (!citationList) {\n citationList = dom.createElement('d-citation-list');\n appendix.appendChild(citationList);\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const frontMatter = new FrontMatter();\n\n const Controller = {\n frontMatter: frontMatter,\n waitingOn: {\n bibliography: [],\n citations: []\n },\n listeners: {\n onCiteKeyCreated(event) {\n const [citeTag, keys] = event.detail;\n\n // ensure we have citations\n if (!frontMatter.citationsCollected) {\n // console.debug('onCiteKeyCreated, but unresolved dependency (""citations""). Enqueing.');\n Controller.waitingOn.citations.push(() =>\n Controller.listeners.onCiteKeyCreated(event)\n );\n return;\n }\n\n // ensure we have a loaded bibliography\n if (!frontMatter.bibliographyParsed) {\n // console.debug('onCiteKeyCreated, but unresolved dependency (""bibliography""). Enqueing.');\n Controller.waitingOn.bibliography.push(() =>\n Controller.listeners.onCiteKeyCreated(event)\n );\n return;\n }\n\n const numbers = keys.map(key => frontMatter.citations.indexOf(key));\n citeTag.numbers = numbers;\n const entries = keys.map(key => frontMatter.bibliography.get(key));\n citeTag.entries = entries;\n },\n\n onCiteKeyChanged() {\n // const [citeTag, keys] = event.detail;\n\n // update citations\n frontMatter.citations = collect_citations();\n frontMatter.citationsCollected = true;\n for (const waitingCallback of Controller.waitingOn.citations.slice()) {\n waitingCallback();\n }\n\n // update bibliography\n const citationListTag = document.querySelector(""d-citation-list"");\n const bibliographyEntries = new Map(\n frontMatter.citations.map(citationKey => {\n return [citationKey, frontMatter.bibliography.get(citationKey)];\n })\n );\n citationListTag.citations = bibliographyEntries;\n\n const citeTags = document.querySelectorAll(""d-cite"");\n for (const citeTag of citeTags) {\n console.log(citeTag);\n const keys = citeTag.keys;\n const numbers = keys.map(key => frontMatter.citations.indexOf(key));\n citeTag.numbers = numbers;\n const entries = keys.map(key => frontMatter.bibliography.get(key));\n citeTag.entries = entries;\n }\n },\n\n onCiteKeyRemoved(event) {\n Controller.listeners.onCiteKeyChanged(event);\n },\n\n onBibliographyChanged(event) {\n const citationListTag = document.querySelector(""d-citation-list"");\n\n const bibliography = event.detail;\n\n frontMatter.bibliography = bibliography;\n frontMatter.bibliographyParsed = true;\n for (const waitingCallback of Controller.waitingOn.bibliography.slice()) {\n waitingCallback();\n }\n\n // ensure we have citations\n if (!frontMatter.citationsCollected) {\n Controller.waitingOn.citations.push(function() {\n Controller.listeners.onBibliographyChanged({\n target: event.target,\n detail: event.detail\n });\n });\n return;\n }\n\n if (citationListTag.hasAttribute(""distill-prerendered"")) {\n console.debug(""Citation list was prerendered; not updating it."");\n } else {\n const entries = new Map(\n frontMatter.citations.map(citationKey => {\n return [citationKey, frontMatter.bibliography.get(citationKey)];\n })\n );\n citationListTag.citations = entries;\n }\n },\n\n onFootnoteChanged() {\n // const footnote = event.detail;\n //TODO: optimize to only update current footnote\n const footnotesList = document.querySelector(""d-footnote-list"");\n if (footnotesList) {\n const footnotes = document.querySelectorAll(""d-footnote"");\n footnotesList.footnotes = footnotes;\n }\n },\n\n onFrontMatterChanged(event) {\n const data = event.detail;\n mergeFromYMLFrontmatter(frontMatter, data);\n\n const interstitial = document.querySelector(""d-interstitial"");\n if (interstitial) {\n if (typeof frontMatter.password !== ""undefined"") {\n interstitial.password = frontMatter.password;\n } else {\n interstitial.parentElement.removeChild(interstitial);\n }\n }\n\n const prerendered = document.body.hasAttribute(""distill-prerendered"");\n if (!prerendered && domContentLoaded()) {\n optionalComponents(document, frontMatter);\n\n const appendix = document.querySelector(""distill-appendix"");\n if (appendix) {\n appendix.frontMatter = frontMatter;\n }\n\n const byline = document.querySelector(""d-byline"");\n if (byline) {\n byline.frontMatter = frontMatter;\n }\n\n if (data.katex) {\n DMath.katexOptions = data.katex;\n }\n }\n },\n\n DOMContentLoaded() {\n if (Controller.loaded) {\n console.warn(\n ""Controller received DOMContentLoaded but was already loaded!""\n );\n return;\n } else if (!domContentLoaded()) {\n console.warn(\n ""Controller received DOMContentLoaded at document.readyState: "" +\n document.readyState +\n ""!""\n );\n return;\n } else {\n Controller.loaded = true;\n console.debug(""Runlevel 4: Controller running DOMContentLoaded"");\n }\n\n const frontMatterTag = document.querySelector(""d-front-matter"");\n if (frontMatterTag) {\n const data = parseFrontmatter(frontMatterTag);\n Controller.listeners.onFrontMatterChanged({ detail: data });\n }\n\n // Resolving ""citations"" dependency due to initial DOM load\n frontMatter.citations = collect_citations();\n frontMatter.citationsCollected = true;\n for (const waitingCallback of Controller.waitingOn.citations.slice()) {\n waitingCallback();\n }\n\n if (frontMatter.bibliographyParsed) {\n for (const waitingCallback of Controller.waitingOn.bibliography.slice()) {\n waitingCallback();\n }\n }\n\n const footnotesList = document.querySelector(""d-footnote-list"");\n if (footnotesList) {\n const footnotes = document.querySelectorAll(""d-footnote"");\n footnotesList.footnotes = footnotes;\n }\n }\n } // listeners\n }; // Controller\n\n var base = ""/*\n * Copyright 2018 The Distill Template Authors\n *\n * Licensed under the Apache License, Version 2.0 (the \""License\"");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \""AS IS\"" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n */\n\nhtml {\n font-size: 14px;\n\tline-height: 1.6em;\n /* font-family: \""Libre Franklin\"", \""Helvetica Neue\"", sans-serif; */\n font-family: -apple-system, BlinkMacSystemFont, \""Segoe UI\"", Roboto, Oxygen, Ubuntu, Cantarell, \""Fira Sans\"", \""Droid Sans\"", \""Helvetica Neue\"", Arial, sans-serif;\n /*, \""Apple Color Emoji\"", \""Segoe UI Emoji\"", \""Segoe UI Symbol\"";*/\n text-size-adjust: 100%;\n -ms-text-size-adjust: 100%;\n -webkit-text-size-adjust: 100%;\n}\n\n@media(min-width: 768px) {\n html {\n font-size: 16px;\n }\n}\n\nbody {\n margin: 0;\n}\n\na {\n color: #004276;\n}\n\nfigure {\n margin: 0;\n}\n\ntable {\n\tborder-collapse: collapse;\n\tborder-spacing: 0;\n}\n\ntable th {\n\ttext-align: left;\n}\n\ntable thead {\n border-bottom: 1px solid rgba(0, 0, 0, 0.05);\n}\n\ntable thead th {\n padding-bottom: 0.5em;\n}\n\ntable tbody :first-child td {\n padding-top: 0.5em;\n}\n\npre {\n overflow: auto;\n max-width: 100%;\n}\n\np {\n margin-top: 0;\n margin-bottom: 1em;\n}\n\nsup, sub {\n vertical-align: baseline;\n position: relative;\n top: -0.4em;\n line-height: 1em;\n}\n\nsub {\n top: 0.4em;\n}\n\n.kicker,\n.marker {\n font-size: 15px;\n font-weight: 600;\n color: rgba(0, 0, 0, 0.5);\n}\n\n\n/* Headline */\n\n@media(min-width: 1024px) {\n d-title h1 span {\n display: block;\n }\n}\n\n/* Figure */\n\nfigure {\n position: relative;\n margin-bottom: 2.5em;\n margin-top: 1.5em;\n}\n\nfigcaption+figure {\n\n}\n\nfigure img {\n width: 100%;\n}\n\nfigure svg text,\nfigure svg tspan {\n}\n\nfigcaption,\n.figcaption {\n color: rgba(0, 0, 0, 0.6);\n font-size: 12px;\n line-height: 1.5em;\n}\n\n@media(min-width: 1024px) {\nfigcaption,\n.figcaption {\n font-size: 13px;\n }\n}\n\nfigure.external img {\n background: white;\n border: 1px solid rgba(0, 0, 0, 0.1);\n box-shadow: 0 1px 8px rgba(0, 0, 0, 0.1);\n padding: 18px;\n box-sizing: border-box;\n}\n\nfigcaption a {\n color: rgba(0, 0, 0, 0.6);\n}\n\nfigcaption b,\nfigcaption strong, {\n font-weight: 600;\n color: rgba(0, 0, 0, 1.0);\n}\n"";\n\n var layout = ""/*\n * Copyright 2018 The Distill Template Authors\n *\n * Licensed under the Apache License, Version 2.0 (the \""License\"");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \""AS IS\"" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n */\n\n@supports not (display: grid) {\n .base-grid,\n distill-header,\n d-title,\n d-abstract,\n d-article,\n d-appendix,\n distill-appendix,\n d-byline,\n d-footnote-list,\n d-citation-list,\n distill-footer {\n display: block;\n padding: 8px;\n }\n}\n\n.base-grid,\ndistill-header,\nd-title,\nd-abstract,\nd-article,\nd-appendix,\ndistill-appendix,\nd-byline,\nd-footnote-list,\nd-citation-list,\ndistill-footer {\n display: grid;\n justify-items: stretch;\n grid-template-columns: [screen-start] 8px [page-start kicker-start text-start gutter-start middle-start] 1fr 1fr 1fr 1fr 1fr 1fr 1fr 1fr [text-end page-end gutter-end kicker-end middle-end] 8px [screen-end];\n grid-column-gap: 8px;\n}\n\n.grid {\n display: grid;\n grid-column-gap: 8px;\n}\n\n@media(min-width: 768px) {\n .base-grid,\n distill-header,\n d-title,\n d-abstract,\n d-article,\n d-appendix,\n distill-appendix,\n d-byline,\n d-footnote-list,\n d-citation-list,\n distill-footer {\n grid-template-columns: [screen-start] 1fr [page-start kicker-start middle-start text-start] 45px 45px 45px 45px 45px 45px 45px 45px [ kicker-end text-end gutter-start] 45px [middle-end] 45px [page-end gutter-end] 1fr [screen-end];\n grid-column-gap: 16px;\n }\n\n .grid {\n grid-column-gap: 16px;\n }\n}\n\n@media(min-width: 1000px) {\n .base-grid,\n distill-header,\n d-title,\n d-abstract,\n d-article,\n d-appendix,\n distill-appendix,\n d-byline,\n d-footnote-list,\n d-citation-list,\n distill-footer {\n grid-template-columns: [screen-start] 1fr [page-start kicker-start] 50px [middle-start] 50px [text-start kicker-end] 50px 50px 50px 50px 50px 50px 50px 50px [text-end gutter-start] 50px [middle-end] 50px [page-end gutter-end] 1fr [screen-end];\n grid-column-gap: 16px;\n }\n\n .grid {\n grid-column-gap: 16px;\n }\n}\n\n@media(min-width: 1180px) {\n .base-grid,\n distill-header,\n d-title,\n d-abstract,\n d-article,\n d-appendix,\n distill-appendix,\n d-byline,\n d-footnote-list,\n d-citation-list,\n distill-footer {\n grid-template-columns: [screen-start] 1fr [page-start kicker-start] 60px [middle-start] 60px [text-start kicker-end] 60px 60px 60px 60px 60px 60px 60px 60px [text-end gutter-start] 60px [middle-end] 60px [page-end gutter-end] 1fr [screen-end];\n grid-column-gap: 32px;\n }\n\n .grid {\n grid-column-gap: 32px;\n }\n}\n\n\n\n\n.base-grid {\n grid-column: screen;\n}\n\n/* .l-body,\nd-article > * {\n grid-column: text;\n}\n\n.l-page,\nd-title > *,\nd-figure {\n grid-column: page;\n} */\n\n.l-gutter {\n grid-column: gutter;\n}\n\n.l-text,\n.l-body {\n grid-column: text;\n}\n\n.l-page {\n grid-column: page;\n}\n\n.l-body-outset {\n grid-column: middle;\n}\n\n.l-page-outset {\n grid-column: page;\n}\n\n.l-screen {\n grid-column: screen;\n}\n\n.l-screen-inset {\n grid-column: screen;\n padding-left: 16px;\n padding-left: 16px;\n}\n\n\n/* Aside */\n\nd-article aside {\n grid-column: gutter;\n font-size: 12px;\n line-height: 1.6em;\n color: rgba(0, 0, 0, 0.6)\n}\n\n@media(min-width: 768px) {\n aside {\n grid-column: gutter;\n }\n\n .side {\n grid-column: gutter;\n }\n}\n"";\n\n var print = ""/*\n * Copyright 2018 The Distill Template Authors\n *\n * Licensed under the Apache License, Version 2.0 (the \""License\"");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \""AS IS\"" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n */\n\n@media print {\n\n @page {\n size: 8in 11in;\n @bottom-right {\n content: counter(page) \"" of \"" counter(pages);\n }\n }\n\n html {\n /* no general margins -- CSS Grid takes care of those */\n }\n\n p, code {\n page-break-inside: avoid;\n }\n\n h2, h3 {\n page-break-after: avoid;\n }\n\n d-header {\n visibility: hidden;\n }\n\n d-footer {\n display: none!important;\n }\n\n}\n"";\n\n var byline = ""/*\n * Copyright 2018 The Distill Template Authors\n *\n * Licensed under the Apache License, Version 2.0 (the \""License\"");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \""AS IS\"" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n */\n\nd-byline {\n contain: style;\n overflow: hidden;\n border-top: 1px solid rgba(0, 0, 0, 0.1);\n font-size: 0.8rem;\n line-height: 1.8em;\n padding: 1.5rem 0;\n min-height: 1.8em;\n}\n\n\nd-byline .byline {\n grid-template-columns: 1fr 1fr;\n grid-column: text;\n}\n\n@media(min-width: 768px) {\n d-byline .byline {\n grid-template-columns: 1fr 1fr 1fr 1fr;\n }\n}\n\nd-byline .authors-affiliations {\n grid-column-end: span 2;\n grid-template-columns: 1fr 1fr;\n margin-bottom: 1em;\n}\n\n@media(min-width: 768px) {\n d-byline .authors-affiliations {\n margin-bottom: 0;\n }\n}\n\nd-byline h3 {\n font-size: 0.6rem;\n font-weight: 400;\n color: rgba(0, 0, 0, 0.5);\n margin: 0;\n text-transform: uppercase;\n}\n\nd-byline p {\n margin: 0;\n}\n\nd-byline a,\nd-article d-byline a {\n color: rgba(0, 0, 0, 0.8);\n text-decoration: none;\n border-bottom: none;\n}\n\nd-article d-byline a:hover {\n text-decoration: underline;\n border-bottom: none;\n}\n\nd-byline p.author {\n font-weight: 500;\n}\n\nd-byline .affiliations {\n\n}\n"";\n\n var article = ""/*\n * Copyright 2018 The Distill Template Authors\n *\n * Licensed under the Apache License, Version 2.0 (the \""License\"");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \""AS IS\"" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n */\n\nd-article {\n contain: layout style;\n overflow-x: hidden;\n border-top: 1px solid rgba(0, 0, 0, 0.1);\n padding-top: 2rem;\n color: rgba(0, 0, 0, 0.8);\n}\n\nd-article > * {\n grid-column: text;\n}\n\n@media(min-width: 768px) {\n d-article {\n font-size: 16px;\n }\n}\n\n@media(min-width: 1024px) {\n d-article {\n font-size: 1.06rem;\n line-height: 1.7em;\n }\n}\n\n\n/* H2 */\n\n\nd-article .marker {\n text-decoration: none;\n border: none;\n counter-reset: section;\n grid-column: kicker;\n line-height: 1.7em;\n}\n\nd-article .marker:hover {\n border: none;\n}\n\nd-article .marker span {\n padding: 0 3px 4px;\n border-bottom: 1px solid rgba(0, 0, 0, 0.2);\n position: relative;\n top: 4px;\n}\n\nd-article .marker:hover span {\n color: rgba(0, 0, 0, 0.7);\n border-bottom: 1px solid rgba(0, 0, 0, 0.7);\n}\n\nd-article h2 {\n font-weight: 600;\n font-size: 24px;\n line-height: 1.25em;\n margin: 2rem 0 1.5rem 0;\n border-bottom: 1px solid rgba(0, 0, 0, 0.1);\n padding-bottom: 1rem;\n}\n\n@media(min-width: 1024px) {\n d-article h2 {\n font-size: 36px;\n }\n}\n\n/* H3 */\n\nd-article h3 {\n font-weight: 700;\n font-size: 18px;\n line-height: 1.4em;\n margin-bottom: 1em;\n margin-top: 2em;\n}\n\n@media(min-width: 1024px) {\n d-article h3 {\n font-size: 20px;\n }\n}\n\n/* H4 */\n\nd-article h4 {\n font-weight: 600;\n text-transform: uppercase;\n font-size: 14px;\n line-height: 1.4em;\n}\n\nd-article a {\n color: inherit;\n}\n\nd-article p,\nd-article ul,\nd-article ol,\nd-article blockquote {\n margin-top: 0;\n margin-bottom: 1em;\n margin-left: 0;\n margin-right: 0;\n}\n\nd-article blockquote {\n border-left: 2px solid rgba(0, 0, 0, 0.2);\n padding-left: 2em;\n font-style: italic;\n color: rgba(0, 0, 0, 0.6);\n}\n\nd-article a {\n border-bottom: 1px solid rgba(0, 0, 0, 0.4);\n text-decoration: none;\n}\n\nd-article a:hover {\n border-bottom: 1px solid rgba(0, 0, 0, 0.8);\n}\n\nd-article .link {\n text-decoration: underline;\n cursor: pointer;\n}\n\nd-article ul,\nd-article ol {\n padding-left: 24px;\n}\n\nd-article li {\n margin-bottom: 1em;\n margin-left: 0;\n padding-left: 0;\n}\n\nd-article li:last-child {\n margin-bottom: 0;\n}\n\nd-article pre {\n font-size: 14px;\n margin-bottom: 20px;\n}\n\nd-article hr {\n grid-column: screen;\n width: 100%;\n border: none;\n border-bottom: 1px solid rgba(0, 0, 0, 0.1);\n margin-top: 60px;\n margin-bottom: 60px;\n}\n\nd-article section {\n margin-top: 60px;\n margin-bottom: 60px;\n}\n\nd-article span.equation-mimic {\n font-family: georgia;\n font-size: 115%;\n font-style: italic;\n}\n\nd-article > d-code,\nd-article section > d-code {\n display: block;\n}\n\nd-article > d-math[block],\nd-article section > d-math[block] {\n display: block;\n}\n\n@media (max-width: 768px) {\n d-article > d-code,\n d-article section > d-code,\n d-article > d-math[block],\n d-article section > d-math[block] {\n overflow-x: scroll;\n -ms-overflow-style: none; // IE 10+\n overflow: -moz-scrollbars-none; // Firefox\n }\n\n d-article > d-code::-webkit-scrollbar,\n d-article section > d-code::-webkit-scrollbar,\n d-article > d-math[block]::-webkit-scrollbar,\n d-article section > d-math[block]::-webkit-scrollbar {\n display: none; // Safari and Chrome\n }\n}\n\nd-article .citation {\n color: #668;\n cursor: pointer;\n}\n\nd-include {\n width: auto;\n display: block;\n}\n\nd-figure {\n contain: layout style;\n}\n\n/* KaTeX */\n\n.katex, .katex-prerendered {\n contain: style;\n display: inline-block;\n}\n\n/* Tables */\n\nd-article table {\n border-collapse: collapse;\n margin-bottom: 1.5rem;\n border-bottom: 1px solid rgba(0, 0, 0, 0.2);\n}\n\nd-article table th {\n border-bottom: 1px solid rgba(0, 0, 0, 0.2);\n}\n\nd-article table td {\n border-bottom: 1px solid rgba(0, 0, 0, 0.05);\n}\n\nd-article table tr:last-of-type td {\n border-bottom: none;\n}\n\nd-article table th,\nd-article table td {\n font-size: 15px;\n padding: 2px 8px;\n}\n\nd-article table tbody :first-child td {\n padding-top: 2px;\n}\n"";\n\n var title = ""/*\n * Copyright 2018 The Distill Template Authors\n *\n * Licensed under the Apache License, Version 2.0 (the \""License\"");\n * you may not use this file except in compliance with the License.\n * You may obtain a copy of the License at\n *\n * http://www.apache.org/licenses/LICENSE-2.0\n *\n * Unless required by applicable law or agreed to in writing, software\n * distributed under the License is distributed on an \""AS IS\"" BASIS,\n * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n * See the License for the specific language governing permissions and\n * limitations under the License.\n */\n\nd-title {\n padding: 2rem 0 1.5rem;\n contain: layout style;\n overflow-x: hidden;\n}\n\n@media(min-width: 768px) {\n d-title {\n padding: 4rem 0 1.5rem;\n }\n}\n\nd-title h1 {\n grid-column: text;\n font-size: 40px;\n font-weight: 700;\n line-height: 1.1em;\n margin: 0 0 0.5rem;\n}\n\n@media(min-width: 768px) {\n d-title h1 {\n font-size: 50px;\n }\n}\n\nd-title p {\n font-weight: 300;\n font-size: 1.2rem;\n line-height: 1.55em;\n grid-column: text;\n}\n\nd-title .status {\n margin-top: 0px;\n font-size: 12px;\n color: #009688;\n opacity: 0.8;\n grid-column: kicker;\n}\n\nd-title .status span {\n line-height: 1;\n display: inline-block;\n padding: 6px 0;\n border-bottom: 1px solid #80cbc4;\n font-size: 11px;\n text-transform: uppercase;\n}\n"";\n\n // Copyright 2018 The Distill Template Authors\n\n const styles = base + layout + title + byline + article + math + print;\n\n function makeStyleTag(dom) {\n\n const styleTagId = 'distill-prerendered-styles';\n const prerenderedTag = dom.getElementById(styleTagId);\n if (!prerenderedTag) {\n const styleTag = dom.createElement('style');\n styleTag.id = styleTagId;\n styleTag.type = 'text/css';\n const cssTextTag = dom.createTextNode(styles);\n styleTag.appendChild(cssTextTag);\n const firstScriptTag = dom.head.querySelector('script');\n dom.head.insertBefore(styleTag, firstScriptTag);\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n function addPolyfill(polyfill, polyfillLoadedCallback) {\n console.debug('Runlevel 0: Polyfill required: ' + polyfill.name);\n const script = document.createElement('script');\n script.src = polyfill.url;\n script.async = false;\n if (polyfillLoadedCallback) {\n script.onload = function() { polyfillLoadedCallback(polyfill); };\n }\n script.onerror = function() {\n new Error('Runlevel 0: Polyfills failed to load script ' + polyfill.name);\n };\n document.head.appendChild(script);\n }\n\n const polyfills = [\n {\n name: 'WebComponents',\n support: function() {\n return 'customElements' in window &&\n 'attachShadow' in Element.prototype &&\n 'getRootNode' in Element.prototype &&\n 'content' in document.createElement('template') &&\n 'Promise' in window &&\n 'from' in Array;\n },\n url: 'https://distill.pub/third-party/polyfills/webcomponents-lite.js'\n }, {\n name: 'IntersectionObserver',\n support: function() {\n return 'IntersectionObserver' in window &&\n 'IntersectionObserverEntry' in window;\n },\n url: 'https://distill.pub/third-party/polyfills/intersection-observer.js'\n },\n ];\n\n class Polyfills {\n\n static browserSupportsAllFeatures() {\n return polyfills.every((poly) => poly.support());\n }\n\n static load(callback) {\n // Define an intermediate callback that checks if all is loaded.\n const polyfillLoaded = function(polyfill) {\n polyfill.loaded = true;\n console.debug('Runlevel 0: Polyfill has finished loading: ' + polyfill.name);\n // console.debug(window[polyfill.name]);\n if (Polyfills.neededPolyfills.every((poly) => poly.loaded)) {\n console.debug('Runlevel 0: All required polyfills have finished loading.');\n console.debug('Runlevel 0->1.');\n window.distillRunlevel = 1;\n callback();\n }\n };\n // Add polyfill script tags\n for (const polyfill of Polyfills.neededPolyfills) {\n addPolyfill(polyfill, polyfillLoaded);\n }\n }\n\n static get neededPolyfills() {\n if (!Polyfills._neededPolyfills) {\n Polyfills._neededPolyfills = polyfills.filter((poly) => !poly.support());\n }\n return Polyfills._neededPolyfills;\n }\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n // const marginSmall = 16;\n // const marginLarge = 3 * marginSmall;\n // const margin = marginSmall + marginLarge;\n // const gutter = marginSmall;\n // const outsetAmount = margin / 2;\n // const numCols = 4;\n // const numGutters = numCols - 1;\n // const columnWidth = (768 - 2 * marginLarge - numGutters * gutter) / numCols;\n //\n // const screenwidth = 768;\n // const pageWidth = screenwidth - 2 * marginLarge;\n // const bodyWidth = pageWidth - columnWidth - gutter;\n\n function body(selector) {\n return `${selector} {\n grid-column: left / text;\n }\n `;\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const T$1 = Template('d-abstract', `\n\n\n\n`);\n\n class Abstract extends T$1(HTMLElement) {\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const T$2 = Template('d-appendix', `\n\n\n`, false);\n\n class Appendix extends T$2(HTMLElement) {\n\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n // import { Template } from '../mixins/template';\n // import { Controller } from '../controller';\n\n const isOnlyWhitespace = /^\s*$/;\n\n class Article extends HTMLElement {\n static get is() { return 'd-article'; }\n\n constructor() {\n super();\n\n new MutationObserver( (mutations) => {\n for (const mutation of mutations) {\n for (const addedNode of mutation.addedNodes) {\n switch (addedNode.nodeName) {\n case '#text': { // usually text nodes are only linebreaks.\n const text = addedNode.nodeValue;\n if (!isOnlyWhitespace.test(text)) {\n console.warn('Use of unwrapped text in distill articles is discouraged as it breaks layout! Please wrap any text in a or

tag. We found the following text: ' + text);\n const wrapper = document.createElement('span');\n wrapper.innerHTML = addedNode.nodeValue;\n addedNode.parentNode.insertBefore(wrapper, addedNode);\n addedNode.parentNode.removeChild(addedNode);\n }\n } break;\n }\n }\n }\n }).observe(this, {childList: true});\n }\n\n }\n\n var commonjsGlobal = typeof globalThis !== 'undefined' ? globalThis : typeof window !== 'undefined' ? window : typeof global !== 'undefined' ? global : typeof self !== 'undefined' ? self : {};\n\n function createCommonjsModule(fn, module) {\n \treturn module = { exports: {} }, fn(module, module.exports), module.exports;\n }\n\n var bibtexParse = createCommonjsModule(function (module, exports) {\n /* start bibtexParse 0.0.22 */\n\n //Original work by Henrik Muehe (c) 2010\n //\n //CommonJS port by Mikola Lysenko 2013\n //\n //Port to Browser lib by ORCID / RCPETERS\n //\n //Issues:\n //no comment handling within strings\n //no string concatenation\n //no variable values yet\n //Grammar implemented here:\n //bibtex -> (string | preamble | comment | entry)*;\n //string -> '@STRING' '{' key_equals_value '}';\n //preamble -> '@PREAMBLE' '{' value '}';\n //comment -> '@COMMENT' '{' value '}';\n //entry -> '@' key '{' key ',' key_value_list '}';\n //key_value_list -> key_equals_value (',' key_equals_value)*;\n //key_equals_value -> key '=' value;\n //value -> value_quotes | value_braces | key;\n //value_quotes -> '""' .*? '""'; // not quite\n //value_braces -> '{' .*? '""'; // not quite\n (function(exports) {\n\n function BibtexParser() {\n \n this.months = [""jan"", ""feb"", ""mar"", ""apr"", ""may"", ""jun"", ""jul"", ""aug"", ""sep"", ""oct"", ""nov"", ""dec""];\n this.notKey = [',','{','}',' ','='];\n this.pos = 0;\n this.input = """";\n this.entries = new Array();\n\n this.currentEntry = """";\n\n this.setInput = function(t) {\n this.input = t;\n };\n\n this.getEntries = function() {\n return this.entries;\n };\n\n this.isWhitespace = function(s) {\n return (s == ' ' || s == '\r' || s == '\t' || s == '\n');\n };\n\n this.match = function(s, canCommentOut) {\n if (canCommentOut == undefined || canCommentOut == null)\n canCommentOut = true;\n this.skipWhitespace(canCommentOut);\n if (this.input.substring(this.pos, this.pos + s.length) == s) {\n this.pos += s.length;\n } else {\n throw ""Token mismatch, expected "" + s + "", found ""\n + this.input.substring(this.pos);\n } this.skipWhitespace(canCommentOut);\n };\n\n this.tryMatch = function(s, canCommentOut) {\n if (canCommentOut == undefined || canCommentOut == null)\n canCommentOut = true;\n this.skipWhitespace(canCommentOut);\n if (this.input.substring(this.pos, this.pos + s.length) == s) {\n return true;\n } else {\n return false;\n } };\n\n /* when search for a match all text can be ignored, not just white space */\n this.matchAt = function() {\n while (this.input.length > this.pos && this.input[this.pos] != '@') {\n this.pos++;\n }\n if (this.input[this.pos] == '@') {\n return true;\n } return false;\n };\n\n this.skipWhitespace = function(canCommentOut) {\n while (this.isWhitespace(this.input[this.pos])) {\n this.pos++;\n } if (this.input[this.pos] == ""%"" && canCommentOut == true) {\n while (this.input[this.pos] != ""\n"") {\n this.pos++;\n } this.skipWhitespace(canCommentOut);\n } };\n\n this.value_braces = function() {\n var bracecount = 0;\n this.match(""{"", false);\n var start = this.pos;\n var escaped = false;\n while (true) {\n if (!escaped) {\n if (this.input[this.pos] == '}') {\n if (bracecount > 0) {\n bracecount--;\n } else {\n var end = this.pos;\n this.match(""}"", false);\n return this.input.substring(start, end);\n } } else if (this.input[this.pos] == '{') {\n bracecount++;\n } else if (this.pos >= this.input.length - 1) {\n throw ""Unterminated value"";\n } } if (this.input[this.pos] == '\\' && escaped == false)\n escaped = true;\n else\n escaped = false;\n this.pos++;\n } };\n\n this.value_comment = function() {\n var str = '';\n var brcktCnt = 0;\n while (!(this.tryMatch(""}"", false) && brcktCnt == 0)) {\n str = str + this.input[this.pos];\n if (this.input[this.pos] == '{')\n brcktCnt++;\n if (this.input[this.pos] == '}')\n brcktCnt--;\n if (this.pos >= this.input.length - 1) {\n throw ""Unterminated value:"" + this.input.substring(start);\n } this.pos++;\n } return str;\n };\n\n this.value_quotes = function() {\n this.match('""', false);\n var start = this.pos;\n var escaped = false;\n while (true) {\n if (!escaped) {\n if (this.input[this.pos] == '""') {\n var end = this.pos;\n this.match('""', false);\n return this.input.substring(start, end);\n } else if (this.pos >= this.input.length - 1) {\n throw ""Unterminated value:"" + this.input.substring(start);\n } }\n if (this.input[this.pos] == '\\' && escaped == false)\n escaped = true;\n else\n escaped = false;\n this.pos++;\n } };\n\n this.single_value = function() {\n var start = this.pos;\n if (this.tryMatch(""{"")) {\n return this.value_braces();\n } else if (this.tryMatch('""')) {\n return this.value_quotes();\n } else {\n var k = this.key();\n if (k.match(""^[0-9]+$""))\n return k;\n else if (this.months.indexOf(k.toLowerCase()) >= 0)\n return k.toLowerCase();\n else\n throw ""Value expected:"" + this.input.substring(start) + ' for key: ' + k;\n \n } };\n\n this.value = function() {\n var values = [];\n values.push(this.single_value());\n while (this.tryMatch(""#"")) {\n this.match(""#"");\n values.push(this.single_value());\n } return values.join("""");\n };\n\n this.key = function() {\n var start = this.pos;\n while (true) {\n if (this.pos >= this.input.length) {\n throw ""Runaway key"";\n } // а-яА-Я is Cyrillic\n //console.log(this.input[this.pos]);\n if (this.notKey.indexOf(this.input[this.pos]) >= 0) {\n return this.input.substring(start, this.pos);\n } else {\n this.pos++;\n \n } } };\n\n this.key_equals_value = function() {\n var key = this.key();\n if (this.tryMatch(""="")) {\n this.match(""="");\n var val = this.value();\n return [ key, val ];\n } else {\n throw ""... = value expected, equals sign missing:""\n + this.input.substring(this.pos);\n } };\n\n this.key_value_list = function() {\n var kv = this.key_equals_value();\n this.currentEntry['entryTags'] = {};\n this.currentEntry['entryTags'][kv[0]] = kv[1];\n while (this.tryMatch("","")) {\n this.match("","");\n // fixes problems with commas at the end of a list\n if (this.tryMatch(""}"")) {\n break;\n }\n kv = this.key_equals_value();\n this.currentEntry['entryTags'][kv[0]] = kv[1];\n } };\n\n this.entry_body = function(d) {\n this.currentEntry = {};\n this.currentEntry['citationKey'] = this.key();\n this.currentEntry['entryType'] = d.substring(1);\n this.match("","");\n this.key_value_list();\n this.entries.push(this.currentEntry);\n };\n\n this.directive = function() {\n this.match(""@"");\n return ""@"" + this.key();\n };\n\n this.preamble = function() {\n this.currentEntry = {};\n this.currentEntry['entryType'] = 'PREAMBLE';\n this.currentEntry['entry'] = this.value_comment();\n this.entries.push(this.currentEntry);\n };\n\n this.comment = function() {\n this.currentEntry = {};\n this.currentEntry['entryType'] = 'COMMENT';\n this.currentEntry['entry'] = this.value_comment();\n this.entries.push(this.currentEntry);\n };\n\n this.entry = function(d) {\n this.entry_body(d);\n };\n\n this.bibtex = function() {\n while (this.matchAt()) {\n var d = this.directive();\n this.match(""{"");\n if (d == ""@STRING"") {\n this.string();\n } else if (d == ""@PREAMBLE"") {\n this.preamble();\n } else if (d == ""@COMMENT"") {\n this.comment();\n } else {\n this.entry(d);\n }\n this.match(""}"");\n } };\n } \n exports.toJSON = function(bibtex) {\n var b = new BibtexParser();\n b.setInput(bibtex);\n b.bibtex();\n return b.entries;\n };\n\n /* added during hackathon don't hate on me */\n exports.toBibtex = function(json) {\n var out = '';\n for ( var i in json) {\n out += ""@"" + json[i].entryType;\n out += '{';\n if (json[i].citationKey)\n out += json[i].citationKey + ', ';\n if (json[i].entry)\n out += json[i].entry ;\n if (json[i].entryTags) {\n var tags = '';\n for (var jdx in json[i].entryTags) {\n if (tags.length != 0)\n tags += ', ';\n tags += jdx + '= {' + json[i].entryTags[jdx] + '}';\n }\n out += tags;\n }\n out += '}\n\n';\n }\n return out;\n \n };\n\n })( exports);\n\n /* end bibtexParse */\n });\n\n // Copyright 2018 The Distill Template Authors\n\n function normalizeTag(string) {\n return string\n .replace(/[\t\n ]+/g, ' ')\n .replace(/{\\[""^`.'acu~Hvs]( )?([a-zA-Z])}/g, (full, x, char) => char)\n .replace(/{\\([a-zA-Z])}/g, (full, char) => char)\n .replace(/[{}]/gi,''); // Replace curly braces forcing plaintext in latex.\n }\n\n function parseBibtex(bibtex) {\n const bibliography = new Map();\n const parsedEntries = bibtexParse.toJSON(bibtex);\n for (const entry of parsedEntries) {\n // normalize tags; note entryTags is an object, not Map\n for (const [key, value] of Object.entries(entry.entryTags)) {\n entry.entryTags[key.toLowerCase()] = normalizeTag(value);\n }\n entry.entryTags.type = entry.entryType;\n // add to bibliography\n bibliography.set(entry.citationKey, entry.entryTags);\n }\n return bibliography;\n }\n\n function serializeFrontmatterToBibtex(frontMatter) {\n return `@article{${frontMatter.slug},\n author = {${frontMatter.bibtexAuthors}},\n title = {${frontMatter.title}},\n journal = {${frontMatter.journal.title}},\n year = {${frontMatter.publishedYear}},\n note = {${frontMatter.url}}\n}`;\n }\n\n // Copyright 2018 The Distill Template Authors\n\n class Bibliography extends HTMLElement {\n\n static get is() { return 'd-bibliography'; }\n\n constructor() {\n super();\n\n // set up mutation observer\n const options = {childList: true, characterData: true, subtree: true};\n const observer = new MutationObserver( (entries) => {\n for (const entry of entries) {\n if (entry.target.nodeName === 'SCRIPT' || entry.type === 'characterData') {\n this.parseIfPossible();\n }\n }\n });\n observer.observe(this, options);\n }\n\n connectedCallback() {\n requestAnimationFrame(() => {\n this.parseIfPossible();\n });\n }\n\n parseIfPossible() {\n const scriptTag = this.querySelector('script');\n if (!scriptTag) return;\n if (scriptTag.type == 'text/bibtex') {\n const newBibtex = scriptTag.textContent;\n if (this.bibtex !== newBibtex) {\n this.bibtex = newBibtex;\n const bibliography = parseBibtex(this.bibtex);\n this.notify(bibliography);\n }\n } else if (scriptTag.type == 'text/json') {\n const bibliography = new Map(JSON.parse(scriptTag.textContent));\n this.notify(bibliography);\n } else {\n console.warn('Unsupported bibliography script tag type: ' + scriptTag.type);\n }\n }\n\n notify(bibliography) {\n const options = { detail: bibliography, bubbles: true };\n const event = new CustomEvent('onBibliographyChanged', options);\n this.dispatchEvent(event);\n }\n\n /* observe 'src' attribute */\n\n static get observedAttributes() {\n return ['src'];\n }\n\n receivedBibtex(event) {\n const bibliography = parseBibtex(event.target.response);\n this.notify(bibliography);\n }\n\n attributeChangedCallback(name, oldValue, newValue) {\n var oReq = new XMLHttpRequest();\n oReq.onload = (e) => this.receivedBibtex(e);\n oReq.onerror = () => console.warn(`Could not load Bibtex! (tried ${newValue})`);\n oReq.responseType = 'text';\n oReq.open('GET', newValue, true);\n oReq.send();\n }\n\n\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n // import style from '../styles/d-byline.css';\n\n function bylineTemplate(frontMatter) {\n return `\n

\n
\n

Authors

\n

Affiliations

\n ${frontMatter.authors.map(author => `\n

\n ${author.personalURL ? `\n ${author.name}` : `\n ${author.name}`}\n

\n

\n ${author.affiliations.map(affiliation =>\n affiliation.url ? `${affiliation.name}` : `${affiliation.name}`\n ).join(', ')}\n

\n `).join('')}\n
\n
\n

Published

\n ${frontMatter.publishedDate ? `\n

${frontMatter.publishedMonth} ${frontMatter.publishedDay}, ${frontMatter.publishedYear}

` : `\n

Not published yet.

`}\n
\n
\n

DOI

\n ${frontMatter.doi ? `\n

${frontMatter.doi}

` : `\n

No DOI yet.

`}\n
\n
\n`;\n }\n\n class Byline extends HTMLElement {\n\n static get is() { return 'd-byline'; }\n\n set frontMatter(frontMatter) {\n this.innerHTML = bylineTemplate(frontMatter);\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const T$3 = Template(\n ""d-cite"",\n `\n\n\n\n\n
\n \n
\n`\n );\n\n class Cite extends T$3(HTMLElement) {\n /* Lifecycle */\n constructor() {\n super();\n this._numbers = [];\n this._entries = [];\n }\n\n connectedCallback() {\n this.outerSpan = this.root.querySelector(""#citation-"");\n this.innerSpan = this.root.querySelector("".citation-number"");\n this.hoverBox = this.root.querySelector(""d-hover-box"");\n window.customElements.whenDefined(""d-hover-box"").then(() => {\n this.hoverBox.listen(this);\n });\n // in case this component got connected after values were set\n if (this.numbers) {\n this.displayNumbers(this.numbers);\n }\n if (this.entries) {\n this.displayEntries(this.entries);\n }\n }\n\n //TODO This causes an infinite loop on firefox with polyfills.\n // This is only needed for interactive editing so no priority.\n // disconnectedCallback() {\n // const options = { detail: [this, this.keys], bubbles: true };\n // const event = new CustomEvent('onCiteKeyRemoved', options);\n // document.dispatchEvent(event);\n // }\n\n /* observe 'key' attribute */\n\n static get observedAttributes() {\n return [""key"", ""bibtex-key""];\n }\n\n attributeChangedCallback(name, oldValue, newValue) {\n const eventName = oldValue ? ""onCiteKeyChanged"" : ""onCiteKeyCreated"";\n const keys = newValue.split("","").map(k => k.trim());\n const options = { detail: [this, keys], bubbles: true };\n const event = new CustomEvent(eventName, options);\n document.dispatchEvent(event);\n }\n\n set key(value) {\n this.setAttribute(""key"", value);\n }\n\n get key() {\n return this.getAttribute(""key"") || this.getAttribute(""bibtex-key"");\n }\n\n get keys() {\n const result = this.key.split("","");\n console.log(result);\n return result;\n }\n\n /* Setters & Rendering */\n\n set numbers(numbers) {\n this._numbers = numbers;\n this.displayNumbers(numbers);\n }\n\n get numbers() {\n return this._numbers;\n }\n\n displayNumbers(numbers) {\n if (!this.innerSpan) return;\n const numberStrings = numbers.map(index => {\n return index == -1 ? ""?"" : index + 1 + """";\n });\n const textContent = ""["" + numberStrings.join("", "") + ""]"";\n this.innerSpan.textContent = textContent;\n }\n\n set entries(entries) {\n this._entries = entries;\n this.displayEntries(entries);\n }\n\n get entries() {\n return this._entries;\n }\n\n displayEntries(entries) {\n if (!this.hoverBox) return;\n this.hoverBox.innerHTML = `
    \n ${entries\n .map(hover_cite)\n .map(html => `
  • ${html}
    • `)\n .join(""\n"")}\n
`;\n }\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const styles$1 = `\nd-citation-list {\n contain: style;\n}\n\nd-citation-list .references {\n grid-column: text;\n}\n\nd-citation-list .references .title {\n font-weight: 500;\n}\n`;\n\n function renderCitationList(element, entries, dom=document) {\n if (entries.size > 0) {\n element.style.display = '';\n let list = element.querySelector('.references');\n if (list) {\n list.innerHTML = '';\n } else {\n const stylesTag = dom.createElement('style');\n stylesTag.innerHTML = styles$1;\n element.appendChild(stylesTag);\n\n const heading = dom.createElement('h3');\n heading.id = 'references';\n heading.textContent = 'References';\n element.appendChild(heading);\n\n list = dom.createElement('ol');\n list.id = 'references-list';\n list.className = 'references';\n element.appendChild(list);\n }\n\n for (const [key, entry] of entries) {\n const listItem = dom.createElement('li');\n listItem.id = key;\n listItem.innerHTML = bibliography_cite(entry);\n list.appendChild(listItem);\n }\n } else {\n element.style.display = 'none';\n }\n }\n\n class CitationList extends HTMLElement {\n\n static get is() { return 'd-citation-list'; }\n\n connectedCallback() {\n if (!this.hasAttribute('distill-prerendered')) {\n this.style.display = 'none';\n }\n }\n\n set citations(citations) {\n renderCitationList(this, citations);\n }\n\n }\n\n var prism = createCommonjsModule(function (module) {\n /* **********************************************\n Begin prism-core.js\n ********************************************** */\n\n var _self = (typeof window !== 'undefined')\n \t? window // if in browser\n \t: (\n \t\t(typeof WorkerGlobalScope !== 'undefined' && self instanceof WorkerGlobalScope)\n \t\t? self // if in worker\n \t\t: {} // if in node js\n \t);\n\n /**\n * Prism: Lightweight, robust, elegant syntax highlighting\n * MIT license http://www.opensource.org/licenses/mit-license.php/\n * @author Lea Verou http://lea.verou.me\n */\n\n var Prism = (function (_self){\n\n // Private helper vars\n var lang = /\blang(?:uage)?-([\w-]+)\b/i;\n var uniqueId = 0;\n\n\n var _ = {\n \tmanual: _self.Prism && _self.Prism.manual,\n \tdisableWorkerMessageHandler: _self.Prism && _self.Prism.disableWorkerMessageHandler,\n \tutil: {\n \t\tencode: function encode(tokens) {\n \t\t\tif (tokens instanceof Token) {\n \t\t\t\treturn new Token(tokens.type, encode(tokens.content), tokens.alias);\n \t\t\t} else if (Array.isArray(tokens)) {\n \t\t\t\treturn tokens.map(encode);\n \t\t\t} else {\n \t\t\t\treturn tokens.replace(/&/g, '&').replace(/' + env.content + '';\n };\n\n /**\n * @param {string} text\n * @param {LinkedList} tokenList\n * @param {any} grammar\n * @param {LinkedListNode} startNode\n * @param {number} startPos\n * @param {boolean} [oneshot=false]\n * @param {string} [target]\n */\n function matchGrammar(text, tokenList, grammar, startNode, startPos, oneshot, target) {\n \tfor (var token in grammar) {\n \t\tif (!grammar.hasOwnProperty(token) || !grammar[token]) {\n \t\t\tcontinue;\n \t\t}\n\n \t\tvar patterns = grammar[token];\n \t\tpatterns = Array.isArray(patterns) ? patterns : [patterns];\n\n \t\tfor (var j = 0; j < patterns.length; ++j) {\n \t\t\tif (target && target == token + ',' + j) {\n \t\t\t\treturn;\n \t\t\t}\n\n \t\t\tvar pattern = patterns[j],\n \t\t\t\tinside = pattern.inside,\n \t\t\t\tlookbehind = !!pattern.lookbehind,\n \t\t\t\tgreedy = !!pattern.greedy,\n \t\t\t\tlookbehindLength = 0,\n \t\t\t\talias = pattern.alias;\n\n \t\t\tif (greedy && !pattern.pattern.global) {\n \t\t\t\t// Without the global flag, lastIndex won't work\n \t\t\t\tvar flags = pattern.pattern.toString().match(/[imsuy]*$/)[0];\n \t\t\t\tpattern.pattern = RegExp(pattern.pattern.source, flags + 'g');\n \t\t\t}\n\n \t\t\tpattern = pattern.pattern || pattern;\n\n \t\t\tfor ( // iterate the token list and keep track of the current token/string position\n \t\t\t\tvar currentNode = startNode.next, pos = startPos;\n \t\t\t\tcurrentNode !== tokenList.tail;\n \t\t\t\tpos += currentNode.value.length, currentNode = currentNode.next\n \t\t\t) {\n\n \t\t\t\tvar str = currentNode.value;\n\n \t\t\t\tif (tokenList.length > text.length) {\n \t\t\t\t\t// Something went terribly wrong, ABORT, ABORT!\n \t\t\t\t\treturn;\n \t\t\t\t}\n\n \t\t\t\tif (str instanceof Token) {\n \t\t\t\t\tcontinue;\n \t\t\t\t}\n\n \t\t\t\tvar removeCount = 1; // this is the to parameter of removeBetween\n\n \t\t\t\tif (greedy && currentNode != tokenList.tail.prev) {\n \t\t\t\t\tpattern.lastIndex = pos;\n \t\t\t\t\tvar match = pattern.exec(text);\n \t\t\t\t\tif (!match) {\n \t\t\t\t\t\tbreak;\n \t\t\t\t\t}\n\n \t\t\t\t\tvar from = match.index + (lookbehind && match[1] ? match[1].length : 0);\n \t\t\t\t\tvar to = match.index + match[0].length;\n \t\t\t\t\tvar p = pos;\n\n \t\t\t\t\t// find the node that contains the match\n \t\t\t\t\tp += currentNode.value.length;\n \t\t\t\t\twhile (from >= p) {\n \t\t\t\t\t\tcurrentNode = currentNode.next;\n \t\t\t\t\t\tp += currentNode.value.length;\n \t\t\t\t\t}\n \t\t\t\t\t// adjust pos (and p)\n \t\t\t\t\tp -= currentNode.value.length;\n \t\t\t\t\tpos = p;\n\n \t\t\t\t\t// the current node is a Token, then the match starts inside another Token, which is invalid\n \t\t\t\t\tif (currentNode.value instanceof Token) {\n \t\t\t\t\t\tcontinue;\n \t\t\t\t\t}\n\n \t\t\t\t\t// find the last node which is affected by this match\n \t\t\t\t\tfor (\n \t\t\t\t\t\tvar k = currentNode;\n \t\t\t\t\t\tk !== tokenList.tail && (p < to || (typeof k.value === 'string' && !k.prev.value.greedy));\n \t\t\t\t\t\tk = k.next\n \t\t\t\t\t) {\n \t\t\t\t\t\tremoveCount++;\n \t\t\t\t\t\tp += k.value.length;\n \t\t\t\t\t}\n \t\t\t\t\tremoveCount--;\n\n \t\t\t\t\t// replace with the new match\n \t\t\t\t\tstr = text.slice(pos, p);\n \t\t\t\t\tmatch.index -= pos;\n \t\t\t\t} else {\n \t\t\t\t\tpattern.lastIndex = 0;\n\n \t\t\t\t\tvar match = pattern.exec(str);\n \t\t\t\t}\n\n \t\t\t\tif (!match) {\n \t\t\t\t\tif (oneshot) {\n \t\t\t\t\t\tbreak;\n \t\t\t\t\t}\n\n \t\t\t\t\tcontinue;\n \t\t\t\t}\n\n \t\t\t\tif (lookbehind) {\n \t\t\t\t\tlookbehindLength = match[1] ? match[1].length : 0;\n \t\t\t\t}\n\n \t\t\t\tvar from = match.index + lookbehindLength,\n \t\t\t\t\tmatch = match[0].slice(lookbehindLength),\n \t\t\t\t\tto = from + match.length,\n \t\t\t\t\tbefore = str.slice(0, from),\n \t\t\t\t\tafter = str.slice(to);\n\n \t\t\t\tvar removeFrom = currentNode.prev;\n\n \t\t\t\tif (before) {\n \t\t\t\t\tremoveFrom = addAfter(tokenList, removeFrom, before);\n \t\t\t\t\tpos += before.length;\n \t\t\t\t}\n\n \t\t\t\tremoveRange(tokenList, removeFrom, removeCount);\n\n \t\t\t\tvar wrapped = new Token(token, inside ? _.tokenize(match, inside) : match, alias, match, greedy);\n \t\t\t\tcurrentNode = addAfter(tokenList, removeFrom, wrapped);\n\n \t\t\t\tif (after) {\n \t\t\t\t\taddAfter(tokenList, currentNode, after);\n \t\t\t\t}\n\n\n \t\t\t\tif (removeCount > 1)\n \t\t\t\t\tmatchGrammar(text, tokenList, grammar, currentNode.prev, pos, true, token + ',' + j);\n\n \t\t\t\tif (oneshot)\n \t\t\t\t\tbreak;\n \t\t\t}\n \t\t}\n \t}\n }\n\n /**\n * @typedef LinkedListNode\n * @property {T} value\n * @property {LinkedListNode | null} prev The previous node.\n * @property {LinkedListNode | null} next The next node.\n * @template T\n */\n\n /**\n * @template T\n */\n function LinkedList() {\n \t/** @type {LinkedListNode} */\n \tvar head = { value: null, prev: null, next: null };\n \t/** @type {LinkedListNode} */\n \tvar tail = { value: null, prev: head, next: null };\n \thead.next = tail;\n\n \t/** @type {LinkedListNode} */\n \tthis.head = head;\n \t/** @type {LinkedListNode} */\n \tthis.tail = tail;\n \tthis.length = 0;\n }\n\n /**\n * Adds a new node with the given value to the list.\n * @param {LinkedList} list\n * @param {LinkedListNode} node\n * @param {T} value\n * @returns {LinkedListNode} The added node.\n * @template T\n */\n function addAfter(list, node, value) {\n \t// assumes that node != list.tail && values.length >= 0\n \tvar next = node.next;\n\n \tvar newNode = { value: value, prev: node, next: next };\n \tnode.next = newNode;\n \tnext.prev = newNode;\n \tlist.length++;\n\n \treturn newNode;\n }\n /**\n * Removes `count` nodes after the given node. The given node will not be removed.\n * @param {LinkedList} list\n * @param {LinkedListNode} node\n * @param {number} count\n * @template T\n */\n function removeRange(list, node, count) {\n \tvar next = node.next;\n \tfor (var i = 0; i < count && next !== list.tail; i++) {\n \t\tnext = next.next;\n \t}\n \tnode.next = next;\n \tnext.prev = node;\n \tlist.length -= i;\n }\n /**\n * @param {LinkedList} list\n * @returns {T[]}\n * @template T\n */\n function toArray(list) {\n \tvar array = [];\n \tvar node = list.head.next;\n \twhile (node !== list.tail) {\n \t\tarray.push(node.value);\n \t\tnode = node.next;\n \t}\n \treturn array;\n }\n\n\n if (!_self.document) {\n \tif (!_self.addEventListener) {\n \t\t// in Node.js\n \t\treturn _;\n \t}\n\n \tif (!_.disableWorkerMessageHandler) {\n \t\t// In worker\n \t\t_self.addEventListener('message', function (evt) {\n \t\t\tvar message = JSON.parse(evt.data),\n \t\t\t\tlang = message.language,\n \t\t\t\tcode = message.code,\n \t\t\t\timmediateClose = message.immediateClose;\n\n \t\t\t_self.postMessage(_.highlight(code, _.languages[lang], lang));\n \t\t\tif (immediateClose) {\n \t\t\t\t_self.close();\n \t\t\t}\n \t\t}, false);\n \t}\n\n \treturn _;\n }\n\n //Get current script and highlight\n var script = _.util.currentScript();\n\n if (script) {\n \t_.filename = script.src;\n\n \tif (script.hasAttribute('data-manual')) {\n \t\t_.manual = true;\n \t}\n }\n\n function highlightAutomaticallyCallback() {\n \tif (!_.manual) {\n \t\t_.highlightAll();\n \t}\n }\n\n if (!_.manual) {\n \t// If the document state is ""loading"", then we'll use DOMContentLoaded.\n \t// If the document state is ""interactive"" and the prism.js script is deferred, then we'll also use the\n \t// DOMContentLoaded event because there might be some plugins or languages which have also been deferred and they\n \t// might take longer one animation frame to execute which can create a race condition where only some plugins have\n \t// been loaded when Prism.highlightAll() is executed, depending on how fast resources are loaded.\n \t// See https://github.com/PrismJS/prism/issues/2102\n \tvar readyState = document.readyState;\n \tif (readyState === 'loading' || readyState === 'interactive' && script && script.defer) {\n \t\tdocument.addEventListener('DOMContentLoaded', highlightAutomaticallyCallback);\n \t} else {\n \t\tif (window.requestAnimationFrame) {\n \t\t\twindow.requestAnimationFrame(highlightAutomaticallyCallback);\n \t\t} else {\n \t\t\twindow.setTimeout(highlightAutomaticallyCallback, 16);\n \t\t}\n \t}\n }\n\n return _;\n\n })(_self);\n\n if ( module.exports) {\n \tmodule.exports = Prism;\n }\n\n // hack for components to work correctly in node.js\n if (typeof commonjsGlobal !== 'undefined') {\n \tcommonjsGlobal.Prism = Prism;\n }\n\n\n /* **********************************************\n Begin prism-markup.js\n ********************************************** */\n\n Prism.languages.markup = {\n \t'comment': //,\n \t'prolog': /<\?[\s\S]+?\?>/,\n \t'doctype': {\n \t\tpattern: /""'[\]]|""[^""]*""|'[^']*')+(?:\[(?:(?!)*\]\s*)?>/i,\n \t\tgreedy: true\n \t},\n \t'cdata': //i,\n \t'tag': {\n \t\tpattern: /<\/?(?!\d)[^\s>\/=$<%]+(?:\s(?:\s*[^\s>\/=]+(?:\s*=\s*(?:""[^""]*""|'[^']*'|[^\s'"">=]+(?=[\s>]))|(?=[\s/>])))+)?\s*\/?>/i,\n \t\tgreedy: true,\n \t\tinside: {\n \t\t\t'tag': {\n \t\t\t\tpattern: /^<\/?[^\s>\/]+/i,\n \t\t\t\tinside: {\n \t\t\t\t\t'punctuation': /^<\/?/,\n \t\t\t\t\t'namespace': /^[^\s>\/:]+:/\n \t\t\t\t}\n \t\t\t},\n \t\t\t'attr-value': {\n \t\t\t\tpattern: /=\s*(?:""[^""]*""|'[^']*'|[^\s'"">=]+)/i,\n \t\t\t\tinside: {\n \t\t\t\t\t'punctuation': [\n \t\t\t\t\t\t/^=/,\n \t\t\t\t\t\t{\n \t\t\t\t\t\t\tpattern: /^(\s*)[""']|[""']$/,\n \t\t\t\t\t\t\tlookbehind: true\n \t\t\t\t\t\t}\n \t\t\t\t\t]\n \t\t\t\t}\n \t\t\t},\n \t\t\t'punctuation': /\/?>/,\n \t\t\t'attr-name': {\n \t\t\t\tpattern: /[^\s>\/]+/,\n \t\t\t\tinside: {\n \t\t\t\t\t'namespace': /^[^\s>\/:]+:/\n \t\t\t\t}\n \t\t\t}\n\n \t\t}\n \t},\n \t'entity': /&#?[\da-z]{1,8};/i\n };\n\n Prism.languages.markup['tag'].inside['attr-value'].inside['entity'] =\n \tPrism.languages.markup['entity'];\n\n // Plugin to make entity title show the real entity, idea by Roman Komarov\n Prism.hooks.add('wrap', function(env) {\n\n \tif (env.type === 'entity') {\n \t\tenv.attributes['title'] = env.content.replace(/&/, '&');\n \t}\n });\n\n Object.defineProperty(Prism.languages.markup.tag, 'addInlined', {\n \t/**\n \t * Adds an inlined language to markup.\n \t *\n \t * An example of an inlined language is CSS with `\n\n\n\n`);\n\n class Code extends Mutating(T$4(HTMLElement)) {\n\n renderContent() {\n\n // check if language can be highlighted\n this.languageName = this.getAttribute('language');\n if (!this.languageName) {\n console.warn('You need to provide a language attribute to your block to let us know how to highlight your code; e.g.:\n zeros = np.zeros(shape).');\n return;\n }\n const language = prism.languages[this.languageName];\n if (language == undefined) {\n console.warn(`Distill does not yet support highlighting your code block in ""${this.languageName}'.`);\n return;\n }\n\n let content = this.textContent;\n const codeTag = this.shadowRoot.querySelector('#code-container');\n\n if (this.hasAttribute('block')) {\n // normalize the tab indents\n content = content.replace(/\n/, '');\n const tabs = content.match(/\s*/);\n content = content.replace(new RegExp('\n' + tabs, 'g'), '\n');\n content = content.trim();\n // wrap code block in pre tag if needed\n if (codeTag.parentNode instanceof ShadowRoot) {\n const preTag = document.createElement('pre');\n this.shadowRoot.removeChild(codeTag);\n preTag.appendChild(codeTag);\n this.shadowRoot.appendChild(preTag);\n }\n\n }\n\n codeTag.className = `language-${this.languageName}`;\n codeTag.innerHTML = prism.highlight(content, language);\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const T$5 = Template('d-footnote', `\n\n\n\n
\n \n
\n\n\n\n \n\n\n`);\n\n class Footnote extends T$5(HTMLElement) {\n\n constructor() {\n super();\n\n const options = {childList: true, characterData: true, subtree: true};\n const observer = new MutationObserver(this.notify);\n observer.observe(this, options);\n }\n\n notify() {\n const options = { detail: this, bubbles: true };\n const event = new CustomEvent('onFootnoteChanged', options);\n document.dispatchEvent(event);\n }\n\n connectedCallback() {\n // listen and notify about changes to slotted content\n // const slot = this.shadowRoot.querySelector('#slot');\n // console.warn(slot.textContent);\n // slot.addEventListener('slotchange', this.notify);\n this.hoverBox = this.root.querySelector('d-hover-box');\n window.customElements.whenDefined('d-hover-box').then(() => {\n this.hoverBox.listen(this);\n });\n // create numeric ID\n Footnote.currentFootnoteId += 1;\n const IdString = Footnote.currentFootnoteId.toString();\n this.root.host.id = 'd-footnote-' + IdString;\n\n // set up hidden hover box\n const id = 'dt-fn-hover-box-' + IdString;\n this.hoverBox.id = id;\n\n // set up visible footnote marker\n const span = this.root.querySelector('#fn-');\n span.setAttribute('id', 'fn-' + IdString);\n span.setAttribute('data-hover-ref', id);\n span.textContent = IdString;\n }\n\n }\n\n Footnote.currentFootnoteId = 0;\n\n // Copyright 2018 The Distill Template Authors\n\n const T$6 = Template('d-footnote-list', `\n\n\n

Footnotes

\n
    \n`, false);\n\n class FootnoteList extends T$6(HTMLElement) {\n\n connectedCallback() {\n super.connectedCallback();\n\n this.list = this.root.querySelector('ol');\n // footnotes list is initially hidden\n this.root.style.display = 'none';\n // look through document and register existing footnotes\n // Store.subscribeTo('footnotes', (footnote) => {\n // this.renderFootnote(footnote);\n // });\n }\n\n // TODO: could optimize this to accept individual footnotes?\n set footnotes(footnotes) {\n this.list.innerHTML = '';\n if (footnotes.length) {\n // ensure footnote list is visible\n this.root.style.display = '';\n\n for (const footnote of footnotes) {\n // construct and append list item to show footnote\n const listItem = document.createElement('li');\n listItem.id = footnote.id + '-listing';\n listItem.innerHTML = footnote.innerHTML;\n\n const backlink = document.createElement('a');\n backlink.setAttribute('class', 'footnote-backlink');\n backlink.textContent = '[↩]';\n backlink.href = '#' + footnote.id;\n\n listItem.appendChild(backlink);\n this.list.appendChild(listItem);\n }\n } else {\n // ensure footnote list is invisible\n this.root.style.display = 'none';\n }\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const T$7 = Template('d-hover-box', `\n\n\n
    \n
    \n \n
    \n
    \n`);\n\n class HoverBox extends T$7(HTMLElement) {\n\n constructor() {\n super();\n }\n\n connectedCallback() {\n\n }\n\n listen(element) {\n // console.log(element)\n this.bindDivEvents(this);\n this.bindTriggerEvents(element);\n // this.style.display = ""block"";\n }\n\n bindDivEvents(element) {\n // For mice, same behavior as hovering on links\n element.addEventListener('mouseover', () => {\n if (!this.visible) this.showAtNode(element);\n this.stopTimeout();\n });\n element.addEventListener('mouseout', () => {\n this.extendTimeout(500);\n });\n // Don't trigger body touchstart event when touching within box\n element.addEventListener('touchstart', (event) => {\n event.stopPropagation();\n }, {passive: true});\n // Close box when touching outside box\n document.body.addEventListener('touchstart', () => {\n this.hide();\n }, {passive: true});\n }\n\n bindTriggerEvents(node) {\n node.addEventListener('mouseover', () => {\n if (!this.visible) {\n this.showAtNode(node);\n }\n this.stopTimeout();\n });\n\n node.addEventListener('mouseout', () => {\n this.extendTimeout(300);\n });\n\n node.addEventListener('touchstart', (event) => {\n if (this.visible) {\n this.hide();\n } else {\n this.showAtNode(node);\n }\n // Don't trigger body touchstart event when touching link\n event.stopPropagation();\n }, {passive: true});\n }\n\n show(position) {\n this.visible = true;\n this.style.display = 'block';\n // 10px extra offset from element\n this.style.top = Math.round(position[1] + 10) + 'px';\n }\n\n showAtNode(node) {\n // https://developer.mozilla.org/en-US/docs/Web/API/HTMLElement/offsetTop\n const bbox = node.getBoundingClientRect();\n this.show([node.offsetLeft + bbox.width, node.offsetTop + bbox.height]);\n }\n\n hide() {\n this.visible = false;\n this.style.display = 'none';\n this.stopTimeout();\n }\n\n stopTimeout() {\n if (this.timeout) {\n clearTimeout(this.timeout);\n }\n }\n\n extendTimeout(time) {\n this.stopTimeout();\n this.timeout = setTimeout(() => {\n this.hide();\n }, time);\n }\n\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n class Title extends HTMLElement {\n static get is() { return 'd-title'; }\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const T$8 = Template('d-references', `\n\n`, false);\n\n class References extends T$8(HTMLElement) {\n\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n class TOC extends HTMLElement {\n\n static get is() { return 'd-toc'; }\n\n connectedCallback() {\n if (!this.getAttribute('prerendered')) {\n window.onload = () => {\n const article = document.querySelector('d-article');\n const headings = article.querySelectorAll('h2, h3');\n renderTOC(this, headings);\n };\n }\n }\n\n }\n\n function renderTOC(element, headings) {\n\n let ToC =`\n \n \n

    Table of contents

    \n
      `;\n\n for (const el of headings) {\n // should element be included in TOC?\n const isInTitle = el.parentElement.tagName == 'D-TITLE';\n const isException = el.getAttribute('no-toc');\n if (isInTitle || isException) continue;\n // create TOC entry\n const title = el.textContent;\n const link = '#' + el.getAttribute('id');\n\n let newLine = '
    • ' + '' + title + '' + '
      • ';\n if (el.tagName == 'H3') {\n newLine = '
        ' + newLine + '
      ';\n } else {\n newLine += '
      ';\n }\n ToC += newLine;\n\n }\n\n ToC += '
    ';\n element.innerHTML = ToC;\n }\n\n // Copyright 2018 The Distill Template Authors\n //\n // Licensed under the Apache License, Version 2.0 (the ""License"");\n // you may not use this file except in compliance with the License.\n // You may obtain a copy of the License at\n //\n // http://www.apache.org/licenses/LICENSE-2.0\n //\n // Unless required by applicable law or agreed to in writing, software\n // distributed under the License is distributed on an ""AS IS"" BASIS,\n // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n // See the License for the specific language governing permissions and\n // limitations under the License.\n\n // Figure\n //\n // d-figure provides a state-machine of visibility events:\n //\n // scroll out of view\n // +----------------+\n // *do work here* | |\n // +----------------+ +-+---------+ +-v---------+\n // | ready +----> onscreen | | offscreen |\n // +----------------+ +---------^-+ +---------+-+\n // | |\n // +----------------+\n // scroll into view\n //\n\n class Figure extends HTMLElement {\n\n static get is() { return 'd-figure'; }\n\n static get readyQueue() {\n if (!Figure._readyQueue) {\n Figure._readyQueue = [];\n }\n return Figure._readyQueue;\n }\n\n static addToReadyQueue(figure) {\n if (Figure.readyQueue.indexOf(figure) === -1) {\n Figure.readyQueue.push(figure);\n Figure.runReadyQueue();\n }\n }\n\n static runReadyQueue() {\n // console.log(""Checking to run readyQueue, length: "" + Figure.readyQueue.length + "", scrolling: "" + Figure.isScrolling);\n // if (Figure.isScrolling) return;\n // console.log(""Running ready Queue"");\n const figure = Figure.readyQueue\n .sort((a,b) => a._seenOnScreen - b._seenOnScreen )\n .filter((figure) => !figure._ready)\n .pop();\n if (figure) {\n figure.ready();\n requestAnimationFrame(Figure.runReadyQueue);\n }\n\n }\n\n constructor() {\n super();\n // debugger\n this._ready = false;\n this._onscreen = false;\n this._offscreen = true;\n }\n\n connectedCallback() {\n this.loadsWhileScrolling = this.hasAttribute('loadsWhileScrolling');\n Figure.marginObserver.observe(this);\n Figure.directObserver.observe(this);\n }\n\n disconnectedCallback() {\n Figure.marginObserver.unobserve(this);\n Figure.directObserver.unobserve(this);\n }\n\n // We use two separate observers:\n // One with an extra 1000px margin to warn if the viewpoint gets close,\n // And one for the actual on/off screen events\n\n static get marginObserver() {\n if (!Figure._marginObserver) {\n // if (!('IntersectionObserver' in window)) {\n // throw new Error('no interscetionobbserver!');\n // }\n const viewportHeight = window.innerHeight;\n const margin = Math.floor(2 * viewportHeight);\n const options = {rootMargin: margin + 'px 0px ' + margin + 'px 0px', threshold: 0.01};\n const callback = Figure.didObserveMarginIntersection;\n const observer = new IntersectionObserver(callback, options);\n Figure._marginObserver = observer;\n }\n return Figure._marginObserver;\n }\n\n static didObserveMarginIntersection(entries) {\n for (const entry of entries) {\n const figure = entry.target;\n if (entry.isIntersecting && !figure._ready) {\n Figure.addToReadyQueue(figure);\n }\n }\n }\n\n static get directObserver() {\n if (!Figure._directObserver) {\n Figure._directObserver = new IntersectionObserver(\n Figure.didObserveDirectIntersection, {\n rootMargin: '0px', threshold: [0, 1.0],\n }\n );\n }\n return Figure._directObserver;\n }\n\n static didObserveDirectIntersection(entries) {\n for (const entry of entries) {\n const figure = entry.target;\n if (entry.isIntersecting) {\n figure._seenOnScreen = new Date();\n // if (!figure._ready) { figure.ready(); }\n if (figure._offscreen) { figure.onscreen(); }\n } else {\n if (figure._onscreen) { figure.offscreen(); }\n }\n }\n }\n\n // Notify listeners that registered late, too:\n\n addEventListener(eventName, callback) {\n super.addEventListener(eventName, callback);\n // if we had already dispatched something while presumingly no one was listening, we do so again\n // debugger\n if (eventName === 'ready') {\n if (Figure.readyQueue.indexOf(this) !== -1) {\n this._ready = false;\n Figure.runReadyQueue();\n }\n }\n if (eventName === 'onscreen') {\n this.onscreen();\n }\n }\n\n // Custom Events\n\n ready() {\n // debugger\n this._ready = true;\n Figure.marginObserver.unobserve(this);\n const event = new CustomEvent('ready');\n this.dispatchEvent(event);\n }\n\n onscreen() {\n this._onscreen = true;\n this._offscreen = false;\n const event = new CustomEvent('onscreen');\n this.dispatchEvent(event);\n }\n\n offscreen() {\n this._onscreen = false;\n this._offscreen = true;\n const event = new CustomEvent('offscreen');\n this.dispatchEvent(event);\n }\n\n }\n\n if (typeof window !== 'undefined') {\n\n Figure.isScrolling = false;\n let timeout;\n const resetTimer = () => {\n Figure.isScrolling = true;\n clearTimeout(timeout);\n timeout = setTimeout(() => {\n Figure.isScrolling = false;\n Figure.runReadyQueue();\n }, 500);\n };\n window.addEventListener('scroll', resetTimer, true);\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n // This overlay is not secure.\n // It is only meant as a social deterrent.\n\n const productionHostname = 'distill.pub';\n const T$9 = Template('d-interstitial', `\n\n\n
    \n
    \n

    This article is in review.

    \n

    Do not share this URL or the contents of this article. Thank you!

    \n \n

    Enter the password we shared with you as part of the review process to view the article.

    \n
    \n
    \n`);\n\n class Interstitial extends T$9(HTMLElement) {\n\n connectedCallback() {\n if (this.shouldRemoveSelf()) {\n this.parentElement.removeChild(this);\n } else {\n const passwordInput = this.root.querySelector('#interstitial-password-input');\n passwordInput.oninput = (event) => this.passwordChanged(event);\n }\n }\n\n passwordChanged(event) {\n const entered = event.target.value;\n if (entered === this.password) {\n console.log('Correct password entered.');\n this.parentElement.removeChild(this);\n if (typeof(Storage) !== 'undefined') {\n console.log('Saved that correct password was entered.');\n localStorage.setItem(this.localStorageIdentifier(), 'true');\n }\n }\n }\n\n shouldRemoveSelf() {\n // should never be visible in production\n if (window && window.location.hostname === productionHostname) {\n console.warn('Interstitial found on production, hiding it.');\n return true\n }\n // should only have to enter password once\n if (typeof(Storage) !== 'undefined') {\n if (localStorage.getItem(this.localStorageIdentifier()) === 'true') {\n console.log('Loaded that correct password was entered before; skipping interstitial.');\n return true;\n }\n }\n // otherwise, leave visible\n return false;\n }\n\n localStorageIdentifier() {\n const prefix = 'distill-drafts';\n const suffix = 'interstitial-password-correct';\n return prefix + (window ? window.location.pathname : '-') + suffix\n }\n\n }\n\n function ascending(a, b) {\n return a < b ? -1 : a > b ? 1 : a >= b ? 0 : NaN;\n }\n\n function bisector(compare) {\n if (compare.length === 1) compare = ascendingComparator(compare);\n return {\n left: function(a, x, lo, hi) {\n if (lo == null) lo = 0;\n if (hi == null) hi = a.length;\n while (lo < hi) {\n var mid = lo + hi >>> 1;\n if (compare(a[mid], x) < 0) lo = mid + 1;\n else hi = mid;\n }\n return lo;\n },\n right: function(a, x, lo, hi) {\n if (lo == null) lo = 0;\n if (hi == null) hi = a.length;\n while (lo < hi) {\n var mid = lo + hi >>> 1;\n if (compare(a[mid], x) > 0) hi = mid;\n else lo = mid + 1;\n }\n return lo;\n }\n };\n }\n\n function ascendingComparator(f) {\n return function(d, x) {\n return ascending(f(d), x);\n };\n }\n\n var ascendingBisect = bisector(ascending);\n var bisectRight = ascendingBisect.right;\n\n function range(start, stop, step) {\n start = +start, stop = +stop, step = (n = arguments.length) < 2 ? (stop = start, start = 0, 1) : n < 3 ? 1 : +step;\n\n var i = -1,\n n = Math.max(0, Math.ceil((stop - start) / step)) | 0,\n range = new Array(n);\n\n while (++i < n) {\n range[i] = start + i * step;\n }\n\n return range;\n }\n\n var e10 = Math.sqrt(50),\n e5 = Math.sqrt(10),\n e2 = Math.sqrt(2);\n\n function ticks(start, stop, count) {\n var reverse,\n i = -1,\n n,\n ticks,\n step;\n\n stop = +stop, start = +start, count = +count;\n if (start === stop && count > 0) return [start];\n if (reverse = stop < start) n = start, start = stop, stop = n;\n if ((step = tickIncrement(start, stop, count)) === 0 || !isFinite(step)) return [];\n\n if (step > 0) {\n start = Math.ceil(start / step);\n stop = Math.floor(stop / step);\n ticks = new Array(n = Math.ceil(stop - start + 1));\n while (++i < n) ticks[i] = (start + i) * step;\n } else {\n start = Math.floor(start * step);\n stop = Math.ceil(stop * step);\n ticks = new Array(n = Math.ceil(start - stop + 1));\n while (++i < n) ticks[i] = (start - i) / step;\n }\n\n if (reverse) ticks.reverse();\n\n return ticks;\n }\n\n function tickIncrement(start, stop, count) {\n var step = (stop - start) / Math.max(0, count),\n power = Math.floor(Math.log(step) / Math.LN10),\n error = step / Math.pow(10, power);\n return power >= 0\n ? (error >= e10 ? 10 : error >= e5 ? 5 : error >= e2 ? 2 : 1) * Math.pow(10, power)\n : -Math.pow(10, -power) / (error >= e10 ? 10 : error >= e5 ? 5 : error >= e2 ? 2 : 1);\n }\n\n function tickStep(start, stop, count) {\n var step0 = Math.abs(stop - start) / Math.max(0, count),\n step1 = Math.pow(10, Math.floor(Math.log(step0) / Math.LN10)),\n error = step0 / step1;\n if (error >= e10) step1 *= 10;\n else if (error >= e5) step1 *= 5;\n else if (error >= e2) step1 *= 2;\n return stop < start ? -step1 : step1;\n }\n\n function initRange(domain, range) {\n switch (arguments.length) {\n case 0: break;\n case 1: this.range(domain); break;\n default: this.range(range).domain(domain); break;\n }\n return this;\n }\n\n function define(constructor, factory, prototype) {\n constructor.prototype = factory.prototype = prototype;\n prototype.constructor = constructor;\n }\n\n function extend(parent, definition) {\n var prototype = Object.create(parent.prototype);\n for (var key in definition) prototype[key] = definition[key];\n return prototype;\n }\n\n function Color() {}\n\n var darker = 0.7;\n var brighter = 1 / darker;\n\n var reI = ""\\s*([+-]?\\d+)\\s*"",\n reN = ""\\s*([+-]?\\d*\\.?\\d+(?:[eE][+-]?\\d+)?)\\s*"",\n reP = ""\\s*([+-]?\\d*\\.?\\d+(?:[eE][+-]?\\d+)?)%\\s*"",\n reHex = /^#([0-9a-f]{3,8})$/,\n reRgbInteger = new RegExp(""^rgb\\("" + [reI, reI, reI] + ""\\)$""),\n reRgbPercent = new RegExp(""^rgb\\("" + [reP, reP, reP] + ""\\)$""),\n reRgbaInteger = new RegExp(""^rgba\\("" + [reI, reI, reI, reN] + ""\\)$""),\n reRgbaPercent = new RegExp(""^rgba\\("" + [reP, reP, reP, reN] + ""\\)$""),\n reHslPercent = new RegExp(""^hsl\\("" + [reN, reP, reP] + ""\\)$""),\n reHslaPercent = new RegExp(""^hsla\\("" + [reN, reP, reP, reN] + ""\\)$"");\n\n var named = {\n aliceblue: 0xf0f8ff,\n antiquewhite: 0xfaebd7,\n aqua: 0x00ffff,\n aquamarine: 0x7fffd4,\n azure: 0xf0ffff,\n beige: 0xf5f5dc,\n bisque: 0xffe4c4,\n black: 0x000000,\n blanchedalmond: 0xffebcd,\n blue: 0x0000ff,\n blueviolet: 0x8a2be2,\n brown: 0xa52a2a,\n burlywood: 0xdeb887,\n cadetblue: 0x5f9ea0,\n chartreuse: 0x7fff00,\n chocolate: 0xd2691e,\n coral: 0xff7f50,\n cornflowerblue: 0x6495ed,\n cornsilk: 0xfff8dc,\n crimson: 0xdc143c,\n cyan: 0x00ffff,\n darkblue: 0x00008b,\n darkcyan: 0x008b8b,\n darkgoldenrod: 0xb8860b,\n darkgray: 0xa9a9a9,\n darkgreen: 0x006400,\n darkgrey: 0xa9a9a9,\n darkkhaki: 0xbdb76b,\n darkmagenta: 0x8b008b,\n darkolivegreen: 0x556b2f,\n darkorange: 0xff8c00,\n darkorchid: 0x9932cc,\n darkred: 0x8b0000,\n darksalmon: 0xe9967a,\n darkseagreen: 0x8fbc8f,\n darkslateblue: 0x483d8b,\n darkslategray: 0x2f4f4f,\n darkslategrey: 0x2f4f4f,\n darkturquoise: 0x00ced1,\n darkviolet: 0x9400d3,\n deeppink: 0xff1493,\n deepskyblue: 0x00bfff,\n dimgray: 0x696969,\n dimgrey: 0x696969,\n dodgerblue: 0x1e90ff,\n firebrick: 0xb22222,\n floralwhite: 0xfffaf0,\n forestgreen: 0x228b22,\n fuchsia: 0xff00ff,\n gainsboro: 0xdcdcdc,\n ghostwhite: 0xf8f8ff,\n gold: 0xffd700,\n goldenrod: 0xdaa520,\n gray: 0x808080,\n green: 0x008000,\n greenyellow: 0xadff2f,\n grey: 0x808080,\n honeydew: 0xf0fff0,\n hotpink: 0xff69b4,\n indianred: 0xcd5c5c,\n indigo: 0x4b0082,\n ivory: 0xfffff0,\n khaki: 0xf0e68c,\n lavender: 0xe6e6fa,\n lavenderblush: 0xfff0f5,\n lawngreen: 0x7cfc00,\n lemonchiffon: 0xfffacd,\n lightblue: 0xadd8e6,\n lightcoral: 0xf08080,\n lightcyan: 0xe0ffff,\n lightgoldenrodyellow: 0xfafad2,\n lightgray: 0xd3d3d3,\n lightgreen: 0x90ee90,\n lightgrey: 0xd3d3d3,\n lightpink: 0xffb6c1,\n lightsalmon: 0xffa07a,\n lightseagreen: 0x20b2aa,\n lightskyblue: 0x87cefa,\n lightslategray: 0x778899,\n lightslategrey: 0x778899,\n lightsteelblue: 0xb0c4de,\n lightyellow: 0xffffe0,\n lime: 0x00ff00,\n limegreen: 0x32cd32,\n linen: 0xfaf0e6,\n magenta: 0xff00ff,\n maroon: 0x800000,\n mediumaquamarine: 0x66cdaa,\n mediumblue: 0x0000cd,\n mediumorchid: 0xba55d3,\n mediumpurple: 0x9370db,\n mediumseagreen: 0x3cb371,\n mediumslateblue: 0x7b68ee,\n mediumspringgreen: 0x00fa9a,\n mediumturquoise: 0x48d1cc,\n mediumvioletred: 0xc71585,\n midnightblue: 0x191970,\n mintcream: 0xf5fffa,\n mistyrose: 0xffe4e1,\n moccasin: 0xffe4b5,\n navajowhite: 0xffdead,\n navy: 0x000080,\n oldlace: 0xfdf5e6,\n olive: 0x808000,\n olivedrab: 0x6b8e23,\n orange: 0xffa500,\n orangered: 0xff4500,\n orchid: 0xda70d6,\n palegoldenrod: 0xeee8aa,\n palegreen: 0x98fb98,\n paleturquoise: 0xafeeee,\n palevioletred: 0xdb7093,\n papayawhip: 0xffefd5,\n peachpuff: 0xffdab9,\n peru: 0xcd853f,\n pink: 0xffc0cb,\n plum: 0xdda0dd,\n powderblue: 0xb0e0e6,\n purple: 0x800080,\n rebeccapurple: 0x663399,\n red: 0xff0000,\n rosybrown: 0xbc8f8f,\n royalblue: 0x4169e1,\n saddlebrown: 0x8b4513,\n salmon: 0xfa8072,\n sandybrown: 0xf4a460,\n seagreen: 0x2e8b57,\n seashell: 0xfff5ee,\n sienna: 0xa0522d,\n silver: 0xc0c0c0,\n skyblue: 0x87ceeb,\n slateblue: 0x6a5acd,\n slategray: 0x708090,\n slategrey: 0x708090,\n snow: 0xfffafa,\n springgreen: 0x00ff7f,\n steelblue: 0x4682b4,\n tan: 0xd2b48c,\n teal: 0x008080,\n thistle: 0xd8bfd8,\n tomato: 0xff6347,\n turquoise: 0x40e0d0,\n violet: 0xee82ee,\n wheat: 0xf5deb3,\n white: 0xffffff,\n whitesmoke: 0xf5f5f5,\n yellow: 0xffff00,\n yellowgreen: 0x9acd32\n };\n\n define(Color, color, {\n copy: function(channels) {\n return Object.assign(new this.constructor, this, channels);\n },\n displayable: function() {\n return this.rgb().displayable();\n },\n hex: color_formatHex, // Deprecated! Use color.formatHex.\n formatHex: color_formatHex,\n formatHsl: color_formatHsl,\n formatRgb: color_formatRgb,\n toString: color_formatRgb\n });\n\n function color_formatHex() {\n return this.rgb().formatHex();\n }\n\n function color_formatHsl() {\n return hslConvert(this).formatHsl();\n }\n\n function color_formatRgb() {\n return this.rgb().formatRgb();\n }\n\n function color(format) {\n var m, l;\n format = (format + """").trim().toLowerCase();\n return (m = reHex.exec(format)) ? (l = m[1].length, m = parseInt(m[1], 16), l === 6 ? rgbn(m) // #ff0000\n : l === 3 ? new Rgb((m >> 8 & 0xf) | (m >> 4 & 0xf0), (m >> 4 & 0xf) | (m & 0xf0), ((m & 0xf) << 4) | (m & 0xf), 1) // #f00\n : l === 8 ? rgba(m >> 24 & 0xff, m >> 16 & 0xff, m >> 8 & 0xff, (m & 0xff) / 0xff) // #ff000000\n : l === 4 ? rgba((m >> 12 & 0xf) | (m >> 8 & 0xf0), (m >> 8 & 0xf) | (m >> 4 & 0xf0), (m >> 4 & 0xf) | (m & 0xf0), (((m & 0xf) << 4) | (m & 0xf)) / 0xff) // #f000\n : null) // invalid hex\n : (m = reRgbInteger.exec(format)) ? new Rgb(m[1], m[2], m[3], 1) // rgb(255, 0, 0)\n : (m = reRgbPercent.exec(format)) ? new Rgb(m[1] * 255 / 100, m[2] * 255 / 100, m[3] * 255 / 100, 1) // rgb(100%, 0%, 0%)\n : (m = reRgbaInteger.exec(format)) ? rgba(m[1], m[2], m[3], m[4]) // rgba(255, 0, 0, 1)\n : (m = reRgbaPercent.exec(format)) ? rgba(m[1] * 255 / 100, m[2] * 255 / 100, m[3] * 255 / 100, m[4]) // rgb(100%, 0%, 0%, 1)\n : (m = reHslPercent.exec(format)) ? hsla(m[1], m[2] / 100, m[3] / 100, 1) // hsl(120, 50%, 50%)\n : (m = reHslaPercent.exec(format)) ? hsla(m[1], m[2] / 100, m[3] / 100, m[4]) // hsla(120, 50%, 50%, 1)\n : named.hasOwnProperty(format) ? rgbn(named[format]) // eslint-disable-line no-prototype-builtins\n : format === ""transparent"" ? new Rgb(NaN, NaN, NaN, 0)\n : null;\n }\n\n function rgbn(n) {\n return new Rgb(n >> 16 & 0xff, n >> 8 & 0xff, n & 0xff, 1);\n }\n\n function rgba(r, g, b, a) {\n if (a <= 0) r = g = b = NaN;\n return new Rgb(r, g, b, a);\n }\n\n function rgbConvert(o) {\n if (!(o instanceof Color)) o = color(o);\n if (!o) return new Rgb;\n o = o.rgb();\n return new Rgb(o.r, o.g, o.b, o.opacity);\n }\n\n function rgb(r, g, b, opacity) {\n return arguments.length === 1 ? rgbConvert(r) : new Rgb(r, g, b, opacity == null ? 1 : opacity);\n }\n\n function Rgb(r, g, b, opacity) {\n this.r = +r;\n this.g = +g;\n this.b = +b;\n this.opacity = +opacity;\n }\n\n define(Rgb, rgb, extend(Color, {\n brighter: function(k) {\n k = k == null ? brighter : Math.pow(brighter, k);\n return new Rgb(this.r * k, this.g * k, this.b * k, this.opacity);\n },\n darker: function(k) {\n k = k == null ? darker : Math.pow(darker, k);\n return new Rgb(this.r * k, this.g * k, this.b * k, this.opacity);\n },\n rgb: function() {\n return this;\n },\n displayable: function() {\n return (-0.5 <= this.r && this.r < 255.5)\n && (-0.5 <= this.g && this.g < 255.5)\n && (-0.5 <= this.b && this.b < 255.5)\n && (0 <= this.opacity && this.opacity <= 1);\n },\n hex: rgb_formatHex, // Deprecated! Use color.formatHex.\n formatHex: rgb_formatHex,\n formatRgb: rgb_formatRgb,\n toString: rgb_formatRgb\n }));\n\n function rgb_formatHex() {\n return ""#"" + hex(this.r) + hex(this.g) + hex(this.b);\n }\n\n function rgb_formatRgb() {\n var a = this.opacity; a = isNaN(a) ? 1 : Math.max(0, Math.min(1, a));\n return (a === 1 ? ""rgb("" : ""rgba("")\n + Math.max(0, Math.min(255, Math.round(this.r) || 0)) + "", ""\n + Math.max(0, Math.min(255, Math.round(this.g) || 0)) + "", ""\n + Math.max(0, Math.min(255, Math.round(this.b) || 0))\n + (a === 1 ? "")"" : "", "" + a + "")"");\n }\n\n function hex(value) {\n value = Math.max(0, Math.min(255, Math.round(value) || 0));\n return (value < 16 ? ""0"" : """") + value.toString(16);\n }\n\n function hsla(h, s, l, a) {\n if (a <= 0) h = s = l = NaN;\n else if (l <= 0 || l >= 1) h = s = NaN;\n else if (s <= 0) h = NaN;\n return new Hsl(h, s, l, a);\n }\n\n function hslConvert(o) {\n if (o instanceof Hsl) return new Hsl(o.h, o.s, o.l, o.opacity);\n if (!(o instanceof Color)) o = color(o);\n if (!o) return new Hsl;\n if (o instanceof Hsl) return o;\n o = o.rgb();\n var r = o.r / 255,\n g = o.g / 255,\n b = o.b / 255,\n min = Math.min(r, g, b),\n max = Math.max(r, g, b),\n h = NaN,\n s = max - min,\n l = (max + min) / 2;\n if (s) {\n if (r === max) h = (g - b) / s + (g < b) * 6;\n else if (g === max) h = (b - r) / s + 2;\n else h = (r - g) / s + 4;\n s /= l < 0.5 ? max + min : 2 - max - min;\n h *= 60;\n } else {\n s = l > 0 && l < 1 ? 0 : h;\n }\n return new Hsl(h, s, l, o.opacity);\n }\n\n function hsl(h, s, l, opacity) {\n return arguments.length === 1 ? hslConvert(h) : new Hsl(h, s, l, opacity == null ? 1 : opacity);\n }\n\n function Hsl(h, s, l, opacity) {\n this.h = +h;\n this.s = +s;\n this.l = +l;\n this.opacity = +opacity;\n }\n\n define(Hsl, hsl, extend(Color, {\n brighter: function(k) {\n k = k == null ? brighter : Math.pow(brighter, k);\n return new Hsl(this.h, this.s, this.l * k, this.opacity);\n },\n darker: function(k) {\n k = k == null ? darker : Math.pow(darker, k);\n return new Hsl(this.h, this.s, this.l * k, this.opacity);\n },\n rgb: function() {\n var h = this.h % 360 + (this.h < 0) * 360,\n s = isNaN(h) || isNaN(this.s) ? 0 : this.s,\n l = this.l,\n m2 = l + (l < 0.5 ? l : 1 - l) * s,\n m1 = 2 * l - m2;\n return new Rgb(\n hsl2rgb(h >= 240 ? h - 240 : h + 120, m1, m2),\n hsl2rgb(h, m1, m2),\n hsl2rgb(h < 120 ? h + 240 : h - 120, m1, m2),\n this.opacity\n );\n },\n displayable: function() {\n return (0 <= this.s && this.s <= 1 || isNaN(this.s))\n && (0 <= this.l && this.l <= 1)\n && (0 <= this.opacity && this.opacity <= 1);\n },\n formatHsl: function() {\n var a = this.opacity; a = isNaN(a) ? 1 : Math.max(0, Math.min(1, a));\n return (a === 1 ? ""hsl("" : ""hsla("")\n + (this.h || 0) + "", ""\n + (this.s || 0) * 100 + ""%, ""\n + (this.l || 0) * 100 + ""%""\n + (a === 1 ? "")"" : "", "" + a + "")"");\n }\n }));\n\n /* From FvD 13.37, CSS Color Module Level 3 */\n function hsl2rgb(h, m1, m2) {\n return (h < 60 ? m1 + (m2 - m1) * h / 60\n : h < 180 ? m2\n : h < 240 ? m1 + (m2 - m1) * (240 - h) / 60\n : m1) * 255;\n }\n\n var deg2rad = Math.PI / 180;\n var rad2deg = 180 / Math.PI;\n\n // https://observablehq.com/@mbostock/lab-and-rgb\n var K = 18,\n Xn = 0.96422,\n Yn = 1,\n Zn = 0.82521,\n t0 = 4 / 29,\n t1 = 6 / 29,\n t2 = 3 * t1 * t1,\n t3 = t1 * t1 * t1;\n\n function labConvert(o) {\n if (o instanceof Lab) return new Lab(o.l, o.a, o.b, o.opacity);\n if (o instanceof Hcl) return hcl2lab(o);\n if (!(o instanceof Rgb)) o = rgbConvert(o);\n var r = rgb2lrgb(o.r),\n g = rgb2lrgb(o.g),\n b = rgb2lrgb(o.b),\n y = xyz2lab((0.2225045 * r + 0.7168786 * g + 0.0606169 * b) / Yn), x, z;\n if (r === g && g === b) x = z = y; else {\n x = xyz2lab((0.4360747 * r + 0.3850649 * g + 0.1430804 * b) / Xn);\n z = xyz2lab((0.0139322 * r + 0.0971045 * g + 0.7141733 * b) / Zn);\n }\n return new Lab(116 * y - 16, 500 * (x - y), 200 * (y - z), o.opacity);\n }\n\n function lab(l, a, b, opacity) {\n return arguments.length === 1 ? labConvert(l) : new Lab(l, a, b, opacity == null ? 1 : opacity);\n }\n\n function Lab(l, a, b, opacity) {\n this.l = +l;\n this.a = +a;\n this.b = +b;\n this.opacity = +opacity;\n }\n\n define(Lab, lab, extend(Color, {\n brighter: function(k) {\n return new Lab(this.l + K * (k == null ? 1 : k), this.a, this.b, this.opacity);\n },\n darker: function(k) {\n return new Lab(this.l - K * (k == null ? 1 : k), this.a, this.b, this.opacity);\n },\n rgb: function() {\n var y = (this.l + 16) / 116,\n x = isNaN(this.a) ? y : y + this.a / 500,\n z = isNaN(this.b) ? y : y - this.b / 200;\n x = Xn * lab2xyz(x);\n y = Yn * lab2xyz(y);\n z = Zn * lab2xyz(z);\n return new Rgb(\n lrgb2rgb( 3.1338561 * x - 1.6168667 * y - 0.4906146 * z),\n lrgb2rgb(-0.9787684 * x + 1.9161415 * y + 0.0334540 * z),\n lrgb2rgb( 0.0719453 * x - 0.2289914 * y + 1.4052427 * z),\n this.opacity\n );\n }\n }));\n\n function xyz2lab(t) {\n return t > t3 ? Math.pow(t, 1 / 3) : t / t2 + t0;\n }\n\n function lab2xyz(t) {\n return t > t1 ? t * t * t : t2 * (t - t0);\n }\n\n function lrgb2rgb(x) {\n return 255 * (x <= 0.0031308 ? 12.92 * x : 1.055 * Math.pow(x, 1 / 2.4) - 0.055);\n }\n\n function rgb2lrgb(x) {\n return (x /= 255) <= 0.04045 ? x / 12.92 : Math.pow((x + 0.055) / 1.055, 2.4);\n }\n\n function hclConvert(o) {\n if (o instanceof Hcl) return new Hcl(o.h, o.c, o.l, o.opacity);\n if (!(o instanceof Lab)) o = labConvert(o);\n if (o.a === 0 && o.b === 0) return new Hcl(NaN, 0 < o.l && o.l < 100 ? 0 : NaN, o.l, o.opacity);\n var h = Math.atan2(o.b, o.a) * rad2deg;\n return new Hcl(h < 0 ? h + 360 : h, Math.sqrt(o.a * o.a + o.b * o.b), o.l, o.opacity);\n }\n\n function hcl(h, c, l, opacity) {\n return arguments.length === 1 ? hclConvert(h) : new Hcl(h, c, l, opacity == null ? 1 : opacity);\n }\n\n function Hcl(h, c, l, opacity) {\n this.h = +h;\n this.c = +c;\n this.l = +l;\n this.opacity = +opacity;\n }\n\n function hcl2lab(o) {\n if (isNaN(o.h)) return new Lab(o.l, 0, 0, o.opacity);\n var h = o.h * deg2rad;\n return new Lab(o.l, Math.cos(h) * o.c, Math.sin(h) * o.c, o.opacity);\n }\n\n define(Hcl, hcl, extend(Color, {\n brighter: function(k) {\n return new Hcl(this.h, this.c, this.l + K * (k == null ? 1 : k), this.opacity);\n },\n darker: function(k) {\n return new Hcl(this.h, this.c, this.l - K * (k == null ? 1 : k), this.opacity);\n },\n rgb: function() {\n return hcl2lab(this).rgb();\n }\n }));\n\n var A = -0.14861,\n B = +1.78277,\n C = -0.29227,\n D = -0.90649,\n E = +1.97294,\n ED = E * D,\n EB = E * B,\n BC_DA = B * C - D * A;\n\n function cubehelixConvert(o) {\n if (o instanceof Cubehelix) return new Cubehelix(o.h, o.s, o.l, o.opacity);\n if (!(o instanceof Rgb)) o = rgbConvert(o);\n var r = o.r / 255,\n g = o.g / 255,\n b = o.b / 255,\n l = (BC_DA * b + ED * r - EB * g) / (BC_DA + ED - EB),\n bl = b - l,\n k = (E * (g - l) - C * bl) / D,\n s = Math.sqrt(k * k + bl * bl) / (E * l * (1 - l)), // NaN if l=0 or l=1\n h = s ? Math.atan2(k, bl) * rad2deg - 120 : NaN;\n return new Cubehelix(h < 0 ? h + 360 : h, s, l, o.opacity);\n }\n\n function cubehelix(h, s, l, opacity) {\n return arguments.length === 1 ? cubehelixConvert(h) : new Cubehelix(h, s, l, opacity == null ? 1 : opacity);\n }\n\n function Cubehelix(h, s, l, opacity) {\n this.h = +h;\n this.s = +s;\n this.l = +l;\n this.opacity = +opacity;\n }\n\n define(Cubehelix, cubehelix, extend(Color, {\n brighter: function(k) {\n k = k == null ? brighter : Math.pow(brighter, k);\n return new Cubehelix(this.h, this.s, this.l * k, this.opacity);\n },\n darker: function(k) {\n k = k == null ? darker : Math.pow(darker, k);\n return new Cubehelix(this.h, this.s, this.l * k, this.opacity);\n },\n rgb: function() {\n var h = isNaN(this.h) ? 0 : (this.h + 120) * deg2rad,\n l = +this.l,\n a = isNaN(this.s) ? 0 : this.s * l * (1 - l),\n cosh = Math.cos(h),\n sinh = Math.sin(h);\n return new Rgb(\n 255 * (l + a * (A * cosh + B * sinh)),\n 255 * (l + a * (C * cosh + D * sinh)),\n 255 * (l + a * (E * cosh)),\n this.opacity\n );\n }\n }));\n\n function constant(x) {\n return function() {\n return x;\n };\n }\n\n function linear(a, d) {\n return function(t) {\n return a + t * d;\n };\n }\n\n function exponential(a, b, y) {\n return a = Math.pow(a, y), b = Math.pow(b, y) - a, y = 1 / y, function(t) {\n return Math.pow(a + t * b, y);\n };\n }\n\n function gamma(y) {\n return (y = +y) === 1 ? nogamma : function(a, b) {\n return b - a ? exponential(a, b, y) : constant(isNaN(a) ? b : a);\n };\n }\n\n function nogamma(a, b) {\n var d = b - a;\n return d ? linear(a, d) : constant(isNaN(a) ? b : a);\n }\n\n var rgb$1 = (function rgbGamma(y) {\n var color = gamma(y);\n\n function rgb$1(start, end) {\n var r = color((start = rgb(start)).r, (end = rgb(end)).r),\n g = color(start.g, end.g),\n b = color(start.b, end.b),\n opacity = nogamma(start.opacity, end.opacity);\n return function(t) {\n start.r = r(t);\n start.g = g(t);\n start.b = b(t);\n start.opacity = opacity(t);\n return start + """";\n };\n }\n\n rgb$1.gamma = rgbGamma;\n\n return rgb$1;\n })(1);\n\n function numberArray(a, b) {\n if (!b) b = [];\n var n = a ? Math.min(b.length, a.length) : 0,\n c = b.slice(),\n i;\n return function(t) {\n for (i = 0; i < n; ++i) c[i] = a[i] * (1 - t) + b[i] * t;\n return c;\n };\n }\n\n function isNumberArray(x) {\n return ArrayBuffer.isView(x) && !(x instanceof DataView);\n }\n\n function genericArray(a, b) {\n var nb = b ? b.length : 0,\n na = a ? Math.min(nb, a.length) : 0,\n x = new Array(na),\n c = new Array(nb),\n i;\n\n for (i = 0; i < na; ++i) x[i] = interpolate(a[i], b[i]);\n for (; i < nb; ++i) c[i] = b[i];\n\n return function(t) {\n for (i = 0; i < na; ++i) c[i] = x[i](t);\n return c;\n };\n }\n\n function date(a, b) {\n var d = new Date;\n return a = +a, b = +b, function(t) {\n return d.setTime(a * (1 - t) + b * t), d;\n };\n }\n\n function interpolateNumber(a, b) {\n return a = +a, b = +b, function(t) {\n return a * (1 - t) + b * t;\n };\n }\n\n function object(a, b) {\n var i = {},\n c = {},\n k;\n\n if (a === null || typeof a !== ""object"") a = {};\n if (b === null || typeof b !== ""object"") b = {};\n\n for (k in b) {\n if (k in a) {\n i[k] = interpolate(a[k], b[k]);\n } else {\n c[k] = b[k];\n }\n }\n\n return function(t) {\n for (k in i) c[k] = i[k](t);\n return c;\n };\n }\n\n var reA = /[-+]?(?:\d+\.?\d*|\.?\d+)(?:[eE][-+]?\d+)?/g,\n reB = new RegExp(reA.source, ""g"");\n\n function zero(b) {\n return function() {\n return b;\n };\n }\n\n function one(b) {\n return function(t) {\n return b(t) + """";\n };\n }\n\n function string(a, b) {\n var bi = reA.lastIndex = reB.lastIndex = 0, // scan index for next number in b\n am, // current match in a\n bm, // current match in b\n bs, // string preceding current number in b, if any\n i = -1, // index in s\n s = [], // string constants and placeholders\n q = []; // number interpolators\n\n // Coerce inputs to strings.\n a = a + """", b = b + """";\n\n // Interpolate pairs of numbers in a & b.\n while ((am = reA.exec(a))\n && (bm = reB.exec(b))) {\n if ((bs = bm.index) > bi) { // a string precedes the next number in b\n bs = b.slice(bi, bs);\n if (s[i]) s[i] += bs; // coalesce with previous string\n else s[++i] = bs;\n }\n if ((am = am[0]) === (bm = bm[0])) { // numbers in a & b match\n if (s[i]) s[i] += bm; // coalesce with previous string\n else s[++i] = bm;\n } else { // interpolate non-matching numbers\n s[++i] = null;\n q.push({i: i, x: interpolateNumber(am, bm)});\n }\n bi = reB.lastIndex;\n }\n\n // Add remains of b.\n if (bi < b.length) {\n bs = b.slice(bi);\n if (s[i]) s[i] += bs; // coalesce with previous string\n else s[++i] = bs;\n }\n\n // Special optimization for only a single match.\n // Otherwise, interpolate each of the numbers and rejoin the string.\n return s.length < 2 ? (q[0]\n ? one(q[0].x)\n : zero(b))\n : (b = q.length, function(t) {\n for (var i = 0, o; i < b; ++i) s[(o = q[i]).i] = o.x(t);\n return s.join("""");\n });\n }\n\n function interpolate(a, b) {\n var t = typeof b, c;\n return b == null || t === ""boolean"" ? constant(b)\n : (t === ""number"" ? interpolateNumber\n : t === ""string"" ? ((c = color(b)) ? (b = c, rgb$1) : string)\n : b instanceof color ? rgb$1\n : b instanceof Date ? date\n : isNumberArray(b) ? numberArray\n : Array.isArray(b) ? genericArray\n : typeof b.valueOf !== ""function"" && typeof b.toString !== ""function"" || isNaN(b) ? object\n : interpolateNumber)(a, b);\n }\n\n function interpolateRound(a, b) {\n return a = +a, b = +b, function(t) {\n return Math.round(a * (1 - t) + b * t);\n };\n }\n\n function constant$1(x) {\n return function() {\n return x;\n };\n }\n\n function number(x) {\n return +x;\n }\n\n var unit = [0, 1];\n\n function identity(x) {\n return x;\n }\n\n function normalize(a, b) {\n return (b -= (a = +a))\n ? function(x) { return (x - a) / b; }\n : constant$1(isNaN(b) ? NaN : 0.5);\n }\n\n function clamper(a, b) {\n var t;\n if (a > b) t = a, a = b, b = t;\n return function(x) { return Math.max(a, Math.min(b, x)); };\n }\n\n // normalize(a, b)(x) takes a domain value x in [a,b] and returns the corresponding parameter t in [0,1].\n // interpolate(a, b)(t) takes a parameter t in [0,1] and returns the corresponding range value x in [a,b].\n function bimap(domain, range, interpolate) {\n var d0 = domain[0], d1 = domain[1], r0 = range[0], r1 = range[1];\n if (d1 < d0) d0 = normalize(d1, d0), r0 = interpolate(r1, r0);\n else d0 = normalize(d0, d1), r0 = interpolate(r0, r1);\n return function(x) { return r0(d0(x)); };\n }\n\n function polymap(domain, range, interpolate) {\n var j = Math.min(domain.length, range.length) - 1,\n d = new Array(j),\n r = new Array(j),\n i = -1;\n\n // Reverse descending domains.\n if (domain[j] < domain[0]) {\n domain = domain.slice().reverse();\n range = range.slice().reverse();\n }\n\n while (++i < j) {\n d[i] = normalize(domain[i], domain[i + 1]);\n r[i] = interpolate(range[i], range[i + 1]);\n }\n\n return function(x) {\n var i = bisectRight(domain, x, 1, j) - 1;\n return r[i](d[i](x));\n };\n }\n\n function copy(source, target) {\n return target\n .domain(source.domain())\n .range(source.range())\n .interpolate(source.interpolate())\n .clamp(source.clamp())\n .unknown(source.unknown());\n }\n\n function transformer() {\n var domain = unit,\n range = unit,\n interpolate$1 = interpolate,\n transform,\n untransform,\n unknown,\n clamp = identity,\n piecewise,\n output,\n input;\n\n function rescale() {\n var n = Math.min(domain.length, range.length);\n if (clamp !== identity) clamp = clamper(domain[0], domain[n - 1]);\n piecewise = n > 2 ? polymap : bimap;\n output = input = null;\n return scale;\n }\n\n function scale(x) {\n return isNaN(x = +x) ? unknown : (output || (output = piecewise(domain.map(transform), range, interpolate$1)))(transform(clamp(x)));\n }\n\n scale.invert = function(y) {\n return clamp(untransform((input || (input = piecewise(range, domain.map(transform), interpolateNumber)))(y)));\n };\n\n scale.domain = function(_) {\n return arguments.length ? (domain = Array.from(_, number), rescale()) : domain.slice();\n };\n\n scale.range = function(_) {\n return arguments.length ? (range = Array.from(_), rescale()) : range.slice();\n };\n\n scale.rangeRound = function(_) {\n return range = Array.from(_), interpolate$1 = interpolateRound, rescale();\n };\n\n scale.clamp = function(_) {\n return arguments.length ? (clamp = _ ? true : identity, rescale()) : clamp !== identity;\n };\n\n scale.interpolate = function(_) {\n return arguments.length ? (interpolate$1 = _, rescale()) : interpolate$1;\n };\n\n scale.unknown = function(_) {\n return arguments.length ? (unknown = _, scale) : unknown;\n };\n\n return function(t, u) {\n transform = t, untransform = u;\n return rescale();\n };\n }\n\n function continuous() {\n return transformer()(identity, identity);\n }\n\n // Computes the decimal coefficient and exponent of the specified number x with\n // significant digits p, where x is positive and p is in [1, 21] or undefined.\n // For example, formatDecimal(1.23) returns [""123"", 0].\n function formatDecimal(x, p) {\n if ((i = (x = p ? x.toExponential(p - 1) : x.toExponential()).indexOf(""e"")) < 0) return null; // NaN, ±Infinity\n var i, coefficient = x.slice(0, i);\n\n // The string returned by toExponential either has the form \d\.\d+e[-+]\d+\n // (e.g., 1.2e+3) or the form \de[-+]\d+ (e.g., 1e+3).\n return [\n coefficient.length > 1 ? coefficient[0] + coefficient.slice(2) : coefficient,\n +x.slice(i + 1)\n ];\n }\n\n function exponent(x) {\n return x = formatDecimal(Math.abs(x)), x ? x[1] : NaN;\n }\n\n function formatGroup(grouping, thousands) {\n return function(value, width) {\n var i = value.length,\n t = [],\n j = 0,\n g = grouping[0],\n length = 0;\n\n while (i > 0 && g > 0) {\n if (length + g + 1 > width) g = Math.max(1, width - length);\n t.push(value.substring(i -= g, i + g));\n if ((length += g + 1) > width) break;\n g = grouping[j = (j + 1) % grouping.length];\n }\n\n return t.reverse().join(thousands);\n };\n }\n\n function formatNumerals(numerals) {\n return function(value) {\n return value.replace(/[0-9]/g, function(i) {\n return numerals[+i];\n });\n };\n }\n\n // [[fill]align][sign][symbol][0][width][,][.precision][~][type]\n var re = /^(?:(.)?([<>=^]))?([+\-( ])?([$#])?(0)?(\d+)?(,)?(\.\d+)?(~)?([a-z%])?$/i;\n\n function formatSpecifier(specifier) {\n if (!(match = re.exec(specifier))) throw new Error(""invalid format: "" + specifier);\n var match;\n return new FormatSpecifier({\n fill: match[1],\n align: match[2],\n sign: match[3],\n symbol: match[4],\n zero: match[5],\n width: match[6],\n comma: match[7],\n precision: match[8] && match[8].slice(1),\n trim: match[9],\n type: match[10]\n });\n }\n\n formatSpecifier.prototype = FormatSpecifier.prototype; // instanceof\n\n function FormatSpecifier(specifier) {\n this.fill = specifier.fill === undefined ? "" "" : specifier.fill + """";\n this.align = specifier.align === undefined ? "">"" : specifier.align + """";\n this.sign = specifier.sign === undefined ? ""-"" : specifier.sign + """";\n this.symbol = specifier.symbol === undefined ? """" : specifier.symbol + """";\n this.zero = !!specifier.zero;\n this.width = specifier.width === undefined ? undefined : +specifier.width;\n this.comma = !!specifier.comma;\n this.precision = specifier.precision === undefined ? undefined : +specifier.precision;\n this.trim = !!specifier.trim;\n this.type = specifier.type === undefined ? """" : specifier.type + """";\n }\n\n FormatSpecifier.prototype.toString = function() {\n return this.fill\n + this.align\n + this.sign\n + this.symbol\n + (this.zero ? ""0"" : """")\n + (this.width === undefined ? """" : Math.max(1, this.width | 0))\n + (this.comma ? "","" : """")\n + (this.precision === undefined ? """" : ""."" + Math.max(0, this.precision | 0))\n + (this.trim ? ""~"" : """")\n + this.type;\n };\n\n // Trims insignificant zeros, e.g., replaces 1.2000k with 1.2k.\n function formatTrim(s) {\n out: for (var n = s.length, i = 1, i0 = -1, i1; i < n; ++i) {\n switch (s[i]) {\n case ""."": i0 = i1 = i; break;\n case ""0"": if (i0 === 0) i0 = i; i1 = i; break;\n default: if (!+s[i]) break out; if (i0 > 0) i0 = 0; break;\n }\n }\n return i0 > 0 ? s.slice(0, i0) + s.slice(i1 + 1) : s;\n }\n\n var prefixExponent;\n\n function formatPrefixAuto(x, p) {\n var d = formatDecimal(x, p);\n if (!d) return x + """";\n var coefficient = d[0],\n exponent = d[1],\n i = exponent - (prefixExponent = Math.max(-8, Math.min(8, Math.floor(exponent / 3))) * 3) + 1,\n n = coefficient.length;\n return i === n ? coefficient\n : i > n ? coefficient + new Array(i - n + 1).join(""0"")\n : i > 0 ? coefficient.slice(0, i) + ""."" + coefficient.slice(i)\n : ""0."" + new Array(1 - i).join(""0"") + formatDecimal(x, Math.max(0, p + i - 1))[0]; // less than 1y!\n }\n\n function formatRounded(x, p) {\n var d = formatDecimal(x, p);\n if (!d) return x + """";\n var coefficient = d[0],\n exponent = d[1];\n return exponent < 0 ? ""0."" + new Array(-exponent).join(""0"") + coefficient\n : coefficient.length > exponent + 1 ? coefficient.slice(0, exponent + 1) + ""."" + coefficient.slice(exponent + 1)\n : coefficient + new Array(exponent - coefficient.length + 2).join(""0"");\n }\n\n var formatTypes = {\n ""%"": function(x, p) { return (x * 100).toFixed(p); },\n ""b"": function(x) { return Math.round(x).toString(2); },\n ""c"": function(x) { return x + """"; },\n ""d"": function(x) { return Math.round(x).toString(10); },\n ""e"": function(x, p) { return x.toExponential(p); },\n ""f"": function(x, p) { return x.toFixed(p); },\n ""g"": function(x, p) { return x.toPrecision(p); },\n ""o"": function(x) { return Math.round(x).toString(8); },\n ""p"": function(x, p) { return formatRounded(x * 100, p); },\n ""r"": formatRounded,\n ""s"": formatPrefixAuto,\n ""X"": function(x) { return Math.round(x).toString(16).toUpperCase(); },\n ""x"": function(x) { return Math.round(x).toString(16); }\n };\n\n function identity$1(x) {\n return x;\n }\n\n var map = Array.prototype.map,\n prefixes = [""y"",""z"",""a"",""f"",""p"",""n"",""µ"",""m"","""",""k"",""M"",""G"",""T"",""P"",""E"",""Z"",""Y""];\n\n function formatLocale(locale) {\n var group = locale.grouping === undefined || locale.thousands === undefined ? identity$1 : formatGroup(map.call(locale.grouping, Number), locale.thousands + """"),\n currencyPrefix = locale.currency === undefined ? """" : locale.currency[0] + """",\n currencySuffix = locale.currency === undefined ? """" : locale.currency[1] + """",\n decimal = locale.decimal === undefined ? ""."" : locale.decimal + """",\n numerals = locale.numerals === undefined ? identity$1 : formatNumerals(map.call(locale.numerals, String)),\n percent = locale.percent === undefined ? ""%"" : locale.percent + """",\n minus = locale.minus === undefined ? ""-"" : locale.minus + """",\n nan = locale.nan === undefined ? ""NaN"" : locale.nan + """";\n\n function newFormat(specifier) {\n specifier = formatSpecifier(specifier);\n\n var fill = specifier.fill,\n align = specifier.align,\n sign = specifier.sign,\n symbol = specifier.symbol,\n zero = specifier.zero,\n width = specifier.width,\n comma = specifier.comma,\n precision = specifier.precision,\n trim = specifier.trim,\n type = specifier.type;\n\n // The ""n"" type is an alias for "",g"".\n if (type === ""n"") comma = true, type = ""g"";\n\n // The """" type, and any invalid type, is an alias for "".12~g"".\n else if (!formatTypes[type]) precision === undefined && (precision = 12), trim = true, type = ""g"";\n\n // If zero fill is specified, padding goes after sign and before digits.\n if (zero || (fill === ""0"" && align === ""="")) zero = true, fill = ""0"", align = ""="";\n\n // Compute the prefix and suffix.\n // For SI-prefix, the suffix is lazily computed.\n var prefix = symbol === ""$"" ? currencyPrefix : symbol === ""#"" && /[boxX]/.test(type) ? ""0"" + type.toLowerCase() : """",\n suffix = symbol === ""$"" ? currencySuffix : /[%p]/.test(type) ? percent : """";\n\n // What format function should we use?\n // Is this an integer type?\n // Can this type generate exponential notation?\n var formatType = formatTypes[type],\n maybeSuffix = /[defgprs%]/.test(type);\n\n // Set the default precision if not specified,\n // or clamp the specified precision to the supported range.\n // For significant precision, it must be in [1, 21].\n // For fixed precision, it must be in [0, 20].\n precision = precision === undefined ? 6\n : /[gprs]/.test(type) ? Math.max(1, Math.min(21, precision))\n : Math.max(0, Math.min(20, precision));\n\n function format(value) {\n var valuePrefix = prefix,\n valueSuffix = suffix,\n i, n, c;\n\n if (type === ""c"") {\n valueSuffix = formatType(value) + valueSuffix;\n value = """";\n } else {\n value = +value;\n\n // Determine the sign. -0 is not less than 0, but 1 / -0 is!\n var valueNegative = value < 0 || 1 / value < 0;\n\n // Perform the initial formatting.\n value = isNaN(value) ? nan : formatType(Math.abs(value), precision);\n\n // Trim insignificant zeros.\n if (trim) value = formatTrim(value);\n\n // If a negative value rounds to zero after formatting, and no explicit positive sign is requested, hide the sign.\n if (valueNegative && +value === 0 && sign !== ""+"") valueNegative = false;\n\n // Compute the prefix and suffix.\n valuePrefix = (valueNegative ? (sign === ""("" ? sign : minus) : sign === ""-"" || sign === ""("" ? """" : sign) + valuePrefix;\n valueSuffix = (type === ""s"" ? prefixes[8 + prefixExponent / 3] : """") + valueSuffix + (valueNegative && sign === ""("" ? "")"" : """");\n\n // Break the formatted value into the integer “value” part that can be\n // grouped, and fractional or exponential “suffix” part that is not.\n if (maybeSuffix) {\n i = -1, n = value.length;\n while (++i < n) {\n if (c = value.charCodeAt(i), 48 > c || c > 57) {\n valueSuffix = (c === 46 ? decimal + value.slice(i + 1) : value.slice(i)) + valueSuffix;\n value = value.slice(0, i);\n break;\n }\n }\n }\n }\n\n // If the fill character is not ""0"", grouping is applied before padding.\n if (comma && !zero) value = group(value, Infinity);\n\n // Compute the padding.\n var length = valuePrefix.length + value.length + valueSuffix.length,\n padding = length < width ? new Array(width - length + 1).join(fill) : """";\n\n // If the fill character is ""0"", grouping is applied after padding.\n if (comma && zero) value = group(padding + value, padding.length ? width - valueSuffix.length : Infinity), padding = """";\n\n // Reconstruct the final output based on the desired alignment.\n switch (align) {\n case ""<"": value = valuePrefix + value + valueSuffix + padding; break;\n case ""="": value = valuePrefix + padding + value + valueSuffix; break;\n case ""^"": value = padding.slice(0, length = padding.length >> 1) + valuePrefix + value + valueSuffix + padding.slice(length); break;\n default: value = padding + valuePrefix + value + valueSuffix; break;\n }\n\n return numerals(value);\n }\n\n format.toString = function() {\n return specifier + """";\n };\n\n return format;\n }\n\n function formatPrefix(specifier, value) {\n var f = newFormat((specifier = formatSpecifier(specifier), specifier.type = ""f"", specifier)),\n e = Math.max(-8, Math.min(8, Math.floor(exponent(value) / 3))) * 3,\n k = Math.pow(10, -e),\n prefix = prefixes[8 + e / 3];\n return function(value) {\n return f(k * value) + prefix;\n };\n }\n\n return {\n format: newFormat,\n formatPrefix: formatPrefix\n };\n }\n\n var locale;\n var format;\n var formatPrefix;\n\n defaultLocale({\n decimal: ""."",\n thousands: "","",\n grouping: [3],\n currency: [""$"", """"],\n minus: ""-""\n });\n\n function defaultLocale(definition) {\n locale = formatLocale(definition);\n format = locale.format;\n formatPrefix = locale.formatPrefix;\n return locale;\n }\n\n function precisionFixed(step) {\n return Math.max(0, -exponent(Math.abs(step)));\n }\n\n function precisionPrefix(step, value) {\n return Math.max(0, Math.max(-8, Math.min(8, Math.floor(exponent(value) / 3))) * 3 - exponent(Math.abs(step)));\n }\n\n function precisionRound(step, max) {\n step = Math.abs(step), max = Math.abs(max) - step;\n return Math.max(0, exponent(max) - exponent(step)) + 1;\n }\n\n function tickFormat(start, stop, count, specifier) {\n var step = tickStep(start, stop, count),\n precision;\n specifier = formatSpecifier(specifier == null ? "",f"" : specifier);\n switch (specifier.type) {\n case ""s"": {\n var value = Math.max(Math.abs(start), Math.abs(stop));\n if (specifier.precision == null && !isNaN(precision = precisionPrefix(step, value))) specifier.precision = precision;\n return formatPrefix(specifier, value);\n }\n case """":\n case ""e"":\n case ""g"":\n case ""p"":\n case ""r"": {\n if (specifier.precision == null && !isNaN(precision = precisionRound(step, Math.max(Math.abs(start), Math.abs(stop))))) specifier.precision = precision - (specifier.type === ""e"");\n break;\n }\n case ""f"":\n case ""%"": {\n if (specifier.precision == null && !isNaN(precision = precisionFixed(step))) specifier.precision = precision - (specifier.type === ""%"") * 2;\n break;\n }\n }\n return format(specifier);\n }\n\n function linearish(scale) {\n var domain = scale.domain;\n\n scale.ticks = function(count) {\n var d = domain();\n return ticks(d[0], d[d.length - 1], count == null ? 10 : count);\n };\n\n scale.tickFormat = function(count, specifier) {\n var d = domain();\n return tickFormat(d[0], d[d.length - 1], count == null ? 10 : count, specifier);\n };\n\n scale.nice = function(count) {\n if (count == null) count = 10;\n\n var d = domain(),\n i0 = 0,\n i1 = d.length - 1,\n start = d[i0],\n stop = d[i1],\n step;\n\n if (stop < start) {\n step = start, start = stop, stop = step;\n step = i0, i0 = i1, i1 = step;\n }\n\n step = tickIncrement(start, stop, count);\n\n if (step > 0) {\n start = Math.floor(start / step) * step;\n stop = Math.ceil(stop / step) * step;\n step = tickIncrement(start, stop, count);\n } else if (step < 0) {\n start = Math.ceil(start * step) / step;\n stop = Math.floor(stop * step) / step;\n step = tickIncrement(start, stop, count);\n }\n\n if (step > 0) {\n d[i0] = Math.floor(start / step) * step;\n d[i1] = Math.ceil(stop / step) * step;\n domain(d);\n } else if (step < 0) {\n d[i0] = Math.ceil(start * step) / step;\n d[i1] = Math.floor(stop * step) / step;\n domain(d);\n }\n\n return scale;\n };\n\n return scale;\n }\n\n function linear$1() {\n var scale = continuous();\n\n scale.copy = function() {\n return copy(scale, linear$1());\n };\n\n initRange.apply(scale, arguments);\n\n return linearish(scale);\n }\n\n var t0$1 = new Date,\n t1$1 = new Date;\n\n function newInterval(floori, offseti, count, field) {\n\n function interval(date) {\n return floori(date = arguments.length === 0 ? new Date : new Date(+date)), date;\n }\n\n interval.floor = function(date) {\n return floori(date = new Date(+date)), date;\n };\n\n interval.ceil = function(date) {\n return floori(date = new Date(date - 1)), offseti(date, 1), floori(date), date;\n };\n\n interval.round = function(date) {\n var d0 = interval(date),\n d1 = interval.ceil(date);\n return date - d0 < d1 - date ? d0 : d1;\n };\n\n interval.offset = function(date, step) {\n return offseti(date = new Date(+date), step == null ? 1 : Math.floor(step)), date;\n };\n\n interval.range = function(start, stop, step) {\n var range = [], previous;\n start = interval.ceil(start);\n step = step == null ? 1 : Math.floor(step);\n if (!(start < stop) || !(step > 0)) return range; // also handles Invalid Date\n do range.push(previous = new Date(+start)), offseti(start, step), floori(start);\n while (previous < start && start < stop);\n return range;\n };\n\n interval.filter = function(test) {\n return newInterval(function(date) {\n if (date >= date) while (floori(date), !test(date)) date.setTime(date - 1);\n }, function(date, step) {\n if (date >= date) {\n if (step < 0) while (++step <= 0) {\n while (offseti(date, -1), !test(date)) {} // eslint-disable-line no-empty\n } else while (--step >= 0) {\n while (offseti(date, +1), !test(date)) {} // eslint-disable-line no-empty\n }\n }\n });\n };\n\n if (count) {\n interval.count = function(start, end) {\n t0$1.setTime(+start), t1$1.setTime(+end);\n floori(t0$1), floori(t1$1);\n return Math.floor(count(t0$1, t1$1));\n };\n\n interval.every = function(step) {\n step = Math.floor(step);\n return !isFinite(step) || !(step > 0) ? null\n : !(step > 1) ? interval\n : interval.filter(field\n ? function(d) { return field(d) % step === 0; }\n : function(d) { return interval.count(0, d) % step === 0; });\n };\n }\n\n return interval;\n }\n\n var millisecond = newInterval(function() {\n // noop\n }, function(date, step) {\n date.setTime(+date + step);\n }, function(start, end) {\n return end - start;\n });\n\n // An optimized implementation for this simple case.\n millisecond.every = function(k) {\n k = Math.floor(k);\n if (!isFinite(k) || !(k > 0)) return null;\n if (!(k > 1)) return millisecond;\n return newInterval(function(date) {\n date.setTime(Math.floor(date / k) * k);\n }, function(date, step) {\n date.setTime(+date + step * k);\n }, function(start, end) {\n return (end - start) / k;\n });\n };\n\n var durationSecond = 1e3;\n var durationMinute = 6e4;\n var durationHour = 36e5;\n var durationDay = 864e5;\n var durationWeek = 6048e5;\n\n var second = newInterval(function(date) {\n date.setTime(date - date.getMilliseconds());\n }, function(date, step) {\n date.setTime(+date + step * durationSecond);\n }, function(start, end) {\n return (end - start) / durationSecond;\n }, function(date) {\n return date.getUTCSeconds();\n });\n\n var minute = newInterval(function(date) {\n date.setTime(date - date.getMilliseconds() - date.getSeconds() * durationSecond);\n }, function(date, step) {\n date.setTime(+date + step * durationMinute);\n }, function(start, end) {\n return (end - start) / durationMinute;\n }, function(date) {\n return date.getMinutes();\n });\n\n var hour = newInterval(function(date) {\n date.setTime(date - date.getMilliseconds() - date.getSeconds() * durationSecond - date.getMinutes() * durationMinute);\n }, function(date, step) {\n date.setTime(+date + step * durationHour);\n }, function(start, end) {\n return (end - start) / durationHour;\n }, function(date) {\n return date.getHours();\n });\n\n var day = newInterval(function(date) {\n date.setHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setDate(date.getDate() + step);\n }, function(start, end) {\n return (end - start - (end.getTimezoneOffset() - start.getTimezoneOffset()) * durationMinute) / durationDay;\n }, function(date) {\n return date.getDate() - 1;\n });\n\n function weekday(i) {\n return newInterval(function(date) {\n date.setDate(date.getDate() - (date.getDay() + 7 - i) % 7);\n date.setHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setDate(date.getDate() + step * 7);\n }, function(start, end) {\n return (end - start - (end.getTimezoneOffset() - start.getTimezoneOffset()) * durationMinute) / durationWeek;\n });\n }\n\n var sunday = weekday(0);\n var monday = weekday(1);\n var tuesday = weekday(2);\n var wednesday = weekday(3);\n var thursday = weekday(4);\n var friday = weekday(5);\n var saturday = weekday(6);\n\n var month = newInterval(function(date) {\n date.setDate(1);\n date.setHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setMonth(date.getMonth() + step);\n }, function(start, end) {\n return end.getMonth() - start.getMonth() + (end.getFullYear() - start.getFullYear()) * 12;\n }, function(date) {\n return date.getMonth();\n });\n\n var year = newInterval(function(date) {\n date.setMonth(0, 1);\n date.setHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setFullYear(date.getFullYear() + step);\n }, function(start, end) {\n return end.getFullYear() - start.getFullYear();\n }, function(date) {\n return date.getFullYear();\n });\n\n // An optimized implementation for this simple case.\n year.every = function(k) {\n return !isFinite(k = Math.floor(k)) || !(k > 0) ? null : newInterval(function(date) {\n date.setFullYear(Math.floor(date.getFullYear() / k) * k);\n date.setMonth(0, 1);\n date.setHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setFullYear(date.getFullYear() + step * k);\n });\n };\n\n var utcMinute = newInterval(function(date) {\n date.setUTCSeconds(0, 0);\n }, function(date, step) {\n date.setTime(+date + step * durationMinute);\n }, function(start, end) {\n return (end - start) / durationMinute;\n }, function(date) {\n return date.getUTCMinutes();\n });\n\n var utcHour = newInterval(function(date) {\n date.setUTCMinutes(0, 0, 0);\n }, function(date, step) {\n date.setTime(+date + step * durationHour);\n }, function(start, end) {\n return (end - start) / durationHour;\n }, function(date) {\n return date.getUTCHours();\n });\n\n var utcDay = newInterval(function(date) {\n date.setUTCHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setUTCDate(date.getUTCDate() + step);\n }, function(start, end) {\n return (end - start) / durationDay;\n }, function(date) {\n return date.getUTCDate() - 1;\n });\n\n function utcWeekday(i) {\n return newInterval(function(date) {\n date.setUTCDate(date.getUTCDate() - (date.getUTCDay() + 7 - i) % 7);\n date.setUTCHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setUTCDate(date.getUTCDate() + step * 7);\n }, function(start, end) {\n return (end - start) / durationWeek;\n });\n }\n\n var utcSunday = utcWeekday(0);\n var utcMonday = utcWeekday(1);\n var utcTuesday = utcWeekday(2);\n var utcWednesday = utcWeekday(3);\n var utcThursday = utcWeekday(4);\n var utcFriday = utcWeekday(5);\n var utcSaturday = utcWeekday(6);\n\n var utcMonth = newInterval(function(date) {\n date.setUTCDate(1);\n date.setUTCHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setUTCMonth(date.getUTCMonth() + step);\n }, function(start, end) {\n return end.getUTCMonth() - start.getUTCMonth() + (end.getUTCFullYear() - start.getUTCFullYear()) * 12;\n }, function(date) {\n return date.getUTCMonth();\n });\n\n var utcYear = newInterval(function(date) {\n date.setUTCMonth(0, 1);\n date.setUTCHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setUTCFullYear(date.getUTCFullYear() + step);\n }, function(start, end) {\n return end.getUTCFullYear() - start.getUTCFullYear();\n }, function(date) {\n return date.getUTCFullYear();\n });\n\n // An optimized implementation for this simple case.\n utcYear.every = function(k) {\n return !isFinite(k = Math.floor(k)) || !(k > 0) ? null : newInterval(function(date) {\n date.setUTCFullYear(Math.floor(date.getUTCFullYear() / k) * k);\n date.setUTCMonth(0, 1);\n date.setUTCHours(0, 0, 0, 0);\n }, function(date, step) {\n date.setUTCFullYear(date.getUTCFullYear() + step * k);\n });\n };\n\n function localDate(d) {\n if (0 <= d.y && d.y < 100) {\n var date = new Date(-1, d.m, d.d, d.H, d.M, d.S, d.L);\n date.setFullYear(d.y);\n return date;\n }\n return new Date(d.y, d.m, d.d, d.H, d.M, d.S, d.L);\n }\n\n function utcDate(d) {\n if (0 <= d.y && d.y < 100) {\n var date = new Date(Date.UTC(-1, d.m, d.d, d.H, d.M, d.S, d.L));\n date.setUTCFullYear(d.y);\n return date;\n }\n return new Date(Date.UTC(d.y, d.m, d.d, d.H, d.M, d.S, d.L));\n }\n\n function newDate(y, m, d) {\n return {y: y, m: m, d: d, H: 0, M: 0, S: 0, L: 0};\n }\n\n function formatLocale$1(locale) {\n var locale_dateTime = locale.dateTime,\n locale_date = locale.date,\n locale_time = locale.time,\n locale_periods = locale.periods,\n locale_weekdays = locale.days,\n locale_shortWeekdays = locale.shortDays,\n locale_months = locale.months,\n locale_shortMonths = locale.shortMonths;\n\n var periodRe = formatRe(locale_periods),\n periodLookup = formatLookup(locale_periods),\n weekdayRe = formatRe(locale_weekdays),\n weekdayLookup = formatLookup(locale_weekdays),\n shortWeekdayRe = formatRe(locale_shortWeekdays),\n shortWeekdayLookup = formatLookup(locale_shortWeekdays),\n monthRe = formatRe(locale_months),\n monthLookup = formatLookup(locale_months),\n shortMonthRe = formatRe(locale_shortMonths),\n shortMonthLookup = formatLookup(locale_shortMonths);\n\n var formats = {\n ""a"": formatShortWeekday,\n ""A"": formatWeekday,\n ""b"": formatShortMonth,\n ""B"": formatMonth,\n ""c"": null,\n ""d"": formatDayOfMonth,\n ""e"": formatDayOfMonth,\n ""f"": formatMicroseconds,\n ""H"": formatHour24,\n ""I"": formatHour12,\n ""j"": formatDayOfYear,\n ""L"": formatMilliseconds,\n ""m"": formatMonthNumber,\n ""M"": formatMinutes,\n ""p"": formatPeriod,\n ""q"": formatQuarter,\n ""Q"": formatUnixTimestamp,\n ""s"": formatUnixTimestampSeconds,\n ""S"": formatSeconds,\n ""u"": formatWeekdayNumberMonday,\n ""U"": formatWeekNumberSunday,\n ""V"": formatWeekNumberISO,\n ""w"": formatWeekdayNumberSunday,\n ""W"": formatWeekNumberMonday,\n ""x"": null,\n ""X"": null,\n ""y"": formatYear,\n ""Y"": formatFullYear,\n ""Z"": formatZone,\n ""%"": formatLiteralPercent\n };\n\n var utcFormats = {\n ""a"": formatUTCShortWeekday,\n ""A"": formatUTCWeekday,\n ""b"": formatUTCShortMonth,\n ""B"": formatUTCMonth,\n ""c"": null,\n ""d"": formatUTCDayOfMonth,\n ""e"": formatUTCDayOfMonth,\n ""f"": formatUTCMicroseconds,\n ""H"": formatUTCHour24,\n ""I"": formatUTCHour12,\n ""j"": formatUTCDayOfYear,\n ""L"": formatUTCMilliseconds,\n ""m"": formatUTCMonthNumber,\n ""M"": formatUTCMinutes,\n ""p"": formatUTCPeriod,\n ""q"": formatUTCQuarter,\n ""Q"": formatUnixTimestamp,\n ""s"": formatUnixTimestampSeconds,\n ""S"": formatUTCSeconds,\n ""u"": formatUTCWeekdayNumberMonday,\n ""U"": formatUTCWeekNumberSunday,\n ""V"": formatUTCWeekNumberISO,\n ""w"": formatUTCWeekdayNumberSunday,\n ""W"": formatUTCWeekNumberMonday,\n ""x"": null,\n ""X"": null,\n ""y"": formatUTCYear,\n ""Y"": formatUTCFullYear,\n ""Z"": formatUTCZone,\n ""%"": formatLiteralPercent\n };\n\n var parses = {\n ""a"": parseShortWeekday,\n ""A"": parseWeekday,\n ""b"": parseShortMonth,\n ""B"": parseMonth,\n ""c"": parseLocaleDateTime,\n ""d"": parseDayOfMonth,\n ""e"": parseDayOfMonth,\n ""f"": parseMicroseconds,\n ""H"": parseHour24,\n ""I"": parseHour24,\n ""j"": parseDayOfYear,\n ""L"": parseMilliseconds,\n ""m"": parseMonthNumber,\n ""M"": parseMinutes,\n ""p"": parsePeriod,\n ""q"": parseQuarter,\n ""Q"": parseUnixTimestamp,\n ""s"": parseUnixTimestampSeconds,\n ""S"": parseSeconds,\n ""u"": parseWeekdayNumberMonday,\n ""U"": parseWeekNumberSunday,\n ""V"": parseWeekNumberISO,\n ""w"": parseWeekdayNumberSunday,\n ""W"": parseWeekNumberMonday,\n ""x"": parseLocaleDate,\n ""X"": parseLocaleTime,\n ""y"": parseYear,\n ""Y"": parseFullYear,\n ""Z"": parseZone,\n ""%"": parseLiteralPercent\n };\n\n // These recursive directive definitions must be deferred.\n formats.x = newFormat(locale_date, formats);\n formats.X = newFormat(locale_time, formats);\n formats.c = newFormat(locale_dateTime, formats);\n utcFormats.x = newFormat(locale_date, utcFormats);\n utcFormats.X = newFormat(locale_time, utcFormats);\n utcFormats.c = newFormat(locale_dateTime, utcFormats);\n\n function newFormat(specifier, formats) {\n return function(date) {\n var string = [],\n i = -1,\n j = 0,\n n = specifier.length,\n c,\n pad,\n format;\n\n if (!(date instanceof Date)) date = new Date(+date);\n\n while (++i < n) {\n if (specifier.charCodeAt(i) === 37) {\n string.push(specifier.slice(j, i));\n if ((pad = pads[c = specifier.charAt(++i)]) != null) c = specifier.charAt(++i);\n else pad = c === ""e"" ? "" "" : ""0"";\n if (format = formats[c]) c = format(date, pad);\n string.push(c);\n j = i + 1;\n }\n }\n\n string.push(specifier.slice(j, i));\n return string.join("""");\n };\n }\n\n function newParse(specifier, Z) {\n return function(string) {\n var d = newDate(1900, undefined, 1),\n i = parseSpecifier(d, specifier, string += """", 0),\n week, day$1;\n if (i != string.length) return null;\n\n // If a UNIX timestamp is specified, return it.\n if (""Q"" in d) return new Date(d.Q);\n if (""s"" in d) return new Date(d.s * 1000 + (""L"" in d ? d.L : 0));\n\n // If this is utcParse, never use the local timezone.\n if (Z && !(""Z"" in d)) d.Z = 0;\n\n // The am-pm flag is 0 for AM, and 1 for PM.\n if (""p"" in d) d.H = d.H % 12 + d.p * 12;\n\n // If the month was not specified, inherit from the quarter.\n if (d.m === undefined) d.m = ""q"" in d ? d.q : 0;\n\n // Convert day-of-week and week-of-year to day-of-year.\n if (""V"" in d) {\n if (d.V < 1 || d.V > 53) return null;\n if (!(""w"" in d)) d.w = 1;\n if (""Z"" in d) {\n week = utcDate(newDate(d.y, 0, 1)), day$1 = week.getUTCDay();\n week = day$1 > 4 || day$1 === 0 ? utcMonday.ceil(week) : utcMonday(week);\n week = utcDay.offset(week, (d.V - 1) * 7);\n d.y = week.getUTCFullYear();\n d.m = week.getUTCMonth();\n d.d = week.getUTCDate() + (d.w + 6) % 7;\n } else {\n week = localDate(newDate(d.y, 0, 1)), day$1 = week.getDay();\n week = day$1 > 4 || day$1 === 0 ? monday.ceil(week) : monday(week);\n week = day.offset(week, (d.V - 1) * 7);\n d.y = week.getFullYear();\n d.m = week.getMonth();\n d.d = week.getDate() + (d.w + 6) % 7;\n }\n } else if (""W"" in d || ""U"" in d) {\n if (!(""w"" in d)) d.w = ""u"" in d ? d.u % 7 : ""W"" in d ? 1 : 0;\n day$1 = ""Z"" in d ? utcDate(newDate(d.y, 0, 1)).getUTCDay() : localDate(newDate(d.y, 0, 1)).getDay();\n d.m = 0;\n d.d = ""W"" in d ? (d.w + 6) % 7 + d.W * 7 - (day$1 + 5) % 7 : d.w + d.U * 7 - (day$1 + 6) % 7;\n }\n\n // If a time zone is specified, all fields are interpreted as UTC and then\n // offset according to the specified time zone.\n if (""Z"" in d) {\n d.H += d.Z / 100 | 0;\n d.M += d.Z % 100;\n return utcDate(d);\n }\n\n // Otherwise, all fields are in local time.\n return localDate(d);\n };\n }\n\n function parseSpecifier(d, specifier, string, j) {\n var i = 0,\n n = specifier.length,\n m = string.length,\n c,\n parse;\n\n while (i < n) {\n if (j >= m) return -1;\n c = specifier.charCodeAt(i++);\n if (c === 37) {\n c = specifier.charAt(i++);\n parse = parses[c in pads ? specifier.charAt(i++) : c];\n if (!parse || ((j = parse(d, string, j)) < 0)) return -1;\n } else if (c != string.charCodeAt(j++)) {\n return -1;\n }\n }\n\n return j;\n }\n\n function parsePeriod(d, string, i) {\n var n = periodRe.exec(string.slice(i));\n return n ? (d.p = periodLookup[n[0].toLowerCase()], i + n[0].length) : -1;\n }\n\n function parseShortWeekday(d, string, i) {\n var n = shortWeekdayRe.exec(string.slice(i));\n return n ? (d.w = shortWeekdayLookup[n[0].toLowerCase()], i + n[0].length) : -1;\n }\n\n function parseWeekday(d, string, i) {\n var n = weekdayRe.exec(string.slice(i));\n return n ? (d.w = weekdayLookup[n[0].toLowerCase()], i + n[0].length) : -1;\n }\n\n function parseShortMonth(d, string, i) {\n var n = shortMonthRe.exec(string.slice(i));\n return n ? (d.m = shortMonthLookup[n[0].toLowerCase()], i + n[0].length) : -1;\n }\n\n function parseMonth(d, string, i) {\n var n = monthRe.exec(string.slice(i));\n return n ? (d.m = monthLookup[n[0].toLowerCase()], i + n[0].length) : -1;\n }\n\n function parseLocaleDateTime(d, string, i) {\n return parseSpecifier(d, locale_dateTime, string, i);\n }\n\n function parseLocaleDate(d, string, i) {\n return parseSpecifier(d, locale_date, string, i);\n }\n\n function parseLocaleTime(d, string, i) {\n return parseSpecifier(d, locale_time, string, i);\n }\n\n function formatShortWeekday(d) {\n return locale_shortWeekdays[d.getDay()];\n }\n\n function formatWeekday(d) {\n return locale_weekdays[d.getDay()];\n }\n\n function formatShortMonth(d) {\n return locale_shortMonths[d.getMonth()];\n }\n\n function formatMonth(d) {\n return locale_months[d.getMonth()];\n }\n\n function formatPeriod(d) {\n return locale_periods[+(d.getHours() >= 12)];\n }\n\n function formatQuarter(d) {\n return 1 + ~~(d.getMonth() / 3);\n }\n\n function formatUTCShortWeekday(d) {\n return locale_shortWeekdays[d.getUTCDay()];\n }\n\n function formatUTCWeekday(d) {\n return locale_weekdays[d.getUTCDay()];\n }\n\n function formatUTCShortMonth(d) {\n return locale_shortMonths[d.getUTCMonth()];\n }\n\n function formatUTCMonth(d) {\n return locale_months[d.getUTCMonth()];\n }\n\n function formatUTCPeriod(d) {\n return locale_periods[+(d.getUTCHours() >= 12)];\n }\n\n function formatUTCQuarter(d) {\n return 1 + ~~(d.getUTCMonth() / 3);\n }\n\n return {\n format: function(specifier) {\n var f = newFormat(specifier += """", formats);\n f.toString = function() { return specifier; };\n return f;\n },\n parse: function(specifier) {\n var p = newParse(specifier += """", false);\n p.toString = function() { return specifier; };\n return p;\n },\n utcFormat: function(specifier) {\n var f = newFormat(specifier += """", utcFormats);\n f.toString = function() { return specifier; };\n return f;\n },\n utcParse: function(specifier) {\n var p = newParse(specifier += """", true);\n p.toString = function() { return specifier; };\n return p;\n }\n };\n }\n\n var pads = {""-"": """", ""_"": "" "", ""0"": ""0""},\n numberRe = /^\s*\d+/, // note: ignores next directive\n percentRe = /^%/,\n requoteRe = /[\\^$*+?|[\]().{}]/g;\n\n function pad(value, fill, width) {\n var sign = value < 0 ? ""-"" : """",\n string = (sign ? -value : value) + """",\n length = string.length;\n return sign + (length < width ? new Array(width - length + 1).join(fill) + string : string);\n }\n\n function requote(s) {\n return s.replace(requoteRe, ""\\$&"");\n }\n\n function formatRe(names) {\n return new RegExp(""^(?:"" + names.map(requote).join(""|"") + "")"", ""i"");\n }\n\n function formatLookup(names) {\n var map = {}, i = -1, n = names.length;\n while (++i < n) map[names[i].toLowerCase()] = i;\n return map;\n }\n\n function parseWeekdayNumberSunday(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 1));\n return n ? (d.w = +n[0], i + n[0].length) : -1;\n }\n\n function parseWeekdayNumberMonday(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 1));\n return n ? (d.u = +n[0], i + n[0].length) : -1;\n }\n\n function parseWeekNumberSunday(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.U = +n[0], i + n[0].length) : -1;\n }\n\n function parseWeekNumberISO(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.V = +n[0], i + n[0].length) : -1;\n }\n\n function parseWeekNumberMonday(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.W = +n[0], i + n[0].length) : -1;\n }\n\n function parseFullYear(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 4));\n return n ? (d.y = +n[0], i + n[0].length) : -1;\n }\n\n function parseYear(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.y = +n[0] + (+n[0] > 68 ? 1900 : 2000), i + n[0].length) : -1;\n }\n\n function parseZone(d, string, i) {\n var n = /^(Z)|([+-]\d\d)(?::?(\d\d))?/.exec(string.slice(i, i + 6));\n return n ? (d.Z = n[1] ? 0 : -(n[2] + (n[3] || ""00"")), i + n[0].length) : -1;\n }\n\n function parseQuarter(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 1));\n return n ? (d.q = n[0] * 3 - 3, i + n[0].length) : -1;\n }\n\n function parseMonthNumber(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.m = n[0] - 1, i + n[0].length) : -1;\n }\n\n function parseDayOfMonth(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.d = +n[0], i + n[0].length) : -1;\n }\n\n function parseDayOfYear(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 3));\n return n ? (d.m = 0, d.d = +n[0], i + n[0].length) : -1;\n }\n\n function parseHour24(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.H = +n[0], i + n[0].length) : -1;\n }\n\n function parseMinutes(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.M = +n[0], i + n[0].length) : -1;\n }\n\n function parseSeconds(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 2));\n return n ? (d.S = +n[0], i + n[0].length) : -1;\n }\n\n function parseMilliseconds(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 3));\n return n ? (d.L = +n[0], i + n[0].length) : -1;\n }\n\n function parseMicroseconds(d, string, i) {\n var n = numberRe.exec(string.slice(i, i + 6));\n return n ? (d.L = Math.floor(n[0] / 1000), i + n[0].length) : -1;\n }\n\n function parseLiteralPercent(d, string, i) {\n var n = percentRe.exec(string.slice(i, i + 1));\n return n ? i + n[0].length : -1;\n }\n\n function parseUnixTimestamp(d, string, i) {\n var n = numberRe.exec(string.slice(i));\n return n ? (d.Q = +n[0], i + n[0].length) : -1;\n }\n\n function parseUnixTimestampSeconds(d, string, i) {\n var n = numberRe.exec(string.slice(i));\n return n ? (d.s = +n[0], i + n[0].length) : -1;\n }\n\n function formatDayOfMonth(d, p) {\n return pad(d.getDate(), p, 2);\n }\n\n function formatHour24(d, p) {\n return pad(d.getHours(), p, 2);\n }\n\n function formatHour12(d, p) {\n return pad(d.getHours() % 12 || 12, p, 2);\n }\n\n function formatDayOfYear(d, p) {\n return pad(1 + day.count(year(d), d), p, 3);\n }\n\n function formatMilliseconds(d, p) {\n return pad(d.getMilliseconds(), p, 3);\n }\n\n function formatMicroseconds(d, p) {\n return formatMilliseconds(d, p) + ""000"";\n }\n\n function formatMonthNumber(d, p) {\n return pad(d.getMonth() + 1, p, 2);\n }\n\n function formatMinutes(d, p) {\n return pad(d.getMinutes(), p, 2);\n }\n\n function formatSeconds(d, p) {\n return pad(d.getSeconds(), p, 2);\n }\n\n function formatWeekdayNumberMonday(d) {\n var day = d.getDay();\n return day === 0 ? 7 : day;\n }\n\n function formatWeekNumberSunday(d, p) {\n return pad(sunday.count(year(d) - 1, d), p, 2);\n }\n\n function formatWeekNumberISO(d, p) {\n var day = d.getDay();\n d = (day >= 4 || day === 0) ? thursday(d) : thursday.ceil(d);\n return pad(thursday.count(year(d), d) + (year(d).getDay() === 4), p, 2);\n }\n\n function formatWeekdayNumberSunday(d) {\n return d.getDay();\n }\n\n function formatWeekNumberMonday(d, p) {\n return pad(monday.count(year(d) - 1, d), p, 2);\n }\n\n function formatYear(d, p) {\n return pad(d.getFullYear() % 100, p, 2);\n }\n\n function formatFullYear(d, p) {\n return pad(d.getFullYear() % 10000, p, 4);\n }\n\n function formatZone(d) {\n var z = d.getTimezoneOffset();\n return (z > 0 ? ""-"" : (z *= -1, ""+""))\n + pad(z / 60 | 0, ""0"", 2)\n + pad(z % 60, ""0"", 2);\n }\n\n function formatUTCDayOfMonth(d, p) {\n return pad(d.getUTCDate(), p, 2);\n }\n\n function formatUTCHour24(d, p) {\n return pad(d.getUTCHours(), p, 2);\n }\n\n function formatUTCHour12(d, p) {\n return pad(d.getUTCHours() % 12 || 12, p, 2);\n }\n\n function formatUTCDayOfYear(d, p) {\n return pad(1 + utcDay.count(utcYear(d), d), p, 3);\n }\n\n function formatUTCMilliseconds(d, p) {\n return pad(d.getUTCMilliseconds(), p, 3);\n }\n\n function formatUTCMicroseconds(d, p) {\n return formatUTCMilliseconds(d, p) + ""000"";\n }\n\n function formatUTCMonthNumber(d, p) {\n return pad(d.getUTCMonth() + 1, p, 2);\n }\n\n function formatUTCMinutes(d, p) {\n return pad(d.getUTCMinutes(), p, 2);\n }\n\n function formatUTCSeconds(d, p) {\n return pad(d.getUTCSeconds(), p, 2);\n }\n\n function formatUTCWeekdayNumberMonday(d) {\n var dow = d.getUTCDay();\n return dow === 0 ? 7 : dow;\n }\n\n function formatUTCWeekNumberSunday(d, p) {\n return pad(utcSunday.count(utcYear(d) - 1, d), p, 2);\n }\n\n function formatUTCWeekNumberISO(d, p) {\n var day = d.getUTCDay();\n d = (day >= 4 || day === 0) ? utcThursday(d) : utcThursday.ceil(d);\n return pad(utcThursday.count(utcYear(d), d) + (utcYear(d).getUTCDay() === 4), p, 2);\n }\n\n function formatUTCWeekdayNumberSunday(d) {\n return d.getUTCDay();\n }\n\n function formatUTCWeekNumberMonday(d, p) {\n return pad(utcMonday.count(utcYear(d) - 1, d), p, 2);\n }\n\n function formatUTCYear(d, p) {\n return pad(d.getUTCFullYear() % 100, p, 2);\n }\n\n function formatUTCFullYear(d, p) {\n return pad(d.getUTCFullYear() % 10000, p, 4);\n }\n\n function formatUTCZone() {\n return ""+0000"";\n }\n\n function formatLiteralPercent() {\n return ""%"";\n }\n\n function formatUnixTimestamp(d) {\n return +d;\n }\n\n function formatUnixTimestampSeconds(d) {\n return Math.floor(+d / 1000);\n }\n\n var locale$1;\n var timeFormat;\n var timeParse;\n var utcFormat;\n var utcParse;\n\n defaultLocale$1({\n dateTime: ""%x, %X"",\n date: ""%-m/%-d/%Y"",\n time: ""%-I:%M:%S %p"",\n periods: [""AM"", ""PM""],\n days: [""Sunday"", ""Monday"", ""Tuesday"", ""Wednesday"", ""Thursday"", ""Friday"", ""Saturday""],\n shortDays: [""Sun"", ""Mon"", ""Tue"", ""Wed"", ""Thu"", ""Fri"", ""Sat""],\n months: [""January"", ""February"", ""March"", ""April"", ""May"", ""June"", ""July"", ""August"", ""September"", ""October"", ""November"", ""December""],\n shortMonths: [""Jan"", ""Feb"", ""Mar"", ""Apr"", ""May"", ""Jun"", ""Jul"", ""Aug"", ""Sep"", ""Oct"", ""Nov"", ""Dec""]\n });\n\n function defaultLocale$1(definition) {\n locale$1 = formatLocale$1(definition);\n timeFormat = locale$1.format;\n timeParse = locale$1.parse;\n utcFormat = locale$1.utcFormat;\n utcParse = locale$1.utcParse;\n return locale$1;\n }\n\n var isoSpecifier = ""%Y-%m-%dT%H:%M:%S.%LZ"";\n\n function formatIsoNative(date) {\n return date.toISOString();\n }\n\n var formatIso = Date.prototype.toISOString\n ? formatIsoNative\n : utcFormat(isoSpecifier);\n\n function parseIsoNative(string) {\n var date = new Date(string);\n return isNaN(date) ? null : date;\n }\n\n var parseIso = +new Date(""2000-01-01T00:00:00.000Z"")\n ? parseIsoNative\n : utcParse(isoSpecifier);\n\n var noop = {value: function() {}};\n\n function dispatch() {\n for (var i = 0, n = arguments.length, _ = {}, t; i < n; ++i) {\n if (!(t = arguments[i] + """") || (t in _) || /[\s.]/.test(t)) throw new Error(""illegal type: "" + t);\n _[t] = [];\n }\n return new Dispatch(_);\n }\n\n function Dispatch(_) {\n this._ = _;\n }\n\n function parseTypenames(typenames, types) {\n return typenames.trim().split(/^|\s+/).map(function(t) {\n var name = """", i = t.indexOf(""."");\n if (i >= 0) name = t.slice(i + 1), t = t.slice(0, i);\n if (t && !types.hasOwnProperty(t)) throw new Error(""unknown type: "" + t);\n return {type: t, name: name};\n });\n }\n\n Dispatch.prototype = dispatch.prototype = {\n constructor: Dispatch,\n on: function(typename, callback) {\n var _ = this._,\n T = parseTypenames(typename + """", _),\n t,\n i = -1,\n n = T.length;\n\n // If no callback was specified, return the callback of the given type and name.\n if (arguments.length < 2) {\n while (++i < n) if ((t = (typename = T[i]).type) && (t = get(_[t], typename.name))) return t;\n return;\n }\n\n // If a type was specified, set the callback for the given type and name.\n // Otherwise, if a null callback was specified, remove callbacks of the given name.\n if (callback != null && typeof callback !== ""function"") throw new Error(""invalid callback: "" + callback);\n while (++i < n) {\n if (t = (typename = T[i]).type) _[t] = set(_[t], typename.name, callback);\n else if (callback == null) for (t in _) _[t] = set(_[t], typename.name, null);\n }\n\n return this;\n },\n copy: function() {\n var copy = {}, _ = this._;\n for (var t in _) copy[t] = _[t].slice();\n return new Dispatch(copy);\n },\n call: function(type, that) {\n if ((n = arguments.length - 2) > 0) for (var args = new Array(n), i = 0, n, t; i < n; ++i) args[i] = arguments[i + 2];\n if (!this._.hasOwnProperty(type)) throw new Error(""unknown type: "" + type);\n for (t = this._[type], i = 0, n = t.length; i < n; ++i) t[i].value.apply(that, args);\n },\n apply: function(type, that, args) {\n if (!this._.hasOwnProperty(type)) throw new Error(""unknown type: "" + type);\n for (var t = this._[type], i = 0, n = t.length; i < n; ++i) t[i].value.apply(that, args);\n }\n };\n\n function get(type, name) {\n for (var i = 0, n = type.length, c; i < n; ++i) {\n if ((c = type[i]).name === name) {\n return c.value;\n }\n }\n }\n\n function set(type, name, callback) {\n for (var i = 0, n = type.length; i < n; ++i) {\n if (type[i].name === name) {\n type[i] = noop, type = type.slice(0, i).concat(type.slice(i + 1));\n break;\n }\n }\n if (callback != null) type.push({name: name, value: callback});\n return type;\n }\n\n var xhtml = ""http://www.w3.org/1999/xhtml"";\n\n var namespaces = {\n svg: ""http://www.w3.org/2000/svg"",\n xhtml: xhtml,\n xlink: ""http://www.w3.org/1999/xlink"",\n xml: ""http://www.w3.org/XML/1998/namespace"",\n xmlns: ""http://www.w3.org/2000/xmlns/""\n };\n\n function namespace(name) {\n var prefix = name += """", i = prefix.indexOf("":"");\n if (i >= 0 && (prefix = name.slice(0, i)) !== ""xmlns"") name = name.slice(i + 1);\n return namespaces.hasOwnProperty(prefix) ? {space: namespaces[prefix], local: name} : name;\n }\n\n function creatorInherit(name) {\n return function() {\n var document = this.ownerDocument,\n uri = this.namespaceURI;\n return uri === xhtml && document.documentElement.namespaceURI === xhtml\n ? document.createElement(name)\n : document.createElementNS(uri, name);\n };\n }\n\n function creatorFixed(fullname) {\n return function() {\n return this.ownerDocument.createElementNS(fullname.space, fullname.local);\n };\n }\n\n function creator(name) {\n var fullname = namespace(name);\n return (fullname.local\n ? creatorFixed\n : creatorInherit)(fullname);\n }\n\n function none() {}\n\n function selector(selector) {\n return selector == null ? none : function() {\n return this.querySelector(selector);\n };\n }\n\n function selection_select(select) {\n if (typeof select !== ""function"") select = selector(select);\n\n for (var groups = this._groups, m = groups.length, subgroups = new Array(m), j = 0; j < m; ++j) {\n for (var group = groups[j], n = group.length, subgroup = subgroups[j] = new Array(n), node, subnode, i = 0; i < n; ++i) {\n if ((node = group[i]) && (subnode = select.call(node, node.__data__, i, group))) {\n if (""__data__"" in node) subnode.__data__ = node.__data__;\n subgroup[i] = subnode;\n }\n }\n }\n\n return new Selection(subgroups, this._parents);\n }\n\n function empty() {\n return [];\n }\n\n function selectorAll(selector) {\n return selector == null ? empty : function() {\n return this.querySelectorAll(selector);\n };\n }\n\n function selection_selectAll(select) {\n if (typeof select !== ""function"") select = selectorAll(select);\n\n for (var groups = this._groups, m = groups.length, subgroups = [], parents = [], j = 0; j < m; ++j) {\n for (var group = groups[j], n = group.length, node, i = 0; i < n; ++i) {\n if (node = group[i]) {\n subgroups.push(select.call(node, node.__data__, i, group));\n parents.push(node);\n }\n }\n }\n\n return new Selection(subgroups, parents);\n }\n\n function matcher(selector) {\n return function() {\n return this.matches(selector);\n };\n }\n\n function selection_filter(match) {\n if (typeof match !== ""function"") match = matcher(match);\n\n for (var groups = this._groups, m = groups.length, subgroups = new Array(m), j = 0; j < m; ++j) {\n for (var group = groups[j], n = group.length, subgroup = subgroups[j] = [], node, i = 0; i < n; ++i) {\n if ((node = group[i]) && match.call(node, node.__data__, i, group)) {\n subgroup.push(node);\n }\n }\n }\n\n return new Selection(subgroups, this._parents);\n }\n\n function sparse(update) {\n return new Array(update.length);\n }\n\n function selection_enter() {\n return new Selection(this._enter || this._groups.map(sparse), this._parents);\n }\n\n function EnterNode(parent, datum) {\n this.ownerDocument = parent.ownerDocument;\n this.namespaceURI = parent.namespaceURI;\n this._next = null;\n this._parent = parent;\n this.__data__ = datum;\n }\n\n EnterNode.prototype = {\n constructor: EnterNode,\n appendChild: function(child) { return this._parent.insertBefore(child, this._next); },\n insertBefore: function(child, next) { return this._parent.insertBefore(child, next); },\n querySelector: function(selector) { return this._parent.querySelector(selector); },\n querySelectorAll: function(selector) { return this._parent.querySelectorAll(selector); }\n };\n\n function constant$2(x) {\n return function() {\n return x;\n };\n }\n\n var keyPrefix = ""$""; // Protect against keys like “__proto__”.\n\n function bindIndex(parent, group, enter, update, exit, data) {\n var i = 0,\n node,\n groupLength = group.length,\n dataLength = data.length;\n\n // Put any non-null nodes that fit into update.\n // Put any null nodes into enter.\n // Put any remaining data into enter.\n for (; i < dataLength; ++i) {\n if (node = group[i]) {\n node.__data__ = data[i];\n update[i] = node;\n } else {\n enter[i] = new EnterNode(parent, data[i]);\n }\n }\n\n // Put any non-null nodes that don’t fit into exit.\n for (; i < groupLength; ++i) {\n if (node = group[i]) {\n exit[i] = node;\n }\n }\n }\n\n function bindKey(parent, group, enter, update, exit, data, key) {\n var i,\n node,\n nodeByKeyValue = {},\n groupLength = group.length,\n dataLength = data.length,\n keyValues = new Array(groupLength),\n keyValue;\n\n // Compute the key for each node.\n // If multiple nodes have the same key, the duplicates are added to exit.\n for (i = 0; i < groupLength; ++i) {\n if (node = group[i]) {\n keyValues[i] = keyValue = keyPrefix + key.call(node, node.__data__, i, group);\n if (keyValue in nodeByKeyValue) {\n exit[i] = node;\n } else {\n nodeByKeyValue[keyValue] = node;\n }\n }\n }\n\n // Compute the key for each datum.\n // If there a node associated with this key, join and add it to update.\n // If there is not (or the key is a duplicate), add it to enter.\n for (i = 0; i < dataLength; ++i) {\n keyValue = keyPrefix + key.call(parent, data[i], i, data);\n if (node = nodeByKeyValue[keyValue]) {\n update[i] = node;\n node.__data__ = data[i];\n nodeByKeyValue[keyValue] = null;\n } else {\n enter[i] = new EnterNode(parent, data[i]);\n }\n }\n\n // Add any remaining nodes that were not bound to data to exit.\n for (i = 0; i < groupLength; ++i) {\n if ((node = group[i]) && (nodeByKeyValue[keyValues[i]] === node)) {\n exit[i] = node;\n }\n }\n }\n\n function selection_data(value, key) {\n if (!value) {\n data = new Array(this.size()), j = -1;\n this.each(function(d) { data[++j] = d; });\n return data;\n }\n\n var bind = key ? bindKey : bindIndex,\n parents = this._parents,\n groups = this._groups;\n\n if (typeof value !== ""function"") value = constant$2(value);\n\n for (var m = groups.length, update = new Array(m), enter = new Array(m), exit = new Array(m), j = 0; j < m; ++j) {\n var parent = parents[j],\n group = groups[j],\n groupLength = group.length,\n data = value.call(parent, parent && parent.__data__, j, parents),\n dataLength = data.length,\n enterGroup = enter[j] = new Array(dataLength),\n updateGroup = update[j] = new Array(dataLength),\n exitGroup = exit[j] = new Array(groupLength);\n\n bind(parent, group, enterGroup, updateGroup, exitGroup, data, key);\n\n // Now connect the enter nodes to their following update node, such that\n // appendChild can insert the materialized enter node before this node,\n // rather than at the end of the parent node.\n for (var i0 = 0, i1 = 0, previous, next; i0 < dataLength; ++i0) {\n if (previous = enterGroup[i0]) {\n if (i0 >= i1) i1 = i0 + 1;\n while (!(next = updateGroup[i1]) && ++i1 < dataLength);\n previous._next = next || null;\n }\n }\n }\n\n update = new Selection(update, parents);\n update._enter = enter;\n update._exit = exit;\n return update;\n }\n\n function selection_exit() {\n return new Selection(this._exit || this._groups.map(sparse), this._parents);\n }\n\n function selection_join(onenter, onupdate, onexit) {\n var enter = this.enter(), update = this, exit = this.exit();\n enter = typeof onenter === ""function"" ? onenter(enter) : enter.append(onenter + """");\n if (onupdate != null) update = onupdate(update);\n if (onexit == null) exit.remove(); else onexit(exit);\n return enter && update ? enter.merge(update).order() : update;\n }\n\n function selection_merge(selection) {\n\n for (var groups0 = this._groups, groups1 = selection._groups, m0 = groups0.length, m1 = groups1.length, m = Math.min(m0, m1), merges = new Array(m0), j = 0; j < m; ++j) {\n for (var group0 = groups0[j], group1 = groups1[j], n = group0.length, merge = merges[j] = new Array(n), node, i = 0; i < n; ++i) {\n if (node = group0[i] || group1[i]) {\n merge[i] = node;\n }\n }\n }\n\n for (; j < m0; ++j) {\n merges[j] = groups0[j];\n }\n\n return new Selection(merges, this._parents);\n }\n\n function selection_order() {\n\n for (var groups = this._groups, j = -1, m = groups.length; ++j < m;) {\n for (var group = groups[j], i = group.length - 1, next = group[i], node; --i >= 0;) {\n if (node = group[i]) {\n if (next && node.compareDocumentPosition(next) ^ 4) next.parentNode.insertBefore(node, next);\n next = node;\n }\n }\n }\n\n return this;\n }\n\n function selection_sort(compare) {\n if (!compare) compare = ascending$1;\n\n function compareNode(a, b) {\n return a && b ? compare(a.__data__, b.__data__) : !a - !b;\n }\n\n for (var groups = this._groups, m = groups.length, sortgroups = new Array(m), j = 0; j < m; ++j) {\n for (var group = groups[j], n = group.length, sortgroup = sortgroups[j] = new Array(n), node, i = 0; i < n; ++i) {\n if (node = group[i]) {\n sortgroup[i] = node;\n }\n }\n sortgroup.sort(compareNode);\n }\n\n return new Selection(sortgroups, this._parents).order();\n }\n\n function ascending$1(a, b) {\n return a < b ? -1 : a > b ? 1 : a >= b ? 0 : NaN;\n }\n\n function selection_call() {\n var callback = arguments[0];\n arguments[0] = this;\n callback.apply(null, arguments);\n return this;\n }\n\n function selection_nodes() {\n var nodes = new Array(this.size()), i = -1;\n this.each(function() { nodes[++i] = this; });\n return nodes;\n }\n\n function selection_node() {\n\n for (var groups = this._groups, j = 0, m = groups.length; j < m; ++j) {\n for (var group = groups[j], i = 0, n = group.length; i < n; ++i) {\n var node = group[i];\n if (node) return node;\n }\n }\n\n return null;\n }\n\n function selection_size() {\n var size = 0;\n this.each(function() { ++size; });\n return size;\n }\n\n function selection_empty() {\n return !this.node();\n }\n\n function selection_each(callback) {\n\n for (var groups = this._groups, j = 0, m = groups.length; j < m; ++j) {\n for (var group = groups[j], i = 0, n = group.length, node; i < n; ++i) {\n if (node = group[i]) callback.call(node, node.__data__, i, group);\n }\n }\n\n return this;\n }\n\n function attrRemove(name) {\n return function() {\n this.removeAttribute(name);\n };\n }\n\n function attrRemoveNS(fullname) {\n return function() {\n this.removeAttributeNS(fullname.space, fullname.local);\n };\n }\n\n function attrConstant(name, value) {\n return function() {\n this.setAttribute(name, value);\n };\n }\n\n function attrConstantNS(fullname, value) {\n return function() {\n this.setAttributeNS(fullname.space, fullname.local, value);\n };\n }\n\n function attrFunction(name, value) {\n return function() {\n var v = value.apply(this, arguments);\n if (v == null) this.removeAttribute(name);\n else this.setAttribute(name, v);\n };\n }\n\n function attrFunctionNS(fullname, value) {\n return function() {\n var v = value.apply(this, arguments);\n if (v == null) this.removeAttributeNS(fullname.space, fullname.local);\n else this.setAttributeNS(fullname.space, fullname.local, v);\n };\n }\n\n function selection_attr(name, value) {\n var fullname = namespace(name);\n\n if (arguments.length < 2) {\n var node = this.node();\n return fullname.local\n ? node.getAttributeNS(fullname.space, fullname.local)\n : node.getAttribute(fullname);\n }\n\n return this.each((value == null\n ? (fullname.local ? attrRemoveNS : attrRemove) : (typeof value === ""function""\n ? (fullname.local ? attrFunctionNS : attrFunction)\n : (fullname.local ? attrConstantNS : attrConstant)))(fullname, value));\n }\n\n function defaultView(node) {\n return (node.ownerDocument && node.ownerDocument.defaultView) // node is a Node\n || (node.document && node) // node is a Window\n || node.defaultView; // node is a Document\n }\n\n function styleRemove(name) {\n return function() {\n this.style.removeProperty(name);\n };\n }\n\n function styleConstant(name, value, priority) {\n return function() {\n this.style.setProperty(name, value, priority);\n };\n }\n\n function styleFunction(name, value, priority) {\n return function() {\n var v = value.apply(this, arguments);\n if (v == null) this.style.removeProperty(name);\n else this.style.setProperty(name, v, priority);\n };\n }\n\n function selection_style(name, value, priority) {\n return arguments.length > 1\n ? this.each((value == null\n ? styleRemove : typeof value === ""function""\n ? styleFunction\n : styleConstant)(name, value, priority == null ? """" : priority))\n : styleValue(this.node(), name);\n }\n\n function styleValue(node, name) {\n return node.style.getPropertyValue(name)\n || defaultView(node).getComputedStyle(node, null).getPropertyValue(name);\n }\n\n function propertyRemove(name) {\n return function() {\n delete this[name];\n };\n }\n\n function propertyConstant(name, value) {\n return function() {\n this[name] = value;\n };\n }\n\n function propertyFunction(name, value) {\n return function() {\n var v = value.apply(this, arguments);\n if (v == null) delete this[name];\n else this[name] = v;\n };\n }\n\n function selection_property(name, value) {\n return arguments.length > 1\n ? this.each((value == null\n ? propertyRemove : typeof value === ""function""\n ? propertyFunction\n : propertyConstant)(name, value))\n : this.node()[name];\n }\n\n function classArray(string) {\n return string.trim().split(/^|\s+/);\n }\n\n function classList(node) {\n return node.classList || new ClassList(node);\n }\n\n function ClassList(node) {\n this._node = node;\n this._names = classArray(node.getAttribute(""class"") || """");\n }\n\n ClassList.prototype = {\n add: function(name) {\n var i = this._names.indexOf(name);\n if (i < 0) {\n this._names.push(name);\n this._node.setAttribute(""class"", this._names.join("" ""));\n }\n },\n remove: function(name) {\n var i = this._names.indexOf(name);\n if (i >= 0) {\n this._names.splice(i, 1);\n this._node.setAttribute(""class"", this._names.join("" ""));\n }\n },\n contains: function(name) {\n return this._names.indexOf(name) >= 0;\n }\n };\n\n function classedAdd(node, names) {\n var list = classList(node), i = -1, n = names.length;\n while (++i < n) list.add(names[i]);\n }\n\n function classedRemove(node, names) {\n var list = classList(node), i = -1, n = names.length;\n while (++i < n) list.remove(names[i]);\n }\n\n function classedTrue(names) {\n return function() {\n classedAdd(this, names);\n };\n }\n\n function classedFalse(names) {\n return function() {\n classedRemove(this, names);\n };\n }\n\n function classedFunction(names, value) {\n return function() {\n (value.apply(this, arguments) ? classedAdd : classedRemove)(this, names);\n };\n }\n\n function selection_classed(name, value) {\n var names = classArray(name + """");\n\n if (arguments.length < 2) {\n var list = classList(this.node()), i = -1, n = names.length;\n while (++i < n) if (!list.contains(names[i])) return false;\n return true;\n }\n\n return this.each((typeof value === ""function""\n ? classedFunction : value\n ? classedTrue\n : classedFalse)(names, value));\n }\n\n function textRemove() {\n this.textContent = """";\n }\n\n function textConstant(value) {\n return function() {\n this.textContent = value;\n };\n }\n\n function textFunction(value) {\n return function() {\n var v = value.apply(this, arguments);\n this.textContent = v == null ? """" : v;\n };\n }\n\n function selection_text(value) {\n return arguments.length\n ? this.each(value == null\n ? textRemove : (typeof value === ""function""\n ? textFunction\n : textConstant)(value))\n : this.node().textContent;\n }\n\n function htmlRemove() {\n this.innerHTML = """";\n }\n\n function htmlConstant(value) {\n return function() {\n this.innerHTML = value;\n };\n }\n\n function htmlFunction(value) {\n return function() {\n var v = value.apply(this, arguments);\n this.innerHTML = v == null ? """" : v;\n };\n }\n\n function selection_html(value) {\n return arguments.length\n ? this.each(value == null\n ? htmlRemove : (typeof value === ""function""\n ? htmlFunction\n : htmlConstant)(value))\n : this.node().innerHTML;\n }\n\n function raise() {\n if (this.nextSibling) this.parentNode.appendChild(this);\n }\n\n function selection_raise() {\n return this.each(raise);\n }\n\n function lower() {\n if (this.previousSibling) this.parentNode.insertBefore(this, this.parentNode.firstChild);\n }\n\n function selection_lower() {\n return this.each(lower);\n }\n\n function selection_append(name) {\n var create = typeof name === ""function"" ? name : creator(name);\n return this.select(function() {\n return this.appendChild(create.apply(this, arguments));\n });\n }\n\n function constantNull() {\n return null;\n }\n\n function selection_insert(name, before) {\n var create = typeof name === ""function"" ? name : creator(name),\n select = before == null ? constantNull : typeof before === ""function"" ? before : selector(before);\n return this.select(function() {\n return this.insertBefore(create.apply(this, arguments), select.apply(this, arguments) || null);\n });\n }\n\n function remove() {\n var parent = this.parentNode;\n if (parent) parent.removeChild(this);\n }\n\n function selection_remove() {\n return this.each(remove);\n }\n\n function selection_cloneShallow() {\n var clone = this.cloneNode(false), parent = this.parentNode;\n return parent ? parent.insertBefore(clone, this.nextSibling) : clone;\n }\n\n function selection_cloneDeep() {\n var clone = this.cloneNode(true), parent = this.parentNode;\n return parent ? parent.insertBefore(clone, this.nextSibling) : clone;\n }\n\n function selection_clone(deep) {\n return this.select(deep ? selection_cloneDeep : selection_cloneShallow);\n }\n\n function selection_datum(value) {\n return arguments.length\n ? this.property(""__data__"", value)\n : this.node().__data__;\n }\n\n var filterEvents = {};\n\n var event = null;\n\n if (typeof document !== ""undefined"") {\n var element = document.documentElement;\n if (!(""onmouseenter"" in element)) {\n filterEvents = {mouseenter: ""mouseover"", mouseleave: ""mouseout""};\n }\n }\n\n function filterContextListener(listener, index, group) {\n listener = contextListener(listener, index, group);\n return function(event) {\n var related = event.relatedTarget;\n if (!related || (related !== this && !(related.compareDocumentPosition(this) & 8))) {\n listener.call(this, event);\n }\n };\n }\n\n function contextListener(listener, index, group) {\n return function(event1) {\n var event0 = event; // Events can be reentrant (e.g., focus).\n event = event1;\n try {\n listener.call(this, this.__data__, index, group);\n } finally {\n event = event0;\n }\n };\n }\n\n function parseTypenames$1(typenames) {\n return typenames.trim().split(/^|\s+/).map(function(t) {\n var name = """", i = t.indexOf(""."");\n if (i >= 0) name = t.slice(i + 1), t = t.slice(0, i);\n return {type: t, name: name};\n });\n }\n\n function onRemove(typename) {\n return function() {\n var on = this.__on;\n if (!on) return;\n for (var j = 0, i = -1, m = on.length, o; j < m; ++j) {\n if (o = on[j], (!typename.type || o.type === typename.type) && o.name === typename.name) {\n this.removeEventListener(o.type, o.listener, o.capture);\n } else {\n on[++i] = o;\n }\n }\n if (++i) on.length = i;\n else delete this.__on;\n };\n }\n\n function onAdd(typename, value, capture) {\n var wrap = filterEvents.hasOwnProperty(typename.type) ? filterContextListener : contextListener;\n return function(d, i, group) {\n var on = this.__on, o, listener = wrap(value, i, group);\n if (on) for (var j = 0, m = on.length; j < m; ++j) {\n if ((o = on[j]).type === typename.type && o.name === typename.name) {\n this.removeEventListener(o.type, o.listener, o.capture);\n this.addEventListener(o.type, o.listener = listener, o.capture = capture);\n o.value = value;\n return;\n }\n }\n this.addEventListener(typename.type, listener, capture);\n o = {type: typename.type, name: typename.name, value: value, listener: listener, capture: capture};\n if (!on) this.__on = [o];\n else on.push(o);\n };\n }\n\n function selection_on(typename, value, capture) {\n var typenames = parseTypenames$1(typename + """"), i, n = typenames.length, t;\n\n if (arguments.length < 2) {\n var on = this.node().__on;\n if (on) for (var j = 0, m = on.length, o; j < m; ++j) {\n for (i = 0, o = on[j]; i < n; ++i) {\n if ((t = typenames[i]).type === o.type && t.name === o.name) {\n return o.value;\n }\n }\n }\n return;\n }\n\n on = value ? onAdd : onRemove;\n if (capture == null) capture = false;\n for (i = 0; i < n; ++i) this.each(on(typenames[i], value, capture));\n return this;\n }\n\n function customEvent(event1, listener, that, args) {\n var event0 = event;\n event1.sourceEvent = event;\n event = event1;\n try {\n return listener.apply(that, args);\n } finally {\n event = event0;\n }\n }\n\n function dispatchEvent(node, type, params) {\n var window = defaultView(node),\n event = window.CustomEvent;\n\n if (typeof event === ""function"") {\n event = new event(type, params);\n } else {\n event = window.document.createEvent(""Event"");\n if (params) event.initEvent(type, params.bubbles, params.cancelable), event.detail = params.detail;\n else event.initEvent(type, false, false);\n }\n\n node.dispatchEvent(event);\n }\n\n function dispatchConstant(type, params) {\n return function() {\n return dispatchEvent(this, type, params);\n };\n }\n\n function dispatchFunction(type, params) {\n return function() {\n return dispatchEvent(this, type, params.apply(this, arguments));\n };\n }\n\n function selection_dispatch(type, params) {\n return this.each((typeof params === ""function""\n ? dispatchFunction\n : dispatchConstant)(type, params));\n }\n\n var root = [null];\n\n function Selection(groups, parents) {\n this._groups = groups;\n this._parents = parents;\n }\n\n function selection() {\n return new Selection([[document.documentElement]], root);\n }\n\n Selection.prototype = selection.prototype = {\n constructor: Selection,\n select: selection_select,\n selectAll: selection_selectAll,\n filter: selection_filter,\n data: selection_data,\n enter: selection_enter,\n exit: selection_exit,\n join: selection_join,\n merge: selection_merge,\n order: selection_order,\n sort: selection_sort,\n call: selection_call,\n nodes: selection_nodes,\n node: selection_node,\n size: selection_size,\n empty: selection_empty,\n each: selection_each,\n attr: selection_attr,\n style: selection_style,\n property: selection_property,\n classed: selection_classed,\n text: selection_text,\n html: selection_html,\n raise: selection_raise,\n lower: selection_lower,\n append: selection_append,\n insert: selection_insert,\n remove: selection_remove,\n clone: selection_clone,\n datum: selection_datum,\n on: selection_on,\n dispatch: selection_dispatch\n };\n\n function select(selector) {\n return typeof selector === ""string""\n ? new Selection([[document.querySelector(selector)]], [document.documentElement])\n : new Selection([[selector]], root);\n }\n\n function sourceEvent() {\n var current = event, source;\n while (source = current.sourceEvent) current = source;\n return current;\n }\n\n function point(node, event) {\n var svg = node.ownerSVGElement || node;\n\n if (svg.createSVGPoint) {\n var point = svg.createSVGPoint();\n point.x = event.clientX, point.y = event.clientY;\n point = point.matrixTransform(node.getScreenCTM().inverse());\n return [point.x, point.y];\n }\n\n var rect = node.getBoundingClientRect();\n return [event.clientX - rect.left - node.clientLeft, event.clientY - rect.top - node.clientTop];\n }\n\n function mouse(node) {\n var event = sourceEvent();\n if (event.changedTouches) event = event.changedTouches[0];\n return point(node, event);\n }\n\n function touch(node, touches, identifier) {\n if (arguments.length < 3) identifier = touches, touches = sourceEvent().changedTouches;\n\n for (var i = 0, n = touches ? touches.length : 0, touch; i < n; ++i) {\n if ((touch = touches[i]).identifier === identifier) {\n return point(node, touch);\n }\n }\n\n return null;\n }\n\n function nopropagation() {\n event.stopImmediatePropagation();\n }\n\n function noevent() {\n event.preventDefault();\n event.stopImmediatePropagation();\n }\n\n function nodrag(view) {\n var root = view.document.documentElement,\n selection = select(view).on(""dragstart.drag"", noevent, true);\n if (""onselectstart"" in root) {\n selection.on(""selectstart.drag"", noevent, true);\n } else {\n root.__noselect = root.style.MozUserSelect;\n root.style.MozUserSelect = ""none"";\n }\n }\n\n function yesdrag(view, noclick) {\n var root = view.document.documentElement,\n selection = select(view).on(""dragstart.drag"", null);\n if (noclick) {\n selection.on(""click.drag"", noevent, true);\n setTimeout(function() { selection.on(""click.drag"", null); }, 0);\n }\n if (""onselectstart"" in root) {\n selection.on(""selectstart.drag"", null);\n } else {\n root.style.MozUserSelect = root.__noselect;\n delete root.__noselect;\n }\n }\n\n function constant$3(x) {\n return function() {\n return x;\n };\n }\n\n function DragEvent(target, type, subject, id, active, x, y, dx, dy, dispatch) {\n this.target = target;\n this.type = type;\n this.subject = subject;\n this.identifier = id;\n this.active = active;\n this.x = x;\n this.y = y;\n this.dx = dx;\n this.dy = dy;\n this._ = dispatch;\n }\n\n DragEvent.prototype.on = function() {\n var value = this._.on.apply(this._, arguments);\n return value === this._ ? this : value;\n };\n\n // Ignore right-click, since that should open the context menu.\n function defaultFilter() {\n return !event.ctrlKey && !event.button;\n }\n\n function defaultContainer() {\n return this.parentNode;\n }\n\n function defaultSubject(d) {\n return d == null ? {x: event.x, y: event.y} : d;\n }\n\n function defaultTouchable() {\n return navigator.maxTouchPoints || (""ontouchstart"" in this);\n }\n\n function drag() {\n var filter = defaultFilter,\n container = defaultContainer,\n subject = defaultSubject,\n touchable = defaultTouchable,\n gestures = {},\n listeners = dispatch(""start"", ""drag"", ""end""),\n active = 0,\n mousedownx,\n mousedowny,\n mousemoving,\n touchending,\n clickDistance2 = 0;\n\n function drag(selection) {\n selection\n .on(""mousedown.drag"", mousedowned)\n .filter(touchable)\n .on(""touchstart.drag"", touchstarted)\n .on(""touchmove.drag"", touchmoved)\n .on(""touchend.drag touchcancel.drag"", touchended)\n .style(""touch-action"", ""none"")\n .style(""-webkit-tap-highlight-color"", ""rgba(0,0,0,0)"");\n }\n\n function mousedowned() {\n if (touchending || !filter.apply(this, arguments)) return;\n var gesture = beforestart(""mouse"", container.apply(this, arguments), mouse, this, arguments);\n if (!gesture) return;\n select(event.view).on(""mousemove.drag"", mousemoved, true).on(""mouseup.drag"", mouseupped, true);\n nodrag(event.view);\n nopropagation();\n mousemoving = false;\n mousedownx = event.clientX;\n mousedowny = event.clientY;\n gesture(""start"");\n }\n\n function mousemoved() {\n noevent();\n if (!mousemoving) {\n var dx = event.clientX - mousedownx, dy = event.clientY - mousedowny;\n mousemoving = dx * dx + dy * dy > clickDistance2;\n }\n gestures.mouse(""drag"");\n }\n\n function mouseupped() {\n select(event.view).on(""mousemove.drag mouseup.drag"", null);\n yesdrag(event.view, mousemoving);\n noevent();\n gestures.mouse(""end"");\n }\n\n function touchstarted() {\n if (!filter.apply(this, arguments)) return;\n var touches = event.changedTouches,\n c = container.apply(this, arguments),\n n = touches.length, i, gesture;\n\n for (i = 0; i < n; ++i) {\n if (gesture = beforestart(touches[i].identifier, c, touch, this, arguments)) {\n nopropagation();\n gesture(""start"");\n }\n }\n }\n\n function touchmoved() {\n var touches = event.changedTouches,\n n = touches.length, i, gesture;\n\n for (i = 0; i < n; ++i) {\n if (gesture = gestures[touches[i].identifier]) {\n noevent();\n gesture(""drag"");\n }\n }\n }\n\n function touchended() {\n var touches = event.changedTouches,\n n = touches.length, i, gesture;\n\n if (touchending) clearTimeout(touchending);\n touchending = setTimeout(function() { touchending = null; }, 500); // Ghost clicks are delayed!\n for (i = 0; i < n; ++i) {\n if (gesture = gestures[touches[i].identifier]) {\n nopropagation();\n gesture(""end"");\n }\n }\n }\n\n function beforestart(id, container, point, that, args) {\n var p = point(container, id), s, dx, dy,\n sublisteners = listeners.copy();\n\n if (!customEvent(new DragEvent(drag, ""beforestart"", s, id, active, p[0], p[1], 0, 0, sublisteners), function() {\n if ((event.subject = s = subject.apply(that, args)) == null) return false;\n dx = s.x - p[0] || 0;\n dy = s.y - p[1] || 0;\n return true;\n })) return;\n\n return function gesture(type) {\n var p0 = p, n;\n switch (type) {\n case ""start"": gestures[id] = gesture, n = active++; break;\n case ""end"": delete gestures[id], --active; // nobreak\n case ""drag"": p = point(container, id), n = active; break;\n }\n customEvent(new DragEvent(drag, type, s, id, n, p[0] + dx, p[1] + dy, p[0] - p0[0], p[1] - p0[1], sublisteners), sublisteners.apply, sublisteners, [type, that, args]);\n };\n }\n\n drag.filter = function(_) {\n return arguments.length ? (filter = typeof _ === ""function"" ? _ : constant$3(!!_), drag) : filter;\n };\n\n drag.container = function(_) {\n return arguments.length ? (container = typeof _ === ""function"" ? _ : constant$3(_), drag) : container;\n };\n\n drag.subject = function(_) {\n return arguments.length ? (subject = typeof _ === ""function"" ? _ : constant$3(_), drag) : subject;\n };\n\n drag.touchable = function(_) {\n return arguments.length ? (touchable = typeof _ === ""function"" ? _ : constant$3(!!_), drag) : touchable;\n };\n\n drag.on = function() {\n var value = listeners.on.apply(listeners, arguments);\n return value === listeners ? drag : value;\n };\n\n drag.clickDistance = function(_) {\n return arguments.length ? (clickDistance2 = (_ = +_) * _, drag) : Math.sqrt(clickDistance2);\n };\n\n return drag;\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const T$a = Template('d-slider', `\n\n\n
    \n
    \n
    \n
    \n
    \n
    \n
    \n
    \n
    \n`);\n\n // ARIA\n // If the slider has a visible label, it is referenced by aria-labelledby on the slider element. Otherwise, the slider element has a label provided by aria-label.\n // If the slider is vertically oriented, it has aria-orientation set to vertical. The default value of aria-orientation for a slider is horizontal.\n\n const keyCodes = {\n left: 37,\n up: 38,\n right: 39,\n down: 40,\n pageUp: 33,\n pageDown: 34,\n end: 35,\n home: 36\n };\n\n class Slider extends T$a(HTMLElement) {\n\n\n connectedCallback() {\n this.connected = true;\n this.setAttribute('role', 'slider');\n // Makes the element tab-able.\n if (!this.hasAttribute('tabindex')) { this.setAttribute('tabindex', 0); }\n\n // Keeps track of keyboard vs. mouse interactions for focus rings\n this.mouseEvent = false;\n\n // Handles to shadow DOM elements\n this.knob = this.root.querySelector('.knob-container');\n this.background = this.root.querySelector('.background');\n this.trackFill = this.root.querySelector('.track-fill');\n this.track = this.root.querySelector('.track');\n\n // Default values for attributes\n this.min = this.min ? this.min : 0;\n this.max = this.max ? this.max : 100;\n this.scale = linear$1().domain([this.min, this.max]).range([0, 1]).clamp(true);\n\n this.origin = this.origin !== undefined ? this.origin : this.min;\n this.step = this.step ? this.step : 1;\n this.update(this.value ? this.value : 0);\n\n this.ticks = this.ticks ? this.ticks : false;\n this.renderTicks();\n\n this.drag = drag()\n .container(this.background)\n .on('start', () => {\n this.mouseEvent = true;\n this.background.classList.add('mousedown');\n this.changeValue = this.value;\n this.dragUpdate();\n })\n .on('drag', () => {\n this.dragUpdate();\n })\n .on('end', () => {\n this.mouseEvent = false;\n this.background.classList.remove('mousedown');\n this.dragUpdate();\n if (this.changeValue !== this.value) this.dispatchChange();\n this.changeValue = this.value;\n });\n this.drag(select(this.background));\n\n this.addEventListener('focusin', () => {\n if(!this.mouseEvent) {\n this.background.classList.add('focus');\n }\n });\n this.addEventListener('focusout', () => {\n this.background.classList.remove('focus');\n });\n this.addEventListener('keydown', this.onKeyDown);\n\n }\n\n static get observedAttributes() {return ['min', 'max', 'value', 'step', 'ticks', 'origin', 'tickValues', 'tickLabels']; }\n\n attributeChangedCallback(attr, oldValue, newValue) {\n if (isNaN(newValue) || newValue === undefined || newValue === null) return;\n if (attr == 'min') {\n this.min = +newValue;\n this.setAttribute('aria-valuemin', this.min);\n }\n if (attr == 'max') {\n this.max = +newValue;\n this.setAttribute('aria-valuemax', this.max);\n }\n if (attr == 'value') {\n this.update(+newValue);\n }\n if (attr == 'origin') {\n this.origin = +newValue;\n // this.update(this.value);\n }\n if (attr == 'step') {\n if (newValue > 0) {\n this.step = +newValue;\n }\n }\n if (attr == 'ticks') {\n this.ticks = (newValue === '' ? true : newValue);\n }\n }\n\n onKeyDown(event) {\n this.changeValue = this.value;\n let stopPropagation = false;\n switch (event.keyCode) {\n case keyCodes.left:\n case keyCodes.down:\n this.update(this.value - this.step);\n stopPropagation = true;\n break;\n case keyCodes.right:\n case keyCodes.up:\n this.update(this.value + this.step);\n stopPropagation = true;\n break;\n case keyCodes.pageUp:\n this.update(this.value + this.step * 10);\n stopPropagation = true;\n break;\n\n case keyCodes.pageDown:\n this.update(this.value + this.step * 10);\n stopPropagation = true;\n break;\n case keyCodes.home:\n this.update(this.min);\n stopPropagation = true;\n break;\n case keyCodes.end:\n this.update(this.max);\n stopPropagation = true;\n break;\n }\n if (stopPropagation) {\n this.background.classList.add('focus');\n event.preventDefault();\n event.stopPropagation();\n if (this.changeValue !== this.value) this.dispatchChange();\n }\n }\n\n validateValueRange(min, max, value) {\n return Math.max(Math.min(max, value), min);\n }\n\n quantizeValue(value, step) {\n return Math.round(value / step) * step;\n }\n\n dragUpdate() {\n const bbox = this.background.getBoundingClientRect();\n const x = event.x;\n const width = bbox.width;\n this.update(this.scale.invert(x / width));\n }\n\n update(value) {\n let v = value;\n if (this.step !== 'any') {\n v = this.quantizeValue(value, this.step);\n }\n v = this.validateValueRange(this.min, this.max, v);\n if (this.connected) {\n this.knob.style.left = this.scale(v) * 100 + '%';\n this.trackFill.style.width = this.scale(this.min + Math.abs(v - this.origin)) * 100 + '%';\n this.trackFill.style.left = this.scale(Math.min(v, this.origin)) * 100 + '%';\n }\n if (this.value !== v) {\n this.value = v;\n this.setAttribute('aria-valuenow', this.value);\n this.dispatchInput();\n }\n }\n\n // Dispatches only on a committed change (basically only on mouseup).\n dispatchChange() {\n const e = new Event('change');\n this.dispatchEvent(e, {});\n }\n\n // Dispatches on each value change.\n dispatchInput() {\n const e = new Event('input');\n this.dispatchEvent(e, {});\n }\n\n renderTicks() {\n const ticksContainer = this.root.querySelector('.ticks');\n if (this.ticks !== false) {\n let tickData = [];\n if (this.ticks > 0) {\n tickData = this.scale.ticks(this.ticks);\n } else if (this.step === 'any') {\n tickData = this.scale.ticks();\n } else {\n tickData = range(this.min, this.max + 1e-6, this.step);\n }\n tickData.forEach(d => {\n const tick = document.createElement('div');\n tick.classList.add('tick');\n tick.style.left = this.scale(d) * 100 + '%';\n ticksContainer.appendChild(tick);\n });\n } else {\n ticksContainer.style.display = 'none';\n }\n }\n }\n\n var img = ""data:image/png;base64,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"";\n\n const headerTemplate = `\n\n\n`;\n\n // Copyright 2018 The Distill Template Authors\n\n const T$b = Template('distill-header', headerTemplate, false);\n\n class DistillHeader extends T$b(HTMLElement) {\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n const styles$2 = `\n\n`;\n\n function appendixTemplate(frontMatter) {\n let html = styles$2;\n frontMatter.journal.title = ""p(doom) blog"";\n frontMatter.journal.url = ""https://pdoom.org/blog.html"";\n\n if (typeof frontMatter.githubUrl !== 'undefined') {\n html += `\n

    Updates and Corrections

    \n

    `;\n if (frontMatter.githubCompareUpdatesUrl) {\n html += `View all changes to this article since it was first published.`;\n }\n html += `\n If you see mistakes or want to suggest changes, please create an issue on GitHub.

    \n `;\n }\n\n const journal = frontMatter.journal;\n if (typeof journal !== 'undefined' && journal.title === 'p(doom) Blog') {\n html += `\n

    Reuse

    \n

    Diagrams and text are licensed under Creative Commons Attribution CC-BY 4.0 with the source available on GitHub, unless noted otherwise. The figures that have been reused from other sources don’t fall under this license and can be recognized by a note in their caption: “Figure from …”.

    \n `;\n }\n\n if (typeof frontMatter.publishedDate !== 'undefined') {\n html += `\n

    Citation

    \n

    For attribution in academic contexts, please cite this work as

    \n 
    ${frontMatter.concatenatedAuthors}, ""${frontMatter.title}"", p(doom), ${frontMatter.publishedYear}.
    \n

    BibTeX citation

    \n 
    ${serializeFrontmatterToBibtex(frontMatter)}
    \n `;\n }\n\n return html;\n }\n\n class DistillAppendix extends HTMLElement {\n\n static get is() { return 'distill-appendix'; }\n\n set frontMatter(frontMatter) {\n this.innerHTML = appendixTemplate(frontMatter);\n }\n\n }\n\n const footerTemplate = `\n\n\n
    \n\n \n \n p(doom)\n is dedicated to truly open research on the path to AGI\n\n
    \n Archive\n GitHub\n Twitter\n
    \n We are thankful to distill.pub for creating the template on which we based this blog.\n
    \n\n
    \n\n`;\n\n // Copyright 2018 The Distill Template Authors\n\n const T$c = Template('distill-footer', footerTemplate);\n\n class DistillFooter extends T$c(HTMLElement) {\n\n }\n\n // Copyright 2018 The Distill Template Authors\n\n let templateIsLoading = false;\n let runlevel = 0;\n const initialize = function() {\n if (window.distill.runlevel < 1) {\n throw new Error(""Insufficient Runlevel for Distill Template!"");\n }\n\n /* 1. Flag that we're being loaded */\n if (""distill"" in window && window.distill.templateIsLoading) {\n throw new Error(\n ""Runlevel 1: Distill Template is getting loaded more than once, aborting!""\n );\n } else {\n window.distill.templateIsLoading = true;\n console.debug(""Runlevel 1: Distill Template has started loading."");\n }\n\n /* 2. Add styles if they weren't added during prerendering */\n makeStyleTag(document);\n console.debug(""Runlevel 1: Static Distill styles have been added."");\n console.debug(""Runlevel 1->2."");\n window.distill.runlevel += 1;\n\n /* 3. Register Controller listener functions */\n /* Needs to happen before components to their connected callbacks have a controller to talk to. */\n for (const [functionName, callback] of Object.entries(Controller.listeners)) {\n if (typeof callback === ""function"") {\n document.addEventListener(functionName, callback);\n } else {\n console.error(""Runlevel 2: Controller listeners need to be functions!"");\n }\n }\n console.debug(""Runlevel 2: We can now listen to controller events."");\n console.debug(""Runlevel 2->3."");\n window.distill.runlevel += 1;\n\n /* 4. Register components */\n const components = [\n Abstract, Appendix, Article, Bibliography, Byline, Cite, CitationList, Code,\n Footnote, FootnoteList, FrontMatter$1, HoverBox, Title, DMath, References, TOC, Figure,\n Slider, Interstitial\n ];\n\n const distillComponents = [DistillHeader, DistillAppendix, DistillFooter];\n\n if (window.distill.runlevel < 2) {\n throw new Error(""Insufficient Runlevel for adding custom elements!"");\n }\n const allComponents = components.concat(distillComponents);\n for (const component of allComponents) {\n console.debug(""Runlevel 2: Registering custom element: "" + component.is);\n customElements.define(component.is, component);\n }\n\n console.debug(\n ""Runlevel 3: Distill Template finished registering custom elements.""\n );\n console.debug(""Runlevel 3->4."");\n window.distill.runlevel += 1;\n\n // If template was added after DOMContentLoaded we may have missed that event.\n // Controller will check for that case, so trigger the event explicitly:\n if (domContentLoaded()) {\n Controller.listeners.DOMContentLoaded();\n }\n\n console.debug(""Runlevel 4: Distill Template initialisation complete."");\n window.distill.templateIsLoading = false;\n window.distill.templateHasLoaded = true;\n };\n\n window.distill = { runlevel, initialize, templateIsLoading };\n\n /* 0. Check browser feature support; synchronously polyfill if needed */\n if (Polyfills.browserSupportsAllFeatures()) {\n console.debug(""Runlevel 0: No need for polyfills."");\n console.debug(""Runlevel 0->1."");\n window.distill.runlevel += 1;\n window.distill.initialize();\n } else {\n console.debug(""Runlevel 0: Distill Template is loading polyfills."");\n Polyfills.load(window.distill.initialize);\n }\n\n})));\n//# sourceMappingURL=template.v2.js.map\n",javascript,tab +2811,86429368,"dist/template.v2.js",46614,14,"distill-footer",javascript,selection_command +2812,86431837,"examples/jasmine.html",0,0,"",html,tab +2813,86434107,"examples/jasmine.html",10378,0,"",html,selection_command +2814,86434358,"examples/jasmine.html",10439,0,"",html,selection_command +2815,86434389,"examples/jasmine.html",10462,0,"",html,selection_command +2816,86434421,"examples/jasmine.html",10486,0,"",html,selection_command +2817,86434454,"examples/jasmine.html",10502,0,"",html,selection_command +2818,86434577,"examples/jasmine.html",10503,0,"",html,selection_command +2819,86435203,"examples/jasmine.html",10503,35," ",html,selection_command +2820,86482237,"examples/jasmine.html",10503,0,"",html,selection_command +2821,86482556,"examples/jasmine.html",0,0,"",html,selection_command +2822,86483889,"examples/jasmine.html",1851,0,"",html,selection_keyboard +2823,86484169,"examples/jasmine.html",0,0,"",html,selection_keyboard +2824,86915232,"examples/jasmine.html",1769,36," ""author"":""Franz Srambical"",\n ""equalContrib"": true,",html,content +2825,86915232,"examples/jasmine.html",1547,34," ""author"":""Alfred Nguyen"",\n ""equalContrib"": true,",html,content +2826,86915232,"examples/jasmine.html",1322,34," ""author"":""Mihir Mahajan"",\n ""equalContrib"": true,",html,content +2827,86973514,"src/front-matter.js",0,0,"// Copyright 2018 The Distill Template Authors\n//\n// Licensed under the Apache License, Version 2.0 (the ""License"");\n// you may not use this file except in compliance with the License.\n// You may obtain a copy of the License at\n//\n// http://www.apache.org/licenses/LICENSE-2.0\n//\n// Unless required by applicable law or agreed to in writing, software\n// distributed under the License is distributed on an ""AS IS"" BASIS,\n// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n// See the License for the specific language governing permissions and\n// limitations under the License.\n\nconst days = ['Sunday', 'Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday'];\nconst months = ['Jan.', 'Feb.', 'March', 'April', 'May', 'June', 'July', 'Aug.', 'Sept.', 'Oct.', 'Nov.', 'Dec.'];\nconst zeroPad = n => n < 10 ? '0' + n : n;\n\nconst RFC = function(date) {\n const day = days[date.getDay()].substring(0, 3);\n const paddedDate = zeroPad(date.getDate());\n const month = months[date.getMonth()].substring(0,3);\n const year = date.getFullYear().toString();\n const hours = date.getUTCHours().toString();\n const minutes = date.getUTCMinutes().toString();\n const seconds = date.getUTCSeconds().toString();\n return `${day}, ${paddedDate} ${month} ${year} ${hours}:${minutes}:${seconds} Z`;\n};\n\nconst objectFromMap = function(map) {\n const object = Array.from(map).reduce((object, [key, value]) => (\n Object.assign(object, { [key]: value }) // Be careful! Maps can have non-String keys; object literals can't.\n ), {});\n return object;\n};\n\nconst mapFromObject = function(object) {\n const map = new Map();\n for (var property in object) {\n if (object.hasOwnProperty(property)) {\n map.set(property, object[property]);\n }\n }\n return map;\n};\n\nclass Author {\n\n // constructor(name='', personalURL='', affiliation='', affiliationURL='') {\n // this.name = name; // 'Chris Olah'\n // this.personalURL = personalURL; // 'https://colah.github.io'\n // this.affiliation = affiliation; // 'Google Brain'\n // this.affiliationURL = affiliationURL; // 'https://g.co/brain'\n // }\n\n constructor(object) {\n const rawName = object.author || ''; // e.g., 'Chris Olah' or 'Chris Olah*'\n // Detect equal contribution via explicit flag or star in the name\n const hasStarInName = rawName.indexOf('*') !== -1;\n this.equalContrib = !!(object.equalContrib || hasStarInName);\n // Sanitize author name to avoid leaking markers into citations/BibTeX\n this.name = rawName.replace(/\*/g, '').trim(); // 'Chris Olah'\n this.personalURL = object.authorURL; // 'https://colah.github.io'\n this.affiliation = object.affiliation; // 'Google Brain'\n this.affiliationURL = object.affiliationURL; // 'https://g.co/brain'\n this.affiliations = object.affiliations || []; // new-style affiliations\n }\n\n // 'Chris'\n get firstName() {\n const names = this.name.split(' ');\n return names.slice(0, names.length - 1).join(' ');\n }\n\n // 'Olah'\n get lastName() {\n const names = this.name.split(' ');\n return names[names.length -1];\n }\n}\n\nexport function mergeFromYMLFrontmatter(target, source) {\n target.title = source.title;\n if (source.published) {\n if (source.published instanceof Date) {\n target.publishedDate = source.published;\n } else if (source.published.constructor === String) {\n target.publishedDate = new Date(source.published);\n }\n }\n if (source.publishedDate) {\n if (source.publishedDate instanceof Date) {\n target.publishedDate = source.publishedDate;\n } else if (source.publishedDate.constructor === String) {\n target.publishedDate = new Date(source.publishedDate);\n } else {\n console.error('Don\'t know what to do with published date: ' + source.publishedDate);\n }\n }\n target.description = source.description;\n target.authors = source.authors.map( (authorObject) => new Author(authorObject));\n target.katex = source.katex;\n target.password = source.password;\n if (source.doi) {\n target.doi = source.doi;\n }\n}\n\nexport class FrontMatter {\n constructor() {\n this.title = 'unnamed article'; // 'Attention and Augmented Recurrent Neural Networks'\n this.description = ''; // 'A visual overview of neural attention...'\n this.authors = []; // Array of Author(s)\n\n this.bibliography = new Map();\n this.bibliographyParsed = false;\n // {\n // 'gregor2015draw': {\n // 'title': 'DRAW: A recurrent neural network for image generation',\n // 'author': 'Gregor, Karol and Danihelka, Ivo and Graves, Alex and Rezende, Danilo Jimenez and Wierstra, Daan',\n // 'journal': 'arXiv preprint arXiv:1502.04623',\n // 'year': '2015',\n // 'url': 'https://arxiv.org/pdf/1502.04623.pdf',\n // 'type': 'article'\n // },\n // }\n\n // Citation keys should be listed in the order that they are appear in the document.\n // Each key refers to a key in the bibliography dictionary.\n this.citations = []; // [ 'gregor2015draw', 'mercier2011humans' ]\n this.citationsCollected = false;\n\n //\n // Assigned from posts.csv\n //\n\n // publishedDate: 2016-09-08T07:00:00.000Z,\n // tags: [ 'rnn' ],\n // distillPath: '2016/augmented-rnns',\n // githubPath: 'distillpub/post--augmented-rnns',\n // doiSuffix: 1,\n\n //\n // Assigned from journal\n //\n this.journal = {};\n // journal: {\n // 'title': 'Distill',\n // 'full_title': 'Distill',\n // 'abbrev_title': 'Distill',\n // 'url': 'http://distill.pub',\n // 'doi': '10.23915/distill',\n // 'publisherName': 'Distill Working Group',\n // 'publisherEmail': 'admin@distill.pub',\n // 'issn': '2476-0757',\n // 'editors': [...],\n // 'committee': [...]\n // }\n // volume: 1,\n // issue: 9,\n\n this.katex = {};\n\n //\n // Assigned from publishing process\n //\n\n // githubCompareUpdatesUrl: 'https://github.com/distillpub/post--augmented-rnns/compare/1596e094d8943d2dc0ea445d92071129c6419c59...3bd9209e0c24d020f87cf6152dcecc6017cbc193',\n // updatedDate: 2017-03-21T07:13:16.000Z,\n // doi: '10.23915/distill.00001',\n this.doi = undefined;\n this.publishedDate = undefined;\n }\n\n // Example:\n // title: Demo Title Attention and Augmented Recurrent Neural Networks\n // published: Jan 10, 2017\n // authors:\n // - Chris Olah:\n // - Shan Carter: http://shancarter.com\n // affiliations:\n // - Google Brain:\n // - Google Brain: http://g.co/brain\n\n //\n // Computed Properties\n //\n\n // 'http://distill.pub/2016/augmented-rnns',\n set url(value) {\n this._url = value;\n }\n get url() {\n if (this._url) {\n return this._url;\n } else if (this.distillPath && this.journal.url) {\n return this.journal.url + '/' + this.distillPath;\n } else if (this.journal.url) {\n return this.journal.url;\n }\n }\n\n // 'https://github.com/distillpub/post--augmented-rnns',\n get githubUrl() {\n if (this.githubPath) {\n return 'https://github.com/' + this.githubPath;\n } else {\n return undefined;\n }\n }\n\n // TODO resolve differences in naming of URL/Url/url.\n // 'http://distill.pub/2016/augmented-rnns/thumbnail.jpg',\n set previewURL(value) {\n this._previewURL = value;\n }\n get previewURL() {\n return this._previewURL ? this._previewURL : this.url + '/thumbnail.jpg';\n }\n\n // 'Thu, 08 Sep 2016 00:00:00 -0700',\n get publishedDateRFC() {\n return RFC(this.publishedDate);\n }\n\n // 'Thu, 08 Sep 2016 00:00:00 -0700',\n get updatedDateRFC() {\n return RFC(this.updatedDate);\n }\n\n // 2016,\n get publishedYear() {\n return this.publishedDate.getFullYear();\n }\n\n // 'Sept',\n get publishedMonth() {\n return months[this.publishedDate.getMonth()];\n }\n\n // 8,\n get publishedDay() {\n return this.publishedDate.getDate();\n }\n\n // '09',\n get publishedMonthPadded() {\n return zeroPad(this.publishedDate.getMonth() + 1);\n }\n\n // '08',\n get publishedDayPadded() {\n return zeroPad(this.publishedDate.getDate());\n }\n\n get publishedISODateOnly() {\n return this.publishedDate.toISOString().split('T')[0];\n }\n\n get volume() {\n const volume = this.publishedYear - 2015;\n if (volume < 1) {\n throw new Error('Invalid publish date detected during computing volume');\n }\n return volume;\n }\n\n get issue() {\n return this.publishedDate.getMonth() + 1;\n }\n\n // 'Olah & Carter',\n get concatenatedAuthors() {\n if (this.authors.length > 2) {\n return this.authors[0].lastName + ', et al.';\n } else if (this.authors.length === 2) {\n return this.authors[0].lastName + ' & ' + this.authors[1].lastName;\n } else if (this.authors.length === 1) {\n return this.authors[0].lastName;\n }\n }\n\n // 'Olah, Chris and Carter, Shan',\n get bibtexAuthors() {\n return this.authors.map(author => {\n return author.lastName + ', ' + author.firstName;\n }).join(' and ');\n }\n\n // 'olah2016attention'\n get slug() {\n let slug = '';\n if (this.authors.length) {\n slug += this.authors[0].lastName.toLowerCase();\n slug += this.publishedYear;\n slug += this.title.split(' ')[0].toLowerCase();\n }\n return slug || 'Untitled';\n }\n\n get bibliographyEntries() {\n return new Map(this.citations.map( citationKey => {\n const entry = this.bibliography.get(citationKey);\n return [citationKey, entry];\n }));\n }\n\n set bibliography(bibliography) {\n if (bibliography instanceof Map) {\n this._bibliography = bibliography;\n } else if (typeof bibliography === 'object') {\n this._bibliography = mapFromObject(bibliography);\n }\n }\n\n get bibliography() {\n return this._bibliography;\n }\n\n static fromObject(source) {\n const frontMatter = new FrontMatter();\n Object.assign(frontMatter, source);\n return frontMatter;\n }\n\n assignToObject(target) {\n Object.assign(target, this);\n target.bibliography = objectFromMap(this.bibliographyEntries);\n target.url = this.url;\n target.doi = this.doi;\n target.githubUrl = this.githubUrl;\n target.previewURL = this.previewURL;\n if (this.publishedDate) {\n target.volume = this.volume;\n target.issue = this.issue;\n target.publishedDateRFC = this.publishedDateRFC;\n target.publishedYear = this.publishedYear;\n target.publishedMonth = this.publishedMonth;\n target.publishedDay = this.publishedDay;\n target.publishedMonthPadded = this.publishedMonthPadded;\n target.publishedDayPadded = this.publishedDayPadded;\n }\n if (this.updatedDate) {\n target.updatedDateRFC = this.updatedDateRFC;\n }\n target.concatenatedAuthors = this.concatenatedAuthors;\n target.bibtexAuthors = this.bibtexAuthors;\n target.slug = this.slug;\n }\n\n}\n",javascript,tab +2828,86973735,"src/front-matter.js",2149,0,"",javascript,selection_command +2829,87019275,"src/front-matter.js",2229,0,"",javascript,selection_command +2830,87019420,"src/front-matter.js",2300,0,"",javascript,selection_command +2831,87019569,"src/front-matter.js",2355,0,"",javascript,selection_command +2832,87032160,"src/front-matter.js",2300,0,"",javascript,selection_command +2833,87032310,"src/front-matter.js",2229,0,"",javascript,selection_command +2834,87032466,"src/front-matter.js",2149,0,"",javascript,selection_command +2835,87034282,"src/front-matter.js",2229,0,"",javascript,selection_command +2836,87038705,"src/front-matter.js",2229,71,"",javascript,content +2837,87038736,"src/front-matter.js",2233,0,"",javascript,selection_command +2838,87039244,"src/front-matter.js",2229,55,"",javascript,content +2839,87039271,"src/front-matter.js",2233,0,"",javascript,selection_command +2840,87039771,"src/front-matter.js",2237,0,"",javascript,selection_command +2841,87039971,"src/front-matter.js",2238,0,"",javascript,selection_command +2842,87040139,"src/front-matter.js",2251,0,"",javascript,selection_command +2843,87040305,"src/front-matter.js",2253,0,"",javascript,selection_command +2844,87040457,"src/front-matter.js",2256,0,"",javascript,selection_command +2845,87040626,"src/front-matter.js",2262,0,"",javascript,selection_command +2846,87040784,"src/front-matter.js",2263,0,"",javascript,selection_command +2847,87040960,"src/front-matter.js",2276,0,"",javascript,selection_command +2848,87041143,"src/front-matter.js",2279,0,"",javascript,selection_command +2849,87041560,"src/front-matter.js",2276,0,"",javascript,selection_command +2850,87042065,"src/front-matter.js",2275,0,"",javascript,selection_command +2851,87042243,"src/front-matter.js",2275,1," ",javascript,selection_command +2852,87042329,"src/front-matter.js",2275,3," ||",javascript,selection_command +2853,87042515,"src/front-matter.js",2275,17," || hasStarInName",javascript,selection_command +2854,87042670,"src/front-matter.js",2275,19," || hasStarInName);",javascript,selection_command +2855,87043061,"src/front-matter.js",2275,18," || hasStarInName)",javascript,selection_command +2856,87043226,"src/front-matter.js",2275,17," || hasStarInName",javascript,selection_command +2857,87043566,"src/front-matter.js",2275,17,"",javascript,content +2858,87044681,"src/front-matter.js",2263,0,"",javascript,selection_command +2859,87044804,"src/front-matter.js",2262,0,"",javascript,selection_command +2860,87044937,"src/front-matter.js",2256,0,"",javascript,selection_command +2861,87045172,"src/front-matter.js",2253,0,"",javascript,selection_command +2862,87045305,"src/front-matter.js",2251,0,"",javascript,selection_command +2863,87045492,"src/front-matter.js",2238,0,"",javascript,selection_command +2864,87045608,"src/front-matter.js",2237,0,"",javascript,selection_command +2865,87045906,"src/front-matter.js",2233,0,"",javascript,selection_command +2866,87046207,"src/front-matter.js",2226,0,"",javascript,selection_command +2867,87046773,"src/front-matter.js",2233,0,"",javascript,selection_command +2868,87137429,"src/front-matter.js",2229,48," this.equalContrib = !!(object.equalContrib);",javascript,selection_command +2869,87137926,"src/front-matter.js",2233,0,"",javascript,selection_command diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-9543d8e2-2376-4957-873e-df7016d502961763465687199-2025_11_18-12.34.49.442/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-9543d8e2-2376-4957-873e-df7016d502961763465687199-2025_11_18-12.34.49.442/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..8e4e99c69cb7b07d55213df13ad5128d3ede8c73 --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-9543d8e2-2376-4957-873e-df7016d502961763465687199-2025_11_18-12.34.49.442/source.csv @@ -0,0 +1,226 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +2,67,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"12:34:49 PM [info] Activating crowd-code\n12:34:49 PM [info] Recording started\n12:34:49 PM [info] Initializing git provider using file system watchers...\n12:34:49 PM [info] Git repository found\n12:34:49 PM [info] Git provider initialized successfully\n",Log,tab +3,116,"extension-output-pdoom-org.crowd-code-#1-crowd-code",250,0,"12:34:49 PM [info] Initial git state: [object Object]\n",Log,content +4,2975,"extension-output-pdoom-org.crowd-code-#1-crowd-code",304,0,"",Log,selection_mouse +5,13030,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,tab +6,20469,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_command +7,20661,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",66,0,"",python,selection_command +8,20694,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",132,0,"",python,selection_command +9,20728,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",199,0,"",python,selection_command +10,20761,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",241,0,"",python,selection_command +11,20794,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",243,0,"",python,selection_command +12,20828,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",292,0,"",python,selection_command +13,20863,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",294,0,"",python,selection_command +14,20904,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",364,0,"",python,selection_command +15,20936,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",432,0,"",python,selection_command +16,20962,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",507,0,"",python,selection_command +17,20996,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",577,0,"",python,selection_command +18,21030,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",610,0,"",python,selection_command +19,21068,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",611,0,"",python,selection_command +20,21097,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",639,0,"",python,selection_command +21,21131,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",640,0,"",python,selection_command +22,21166,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",671,0,"",python,selection_command +23,21203,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",694,0,"",python,selection_command +24,21233,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",707,0,"",python,selection_command +25,21265,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",795,0,"",python,selection_command +26,21297,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",796,0,"",python,selection_command +27,21331,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",852,0,"",python,selection_command +28,21364,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",914,0,"",python,selection_command +29,21397,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",965,0,"",python,selection_command +30,21431,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1047,0,"",python,selection_command +31,21464,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1148,0,"",python,selection_command +32,21497,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1245,0,"",python,selection_command +33,21530,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1319,0,"",python,selection_command +34,21565,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1369,0,"",python,selection_command +35,21598,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1370,0,"",python,selection_command +36,21631,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1393,0,"",python,selection_command +37,21668,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1394,0,"",python,selection_command +38,21701,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1395,0,"",python,selection_command +39,21890,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1423,0,"",python,selection_command +40,23050,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1427,0,"",python,selection_command +41,23197,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1432,0,"",python,selection_command +42,23489,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1435,0,"",python,selection_command +43,23702,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1438,0,"",python,selection_command +44,23912,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1441,0,"",python,selection_command +45,24119,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1442,0,"",python,selection_command +46,24327,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1448,0,"",python,selection_command +47,24581,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1449,0,"",python,selection_command +48,24945,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1451,0,"",python,selection_command +49,25331,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1452,0,"",python,selection_command +50,26280,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1423,0,"",python,selection_command +51,26667,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1395,0,"",python,selection_command +52,26920,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1394,0,"",python,selection_command +53,26954,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1393,0,"",python,selection_command +54,146021,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +55,146654,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +56,203332,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2838,0,"",python,selection_keyboard +57,203923,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",5124,0,"",python,selection_keyboard +58,204398,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7126,0,"",python,selection_keyboard +59,204829,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9057,0,"",python,selection_keyboard +60,205408,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_keyboard +61,210105,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7721,0,"",python,selection_keyboard +62,210235,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",5922,0,"",python,selection_keyboard +63,210382,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3490,0,"",python,selection_keyboard +64,210771,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1906,0,"",python,selection_keyboard +65,210944,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_keyboard +66,211099,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_keyboard +67,214839,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10,0,"",python,selection_command +68,2267039,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2294,0,"",python,selection_mouse +69,2272461,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2114,0,"",python,selection_mouse +70,3052561,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +71,3052668,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +72,3145153,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_command +73,3148126,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7487,0,"",python,selection_command +74,3149770,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7876,0,"",python,selection_command +75,3150226,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",8235,0,"",python,selection_command +76,3150898,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",7876,0,"",python,selection_command +77,3151966,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",8235,0,"",python,selection_command +78,3161866,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9070,0,"",python,selection_command +79,3164224,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9131,0,"",python,selection_command +80,3166810,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9194,0,"",python,selection_command +81,3167056,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9257,0,"",python,selection_command +82,3167090,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9321,0,"",python,selection_command +83,3167123,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9386,0,"",python,selection_command +84,3167156,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9443,0,"",python,selection_command +85,3167189,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9481,0,"",python,selection_command +86,3167223,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9531,0,"",python,selection_command +87,3168286,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9532,0,"",python,selection_command +88,3168527,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9538,0,"",python,selection_command +89,3168566,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9539,0,"",python,selection_command +90,3168597,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9546,0,"",python,selection_command +91,3168633,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9547,0,"",python,selection_command +92,3168887,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9555,0,"",python,selection_command +93,3169118,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +94,3169292,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9583,0,"",python,selection_command +95,3169900,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +96,3170394,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +97,3171165,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +98,3171795,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +99,3171945,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +100,3172433,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +101,3173957,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +102,3174398,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +103,3174801,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9682,0,"",python,selection_command +104,3175123,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9746,0,"",python,selection_command +105,3175686,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9682,0,"",python,selection_command +106,3175821,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9619,0,"",python,selection_command +107,3175981,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9556,0,"",python,selection_command +108,3240534,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_command +109,3241210,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1898,0,"",python,selection_keyboard +110,3241373,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3420,0,"",python,selection_keyboard +111,3242843,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3370,0,"",python,selection_command +112,3243091,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3298,0,"",python,selection_command +113,3243124,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3288,0,"",python,selection_command +114,3243158,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3287,0,"",python,selection_command +115,3243191,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3211,0,"",python,selection_command +116,3243225,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3137,0,"",python,selection_command +117,3243258,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3058,0,"",python,selection_command +118,3243291,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3057,0,"",python,selection_command +119,3243324,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2999,0,"",python,selection_command +120,3243357,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2991,0,"",python,selection_command +121,3243391,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2973,0,"",python,selection_command +122,3243424,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2956,0,"",python,selection_command +123,3243458,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2932,0,"",python,selection_command +124,3243491,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2915,0,"",python,selection_command +125,3243525,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2898,0,"",python,selection_command +126,3243558,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2876,0,"",python,selection_command +127,3243591,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2854,0,"",python,selection_command +128,3243632,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2838,0,"",python,selection_command +129,3243658,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2817,0,"",python,selection_command +130,3243691,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2793,0,"",python,selection_command +131,3243734,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2767,0,"",python,selection_command +132,3243758,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2756,0,"",python,selection_command +133,3243791,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2721,0,"",python,selection_command +134,3243824,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2688,0,"",python,selection_command +135,3243857,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2655,0,"",python,selection_command +136,3243891,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2621,0,"",python,selection_command +137,3243925,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2585,0,"",python,selection_command +138,3243958,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2547,0,"",python,selection_command +139,3243991,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2512,0,"",python,selection_command +140,3244025,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2483,0,"",python,selection_command +141,3244059,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2451,0,"",python,selection_command +142,3244092,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2427,0,"",python,selection_command +143,3244125,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,0,"",python,selection_command +144,3286103,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2411,0,"",python,selection_command +145,3296751,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,23,"",python,content +146,3298859,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,0," sequence_parallelis",python,content +147,3298862,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2388,0,"",python,selection_command +148,3483131,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +149,3484741,"TERMINAL",0,0,"",,terminal_focus +150,3484752,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +151,3788287,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2427,0,"",python,selection_command +152,3788457,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2428,0,"",python,selection_command +153,3788946,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2429,0,"",python,selection_command +154,3789583,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2427,0,"",python,selection_command +155,4549882,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +156,4550471,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +157,4559897,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +158,4560021,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +159,4560890,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,content +160,4583906,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18686,0,"",python,selection_mouse +161,4583914,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18685,0,"",python,selection_command +162,4584529,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18699,0,"",python,selection_command +163,4585394,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18743,0,"",python,selection_command +164,4585610,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",18752,0,"",python,selection_command +165,4591082,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +166,4591147,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_command +167,4595172,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,10041,"",python,content +168,4597038,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,tab +169,5454427,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",9372,0,"",python,selection_mouse +170,5454796,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,0,"",python,selection_command +171,5455466,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",64,0,"",python,selection_command +172,5455793,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",1906,0,"",python,selection_keyboard +173,5456269,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3490,0,"",python,selection_keyboard +174,5456884,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3420,0,"",python,selection_command +175,5457139,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3370,0,"",python,selection_command +176,5457172,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3298,0,"",python,selection_command +177,5457202,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3288,0,"",python,selection_command +178,5457232,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3287,0,"",python,selection_command +179,5457265,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3211,0,"",python,selection_command +180,5457298,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3137,0,"",python,selection_command +181,5457331,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3058,0,"",python,selection_command +182,5457367,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",3057,0,"",python,selection_command +183,5457402,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2999,0,"",python,selection_command +184,5457431,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2991,0,"",python,selection_command +185,5457465,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2973,0,"",python,selection_command +186,5457497,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2956,0,"",python,selection_command +187,5457533,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2932,0,"",python,selection_command +188,5457565,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2915,0,"",python,selection_command +189,5457598,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2898,0,"",python,selection_command +190,5457631,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2876,0,"",python,selection_command +191,5457665,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2854,0,"",python,selection_command +192,5457699,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2838,0,"",python,selection_command +193,5457732,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2817,0,"",python,selection_command +194,5457764,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2793,0,"",python,selection_command +195,5457799,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2767,0,"",python,selection_command +196,5457833,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2756,0,"",python,selection_command +197,5457866,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2721,0,"",python,selection_command +198,5457899,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2688,0,"",python,selection_command +199,5457931,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2655,0,"",python,selection_command +200,5457964,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2621,0,"",python,selection_command +201,5457999,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2585,0,"",python,selection_command +202,5458033,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2547,0,"",python,selection_command +203,5458067,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2512,0,"",python,selection_command +204,5458102,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2483,0,"",python,selection_command +205,5458137,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2451,0,"",python,selection_command +206,5459402,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2483,0,"",python,selection_command +207,5459593,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2512,0,"",python,selection_command +208,5459748,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2547,0,"",python,selection_command +209,5460152,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2551,0,"",python,selection_command +210,5461248,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2589,0,"",python,selection_command +211,5461431,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2625,0,"",python,selection_command +212,5461588,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2659,0,"",python,selection_command +213,5461782,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2692,0,"",python,selection_command +214,5462187,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2725,0,"",python,selection_command +215,5462694,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2692,0,"",python,selection_command +216,5462868,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2659,0,"",python,selection_command +217,5463112,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2625,0,"",python,selection_command +218,5463146,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2589,0,"",python,selection_command +219,5463184,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2551,0,"",python,selection_command +220,5463214,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2516,0,"",python,selection_command +221,5463329,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2487,0,"",python,selection_command +222,8520791,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",0,10041,"# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom typing import Optional\n\nimport lightning.pytorch as pl\nimport nemo_run as run\nimport torch\nfrom nemo.collections.common.tokenizers.huggingface.auto_tokenizer import AutoTokenizer\n\nfrom nemo.collections.llm.api import finetune, pretrain\nfrom nemo.collections.llm.gpt.data.mock import MockDataModule\nfrom nemo.collections.llm.peft import PEFT_STR2CLS\nfrom nemo.collections.llm.recipes.finetune_default import default_finetune_recipe\nfrom nemo.collections.llm.recipes.log.default import default_log, default_resume, tensorboard_logger\nfrom nemo.collections.llm.recipes.optim.adam import distributed_fused_adam_with_cosine_annealing\nfrom nemo.collections.llm.recipes.qwen3 import qwen3_model, qwen3_trainer\nfrom nemo.utils.exp_manager import TimingCallback\n\nNAME = ""qwen3_30b_a3b""\n\n\n@run.cli.factory(name=NAME)\ndef model() -> run.Config[pl.LightningModule]:\n """"""\n Factory function to create a Qwen3 30B-A3B model configuration.\n This is a MoE (Mixture of Experts) model with 128 experts.\n\n Returns:\n run.Config[pl.LightningModule]: Configuration for the Qwen3 30B-A3B model.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain model=qwen3_30b_a3b ...\n\n Python API usage:\n >>> model_config = model()\n >>> print(model_config)\n """"""\n return qwen3_model(version=NAME)\n\n\n@run.cli.factory(target=pretrain, name=NAME)\ndef pretrain_recipe(\n # General\n dir: Optional[str] = None,\n name: str = ""default"",\n # Trainer\n tensor_parallelism: int = 4, # Default for 30B-A3B model\n pipeline_parallelism: int = 2,\n pipeline_parallelism_type: Optional[torch.dtype] = None,\n virtual_pipeline_parallelism: Optional[int] = None,\n context_parallelism: int = 1,\n expert_parallelism: Optional[int] = 4,\n sequence_parallelism: bool = True,\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n max_steps: int = 300000,\n precision: str = ""bf16-mixed"",\n accumulate_grad_batches: int = 1,\n gradient_clip_val: float = 1.0,\n limit_test_batches: int = 32,\n limit_val_batches: int = 32,\n log_every_n_steps: int = 10,\n val_check_interval: int = 500,\n # Data\n global_batch_size=32,\n micro_batch_size=2,\n seq_length=4096,\n # Optimizer\n warmup_steps=500,\n constant_steps=0,\n min_lr=3e-5,\n max_lr=3e-4,\n # Training function\n fn=pretrain,\n) -> run.Partial:\n """"""\n Create a pre-training recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for pre-training, including\n model, trainer, data, logging, optimization, and resumption settings.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the pre-training run.\n tensor_parallelism (int): Degree of tensor model parallelism.\n pipeline_parallelism (int): Degree of pipeline model parallelism.\n pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.\n virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.\n context_parallelism (int): Degree of context parallelism.\n sequence_parallelism (bool): Whether to use sequence parallelism.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n max_steps (int): Maximum number of training steps.\n precision (str): Precision configuration, one of fp32, 16-mixed or bf16-mixed.\n accumulate_grad_batches (int): Number of steps per gradient accumulation.\n gradient_clip_val (float): Value for gradient clipping.\n limit_test_batches (int): Limit the number of test batches.\n limit_val_batches (int): Limit the number of validation batches.\n log_every_n_steps (int): Log every n steps.\n val_check_interval (int): Run validation every N steps.\n global_batch_size (int): Global batch size.\n micro_batch_size (int): Micro batch size.\n seq_length (int): Sequence length.\n warmup_steps (int): Number of warmup steps.\n constant_steps (int): Number of constant steps.\n min_lr (float): Minimum learning rate.\n max_lr (float): Maximum learning rate.\n fn (Callable): The pre-training function to use.\n\n Returns:\n run.Partial: Partial configuration for pre-training.\n\n Examples:\n CLI usage:\n $ nemo llm pretrain --factory qwen3_30b_a3b\n $ nemo llm pretrain --factory ""qwen3_30b_a3b(num_nodes=1, name='my_qwen3_pretrain')""\n\n Python API usage:\n >>> recipe = pretrain_recipe(name=""qwen3_pretrain"", num_nodes=1)\n >>> print(recipe)\n\n Note:\n This recipe uses a mock dataset, look for the finetune examples to see how to change the dataset.\n """"""\n recipe = run.Partial(\n fn,\n model=model(),\n trainer=qwen3_trainer(\n tensor_parallelism=tensor_parallelism,\n pipeline_parallelism=pipeline_parallelism,\n pipeline_parallelism_type=pipeline_parallelism_type,\n virtual_pipeline_parallelism=virtual_pipeline_parallelism,\n context_parallelism=context_parallelism,\n sequence_parallelism=sequence_parallelism,\n expert_parallelism=expert_parallelism,\n num_nodes=num_nodes,\n num_gpus_per_node=num_gpus_per_node,\n max_steps=max_steps,\n precision=precision,\n accumulate_grad_batches=accumulate_grad_batches,\n limit_test_batches=limit_test_batches,\n limit_val_batches=limit_val_batches,\n log_every_n_steps=log_every_n_steps,\n val_check_interval=val_check_interval,\n callbacks=[run.Config(TimingCallback)],\n ),\n data=run.Config(\n MockDataModule,\n seq_length=seq_length,\n global_batch_size=global_batch_size,\n micro_batch_size=micro_batch_size,\n tokenizer=run.Config(AutoTokenizer, ""Qwen/Qwen3-30B-A3B""),\n ),\n log=default_log(dir=dir, name=name, tensorboard_logger=tensorboard_logger(name=name)),\n optim=distributed_fused_adam_with_cosine_annealing(\n precision=precision,\n warmup_steps=warmup_steps,\n constant_steps=constant_steps,\n min_lr=min_lr,\n max_lr=max_lr,\n clip_grad=gradient_clip_val,\n ),\n resume=default_resume(),\n )\n recipe.model.config.recompute_granularity = ""full""\n recipe.model.config.recompute_method = ""uniform""\n recipe.model.config.recompute_num_layers = 1\n return recipe\n\n\n@run.cli.factory(target=finetune, name=NAME)\ndef finetune_recipe(\n dir: Optional[str] = None,\n name: str = ""default"",\n num_nodes: int = 1,\n num_gpus_per_node: int = 8,\n peft_scheme: Optional[str] = 'lora',\n packed_sequence: bool = False,\n) -> run.Partial:\n """"""\n Create a fine-tuning recipe for Qwen3 30B-A3B model.\n\n This function sets up a complete configuration for fine-tuning, including\n model, trainer, data, logging, optimization, and resumption settings.\n The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None.\n This model uses Mixture of Experts (MoE) architecture with 128 experts.\n\n Args:\n dir (Optional[str]): Directory for saving logs and checkpoints.\n name (str): Name of the fine-tuning run.\n num_nodes (int): Number of compute nodes to use.\n num_gpus_per_node (int): Number of GPUs per node.\n peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning.\n Allowed values: 'lora'/'dora'/'none'/None.\n packed_sequence (Optional[bool]): Packing multiple training sequences into one long sequence for training\n efficiency. Default sequence length is 2048.\n\n Returns:\n run.Partial: Partial configuration for fine-tuning.\n\n Examples:\n CLI usage:\n $ nemo llm finetune --factory qwen3_30b_a3b\n\n Python API usage:\n >>> recipe = finetune_recipe(name=""qwen3_30b_a3b_finetune"", num_nodes=2)\n >>> print(recipe)\n\n Note:\n This recipe uses the SQuAD dataset for fine-tuning.\n """"""\n recipe = default_finetune_recipe(\n model(), ""Qwen/Qwen3-30B-A3B"", dir, name, num_nodes, num_gpus_per_node, packed_sequence\n )\n if peft_scheme is None or peft_scheme.lower() == 'none':\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.pipeline_model_parallel_size = 2\n recipe.trainer.strategy.sequence_parallel = True\n recipe.optim.config.lr = 5e-6\n elif peft_scheme.lower() in ['lora', 'dora']:\n recipe.trainer.strategy.tensor_model_parallel_size = 4\n recipe.trainer.strategy.expert_model_parallel_size = 4\n recipe.trainer.strategy.expert_tensor_parallel_size = 1\n recipe.trainer.strategy.sequence_parallel = True\n recipe.peft = run.Config(PEFT_STR2CLS[peft_scheme.lower()])\n recipe.peft.target_modules = ['linear_qkv', 'linear_proj']\n recipe.optim.config.lr = 1e-4\n else:\n raise ValueError(f""Unrecognized peft scheme: {peft_scheme}"")\n return recipe\n",python,selection_command +223,8520860,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",10041,0,"",python,selection_command +224,8685736,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2797,0,"",python,selection_command +225,8686757,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2813,0,"",python,selection_command +226,8686946,"nemo/collections/llm/recipes/qwen3_30b_a3b.py",2814,0,"",python,selection_command diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-a7b808c2-b1d0-43a0-a38c-8b82cd2886711764488770794-2025_11_30-08.46.18.897/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-a7b808c2-b1d0-43a0-a38c-8b82cd2886711764488770794-2025_11_30-08.46.18.897/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..676fb3dc5a7ae81ddd54419a7b947606d964c631 --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-a7b808c2-b1d0-43a0-a38c-8b82cd2886711764488770794-2025_11_30-08.46.18.897/source.csv @@ -0,0 +1,32 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,13,"Untitled-1",0,0,"",plaintext,tab +2,77,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"8:46:18 AM [info] Activating crowd-code\n8:46:18 AM [info] Recording started\n8:46:18 AM [info] Initializing git provider using file system watchers...\n8:46:18 AM [info] No workspace folder found\n",Log,tab +3,2028,"extension-output-pdoom-org.crowd-code-#1-crowd-code",194,0,"8:46:20 AM [info] Retrying git provider initialization...\n8:46:20 AM [info] No workspace folder found\n",Log,content +4,2227,"Untitled-1",0,0,"",plaintext,tab +5,3588,"TERMINAL",0,0,"Test",,terminal_focus +6,3593,"Untitled-1",0,0,"/* crowd-pilot: insert start */\nline A\nline B\n/* crowd-pilot: insert end */\n",plaintext,content +7,4572,"Untitled-1",76,0,"/* crowd-pilot: replacement */\nREPLACED LINE 1\nREPLACED LINE 2",plaintext,content +8,6374,"Untitled-1",122,0,"",plaintext,selection_command +9,6522,"Untitled-1",91,0,"",plaintext,selection_command +10,6765,"Untitled-1",122,0,"",plaintext,selection_command +11,6932,"Untitled-1",138,0,"",plaintext,selection_command +12,7497,"Untitled-1",123,15,"",plaintext,content +13,8248,"TERMINAL",0,0,"echo ""Hello World""",,terminal_command +14,8249,"TERMINAL",0,0,"]633;CHello World\r\n% \r \r",,terminal_output +15,9755,"Untitled-1",107,0,"",plaintext,selection_command +16,9857,"Untitled-1",76,0,"",plaintext,selection_command +17,10024,"Untitled-1",46,0,"",plaintext,selection_command +18,10659,"Untitled-1",39,0,"",plaintext,selection_command +19,10826,"Untitled-1",32,0,"",plaintext,selection_command +20,10955,"Untitled-1",0,0,"",plaintext,selection_command +21,14633,"Untitled-1",0,31,"/* crowd-pilot: insert start */",plaintext,selection_command +22,15287,"Untitled-1",0,38,"/* crowd-pilot: insert start */\nline A",plaintext,selection_command +23,15426,"Untitled-1",0,45,"/* crowd-pilot: insert start */\nline A\nline B",plaintext,selection_command +24,15551,"Untitled-1",0,75,"/* crowd-pilot: insert start */\nline A\nline B\n/* crowd-pilot: insert end */",plaintext,selection_command +25,15693,"Untitled-1",0,106,"/* crowd-pilot: insert start */\nline A\nline B\n/* crowd-pilot: insert end */\n/* crowd-pilot: replacement */",plaintext,selection_command +26,15820,"Untitled-1",0,122,"/* crowd-pilot: insert start */\nline A\nline B\n/* crowd-pilot: insert end */\n/* crowd-pilot: replacement */\nREPLACED LINE 1",plaintext,selection_command +27,15909,"Untitled-1",0,122,"",plaintext,content +28,16827,"Untitled-1",1,0,"/* crowd-pilot: insert start */\nline A\nline B\n/* crowd-pilot: insert end */\n",plaintext,content +29,19025,"Untitled-1",0,1,"\n",plaintext,selection_command +30,19300,"Untitled-1",0,77,"\n/* crowd-pilot: insert start */\nline A\nline B\n/* crowd-pilot: insert end */\n",plaintext,selection_command +31,19520,"Untitled-1",0,77,"",plaintext,content diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-aeed47b9-f6ef-4272-b0ca-0c15ab4c25021758266694991-2025_09_19-09.25.04.660/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-aeed47b9-f6ef-4272-b0ca-0c15ab4c25021758266694991-2025_09_19-09.25.04.660/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..8d3ca4a299bb019cba34c41e13c7d67a1ec8e7de --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-aeed47b9-f6ef-4272-b0ca-0c15ab4c25021758266694991-2025_09_19-09.25.04.660/source.csv @@ -0,0 +1,151 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,3,"cleanrl/ppo_atari_envpool.py",0,0,"# docs and experiment results can be found at https://docs.cleanrl.dev/rl-algorithms/ppo/#ppo_atari_envpoolpy\nimport os\nimport random\nimport time\nfrom collections import deque\nfrom dataclasses import dataclass\n\nimport envpool\nimport gym\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nimport tyro\nfrom torch.distributions.categorical import Categorical\nfrom torch.utils.tensorboard import SummaryWriter\n\n\n@dataclass\nclass Args:\n exp_name: str = os.path.basename(__file__)[: -len("".py"")]\n """"""the name of this experiment""""""\n seed: int = 1\n """"""seed of the experiment""""""\n torch_deterministic: bool = True\n """"""if toggled, `torch.backends.cudnn.deterministic=False`""""""\n cuda: bool = True\n """"""if toggled, cuda will be enabled by default""""""\n track: bool = False\n """"""if toggled, this experiment will be tracked with Weights and Biases""""""\n wandb_project_name: str = ""cleanRL""\n """"""the wandb's project name""""""\n wandb_entity: str = None\n """"""the entity (team) of wandb's project""""""\n capture_video: bool = False\n """"""whether to capture videos of the agent performances (check out `videos` folder)""""""\n\n # Algorithm specific arguments\n env_id: str = ""Breakout-v5""\n """"""the id of the environment""""""\n total_timesteps: int = 10000000\n """"""total timesteps of the experiments""""""\n learning_rate: float = 2.5e-4\n """"""the learning rate of the optimizer""""""\n num_envs: int = 8\n """"""the number of parallel game environments""""""\n num_steps: int = 128\n """"""the number of steps to run in each environment per policy rollout""""""\n anneal_lr: bool = True\n """"""Toggle learning rate annealing for policy and value networks""""""\n gamma: float = 0.99\n """"""the discount factor gamma""""""\n gae_lambda: float = 0.95\n """"""the lambda for the general advantage estimation""""""\n num_minibatches: int = 4\n """"""the number of mini-batches""""""\n update_epochs: int = 4\n """"""the K epochs to update the policy""""""\n norm_adv: bool = True\n """"""Toggles advantages normalization""""""\n clip_coef: float = 0.1\n """"""the surrogate clipping coefficient""""""\n clip_vloss: bool = True\n """"""Toggles whether or not to use a clipped loss for the value function, as per the paper.""""""\n ent_coef: float = 0.01\n """"""coefficient of the entropy""""""\n vf_coef: float = 0.5\n """"""coefficient of the value function""""""\n max_grad_norm: float = 0.5\n """"""the maximum norm for the gradient clipping""""""\n target_kl: float = None\n """"""the target KL divergence threshold""""""\n\n # to be filled in runtime\n batch_size: int = 0\n """"""the batch size (computed in runtime)""""""\n minibatch_size: int = 0\n """"""the mini-batch size (computed in runtime)""""""\n num_iterations: int = 0\n """"""the number of iterations (computed in runtime)""""""\n\n\nclass RecordEpisodeStatistics(gym.Wrapper):\n def __init__(self, env, deque_size=100):\n super().__init__(env)\n self.num_envs = getattr(env, ""num_envs"", 1)\n self.episode_returns = None\n self.episode_lengths = None\n\n def reset(self, **kwargs):\n observations = super().reset(**kwargs)\n self.episode_returns = np.zeros(self.num_envs, dtype=np.float32)\n self.episode_lengths = np.zeros(self.num_envs, dtype=np.int32)\n self.lives = np.zeros(self.num_envs, dtype=np.int32)\n self.returned_episode_returns = np.zeros(self.num_envs, dtype=np.float32)\n self.returned_episode_lengths = np.zeros(self.num_envs, dtype=np.int32)\n return observations\n\n def step(self, action):\n observations, rewards, dones, infos = super().step(action)\n self.episode_returns += infos[""reward""]\n self.episode_lengths += 1\n self.returned_episode_returns[:] = self.episode_returns\n self.returned_episode_lengths[:] = self.episode_lengths\n self.episode_returns *= 1 - infos[""terminated""]\n self.episode_lengths *= 1 - infos[""terminated""]\n infos[""r""] = self.returned_episode_returns\n infos[""l""] = self.returned_episode_lengths\n return (\n observations,\n rewards,\n dones,\n infos,\n )\n\n\ndef layer_init(layer, std=np.sqrt(2), bias_const=0.0):\n torch.nn.init.orthogonal_(layer.weight, std)\n torch.nn.init.constant_(layer.bias, bias_const)\n return layer\n\n\nclass Agent(nn.Module):\n def __init__(self, envs):\n super().__init__()\n self.network = nn.Sequential(\n layer_init(nn.Conv2d(4, 32, 8, stride=4)),\n nn.ReLU(),\n layer_init(nn.Conv2d(32, 64, 4, stride=2)),\n nn.ReLU(),\n layer_init(nn.Conv2d(64, 64, 3, stride=1)),\n nn.ReLU(),\n nn.Flatten(),\n layer_init(nn.Linear(64 * 7 * 7, 512)),\n nn.ReLU(),\n )\n self.actor = layer_init(nn.Linear(512, envs.single_action_space.n), std=0.01)\n self.critic = layer_init(nn.Linear(512, 1), std=1)\n\n def get_value(self, x):\n return self.critic(self.network(x / 255.0))\n\n def get_action_and_value(self, x, action=None):\n hidden = self.network(x / 255.0)\n logits = self.actor(hidden)\n probs = Categorical(logits=logits)\n if action is None:\n action = probs.sample()\n return action, probs.log_prob(action), probs.entropy(), self.critic(hidden)\n\n\nif __name__ == ""__main__"":\n args = tyro.cli(Args)\n args.batch_size = int(args.num_envs * args.num_steps)\n args.minibatch_size = int(args.batch_size // args.num_minibatches)\n args.num_iterations = args.total_timesteps // args.batch_size\n run_name = f""{args.env_id}__{args.exp_name}__{args.seed}__{int(time.time())}""\n if args.track:\n import wandb\n\n wandb.init(\n project=args.wandb_project_name,\n entity=args.wandb_entity,\n sync_tensorboard=True,\n config=vars(args),\n name=run_name,\n monitor_gym=True,\n save_code=True,\n )\n writer = SummaryWriter(f""runs/{run_name}"")\n writer.add_text(\n ""hyperparameters"",\n ""|param|value|\n|-|-|\n%s"" % (""\n"".join([f""|{key}|{value}|"" for key, value in vars(args).items()])),\n )\n\n # TRY NOT TO MODIFY: seeding\n random.seed(args.seed)\n np.random.seed(args.seed)\n torch.manual_seed(args.seed)\n torch.backends.cudnn.deterministic = args.torch_deterministic\n\n device = torch.device(""cuda"" if torch.cuda.is_available() and args.cuda else ""cpu"")\n\n # env setup\n envs = envpool.make(\n args.env_id,\n env_type=""gym"",\n num_envs=args.num_envs,\n episodic_life=True,\n reward_clip=True,\n seed=args.seed,\n )\n envs.num_envs = args.num_envs\n envs.single_action_space = envs.action_space\n envs.single_observation_space = envs.observation_space\n envs = RecordEpisodeStatistics(envs)\n assert isinstance(envs.action_space, gym.spaces.Discrete), ""only discrete action space is supported""\n\n agent = Agent(envs).to(device)\n optimizer = optim.Adam(agent.parameters(), lr=args.learning_rate, eps=1e-5)\n\n # ALGO Logic: Storage setup\n obs = torch.zeros((args.num_steps, args.num_envs) + envs.single_observation_space.shape).to(device)\n actions = torch.zeros((args.num_steps, args.num_envs) + envs.single_action_space.shape).to(device)\n logprobs = torch.zeros((args.num_steps, args.num_envs)).to(device)\n rewards = torch.zeros((args.num_steps, args.num_envs)).to(device)\n dones = torch.zeros((args.num_steps, args.num_envs)).to(device)\n values = torch.zeros((args.num_steps, args.num_envs)).to(device)\n avg_returns = deque(maxlen=20)\n\n # TRY NOT TO MODIFY: start the game\n global_step = 0\n start_time = time.time()\n next_obs = torch.Tensor(envs.reset()).to(device)\n next_done = torch.zeros(args.num_envs).to(device)\n\n for iteration in range(1, args.num_iterations + 1):\n # Annealing the rate if instructed to do so.\n if args.anneal_lr:\n frac = 1.0 - (iteration - 1.0) / args.num_iterations\n lrnow = frac * args.learning_rate\n optimizer.param_groups[0][""lr""] = lrnow\n\n for step in range(0, args.num_steps):\n global_step += args.num_envs\n obs[step] = next_obs\n dones[step] = next_done\n\n # ALGO LOGIC: action logic\n with torch.no_grad():\n action, logprob, _, value = agent.get_action_and_value(next_obs)\n values[step] = value.flatten()\n actions[step] = action\n logprobs[step] = logprob\n\n # TRY NOT TO MODIFY: execute the game and log data.\n next_obs, reward, next_done, info = envs.step(action.cpu().numpy())\n rewards[step] = torch.tensor(reward).to(device).view(-1)\n next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(next_done).to(device)\n\n for idx, d in enumerate(next_done):\n if d and info[""lives""][idx] == 0:\n print(f""global_step={global_step}, episodic_return={info['r'][idx]}"")\n avg_returns.append(info[""r""][idx])\n writer.add_scalar(""charts/avg_episodic_return"", np.average(avg_returns), global_step)\n writer.add_scalar(""charts/episodic_return"", info[""r""][idx], global_step)\n writer.add_scalar(""charts/episodic_length"", info[""l""][idx], global_step)\n\n # bootstrap value if not done\n with torch.no_grad():\n next_value = agent.get_value(next_obs).reshape(1, -1)\n advantages = torch.zeros_like(rewards).to(device)\n lastgaelam = 0\n for t in reversed(range(args.num_steps)):\n if t == args.num_steps - 1:\n nextnonterminal = 1.0 - next_done\n nextvalues = next_value\n else:\n nextnonterminal = 1.0 - dones[t + 1]\n nextvalues = values[t + 1]\n delta = rewards[t] + args.gamma * nextvalues * nextnonterminal - values[t]\n advantages[t] = lastgaelam = delta + args.gamma * args.gae_lambda * nextnonterminal * lastgaelam\n returns = advantages + values\n\n # flatten the batch\n b_obs = obs.reshape((-1,) + envs.single_observation_space.shape)\n b_logprobs = logprobs.reshape(-1)\n b_actions = actions.reshape((-1,) + envs.single_action_space.shape)\n b_advantages = advantages.reshape(-1)\n b_returns = returns.reshape(-1)\n b_values = values.reshape(-1)\n\n # Optimizing the policy and value network\n b_inds = np.arange(args.batch_size)\n clipfracs = []\n for epoch in range(args.update_epochs):\n np.random.shuffle(b_inds)\n for start in range(0, args.batch_size, args.minibatch_size):\n end = start + args.minibatch_size\n mb_inds = b_inds[start:end]\n\n _, newlogprob, entropy, newvalue = agent.get_action_and_value(b_obs[mb_inds], b_actions.long()[mb_inds])\n logratio = newlogprob - b_logprobs[mb_inds]\n ratio = logratio.exp()\n\n with torch.no_grad():\n # calculate approx_kl http://joschu.net/blog/kl-approx.html\n old_approx_kl = (-logratio).mean()\n approx_kl = ((ratio - 1) - logratio).mean()\n clipfracs += [((ratio - 1.0).abs() > args.clip_coef).float().mean().item()]\n\n mb_advantages = b_advantages[mb_inds]\n if args.norm_adv:\n mb_advantages = (mb_advantages - mb_advantages.mean()) / (mb_advantages.std() + 1e-8)\n\n # Policy loss\n pg_loss1 = -mb_advantages * ratio\n pg_loss2 = -mb_advantages * torch.clamp(ratio, 1 - args.clip_coef, 1 + args.clip_coef)\n pg_loss = torch.max(pg_loss1, pg_loss2).mean()\n\n # Value loss\n newvalue = newvalue.view(-1)\n if args.clip_vloss:\n v_loss_unclipped = (newvalue - b_returns[mb_inds]) ** 2\n v_clipped = b_values[mb_inds] + torch.clamp(\n newvalue - b_values[mb_inds],\n -args.clip_coef,\n args.clip_coef,\n )\n v_loss_clipped = (v_clipped - b_returns[mb_inds]) ** 2\n v_loss_max = torch.max(v_loss_unclipped, v_loss_clipped)\n v_loss = 0.5 * v_loss_max.mean()\n else:\n v_loss = 0.5 * ((newvalue - b_returns[mb_inds]) ** 2).mean()\n\n entropy_loss = entropy.mean()\n loss = pg_loss - args.ent_coef * entropy_loss + v_loss * args.vf_coef\n\n optimizer.zero_grad()\n loss.backward()\n nn.utils.clip_grad_norm_(agent.parameters(), args.max_grad_norm)\n optimizer.step()\n\n if args.target_kl is not None and approx_kl > args.target_kl:\n break\n\n y_pred, y_true = b_values.cpu().numpy(), b_returns.cpu().numpy()\n var_y = np.var(y_true)\n explained_var = np.nan if var_y == 0 else 1 - np.var(y_true - y_pred) / var_y\n\n # TRY NOT TO MODIFY: record rewards for plotting purposes\n writer.add_scalar(""charts/learning_rate"", optimizer.param_groups[0][""lr""], global_step)\n writer.add_scalar(""losses/value_loss"", v_loss.item(), global_step)\n writer.add_scalar(""losses/policy_loss"", pg_loss.item(), global_step)\n writer.add_scalar(""losses/entropy"", entropy_loss.item(), global_step)\n writer.add_scalar(""losses/old_approx_kl"", old_approx_kl.item(), global_step)\n writer.add_scalar(""losses/approx_kl"", approx_kl.item(), global_step)\n writer.add_scalar(""losses/clipfrac"", np.mean(clipfracs), global_step)\n writer.add_scalar(""losses/explained_variance"", explained_var, global_step)\n print(""SPS:"", int(global_step / (time.time() - start_time)))\n writer.add_scalar(""charts/SPS"", int(global_step / (time.time() - start_time)), global_step)\n\n envs.close()\n writer.close()\n",python,tab +2,142,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"9:25:04 AM [info] Activating crowd-code\n9:25:04 AM [info] Recording started\n9:25:04 AM [info] Initializing git provider using file system watchers...\n",Log,tab +3,155,"extension-output-pdoom-org.crowd-code-#1-crowd-code",40,0,"",Log,selection_command +4,197,"extension-output-pdoom-org.crowd-code-#1-crowd-code",150,0,"9:25:04 AM [info] Git repository found\n9:25:04 AM [info] Git provider initialized successfully\n9:25:04 AM [info] Initial git state: [object Object]\n",Log,content +5,2189,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +6,2191,"TERMINAL",0,0,"",,terminal_focus +7,2213,"TERMINAL",0,0,"",,terminal_command +8,9325,"TERMINAL",0,0,"",,terminal_command +9,18113,"TERMINAL",0,0,"source .venv/bin/activate",,terminal_command +10,24112,"TERMINAL",0,0,"python cleanrl/ppo_atari_envpool.py",,terminal_command +11,24163,"TERMINAL",0,0,"]633;C",,terminal_output +12,26014,"TERMINAL",0,0,"/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/treevalue/tree/integration/torch.py:23: FutureWarning: `torch.utils._pytree._register_pytree_node` is deprecated. Please use `torch.utils._pytree.register_pytree_node` instead.\r\n register_for_torch(TreeValue)\r\n/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/treevalue/tree/integration/torch.py:24: FutureWarning: `torch.utils._pytree._register_pytree_node` is deprecated. Please use `torch.utils._pytree.register_pytree_node` instead.\r\n register_for_torch(FastTreeValue)\r\n",,terminal_output +13,26264,"TERMINAL",0,0,"Gym has been unmaintained since 2022 and does not support NumPy 2.0 amongst other critical functionality.\r\nPlease upgrade to Gymnasium, the maintained drop-in replacement of Gym, or contact the authors of your software and request that they upgrade.\r\nSee the migration guide at https://gymnasium.farama.org/introduction/migration_guide/ for additional information.\r\n",,terminal_output +14,26759,"TERMINAL",0,0,"/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/tyro/_parsers.py:379: UserWarning: The field `wandb-entity` is annotated with type ``, but the default value `None` has type ``. We'll try to handle this gracefully, but it may cause unexpected behavior.\r\n warnings.warn(message)\r\n/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/tyro/_parsers.py:379: UserWarning: The field `target-kl` is annotated with type ``, but the default value `None` has type ``. We'll try to handle this gracefully, but it may cause unexpected behavior.\r\n warnings.warn(message)\r\n",,terminal_output +15,28466,"TERMINAL",0,0,"global_step=992, episodic_return=0.0\r\nglobal_step=992, episodic_return=0.0\r\nglobal_step=992, episodic_return=0.0\r\nglobal_step=992, episodic_return=0.0\r\nglobal_step=992, episodic_return=0.0\r\nglobal_step=992, episodic_return=0.0\r\nglobal_step=992, episodic_return=0.0\r\nglobal_step=992, episodic_return=0.0\r\n",,terminal_output +16,28939,"TERMINAL",0,0,"SPS: 1339\r\n",,terminal_output +17,29746,"TERMINAL",0,0,"SPS: 1302\r\n",,terminal_output +18,29996,"TERMINAL",0,0,"global_step=2776, episodic_return=2.0\r\nglobal_step=2776, episodic_return=2.0\r\nglobal_step=2776, episodic_return=2.0\r\nglobal_step=2776, episodic_return=2.0\r\nglobal_step=2776, episodic_return=2.0\r\nglobal_step=2776, episodic_return=2.0\r\nglobal_step=2776, episodic_return=2.0\r\nglobal_step=2776, episodic_return=2.0\r\n",,terminal_output +19,30574,"TERMINAL",0,0,"SPS: 1280\r\n",,terminal_output +20,30805,"TERMINAL",0,0,"global_step=3776, episodic_return=0.0\r\nglobal_step=3776, episodic_return=0.0\r\nglobal_step=3776, episodic_return=0.0\r\nglobal_step=3776, episodic_return=0.0\r\nglobal_step=3776, episodic_return=0.0\r\nglobal_step=3776, episodic_return=0.0\r\nglobal_step=3776, episodic_return=0.0\r\nglobal_step=3776, episodic_return=0.0\r\n",,terminal_output +21,31406,"TERMINAL",0,0,"SPS: 1266\r\n",,terminal_output +22,31746,"TERMINAL",0,0,"global_step=5104, episodic_return=1.0\r\nglobal_step=5104, episodic_return=1.0\r\nglobal_step=5104, episodic_return=1.0\r\nglobal_step=5104, episodic_return=1.0\r\nglobal_step=5104, episodic_return=1.0\r\nglobal_step=5104, episodic_return=1.0\r\nglobal_step=5104, episodic_return=1.0\r\nglobal_step=5104, episodic_return=1.0\r\n",,terminal_output +23,32262,"TERMINAL",0,0,"SPS: 1252\r\n",,terminal_output +24,33098,"TERMINAL",0,0,"SPS: 1247\r\n",,terminal_output +25,33217,"TERMINAL",0,0,"global_step=6496, episodic_return=1.0\r\nglobal_step=6496, episodic_return=1.0\r\nglobal_step=6496, episodic_return=1.0\r\nglobal_step=6496, episodic_return=1.0\r\nglobal_step=6496, episodic_return=1.0\r\nglobal_step=6496, episodic_return=1.0\r\nglobal_step=6496, episodic_return=1.0\r\nglobal_step=6496, episodic_return=1.0\r\n",,terminal_output +26,33964,"TERMINAL",0,0,"SPS: 1237\r\n",,terminal_output +27,34109,"TERMINAL",0,0,"^CTraceback (most recent call last):\r\n File ""/fast/home/franz.srambical/cleanrl/cleanrl/ppo_atari_envpool.py"", line 237, in \r\n next_obs, reward, next_done, info = envs.step(action.cpu().numpy())\r\n File ""/fast/home/franz.srambical/cleanrl/cleanrl/ppo_atari_envpool.py"", line 100, in step\r\n observations, rewards, dones, infos = super().step(action)\r\n File ""/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/gym/core.py"", line 280, in step\r\n return self.env.step(action)\r\n File ""/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/envpool/python/envpool.py"", line 144, in step\r\n return self.recv(reset=False, return_info=True)\r\n File ""/fast/home/franz.srambical/cleanrl/.venv/lib/python3.10/site-packages/envpool/python/envpool.py"", line 130, in recv\r\n state_list = self._recv()\r\nKeyboardInterrupt\r\n",,terminal_output +28,34254,"TERMINAL",0,0,"^C",,terminal_output +29,34258,"TERMINAL",0,0,"\r\n]0;franz.srambical@hai-login2:~/cleanrl",,terminal_output +30,116742,"TERMINAL",0,0,"",,terminal_command +31,278446,"cleanrl/ppo_atari_envpool.py",236,0,"",python,selection_mouse +32,278448,"cleanrl/ppo_atari_envpool.py",235,0,"",python,selection_command +33,290061,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"""""""\nGenerates a dataset of random-action CoinRun episodes.\nEpisodes are saved individually as memory-mapped files for efficient loading.\n""""""\n\nfrom dataclasses import dataclass\n\nfrom gym3 import types_np\nimport numpy as np\nfrom procgen import ProcgenGym3Env\nimport tyro\nimport json\nimport os\nfrom utils import save_chunks\n\n\n@dataclass\nclass Args:\n num_episodes_train: int = 10000\n num_episodes_val: int = 500\n num_episodes_test: int = 500\n output_dir: str = ""data/coinrun_episodes""\n min_episode_length: int = 1000\n max_episode_length: int = 1000\n chunk_size: int = 100\n chunks_per_file: int = 100\n seed: int = 0\n\n\nargs = tyro.cli(Args)\nassert (\n args.max_episode_length >= args.min_episode_length\n), ""Maximum episode length must be greater than or equal to minimum episode length.""\n\nif args.min_episode_length < args.chunk_size:\n print(\n ""Warning: Minimum episode length is smaller than chunk size. Note that episodes shorter than the chunk size will be discarded.""\n )\n\n\n# --- Generate episodes ---\ndef generate_episodes(num_episodes, split):\n episode_idx = 0\n episode_metadata = []\n obs_chunks = []\n act_chunks = []\n file_idx = 0\n output_dir_split = os.path.join(args.output_dir, split)\n while episode_idx < num_episodes:\n seed = np.random.randint(0, 10000)\n env = ProcgenGym3Env(num=1, env_name=""coinrun"", start_level=seed)\n\n observations_seq = []\n actions_seq = []\n episode_obs_chunks = []\n episode_act_chunks = []\n\n # --- Run episode ---\n step_t = 0\n for step_t in range(args.max_episode_length):\n action = types_np.sample(env.ac_space, bshape=(env.num,))\n env.act(action)\n _, obs, first = env.observe()\n observations_seq.append(obs[""rgb""])\n actions_seq.append(action)\n if len(observations_seq) == args.chunk_size:\n episode_obs_chunks.append(observations_seq)\n episode_act_chunks.append(actions_seq)\n observations_seq = []\n actions_seq = []\n if first:\n break\n\n # --- Save episode ---\n if step_t + 1 >= args.min_episode_length:\n if observations_seq:\n if len(observations_seq) < args.chunk_size:\n print(\n f""Warning: Inconsistent chunk_sizes. Episode has {len(observations_seq)} frames, ""\n f""which is smaller than the requested chunk_size: {args.chunk_size}. ""\n ""This might lead to performance degradation during training.""\n )\n episode_obs_chunks.append(observations_seq)\n episode_act_chunks.append(actions_seq)\n\n obs_chunks_data = [\n np.concatenate(seq, axis=0).astype(np.uint8)\n for seq in episode_obs_chunks\n ]\n act_chunks_data = [\n np.concatenate(act, axis=0) for act in episode_act_chunks\n ]\n obs_chunks.extend(obs_chunks_data)\n act_chunks.extend(act_chunks_data)\n\n ep_metadata, obs_chunks, file_idx, act_chunks = save_chunks(\n obs_chunks, file_idx, args.chunks_per_file, output_dir_split, act_chunks\n )\n episode_metadata.extend(ep_metadata)\n\n print(f""Episode {episode_idx} completed, length: {step_t + 1}."")\n episode_idx += 1\n else:\n print(f""Episode too short ({step_t + 1}), resampling..."")\n\n if len(obs_chunks) > 0:\n print(\n f""Warning: Dropping {len(obs_chunks)} chunks for consistent number of chunks per file."",\n ""Consider changing the chunk_size and chunks_per_file parameters to prevent data-loss."",\n )\n\n print(f""Done generating {split} split"")\n return episode_metadata\n\n\ndef get_action_space():\n env = ProcgenGym3Env(num=1, env_name=""coinrun"", start_level=0)\n return env.ac_space.eltype.n\n\n\ndef main():\n # Set random seed and create dataset directories\n np.random.seed(args.seed)\n # --- Generate episodes ---\n train_episode_metadata = generate_episodes(args.num_episodes_train, ""train"")\n val_episode_metadata = generate_episodes(args.num_episodes_val, ""val"")\n test_episode_metadata = generate_episodes(args.num_episodes_test, ""test"")\n\n # --- Save metadata ---\n metadata = {\n ""env"": ""coinrun"",\n ""num_actions"": get_action_space(),\n ""num_episodes_train"": args.num_episodes_train,\n ""num_episodes_val"": args.num_episodes_val,\n ""num_episodes_test"": args.num_episodes_test,\n ""avg_episode_len_train"": np.mean(\n [ep[""avg_seq_len""] for ep in train_episode_metadata]\n ),\n ""avg_episode_len_val"": np.mean(\n [ep[""avg_seq_len""] for ep in val_episode_metadata]\n ),\n ""avg_episode_len_test"": np.mean(\n [ep[""avg_seq_len""] for ep in test_episode_metadata]\n ),\n ""episode_metadata_train"": train_episode_metadata,\n ""episode_metadata_val"": val_episode_metadata,\n ""episode_metadata_test"": test_episode_metadata,\n }\n with open(os.path.join(args.output_dir, ""metadata.json""), ""w"") as f:\n json.dump(metadata, f)\n\n print(f""Done generating dataset."")\n\n\nif __name__ == ""__main__"":\n main()\n",python,tab +34,299089,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +35,300383,"cleanrl/ppo_atari_envpool.py",318,0,"",python,selection_mouse +36,300384,"cleanrl/ppo_atari_envpool.py",317,0,"",python,selection_command +37,314316,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,tab +38,315490,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +39,316028,"cleanrl/ppo_atari_envpool.py",330,0,"",python,selection_mouse +40,316031,"cleanrl/ppo_atari_envpool.py",329,0,"",python,selection_command +41,316695,"cleanrl/ppo_atari_envpool.py",436,0,"",python,selection_mouse +42,316698,"cleanrl/ppo_atari_envpool.py",435,0,"",python,selection_command +43,317177,"cleanrl/ppo_atari_envpool.py",438,0,"",python,selection_mouse +44,397100,"cleanrl/ppo_atari_envpool.py",2542,0,"",python,selection_mouse +45,402425,"cleanrl/ppo_atari_envpool.py",4160,0,"",python,selection_mouse +46,541375,"TERMINAL",0,0,"",,terminal_focus +47,541842,"TERMINAL",0,0,"source /home/franz.srambical/cleanrl/.venv/bin/activate",,terminal_command +48,544327,"TERMINAL",0,0,"cd ../jafar/",,terminal_command +49,554362,"TERMINAL",0,0,"git checkout gt-actions",,terminal_command +50,554410,"TERMINAL",0,0,"]633;C",,terminal_output +51,554473,"TERMINAL",0,0,"M\t.gitignore\r\nAlready on 'gt-actions'\r\nYour branch is behind 'origin/gt-actions' by 1 commit, and can be fast-forwarded.\r\n (use ""git pull"" to update your local branch)\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +52,555777,"TERMINAL",0,0,"git pull",,terminal_command +53,555825,"TERMINAL",0,0,"]633;C",,terminal_output +54,557974,"TERMINAL",0,0,"Updating 96d560e..1b6b878\r\nFast-forward\r\n input_pipeline/generate_coinrun_dataset.py | 2 +-\r\n 1 file changed, 1 insertion(+), 1 deletion(-)\r\n]0;franz.srambical@hai-login2:~/jafar",,terminal_output +55,560140,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,tab +56,560211,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",563,26," chunk_size: int = 160\n",python,content +57,561853,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +58,570185,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,tab +59,571693,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +60,576013,"cleanrl/ppo_atari_envpool.py",4335,0,"",python,selection_mouse +61,725152,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,tab +62,725643,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",280,0,"",python,selection_mouse +63,725644,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",279,0,"",python,selection_command +64,727140,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",1442,0,"",python,selection_command +65,728476,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +66,782305,"cleanrl/trajectory_saver.py",0,0,"import os\nimport json\nfrom typing import List, Dict, Any\n\nimport numpy as np\n\n\nclass TrajectorySaver:\n """"""\n Collects per-environment observation frames and actions, chunks them,\n and writes chunked files to disk. Designed to be lightweight and only\n enabled when trajectory capture is requested.\n\n Behavior mirrors the high level flow in `generate_coinrun_dataset.py`:\n - Build sequences of length `chunk_size` from steps\n - At episode end, if a partial sequence exists (< chunk_size), include it\n with a warning about inconsistent chunk sizes\n - Group `chunks_per_file` chunks per output file. Any remainder at the\n very end is dropped with a warning\n - Store both observation chunks and action chunks side-by-side\n """"""\n\n def __init__(\n self,\n output_dir: str,\n num_envs: int,\n chunk_size: int = 160,\n chunks_per_file: int = 100,\n ) -> None:\n self.output_dir = output_dir\n self.chunks_dir = os.path.join(output_dir, ""chunks"")\n os.makedirs(self.chunks_dir, exist_ok=True)\n self.num_envs = num_envs\n self.chunk_size = int(chunk_size)\n self.chunks_per_file = int(chunks_per_file)\n\n # Per-env rolling buffers for the current episode\n self.env_buffers: List[Dict[str, Any]] = [\n {""obs_seq"": [], ""act_seq"": [], ""episode_obs_chunks"": [], ""episode_act_chunks"": []}\n for _ in range(self.num_envs)\n ]\n\n # Global chunk buffers waiting to be flushed to disk\n self.obs_chunks: List[np.ndarray] = []\n self.act_chunks: List[np.ndarray] = []\n self.file_idx: int = 0\n\n # Metadata (optional). We keep a minimal JSONL with per-file info\n self.metadata_path = os.path.join(self.output_dir, ""metadata.jsonl"")\n\n def _finalize_episode_for_env(self, env_idx: int) -> None:\n buf = self.env_buffers[env_idx]\n obs_seq: List[np.ndarray] = buf[""obs_seq""]\n act_seq: List[np.ndarray] = buf[""act_seq""]\n\n if len(obs_seq) > 0:\n # Partial chunk at episode end\n if len(obs_seq) < self.chunk_size:\n print(\n f""Warning: Inconsistent chunk sizes. Env {env_idx} episode ended with ""\n f""{len(obs_seq)} frames (< chunk_size={self.chunk_size}). ""\n ""Including the partial chunk may impact training I/O performance.""\n )\n self._append_chunk_from_sequences(env_idx)\n\n # Move episode chunks to global buffers\n if len(buf[""episode_obs_chunks""]) > 0:\n self.obs_chunks.extend(buf[""episode_obs_chunks""])\n self.act_chunks.extend(buf[""episode_act_chunks""])\n buf[""episode_obs_chunks""].clear()\n buf[""episode_act_chunks""].clear()\n\n self._save_if_ready()\n\n def _append_chunk_from_sequences(self, env_idx: int) -> None:\n buf = self.env_buffers[env_idx]\n obs_seq: List[np.ndarray] = buf[""obs_seq""]\n act_seq: List[np.ndarray] = buf[""act_seq""]\n\n if len(obs_seq) == 0:\n return\n\n # Stack along time dimension. Shapes become:\n # obs_chunk: (T, C, H, W) dtype=uint8\n # act_chunk: (T,) dtype=int64 (or int)\n obs_chunk = np.stack(obs_seq, axis=0).astype(np.uint8)\n act_chunk = np.asarray(act_seq, dtype=np.int64)\n\n buf[""episode_obs_chunks""].append(obs_chunk)\n buf[""episode_act_chunks""].append(act_chunk)\n\n # Reset rolling sequences\n buf[""obs_seq""] = []\n buf[""act_seq""] = []\n\n def _save_if_ready(self) -> None:\n while len(self.obs_chunks) >= self.chunks_per_file:\n obs_to_write = self.obs_chunks[: self.chunks_per_file]\n act_to_write = self.act_chunks[: self.chunks_per_file]\n\n # Remove from buffers\n self.obs_chunks = self.obs_chunks[self.chunks_per_file :]\n self.act_chunks = self.act_chunks[self.chunks_per_file :]\n\n file_stem = f""chunks_{self.file_idx:06d}""\n out_path = os.path.join(self.chunks_dir, f""{file_stem}.npz"")\n\n # Write as a .npz with numbered arrays to preserve variable chunk lengths\n save_dict: Dict[str, Any] = {}\n for idx, (o, a) in enumerate(zip(obs_to_write, act_to_write)):\n save_dict[f""obs_{idx:03d}""] = o\n save_dict[f""acts_{idx:03d}""] = a\n\n np.savez_compressed(out_path, **save_dict)\n\n # Minimal metadata per file for bookkeeping\n meta = {\n ""file"": os.path.basename(out_path),\n ""num_chunks"": len(obs_to_write),\n ""chunk_size"": self.chunk_size,\n ""chunk_indices"": [i for i in range(len(obs_to_write))],\n }\n with open(self.metadata_path, ""a"") as f:\n f.write(json.dumps(meta) + ""\n"")\n\n self.file_idx += 1\n\n def add_step(\n self,\n obs_batch: np.ndarray,\n actions_batch: np.ndarray,\n episode_done_batch: np.ndarray,\n ) -> None:\n """"""\n Add a single environment step across all envs.\n\n Args:\n obs_batch: np.ndarray with shape (num_envs, C, H, W). Values should be [0, 255].\n actions_batch: np.ndarray with shape (num_envs,).\n episode_done_batch: np.ndarray with shape (num_envs,), boolean flags indicating end-of-episode.\n """"""\n assert obs_batch.shape[0] == self.num_envs\n assert actions_batch.shape[0] == self.num_envs\n assert episode_done_batch.shape[0] == self.num_envs\n\n for env_idx in range(self.num_envs):\n buf = self.env_buffers[env_idx]\n buf[""obs_seq""].append(obs_batch[env_idx])\n buf[""act_seq""].append(int(actions_batch[env_idx]))\n\n if len(buf[""obs_seq""]) == self.chunk_size:\n self._append_chunk_from_sequences(env_idx)\n\n # Episode ended: finalize the episode\n if bool(episode_done_batch[env_idx]):\n self._finalize_episode_for_env(env_idx)\n\n def close(self) -> None:\n # Drop remainder to keep consistent chunks-per-file contract\n remainder = len(self.obs_chunks)\n if remainder > 0:\n print(\n f""Warning: Dropping {remainder} chunks to keep a consistent number of chunks per file. ""\n ""Consider adjusting chunk_size and chunks_per_file to reduce data loss.""\n )\n # Nothing else to do; per-episode buffers do not carry data unless an episode just ended\n\n\n",python,tab +67,859278,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +68,877159,"cleanrl/ppo_atari_envpool.py",14147,0," if traj_saver is not None:\n traj_saver.close()\n",python,content +69,877159,"cleanrl/ppo_atari_envpool.py",8882,0," if traj_saver is not None:\n # Use true termination flags when available\n terminated = np.asarray(info.get(""terminated"", next_done.detach().cpu().numpy()), dtype=bool)\n traj_saver.add_step(obs_to_save, actions_to_save, terminated)\n\n",python,content +70,877159,"cleanrl/ppo_atari_envpool.py",8628,0," # Capture current observation and action prior to stepping the envs\n if traj_saver is not None:\n obs_to_save = next_obs.detach().cpu().numpy().astype(np.uint8)\n actions_to_save = action.detach().cpu().numpy()\n",python,content +71,877159,"cleanrl/ppo_atari_envpool.py",6967,0," # Optional: trajectory saver\n traj_saver = None\n if args.capture_trajectories:\n traj_saver = TrajectorySaver(\n output_dir=args.trajectories_output_dir,\n num_envs=args.num_envs,\n chunk_size=args.trajectories_chunk_size,\n chunks_per_file=args.trajectories_chunks_per_file,\n )\n\n",python,content +72,877159,"cleanrl/ppo_atari_envpool.py",1167,0," # Trajectory capture\n capture_trajectories: bool = False\n """"""if toggled, save observation frames and actions during training""""""\n trajectories_output_dir: str = ""data/atari_trajectories""\n """"""directory to store trajectory chunks (.npz)""""""\n trajectories_chunk_size: int = 160\n """"""number of time steps per chunk before writing""""""\n trajectories_chunks_per_file: int = 100\n """"""number of chunks grouped per npz file""""""\n\n",python,content +73,877160,"cleanrl/ppo_atari_envpool.py",437,0,"from .trajectory_saver import TrajectorySaver\n",python,content +74,913280,"cleanrl/ppo_atari_envpool.py",15527,58,"",python,content +75,913280,"cleanrl/ppo_atari_envpool.py",9974,288,"",python,content +76,913280,"cleanrl/ppo_atari_envpool.py",9458,262,"",python,content +77,913280,"cleanrl/ppo_atari_envpool.py",7454,343,"",python,content +78,913280,"cleanrl/ppo_atari_envpool.py",1213,441,"",python,content +79,913280,"cleanrl/ppo_atari_envpool.py",437,46,"",python,content +80,913354,"cleanrl/ppo_atari_envpool.py",14147,0," if traj_saver is not None:\n traj_saver.close()\n",python,content +81,913354,"cleanrl/ppo_atari_envpool.py",10671,0," # Initialize for type-checkers\n approx_kl = torch.tensor(0.0)\n old_approx_kl = torch.tensor(0.0)\n v_loss = torch.tensor(0.0)\n pg_loss = torch.tensor(0.0)\n entropy_loss = torch.tensor(0.0)\n",python,content +82,913354,"cleanrl/ppo_atari_envpool.py",8882,0," if traj_saver is not None and obs_to_save is not None and actions_to_save is not None:\n # Use true termination flags when available\n terminated = np.asarray(info.get(""terminated"", next_done.detach().cpu().numpy()), dtype=bool)\n traj_saver.add_step(obs_to_save, actions_to_save, terminated)\n\n",python,content +83,913354,"cleanrl/ppo_atari_envpool.py",8628,0," # Capture current observation and action prior to stepping the envs\n obs_to_save = None\n actions_to_save = None\n if traj_saver is not None:\n obs_to_save = next_obs.detach().cpu().numpy().astype(np.uint8)\n actions_to_save = action.detach().cpu().numpy()\n",python,content +84,913354,"cleanrl/ppo_atari_envpool.py",6861,104," assert isinstance(envs.action_space, spaces.Discrete), ""only discrete action space is supported""\n\n # Optional: trajectory saver\n traj_saver = None\n if args.capture_trajectories:\n traj_saver = TrajectorySaver(\n output_dir=args.trajectories_output_dir,\n num_envs=args.num_envs,\n chunk_size=args.trajectories_chunk_size,\n chunks_per_file=args.trajectories_chunks_per_file,\n )",python,content +85,913354,"cleanrl/ppo_atari_envpool.py",2469,27," target_kl: Optional[float] = None",python,content +86,913355,"cleanrl/ppo_atari_envpool.py",1167,0," # Trajectory capture\n capture_trajectories: bool = False\n """"""if toggled, save observation frames and actions during training""""""\n trajectories_output_dir: str = ""data/atari_trajectories""\n """"""directory to store trajectory chunks (.npz)""""""\n trajectories_chunk_size: int = 160\n """"""number of time steps per chunk before writing""""""\n trajectories_chunks_per_file: int = 100\n """"""number of chunks grouped per npz file""""""\n\n",python,content +87,913355,"cleanrl/ppo_atari_envpool.py",968,28," wandb_entity: Optional[str] = None",python,content +88,913355,"cleanrl/ppo_atari_envpool.py",387,49,"from torch.utils.tensorboard.writer import SummaryWriter\nfrom typing import Optional\nfrom gym import spaces\nfrom .trajectory_saver import TrajectorySaver",python,content +89,961375,"cleanrl/ppo_atari_envpool.py",15958,58,"",python,content +90,961375,"cleanrl/ppo_atari_envpool.py",12251,231,"",python,content +91,961375,"cleanrl/ppo_atari_envpool.py",10114,348,"",python,content +92,961375,"cleanrl/ppo_atari_envpool.py",9532,328,"",python,content +93,961375,"cleanrl/ppo_atari_envpool.py",7426,443," assert isinstance(envs.action_space, gym.spaces.Discrete), ""only discrete action space is supported""",python,content +94,961375,"cleanrl/ppo_atari_envpool.py",3024,37," target_kl: float = None",python,content +95,961375,"cleanrl/ppo_atari_envpool.py",1281,441,"",python,content +96,961375,"cleanrl/ppo_atari_envpool.py",1072,38," wandb_entity: str = None",python,content +97,961375,"cleanrl/ppo_atari_envpool.py",387,153,"from torch.utils.tensorboard import SummaryWriter",python,content +98,961438,"cleanrl/ppo_atari_envpool.py",14147,0," if traj_saver is not None:\n traj_saver.close()\n",python,content +99,961438,"cleanrl/ppo_atari_envpool.py",10671,0," # Initialize for type-checkers\n approx_kl = torch.tensor(0.0)\n old_approx_kl = torch.tensor(0.0)\n v_loss = torch.tensor(0.0)\n pg_loss = torch.tensor(0.0)\n entropy_loss = torch.tensor(0.0)\n",python,content +100,961438,"cleanrl/ppo_atari_envpool.py",8882,0," if traj_saver is not None and obs_to_save is not None and actions_to_save is not None:\n # Use true termination flags when available\n terminated = np.asarray(info.get(""terminated"", next_done.detach().cpu().numpy()), dtype=bool)\n traj_saver.add_step(obs_to_save, actions_to_save, terminated)\n\n",python,content +101,961438,"cleanrl/ppo_atari_envpool.py",8628,0," # Capture current observation and action prior to stepping the envs\n obs_to_save = None\n actions_to_save = None\n if traj_saver is not None:\n obs_to_save = next_obs.detach().cpu().numpy().astype(np.uint8)\n actions_to_save = action.detach().cpu().numpy()\n",python,content +102,961438,"cleanrl/ppo_atari_envpool.py",7002,79," optimizer = Adam(agent.parameters(), lr=args.learning_rate, eps=1e-5)",python,content +103,961438,"cleanrl/ppo_atari_envpool.py",6861,104," assert isinstance(envs.action_space, spaces.Discrete), ""only discrete action space is supported""\n\n # Optional: trajectory saver\n traj_saver = None\n if args.capture_trajectories:\n traj_saver = TrajectorySaver(\n output_dir=args.trajectories_output_dir,\n num_envs=args.num_envs,\n chunk_size=args.trajectories_chunk_size,\n chunks_per_file=args.trajectories_chunks_per_file,\n )",python,content +104,961438,"cleanrl/ppo_atari_envpool.py",2982,71," self.episode_returns = np.zeros(self.num_envs, dtype=np.float32)\n self.episode_lengths = np.zeros(self.num_envs, dtype=np.int32)",python,content +105,961438,"cleanrl/ppo_atari_envpool.py",2469,27," target_kl: Optional[float] = None",python,content +106,961438,"cleanrl/ppo_atari_envpool.py",1167,0," # Trajectory capture\n capture_trajectories: bool = False\n """"""if toggled, save observation frames and actions during training""""""\n trajectories_output_dir: str = ""data/atari_trajectories""\n """"""directory to store trajectory chunks (.npz)""""""\n trajectories_chunk_size: int = 160\n """"""number of time steps per chunk before writing""""""\n trajectories_chunks_per_file: int = 100\n """"""number of chunks grouped per npz file""""""\n\n",python,content +107,961438,"cleanrl/ppo_atari_envpool.py",968,28," wandb_entity: Optional[str] = None",python,content +108,961438,"cleanrl/ppo_atari_envpool.py",387,49,"from torch.utils.tensorboard.writer import SummaryWriter\nfrom typing import Optional\nfrom gym import spaces\nfrom .trajectory_saver import TrajectorySaver",python,content +109,961438,"cleanrl/ppo_atari_envpool.py",291,27,"from torch.optim import Adam",python,content +110,968028,"cleanrl/ppo_atari_envpool.py",268,0,"",python,selection_mouse +111,968030,"cleanrl/ppo_atari_envpool.py",267,0,"",python,selection_command +112,969822,"cleanrl/ppo_atari_envpool.py",280,0,"",python,selection_command +113,974152,"cleanrl/ppo_atari_envpool.py",291,28,"import torch.optim as optim",python,content +114,977518,"cleanrl/ppo_atari_envpool.py",7994,0,"",python,selection_command +115,978956,"cleanrl/ppo_atari_envpool.py",7978,73," optimizer = optim.Adam(agent.parameters(), lr=args.learning_rate, eps=1e-5)",python,content +116,979756,"cleanrl/ppo_atari_envpool.py",8056,0,"",python,selection_command +117,980869,"cleanrl/ppo_atari_envpool.py",0,0,"",python,selection_command +118,1015212,"cleanrl/ppo_atari_envpool.py",12299,0,"",python,selection_mouse +119,1015215,"cleanrl/ppo_atari_envpool.py",12298,0,"",python,selection_command +120,1016996,"cleanrl/ppo_atari_envpool.py",12323,231,"",python,content +121,1021028,"cleanrl/ppo_atari_envpool.py",12205,0,"",python,selection_mouse +122,1027883,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,tab +123,1037912,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",3201,0,"",python,selection_mouse +124,1043895,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,selection_command +125,1063321,"/home/franz.srambical/jafar/input_pipeline/utils.py",0,0,"import os\nimport pickle\nimport numpy as np\nfrom array_record.python.array_record_module import ArrayRecordWriter\n\n\ndef save_chunks(obs_chunks, file_idx, chunks_per_file, output_dir, act_chunks=None):\n os.makedirs(output_dir, exist_ok=True)\n\n metadata = []\n while len(obs_chunks) >= chunks_per_file:\n chunk_batch = obs_chunks[:chunks_per_file]\n obs_chunks = obs_chunks[chunks_per_file:]\n act_chunk_batch = None\n if act_chunks:\n act_chunk_batch = act_chunks[:chunks_per_file]\n act_chunks = act_chunks[chunks_per_file:]\n episode_path = os.path.join(output_dir, f""data_{file_idx:04d}.array_record"")\n writer = ArrayRecordWriter(str(episode_path), ""group_size:1"")\n seq_lens = []\n for idx, chunk in enumerate(chunk_batch):\n seq_len = chunk.shape[0]\n seq_lens.append(seq_len)\n chunk_record = {\n ""raw_video"": chunk.tobytes(),\n ""sequence_length"": seq_len,\n }\n if act_chunk_batch:\n assert len(chunk) == len(\n act_chunk_batch[idx]\n ), f""Observation data length and action sequence length do not match: {len(chunk)} != {len(act_chunk_batch[idx])}""\n chunk_record[""actions""] = act_chunk_batch[idx]\n writer.write(pickle.dumps(chunk_record))\n writer.close()\n file_idx += 1\n metadata.append(\n {\n ""path"": episode_path,\n ""num_chunks"": len(chunk_batch),\n ""avg_seq_len"": np.mean(seq_lens),\n }\n )\n print(f""Created {episode_path} with {len(chunk_batch)} video chunks"")\n\n return metadata, obs_chunks, file_idx, act_chunks\n",python,tab +126,1077626,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,tab +127,1078649,"/home/franz.srambical/jafar/input_pipeline/utils.py",0,0,"",python,tab +128,1084681,"/home/franz.srambical/jafar/input_pipeline/generate_coinrun_dataset.py",0,0,"",python,tab +129,1098108,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +130,1107176,"cleanrl/ppo_atari_envpool.py",716,0,"",python,selection_mouse +131,1107177,"cleanrl/ppo_atari_envpool.py",715,0,"",python,selection_command +132,1163039,"cleanrl/ppo_atari_envpool.py",553,0,"",python,selection_mouse +133,1163040,"cleanrl/ppo_atari_envpool.py",552,0,"",python,selection_command +134,1164422,"cleanrl/ppo_atari_envpool.py",542,0,"",python,selection_mouse +135,1167929,"cleanrl/ppo_atari_envpool.py",565,0,"",python,selection_mouse +136,1167931,"cleanrl/ppo_atari_envpool.py",564,0,"",python,selection_command +137,1169182,"cleanrl/ppo_atari_envpool.py",387,0,"",python,selection_command +138,1170321,"cleanrl/ppo_atari_envpool.py",1072,0,"",python,selection_command +139,1170873,"cleanrl/ppo_atari_envpool.py",387,0,"",python,selection_command +140,1172070,"cleanrl/ppo_atari_envpool.py",1072,0,"",python,selection_command +141,1172368,"cleanrl/ppo_atari_envpool.py",387,0,"",python,selection_command +142,1172644,"cleanrl/ppo_atari_envpool.py",1072,0,"",python,selection_command +143,4767412,"cleanrl/ppo_atari_envpool.py",840,0,"",python,selection_mouse +144,4767415,"cleanrl/ppo_atari_envpool.py",839,0,"",python,selection_command +145,4769432,"cleanrl/trajectory_saver.py",0,0,"",python,tab +146,4770255,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +147,4770257,"cleanrl/ppo_atari_envpool.py",387,0,"",python,selection_command +148,4770719,"cleanrl/trajectory_saver.py",0,0,"",python,tab +149,4771118,"cleanrl/ppo_atari_envpool.py",0,0,"",python,tab +150,5039789,"cleanrl/ppo_atari_envpool.py",210,0,"",python,selection_mouse diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-d9cdf338-0ddd-4679-853a-6d7bdf2b18581751046137722-2025_06_27-10.42.19.354/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-d9cdf338-0ddd-4679-853a-6d7bdf2b18581751046137722-2025_06_27-10.42.19.354/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..e0cff80c22bfd8e05fecca513063072f6928a428 --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-d9cdf338-0ddd-4679-853a-6d7bdf2b18581751046137722-2025_06_27-10.42.19.354/source.csv @@ -0,0 +1,167 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,1,"utils/dataloader.py",0,0,"import functools\nimport jax\n\nimport tensorflow as tf\n\n# reserve GPU memory for JAX only if tensorflow is built with GPU support\ntf.config.experimental.set_visible_devices([], ""GPU"")\n\n\n# --- TensorFlow function for processing: slicing, normalization ---\ndef _tf_process_episode(episode_tensor, seq_len, image_h, image_w, image_c):\n """"""\n Processes a raw episode tensor in TensorFlow.\n Takes a full episode, extracts a random sequence, and normalizes it.\n Args:\n episode_tensor: A TensorFlow tensor representing a full video episode.\n Expected shape: (dynamic_length, image_h, image_w, image_c)\n Expected dtype: e.g., tf.uint8 (raw pixel values)\n seq_len: The desired length of the sub-sequence to extract.\n image_h: The height of each frame.\n image_w: The width of each frame.\n image_c: The number of channels in each frame.\n Returns:\n A TensorFlow tensor representing the processed video sequence.\n Shape: (seq_len, image_h, image_w, image_c)\n Dtype: tf.float32 (normalized pixel values)\n """"""\n current_episode_len = tf.shape(episode_tensor)[0]\n\n max_start_idx = current_episode_len - seq_len\n\n start_idx = tf.random.uniform(\n shape=(), minval=0, maxval=max_start_idx + 1, dtype=tf.int32\n )\n\n seq = episode_tensor[start_idx : start_idx + seq_len]\n\n seq = tf.cast(seq, tf.float32) / 255.0\n\n # Ensure the final shape is statically known for batching.\n # tf.reshape is robust, but tf.ensure_shape or set_shape can also be used if confident.\n processed_sequence = tf.reshape(seq, [seq_len, image_h, image_w, image_c])\n\n return processed_sequence\n\n\ndef _parse_tfrecord_fn(example_proto, image_h, image_w, image_c):\n feature_description = {\n ""height"": tf.io.FixedLenFeature([], tf.int64),\n ""width"": tf.io.FixedLenFeature([], tf.int64),\n ""channels"": tf.io.FixedLenFeature([], tf.int64),\n ""sequence_length"": tf.io.FixedLenFeature([], tf.int64),\n ""raw_video"": tf.io.FixedLenFeature([], tf.string),\n }\n example = tf.io.parse_single_example(example_proto, feature_description)\n\n video_shape = (example[""sequence_length""], image_h, image_w, image_c)\n\n episode_tensor = tf.io.decode_raw(example[""raw_video""], out_type=tf.uint8)\n episode_tensor = tf.reshape(episode_tensor, video_shape)\n\n episode_tensor = tf.ensure_shape(episode_tensor, [None, image_h, image_w, image_c])\n return episode_tensor\n\n\ndef get_dataloader(\n tfrecord_paths: list[str], # List of TFRecord file paths\n seq_len: int,\n global_batch_size: int,\n image_h: int,\n image_w: int,\n image_c: int,\n shuffle_buffer_size: int = 1000,\n num_parallel_calls: int = tf.data.AUTOTUNE,\n seed: int = 42,\n):\n """"""\n Creates a tf.data.Dataset pipeline from TFRecord files.\n """"""\n if not tfrecord_paths:\n raise ValueError(""tfrecord_paths list cannot be empty."")\n\n process_id = jax.process_index()\n num_processes = jax.process_count()\n\n assert (\n global_batch_size % num_processes == 0\n ), ""Global batch size {global_batch_size} \\n must be divisible by the number of JAX processes {num_processes} for proper sharding.""\n per_process_batch_size = global_batch_size // num_processes\n\n dataset = tf.data.TFRecordDataset(\n tfrecord_paths, num_parallel_reads=tf.data.AUTOTUNE\n )\n\n dataset = dataset.shard(num_shards=num_processes, index=process_id)\n\n # (f.srambical) NOTE: For TFRecords, it's often good to have a large shuffle buffer.\n if shuffle_buffer_size > 0:\n dataset = dataset.shuffle(\n buffer_size=shuffle_buffer_size, seed=seed, reshuffle_each_iteration=True\n )\n parse_fn = functools.partial(\n _parse_tfrecord_fn, image_h=image_h, image_w=image_w, image_c=image_c\n )\n dataset = dataset.map(parse_fn, num_parallel_calls=num_parallel_calls)\n\n tf_process_fn = functools.partial(\n _tf_process_episode,\n seq_len=seq_len,\n image_h=image_h,\n image_w=image_w,\n image_c=image_c,\n )\n dataset = dataset.map(tf_process_fn, num_parallel_calls=num_parallel_calls)\n\n dataset = dataset.repeat(None)\n dataset = dataset.batch(per_process_batch_size, drop_remainder=True)\n dataset = dataset.prefetch(tf.data.AUTOTUNE)\n\n return dataset.as_numpy_iterator()\n",python,tab +2,38,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +3,67,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"10:42:19 AM [info] Activating crowd-code\n10:42:19 AM [info] Recording started\n10:42:19 AM [info] Initializing git provider using file system watchers...\n10:42:19 AM [info] Git repository found\n10:42:19 AM [info] Git provider initialized successfully\n10:42:19 AM [info] Initial git state: [object Object]\n",Log,content +4,905,"utils/dataloader.py",0,0,"",python,tab +5,19039,"utils/dataloader.py",2752,0,"",python,selection_command +6,19130,"utils/dataloader.py",2715,0,"",python,selection_command +7,30009,"models/dynamics.py",0,0,"from typing import Dict, Any\n\nimport jax\nimport jax.numpy as jnp\nimport flax.linen as nn\n\nfrom utils.nn import STTransformer\n\n\nclass DynamicsMaskGIT(nn.Module):\n """"""MaskGIT dynamics model""""""\n\n model_dim: int\n num_latents: int\n num_blocks: int\n num_heads: int\n dropout: float\n mask_limit: float\n\n def setup(self):\n self.dynamics = STTransformer(\n self.model_dim,\n self.num_latents,\n self.num_blocks,\n self.num_heads,\n self.dropout,\n )\n self.patch_embed = nn.Embed(self.num_latents, self.model_dim)\n self.mask_token = self.param(\n ""mask_token"",\n nn.initializers.lecun_uniform(),\n (1, 1, 1, self.model_dim),\n )\n self.action_up = nn.Dense(self.model_dim)\n\n def __call__(self, batch: Dict[str, Any], training: bool = True) -> Dict[str, Any]:\n # --- Mask videos ---\n vid_embed = self.patch_embed(batch[""video_tokens""])\n if training:\n rng1, rng2 = jax.random.split(batch[""mask_rng""])\n mask_prob = jax.random.uniform(rng1, minval=self.mask_limit)\n mask = jax.random.bernoulli(rng2, mask_prob, vid_embed.shape[:-1])\n mask = mask.at[:, 0].set(False)\n vid_embed = jnp.where(jnp.expand_dims(mask, -1), self.mask_token, vid_embed)\n else:\n mask = None\n\n # --- Predict transition ---\n act_embed = self.action_up(batch[""latent_actions""])\n vid_embed += jnp.pad(act_embed, ((0, 0), (1, 0), (0, 0), (0, 0)))\n logits = self.dynamics(vid_embed)\n return dict(token_logits=logits, mask=mask)\n",python,tab +8,30937,"models/dynamics.py",416,0,"",python,selection_command +9,31509,"models/dynamics.py",1359,0,"",python,selection_command +10,31790,"models/dynamics.py",1655,0,"",python,selection_command +11,32552,"models/dynamics.py",753,0,"",python,selection_command +12,32710,"models/dynamics.py",165,0,"",python,selection_command +13,32898,"models/dynamics.py",0,0,"",python,selection_command +14,35425,"train_dynamics.py",0,0,"from dataclasses import dataclass, field\nimport os\nimport time\n\nimport einops\nfrom flax.training import orbax_utils\nfrom flax.training.train_state import TrainState\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax\nfrom orbax.checkpoint import PyTreeCheckpointer\nimport numpy as np\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\n\nfrom genie import Genie, restore_genie_components\nfrom models.tokenizer import TokenizerVQVAE\nfrom models.lam import LatentActionModel\nfrom utils.dataloader import get_dataloader\n\nts = int(time.time())\n\n\n@dataclass\nclass Args:\n # Experiment\n num_steps: int = 200_000\n seed: int = 0\n seq_len: int = 16\n image_channels: int = 3\n image_height: int = 90\n image_width: int = 160\n data_dir: str = ""data_tfrecords/coinrun""\n # Optimization\n batch_size: int = 36\n min_lr: float = 3e-6\n max_lr: float = 3e-5\n warmup_steps: int = 5000\n # Tokenizer\n tokenizer_dim: int = 512\n latent_patch_dim: int = 32\n num_patch_latents: int = 1024\n patch_size: int = 4\n tokenizer_num_blocks: int = 8\n tokenizer_num_heads: int = 8\n tokenizer_checkpoint: str = """"\n # LAM\n lam_dim: int = 512\n latent_action_dim: int = 32\n num_latent_actions: int = 6\n lam_patch_size: int = 16\n lam_num_blocks: int = 8\n lam_num_heads: int = 8\n lam_checkpoint: str = """"\n # Dynamics\n dyna_dim: int = 512\n dyna_num_blocks: int = 12\n dyna_num_heads: int = 8\n dropout: float = 0.0\n mask_limit: float = 0.5\n # Logging\n log: bool = False\n entity: str = """"\n project: str = """"\n name: str = ""train_dynamics""\n tags: list[str] = field(default_factory=lambda: [""dynamics""])\n log_interval: int = 5\n log_image_interval: int = 250\n ckpt_dir: str = """"\n log_checkpoint_interval: int = 25000\n log_gradients: bool = False\n\n\nargs = tyro.cli(Args)\n\n\ndef dynamics_loss_fn(params, state, inputs):\n """"""Compute masked dynamics loss""""""\n outputs = state.apply_fn(\n params,\n inputs,\n training=True,\n rngs={""params"": inputs[""rng""], ""dropout"": inputs[""dropout_rng""]},\n )\n mask = outputs[""mask""]\n ce_loss = optax.softmax_cross_entropy_with_integer_labels(\n outputs[""token_logits""], outputs[""video_tokens""]\n )\n ce_loss = (mask * ce_loss).sum() / mask.sum()\n acc = outputs[""token_logits""].argmax(-1) == outputs[""video_tokens""]\n acc = (mask * acc).sum() / mask.sum()\n select_probs = jax.nn.softmax(outputs[""token_logits""])\n metrics = dict(\n cross_entropy_loss=ce_loss,\n masked_token_accuracy=acc,\n select_logit=outputs[""token_logits""].max(-1).mean(),\n select_p=select_probs.max(-1).mean(),\n entropy=jax.scipy.special.entr(select_probs).sum(-1).mean(),\n )\n return ce_loss, (outputs[""recon""], metrics)\n\n\n@jax.jit\ndef train_step(state, inputs):\n """"""Update state and compute metrics""""""\n grad_fn = jax.value_and_grad(dynamics_loss_fn, has_aux=True, allow_int=True)\n (loss, (recon, metrics)), grads = grad_fn(state.params, state, inputs)\n state = state.apply_gradients(grads=grads)\n if args.log_gradients:\n metrics[""gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""dynamics""]\n )\n return state, loss, recon, metrics\n\n\nif __name__ == ""__main__"":\n jax.distributed.initialize()\n num_devices = jax.device_count()\n if num_devices == 0:\n raise ValueError(""No JAX devices found."")\n print(f""Running on {num_devices} devices."")\n\n if args.batch_size % num_devices != 0:\n raise ValueError(\n f""Global batch size {args.batch_size} must be divisible by ""\n f""number of devices {num_devices}.""\n )\n\n per_device_batch_size_for_init = args.batch_size // num_devices\n\n rng = jax.random.PRNGKey(args.seed)\n if args.log and jax.process_index() == 0:\n wandb.init(\n entity=args.entity,\n project=args.project,\n name=args.name,\n tags=args.tags,\n group=""debug"",\n config=args\n )\n\n # --- Initialize model ---\n genie = Genie(\n # Tokenizer\n in_dim=args.image_channels,\n tokenizer_dim=args.tokenizer_dim,\n latent_patch_dim=args.latent_patch_dim,\n num_patch_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n tokenizer_num_blocks=args.tokenizer_num_blocks,\n tokenizer_num_heads=args.tokenizer_num_heads,\n # LAM\n lam_dim=args.lam_dim,\n latent_action_dim=args.latent_action_dim,\n num_latent_actions=args.num_latent_actions,\n lam_patch_size=args.lam_patch_size,\n lam_num_blocks=args.lam_num_blocks,\n lam_num_heads=args.lam_num_heads,\n # Dynamics\n dyna_dim=args.dyna_dim,\n dyna_num_blocks=args.dyna_num_blocks,\n dyna_num_heads=args.dyna_num_heads,\n dropout=args.dropout,\n mask_limit=args.mask_limit,\n )\n rng, _rng = jax.random.split(rng)\n image_shape = (args.image_height, args.image_width, args.image_channels)\n dummy_inputs = dict(\n videos=jnp.zeros(\n (per_device_batch_size_for_init, args.seq_len, *image_shape),\n dtype=jnp.float32,\n ),\n action=jnp.zeros(\n (per_device_batch_size_for_init, args.seq_len), dtype=jnp.float32\n ),\n mask_rng=_rng,\n )\n rng, _rng = jax.random.split(rng)\n init_params = genie.init(_rng, dummy_inputs)\n\n # --- Initialize optimizer ---\n lr_schedule = optax.warmup_cosine_decay_schedule(\n args.min_lr, args.max_lr, args.warmup_steps, args.num_steps\n )\n tx = optax.adamw(learning_rate=lr_schedule, b1=0.9, b2=0.9, weight_decay=1e-4)\n train_state = TrainState.create(apply_fn=genie.apply, params=init_params, tx=tx)\n\n device_mesh_arr = create_device_mesh((num_devices,))\n mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n\n replicated_sharding = NamedSharding(mesh, PartitionSpec())\n train_state = jax.device_put(train_state, replicated_sharding)\n\n # --- Restore checkpoint ---\n train_state = restore_genie_components(\n train_state, replicated_sharding, dummy_inputs, rng, args\n )\n\n # --- TRAIN LOOP ---\n tfrecord_files = [\n os.path.join(args.data_dir, x)\n for x in os.listdir(args.data_dir)\n if x.endswith("".tfrecord"")\n ]\n dataloader = get_dataloader(\n # NOTE: We deliberately pass the global batch size\n # The dataloader shards the dataset across all processes\n tfrecord_files,\n args.seq_len,\n args.batch_size,\n *image_shape,\n )\n step = 0\n while step < args.num_steps:\n for videos in dataloader:\n # --- Train step ---\n rng, _rng, _rng_dropout, _rng_mask = jax.random.split(rng, 4)\n\n videos_sharding = NamedSharding(\n mesh, PartitionSpec(""data"", None, None, None, None)\n )\n videos = jax.make_array_from_process_local_data(videos_sharding, videos)\n\n inputs = dict(\n videos=videos,\n rng=_rng,\n dropout_rng=_rng_dropout,\n mask_rng=_rng_mask,\n )\n start_time = time.time()\n train_state, loss, recon, metrics = train_step(train_state, inputs)\n elapsed_time = (time.time() - start_time) * 1000\n print(f""Step {step}, loss: {loss}, step time: {elapsed_time}ms"")\n step += 1\n\n # --- Logging ---\n if args.log:\n if step % args.log_interval == 0 and jax.process_index() == 0:\n wandb.log(\n {\n ""loss"": loss,\n ""step"": step,\n ""step_time_ms"": elapsed_time,\n **metrics,\n }\n )\n if step % args.log_image_interval == 0:\n gt_seq = inputs[""videos""][0]\n recon_seq = recon[0].clip(0, 1)\n comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n comparison_seq = einops.rearrange(\n comparison_seq * 255, ""t h w c -> h (t w) c""\n )\n if jax.process_index() == 0:\n log_images = dict(\n image=wandb.Image(np.asarray(gt_seq[args.seq_len - 1])),\n recon=wandb.Image(np.asarray(recon_seq[args.seq_len - 1])),\n true_vs_recon=wandb.Image(\n np.asarray(comparison_seq.astype(np.uint8))\n ),\n )\n wandb.log(log_images)\n if step % args.log_checkpoint_interval == 0:\n ckpt = {""model"": train_state}\n orbax_checkpointer = orbax.checkpoint.PyTreeCheckpointer()\n save_args = orbax_utils.save_args_from_target(ckpt)\n orbax_checkpointer.save(\n os.path.join(os.getcwd(), args.ckpt_dir, f""genie_{ts}_{step}""),\n ckpt,\n save_args=save_args,\n )\n if step >= args.num_steps:\n break\n",python,tab +15,39797,"train_dynamics.py",5890,0,"",python,selection_command +16,40207,"train_dynamics.py",5044,0,"",python,selection_command +17,40597,"train_dynamics.py",4217,0,"",python,selection_command +18,40765,"train_dynamics.py",3598,0,"",python,selection_command +19,40946,"train_dynamics.py",2859,0,"",python,selection_command +20,41080,"train_dynamics.py",2006,0,"",python,selection_command +21,41229,"train_dynamics.py",1466,0,"",python,selection_command +22,41446,"train_dynamics.py",882,0,"",python,selection_command +23,41604,"train_dynamics.py",388,0,"",python,selection_command +24,41744,"train_dynamics.py",0,0,"",python,selection_command +25,43268,"train_dynamics.py",617,0,"",python,selection_command +26,46121,"train_dynamics.py",0,0,"",python,selection_command +27,46786,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +28,49091,"train_dynamics.py",0,0,"",python,tab +29,49113,"TERMINAL",0,0,"",,terminal_focus +30,104571,"requirements.txt",0,0,"dm_pix>=0.4.3\neinops>=0.8.0\nflax>=0.8.5\njax[cuda12]>=0.4.30\noptax>=0.2.3\nprocgen>=0.10.7\ntyro>=0.8.5\nwandb>=0.17.4\ntensorflow>=2.1\npre-commit>=4.2.0",pip-requirements,tab +31,106305,"requirements.txt",73,0,"",pip-requirements,selection_command +32,106838,"requirements.txt",73,0,"#",pip-requirements,content +33,106844,"requirements.txt",74,0,"",pip-requirements,selection_keyboard +34,106906,"requirements.txt",74,0," ",pip-requirements,content +35,106910,"requirements.txt",75,0,"",pip-requirements,selection_keyboard +36,107095,"requirements.txt",74,0,"",pip-requirements,selection_command +37,114295,"requirements.txt",61,0,"",pip-requirements,selection_command +38,114428,"requirements.txt",41,0,"",pip-requirements,selection_command +39,114530,"requirements.txt",43,0,"",pip-requirements,selection_command +40,114983,"requirements.txt",43,1,"[",pip-requirements,selection_command +41,115082,"requirements.txt",43,7,"[cuda12",pip-requirements,selection_command +42,115265,"requirements.txt",43,10,"[cuda12]>=",pip-requirements,selection_command +43,115682,"requirements.txt",43,9,"[cuda12]>",pip-requirements,selection_command +44,115832,"requirements.txt",43,8,"[cuda12]",pip-requirements,selection_command +45,116046,"requirements.txt",43,8,"",pip-requirements,content +46,134920,"train_dynamics.py",0,0,"",python,tab +47,141347,"train_dynamics.py",41,0,"",python,selection_command +48,141596,"train_dynamics.py",51,0,"",python,selection_command +49,141626,"train_dynamics.py",63,0,"",python,selection_command +50,141657,"train_dynamics.py",64,0,"",python,selection_command +51,141691,"train_dynamics.py",78,0,"",python,selection_command +52,141894,"train_dynamics.py",116,0,"",python,selection_command +53,141964,"train_dynamics.py",121,0,"",python,selection_command +54,142163,"train_dynamics.py",125,0,"",python,selection_command +55,142229,"train_dynamics.py",87,0,"",python,selection_command +56,145246,"train_dynamics.py",653,0,"",python,selection_command +57,145597,"train_dynamics.py",1241,0,"",python,selection_command +58,148942,"train_dynamics.py",1839,0,"",python,selection_command +59,150414,"train_dynamics.py",2485,0,"",python,selection_command +60,150985,"train_dynamics.py",3341,0,"",python,selection_command +61,153046,"train_dynamics.py",2485,0,"",python,selection_command +62,153663,"train_dynamics.py",2527,0,"",python,selection_command +63,153914,"train_dynamics.py",2586,0,"",python,selection_command +64,153946,"train_dynamics.py",2606,0,"",python,selection_command +65,153981,"train_dynamics.py",2642,0,"",python,selection_command +66,154015,"train_dynamics.py",2677,0,"",python,selection_command +67,154046,"train_dynamics.py",2738,0,"",python,selection_command +68,154081,"train_dynamics.py",2784,0,"",python,selection_command +69,154113,"train_dynamics.py",2853,0,"",python,selection_command +70,154146,"train_dynamics.py",2859,0,"",python,selection_command +71,154180,"train_dynamics.py",2903,0,"",python,selection_command +72,154213,"train_dynamics.py",2904,0,"",python,selection_command +73,154246,"train_dynamics.py",2909,0,"",python,selection_command +74,154280,"train_dynamics.py",2918,0,"",python,selection_command +75,154508,"train_dynamics.py",3753,0,"",python,selection_command +76,155160,"train_dynamics.py",4363,0,"",python,selection_command +77,155948,"train_dynamics.py",5218,0,"",python,selection_command +78,158479,"train_dynamics.py",6074,0,"",python,selection_command +79,161795,"train_dynamics.py",6732,0,"",python,selection_command +80,166966,"train_dynamics.py",7582,0,"",python,selection_command +81,170485,"train_dynamics.py",8668,0,"",python,selection_command +82,175934,"train_dynamics.py",7582,0,"",python,selection_command +83,177106,"train_dynamics.py",7569,0,"",python,selection_command +84,177361,"train_dynamics.py",7559,0,"",python,selection_command +85,177390,"train_dynamics.py",7482,0,"",python,selection_command +86,177422,"train_dynamics.py",7421,0,"",python,selection_command +87,177457,"train_dynamics.py",7341,0,"",python,selection_command +88,177488,"train_dynamics.py",7304,0,"",python,selection_command +89,178494,"train_dynamics.py",7341,0,"",python,selection_command +90,178693,"train_dynamics.py",7421,0,"",python,selection_command +91,178849,"train_dynamics.py",7482,0,"",python,selection_command +92,179027,"train_dynamics.py",7559,0,"",python,selection_command +93,179210,"train_dynamics.py",7569,0,"",python,selection_command +94,179506,"train_dynamics.py",7559,0,"",python,selection_command +95,179674,"train_dynamics.py",7482,0,"",python,selection_command +96,179833,"train_dynamics.py",7421,0,"",python,selection_command +97,179926,"train_dynamics.py",7482,0,"",python,selection_command +98,180126,"train_dynamics.py",7421,0,"",python,selection_command +99,180497,"train_dynamics.py",7482,0,"",python,selection_command +100,181016,"train_dynamics.py",7559,0,"",python,selection_command +101,181259,"train_dynamics.py",7482,0,"",python,selection_command +102,183280,"train_dynamics.py",7421,0,"",python,selection_command +103,183526,"train_dynamics.py",7341,0,"",python,selection_command +104,185378,"train_dynamics.py",7407,0,"",python,selection_command +105,185754,"train_dynamics.py",7401,0,"",python,selection_command +106,185894,"train_dynamics.py",7399,0,"",python,selection_command +107,186075,"train_dynamics.py",7388,0,"",python,selection_command +108,186198,"train_dynamics.py",7387,0,"",python,selection_command +109,186477,"train_dynamics.py",7377,0,"",python,selection_command +110,212701,"train_dynamics.py",2949,0,"",python,selection_command +111,212882,"train_dynamics.py",2992,0,"",python,selection_command +112,212978,"train_dynamics.py",3000,0,"",python,selection_command +113,213228,"train_dynamics.py",3002,0,"",python,selection_command +114,213261,"train_dynamics.py",3005,0,"",python,selection_command +115,213294,"train_dynamics.py",3006,0,"",python,selection_command +116,213327,"train_dynamics.py",3020,0,"",python,selection_command +117,213590,"train_dynamics.py",3021,0,"",python,selection_command +118,214182,"train_dynamics.py",2988,0,"",python,selection_command +119,215003,"train_dynamics.py",2992,0,"",python,selection_command +120,215250,"train_dynamics.py",3000,0,"",python,selection_command +121,215283,"train_dynamics.py",3002,0,"",python,selection_command +122,215321,"train_dynamics.py",3005,0,"",python,selection_command +123,215353,"train_dynamics.py",3006,0,"",python,selection_command +124,215386,"train_dynamics.py",3020,0,"",python,selection_command +125,215420,"train_dynamics.py",3021,0,"",python,selection_command +126,221808,"train_dynamics.py",2988,0,"",python,selection_command +127,221880,"train_dynamics.py",2992,0,"",python,selection_command +128,222128,"train_dynamics.py",3000,0,"",python,selection_command +129,222161,"train_dynamics.py",3002,0,"",python,selection_command +130,222190,"train_dynamics.py",3005,0,"",python,selection_command +131,222224,"train_dynamics.py",3006,0,"",python,selection_command +132,222259,"train_dynamics.py",3020,0,"",python,selection_command +133,222293,"train_dynamics.py",3021,0,"",python,selection_command +134,223174,"train_dynamics.py",2988,0,"",python,selection_command +135,223299,"train_dynamics.py",2992,0,"",python,selection_command +136,223551,"train_dynamics.py",3000,0,"",python,selection_command +137,223581,"train_dynamics.py",3002,0,"",python,selection_command +138,223614,"train_dynamics.py",3005,0,"",python,selection_command +139,223648,"train_dynamics.py",3006,0,"",python,selection_command +140,223681,"train_dynamics.py",3020,0,"",python,selection_command +141,223715,"train_dynamics.py",3021,0,"",python,selection_command +142,223749,"train_dynamics.py",3037,0,"",python,selection_command +143,224380,"train_dynamics.py",3021,0,"",python,selection_command +144,225595,"train_dynamics.py",2988,0,"",python,selection_command +145,225681,"train_dynamics.py",2992,0,"",python,selection_command +146,225938,"train_dynamics.py",3000,0,"",python,selection_command +147,225966,"train_dynamics.py",3002,0,"",python,selection_command +148,225999,"train_dynamics.py",3005,0,"",python,selection_command +149,226037,"train_dynamics.py",3006,0,"",python,selection_command +150,226070,"train_dynamics.py",3020,0,"",python,selection_command +151,226103,"train_dynamics.py",3021,0,"",python,selection_command +152,251190,"train_dynamics.py",2006,0,"",python,selection_command +153,251439,"train_dynamics.py",2045,0,"",python,selection_command +154,251470,"train_dynamics.py",2075,0,"",python,selection_command +155,251506,"train_dynamics.py",2091,0,"",python,selection_command +156,251539,"train_dynamics.py",2107,0,"",python,selection_command +157,251574,"train_dynamics.py",2130,0,"",python,selection_command +158,251760,"train_dynamics.py",2204,0,"",python,selection_command +159,252115,"train_dynamics.py",2210,0,"",python,selection_command +160,256408,"train_dynamics.py",2215,0,"",python,selection_command +161,256583,"train_dynamics.py",2217,0,"",python,selection_command +162,262598,"models/dynamics.py",0,0,"",python,tab +163,263846,"models/dynamics.py",416,0,"",python,selection_keyboard +164,271113,"models/dynamics.py",1359,0,"",python,selection_command +165,273515,"models/dynamics.py",1373,0,"",python,selection_command +166,273673,"models/dynamics.py",1389,0,"",python,selection_command diff --git a/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-ef5ea013-ac2b-459c-8783-a7b025d58a391754900011518-2025_08_11-10.13.34.249/source.csv b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-ef5ea013-ac2b-459c-8783-a7b025d58a391754900011518-2025_08_11-10.13.34.249/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..6776be94f3a0690b98c93b5e1edbee3808bc4bd7 --- /dev/null +++ b/4de8d861ed2563988d5f1871647ebc5fe70861b32d24a4b32f9363518653a328/crowd-code-ef5ea013-ac2b-459c-8783-a7b025d58a391754900011518-2025_08_11-10.13.34.249/source.csv @@ -0,0 +1,3025 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,2,"train_lam.py",0,0,"from dataclasses import dataclass, field\nimport os\nfrom typing import cast\n\nimport einops\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax.checkpoint as ocp\nimport numpy as np\nimport dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\nimport grain\nimport flax.nnx as nnx\n\nfrom models.lam import LatentActionModel\nfrom utils.dataloader import get_dataloader\nfrom utils.lr_utils import get_lr_schedule\nfrom utils.parameter_utils import count_parameters_by_component\n\n\n@dataclass\nclass Args:\n # Experiment\n num_steps: int = 200_000\n seed: int = 0\n seq_len: int = 16\n image_channels: int = 3\n image_height: int = 90\n image_width: int = 160\n data_dir: str = """"\n save_ckpt: bool = False\n restore_ckpt: bool = False\n # Optimization\n batch_size: int = 36\n vq_beta: float = 0.25\n init_lr: float = 0.0\n max_lr: float = 3e-5\n decay_end: float = 0.0\n wsd_decay_steps: int = (\n 10000 # NOTE: wsd_decay_steps will only be used when using a wsd-schedule\n )\n warmup_steps: int = 5000\n lr_schedule: str = ""wsd"" # supported options: wsd, cos\n vq_reset_thresh: int = 50\n # LAM\n model_dim: int = 512\n ffn_dim: int = 2048\n latent_dim: int = 32\n num_latents: int = 6\n patch_size: int = 16\n num_blocks: int = 4\n num_heads: int = 8\n dropout: float = 0.0\n codebook_dropout: float = 0.0\n param_dtype = jnp.float32\n dtype = jnp.bfloat16\n # Logging\n log: bool = False\n entity: str = """"\n project: str = """"\n name: str = ""train_lam""\n tags: list[str] = field(default_factory=lambda: [""lam""])\n log_interval: int = 5\n log_image_interval: int = 250\n ckpt_dir: str = """"\n log_checkpoint_interval: int = 10000\n log_checkpoint_keep_period: int = 20000\n wandb_id: str = """"\n use_flash_attention: bool = True\n\n\nargs = tyro.cli(Args)\n\n\ndef lam_loss_fn(\n model: LatentActionModel, inputs: dict\n) -> tuple[jax.Array, tuple[jax.Array, jax.Array, dict]]:\n # --- Compute loss ---\n gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n inputs[""videos""] = gt.astype(args.dtype)\n model.train()\n outputs = model(inputs, training=True)\n outputs[""recon""] = outputs[""recon""].astype(jnp.float32)\n gt_future_frames = gt[:, 1:]\n mse = jnp.square(gt_future_frames - outputs[""recon""]).mean()\n q_loss = jnp.square(jax.lax.stop_gradient(outputs[""emb""]) - outputs[""z""]).mean()\n commitment_loss = jnp.square(\n outputs[""emb""] - jax.lax.stop_gradient(outputs[""z""])\n ).mean()\n loss = mse + q_loss + args.vq_beta * commitment_loss\n\n # --- Compute validation metrics ---\n gt = gt_future_frames.clip(0, 1).reshape(-1, *gt_future_frames.shape[2:])\n recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n psnr = jnp.asarray(pix.psnr(gt, recon)).mean()\n ssim = jnp.asarray(pix.ssim(gt, recon)).mean()\n count_fn = jax.vmap(lambda i: (outputs[""indices""] == i).sum())\n index_counts = count_fn(jnp.arange(args.num_latents))\n metrics = dict(\n loss=loss,\n mse=mse,\n q_loss=q_loss,\n commitment_loss=commitment_loss,\n psnr=psnr,\n ssim=ssim,\n codebook_usage=(index_counts != 0).mean(),\n )\n return loss, (outputs[""recon""], index_counts, metrics)\n\n\n@nnx.jit\ndef train_step(\n lam: LatentActionModel,\n optimizer: nnx.Optimizer,\n inputs: dict,\n action_last_active: jax.Array,\n rng: jax.Array,\n) -> tuple[jax.Array, jax.Array, jax.Array, dict]:\n def loss_fn(\n model: LatentActionModel,\n ) -> tuple[jax.Array, tuple[jax.Array, jax.Array, dict]]:\n return lam_loss_fn(model, inputs)\n\n # --- Update model ---\n (loss, (recon, idx_counts, metrics)), grads = nnx.value_and_grad(\n loss_fn, has_aux=True\n )(lam)\n optimizer.update(grads)\n\n # --- Reset inactive latent actions ---\n codebook = lam.vq.codebook\n num_codes = len(codebook)\n active_codes = idx_counts != 0.0\n action_last_active = jnp.where(active_codes, 0, action_last_active + 1)\n p_code = active_codes / active_codes.sum()\n reset_idxs = jax.random.choice(rng, num_codes, shape=(num_codes,), p=p_code)\n do_reset = action_last_active >= args.vq_reset_thresh\n new_codebook = jnp.where(\n jnp.expand_dims(do_reset, -1), codebook[reset_idxs], codebook.value\n )\n lam.vq.codebook.value = new_codebook\n action_last_active = jnp.where(do_reset, 0, action_last_active)\n return loss, recon, action_last_active, metrics\n\n\nif __name__ == ""__main__"":\n jax.distributed.initialize()\n num_devices = jax.device_count()\n if num_devices == 0:\n raise ValueError(""No JAX devices found."")\n print(f""Running on {num_devices} devices."")\n\n if args.batch_size % num_devices != 0:\n raise ValueError(\n f""Global batch size {args.batch_size} must be divisible by ""\n f""number of devices {num_devices}.""\n )\n\n per_device_batch_size_for_init = args.batch_size // num_devices\n\n rng = jax.random.key(args.seed)\n\n # --- Initialize model ---\n rng, _rng = jax.random.split(rng)\n rngs = nnx.Rngs(_rng)\n lam = LatentActionModel(\n in_dim=args.image_channels,\n model_dim=args.model_dim,\n ffn_dim=args.ffn_dim,\n latent_dim=args.latent_dim,\n num_latents=args.num_latents,\n patch_size=args.patch_size,\n num_blocks=args.num_blocks,\n num_heads=args.num_heads,\n dropout=args.dropout,\n codebook_dropout=args.codebook_dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n )\n\n # Count parameters\n _, params, _ = nnx.split(lam, nnx.Param, ...)\n param_counts = count_parameters_by_component(params)\n\n if args.log and jax.process_index() == 0:\n wandb_init_kwargs = {\n ""entity"": args.entity,\n ""project"": args.project,\n ""name"": args.name,\n ""tags"": args.tags,\n ""group"": ""debug"",\n ""config"": args,\n }\n\n if args.wandb_id:\n wandb_init_kwargs.update(\n {\n ""id"": args.wandb_id,\n ""resume"": ""allow"",\n }\n )\n wandb.init(**wandb_init_kwargs)\n\n wandb.config.update({""model_param_count"": param_counts})\n\n print(""Parameter counts:"")\n print(param_counts)\n\n # --- Initialize optimizer ---\n lr_schedule = get_lr_schedule(\n args.lr_schedule,\n args.init_lr,\n args.max_lr,\n args.decay_end,\n args.num_steps,\n args.warmup_steps,\n args.wsd_decay_steps,\n )\n tx = optax.adamw(\n learning_rate=lr_schedule,\n b1=0.9,\n b2=0.9,\n weight_decay=1e-4,\n mu_dtype=args.dtype,\n )\n optimizer = nnx.Optimizer(lam, tx)\n\n # FIXME: switch to create_hybrid_device_mesh for runs spanning multiple nodes\n device_mesh_arr = create_device_mesh((num_devices,))\n mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n\n replicated_sharding = NamedSharding(mesh, PartitionSpec())\n videos_sharding = NamedSharding(mesh, PartitionSpec(""data"", None, None, None, None))\n\n model_state = nnx.state(optimizer.model)\n model_sharded_state = jax.lax.with_sharding_constraint(\n model_state, replicated_sharding\n )\n nnx.update(optimizer.model, model_sharded_state)\n optimizer_state = nnx.state(optimizer, nnx.optimizer.OptState)\n optimizer_sharded_state = jax.lax.with_sharding_constraint(\n optimizer_state, replicated_sharding\n )\n nnx.update(optimizer, optimizer_sharded_state)\n\n # --- Initialize checkpoint manager ---\n step = 0\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointSave,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointRestore,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n save_interval_steps=args.log_checkpoint_interval,\n max_to_keep=3,\n keep_period=args.log_checkpoint_keep_period,\n step_format_fixed_length=6,\n cleanup_tmp_directories=True,\n )\n\n checkpoint_manager = ocp.CheckpointManager(\n args.ckpt_dir,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n\n # --- Create DataLoaderIterator from dataloader ---\n image_shape = (args.image_height, args.image_width, args.image_channels)\n array_record_files = [\n os.path.join(args.data_dir, x)\n for x in os.listdir(args.data_dir)\n if x.endswith("".array_record"")\n ]\n grain_dataloader = get_dataloader(\n array_record_files,\n args.seq_len,\n # NOTE: We deliberately pass the global batch size\n # The dataloader shards the dataset across all processes\n args.batch_size,\n *image_shape,\n num_workers=8,\n prefetch_buffer_size=1,\n seed=args.seed,\n )\n initial_state = grain_dataloader._create_initial_state()\n grain_iterator = grain.DataLoaderIterator(grain_dataloader, initial_state)\n\n # --- Restore checkpoint ---\n if args.restore_ckpt:\n abstract_optimizer = nnx.eval_shape(lambda: optimizer)\n abstract_optimizer_state = nnx.state(abstract_optimizer)\n restored = checkpoint_manager.restore(\n checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore(abstract_optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointRestore(grain_iterator), # type: ignore\n ),\n )\n restored_optimizer_state = restored[""model_state""]\n nnx.update(optimizer, restored_optimizer_state)\n grain_iterator = restored[""dataloader_state""]\n step = checkpoint_manager.latest_step() or 0\n print(f""Restored dataloader and model state from step {step}"")\n\n # --- TRAIN LOOP ---\n dataloader = (\n jax.make_array_from_process_local_data(videos_sharding, elem)\n for elem in grain_iterator\n )\n print(f""Starting training from step {step}..."")\n action_last_active = jnp.zeros(args.num_latents, dtype=jnp.int32)\n while step < args.num_steps:\n for videos in dataloader:\n # --- Train step ---\n rng, _rng = jax.random.split(rng)\n\n inputs = dict(videos=videos, rng=_rng)\n rng, _rng = jax.random.split(rng)\n loss, recon, action_last_active, metrics = train_step(\n lam, optimizer, inputs, action_last_active, _rng\n )\n metrics[""lr""] = lr_schedule(step)\n print(f""Step {step}, loss: {loss}"")\n step += 1\n\n # --- Logging ---\n if args.log:\n if step % args.log_interval == 0 and jax.process_index() == 0:\n wandb.log(\n {\n ""loss"": loss,\n ""step"": step,\n **metrics,\n }\n )\n if step % args.log_image_interval == 0:\n gt_seq = inputs[""videos""][0, 1:].astype(jnp.float32) / 255.0\n recon_seq = recon[0].clip(0, 1)\n comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n comparison_seq = einops.rearrange(\n comparison_seq * 255, ""t h w c -> h (t w) c""\n )\n # NOTE: Process-dependent control flow deliberately happens\n # after indexing operation since it must not contain code\n # sections that lead to cross-accelerator communication.\n if jax.process_index() == 0:\n log_images = dict(\n image=wandb.Image(np.asarray(gt_seq[0])),\n recon=wandb.Image(np.asarray(recon_seq[0])),\n true_vs_recon=wandb.Image(\n np.asarray(comparison_seq.astype(np.uint8))\n ),\n )\n wandb.log(log_images)\n # --- Checkpointing ---\n if args.save_ckpt and step % args.log_checkpoint_interval == 0:\n optimizer_state = nnx.state(optimizer)\n checkpoint_manager.save(\n step,\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeSave(optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointSave( # type: ignore\n grain_iterator # type: ignore\n ),\n ),\n )\n print(f""Saved checkpoint at step {step}"")\n if step >= args.num_steps:\n break\n\n checkpoint_manager.close()\n",python,tab +2,60,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"10:13:34 AM [info] Activating crowd-code\n10:13:34 AM [info] Recording started\n10:13:34 AM [info] Initializing git provider using file system watchers...\n10:13:34 AM [info] Git repository found\n10:13:34 AM [info] Git provider initialized successfully\n",Log,tab +3,142,"extension-output-pdoom-org.crowd-code-#1-crowd-code",250,0,"10:13:34 AM [info] Initial git state: [object Object]\n",Log,content +4,1584,"train_lam.py",0,0,"",python,tab +5,12407,"train_tokenizer.py",0,0,"from dataclasses import dataclass, field\nimport os\nfrom typing import cast\n\nimport einops\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax.checkpoint as ocp\nimport numpy as np\nimport dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\nimport grain\nimport flax.nnx as nnx\n\nfrom models.tokenizer import TokenizerVQVAE\nfrom utils.dataloader import get_dataloader\nfrom utils.lr_utils import get_lr_schedule\nfrom utils.parameter_utils import count_parameters_by_component\n\n\n@dataclass\nclass Args:\n # Experiment\n num_steps: int = 300_000\n seed: int = 0\n seq_len: int = 16\n image_channels: int = 3\n image_height: int = 90\n image_width: int = 160\n data_dir: str = """"\n save_ckpt: bool = False\n restore_ckpt: bool = False\n # Optimization\n vq_beta: float = 0.25\n batch_size: int = 48\n init_lr: float = 0.0\n max_lr: float = 3e-4\n decay_end: float = 0.0\n wsd_decay_steps: int = (\n 20000 # NOTE: wsd_decay_steps will only be used when using a wsd-schedule\n )\n lr_schedule: str = ""wsd"" # supported options: wsd, cos\n warmup_steps: int = 10000\n # Tokenizer\n model_dim: int = 512\n ffn_dim: int = 2048\n latent_dim: int = 32\n num_latents: int = 1024\n patch_size: int = 4\n num_blocks: int = 4\n num_heads: int = 8\n dropout: float = 0.0\n codebook_dropout: float = 0.01\n param_dtype = jnp.float32\n dtype = jnp.bfloat16\n # Logging\n log: bool = False\n entity: str = """"\n project: str = """"\n name: str = ""train_tokenizer""\n tags: list[str] = field(default_factory=lambda: [""tokenizer""])\n log_interval: int = 5\n log_image_interval: int = 250\n ckpt_dir: str = """"\n log_checkpoint_interval: int = 10000\n log_checkpoint_keep_period: int = 20000\n log_gradients: bool = False\n wandb_id: str = """"\n use_flash_attention: bool = True\n\n\nargs = tyro.cli(Args)\n\n\ndef tokenizer_loss_fn(\n model: TokenizerVQVAE, inputs: dict\n) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n # --- Compute loss ---\n # FIXME (f.srambical): Can we even do native int8 training without casting the video at all?\n # FIXME (f.srambical): If the tokenizer is the reason for the dynamics model being memory-bound,\n # should we at least train the tokenizer natively in int8?\n gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n inputs[""videos""] = gt.astype(args.dtype)\n model.train()\n outputs = model(inputs, training=True)\n outputs[""recon""] = outputs[""recon""].astype(jnp.float32)\n mse = jnp.square(gt - outputs[""recon""]).mean()\n q_loss = jnp.square(jax.lax.stop_gradient(outputs[""emb""]) - outputs[""z""]).mean()\n commitment_loss = jnp.square(\n outputs[""emb""] - jax.lax.stop_gradient(outputs[""z""])\n ).mean()\n loss = mse + q_loss + args.vq_beta * commitment_loss\n\n # --- Compute validation metrics ---\n gt = gt.clip(0, 1).reshape(-1, *gt.shape[2:])\n recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n psnr = jnp.asarray(pix.psnr(gt, recon)).mean()\n ssim = jnp.asarray(pix.ssim(gt, recon)).mean()\n _, index_counts = jnp.unique_counts(\n jnp.ravel(outputs[""indices""]), size=args.num_latents, fill_value=0\n )\n codebook_usage = (index_counts != 0).mean()\n metrics = dict(\n loss=loss,\n mse=mse,\n q_loss=q_loss,\n commitment_loss=commitment_loss,\n psnr=psnr,\n ssim=ssim,\n codebook_usage=codebook_usage,\n )\n return loss, (outputs[""recon""], metrics)\n\n\n@nnx.jit\ndef train_step(\n tokenizer: TokenizerVQVAE, optimizer: nnx.Optimizer, inputs: dict\n) -> tuple[jax.Array, jax.Array, dict]:\n def loss_fn(model: TokenizerVQVAE) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n return tokenizer_loss_fn(model, inputs)\n\n (loss, (recon, metrics)), grads = nnx.value_and_grad(loss_fn, has_aux=True)(\n tokenizer\n )\n optimizer.update(grads)\n if args.log_gradients:\n metrics[""encoder_gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""encoder""]\n )\n metrics[""vq_gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""vq""]\n )\n metrics[""decoder_gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""decoder""]\n )\n return loss, recon, metrics\n\n\nif __name__ == ""__main__"":\n jax.distributed.initialize()\n num_devices = jax.device_count()\n if num_devices == 0:\n raise ValueError(""No JAX devices found."")\n print(f""Running on {num_devices} devices."")\n\n if args.batch_size % num_devices != 0:\n raise ValueError(\n f""Global batch size {args.batch_size} must be divisible by ""\n f""number of devices {num_devices}.""\n )\n\n per_device_batch_size_for_init = args.batch_size // num_devices\n\n rng = jax.random.key(args.seed)\n\n # --- Initialize model ---\n rng, _rng = jax.random.split(rng)\n rngs = nnx.Rngs(_rng)\n tokenizer = TokenizerVQVAE(\n in_dim=args.image_channels,\n model_dim=args.model_dim,\n ffn_dim=args.ffn_dim,\n latent_dim=args.latent_dim,\n num_latents=args.num_latents,\n patch_size=args.patch_size,\n num_blocks=args.num_blocks,\n num_heads=args.num_heads,\n dropout=args.dropout,\n codebook_dropout=args.codebook_dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n )\n\n _, params, _ = nnx.split(tokenizer, nnx.Param, ...)\n param_counts = count_parameters_by_component(params)\n\n if args.log and jax.process_index() == 0:\n wandb_init_kwargs = {\n ""entity"": args.entity,\n ""project"": args.project,\n ""name"": args.name,\n ""tags"": args.tags,\n ""group"": ""debug"",\n ""config"": args,\n }\n\n if args.wandb_id:\n wandb_init_kwargs.update(\n {\n ""id"": args.wandb_id,\n ""resume"": ""allow"",\n }\n )\n wandb.init(**wandb_init_kwargs)\n\n wandb.config.update({""model_param_count"": param_counts})\n\n print(""Parameter counts:"")\n print(param_counts)\n\n # --- Initialize optimizer ---\n lr_schedule = get_lr_schedule(\n args.lr_schedule,\n args.init_lr,\n args.max_lr,\n args.decay_end,\n args.num_steps,\n args.warmup_steps,\n args.wsd_decay_steps,\n )\n tx = optax.adamw(\n learning_rate=lr_schedule,\n b1=0.9,\n b2=0.9,\n weight_decay=1e-4,\n mu_dtype=args.dtype,\n )\n optimizer = nnx.Optimizer(tokenizer, tx)\n\n # FIXME: switch to create_hybrid_device_mesh for runs spanning multiple nodes\n device_mesh_arr = create_device_mesh((num_devices,))\n mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n\n replicated_sharding = NamedSharding(mesh, PartitionSpec())\n videos_sharding = NamedSharding(mesh, PartitionSpec(""data"", None, None, None, None))\n\n model_state = nnx.state(optimizer.model)\n model_sharded_state = jax.lax.with_sharding_constraint(\n model_state, replicated_sharding\n )\n nnx.update(optimizer.model, model_sharded_state)\n optimizer_state = nnx.state(optimizer, nnx.optimizer.OptState)\n optimizer_sharded_state = jax.lax.with_sharding_constraint(\n optimizer_state, replicated_sharding\n )\n nnx.update(optimizer, optimizer_sharded_state)\n\n # --- Initialize checkpoint manager ---\n step = 0\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointSave,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointRestore,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n save_interval_steps=args.log_checkpoint_interval,\n max_to_keep=3,\n keep_period=args.log_checkpoint_keep_period,\n step_format_fixed_length=6,\n cleanup_tmp_directories=True,\n )\n\n checkpoint_manager = ocp.CheckpointManager(\n args.ckpt_dir,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n\n # --- Create DataLoaderIterator from dataloader ---\n image_shape = (args.image_height, args.image_width, args.image_channels)\n array_record_files = [\n os.path.join(args.data_dir, x)\n for x in os.listdir(args.data_dir)\n if x.endswith("".array_record"")\n ]\n grain_dataloader = get_dataloader(\n array_record_files,\n args.seq_len,\n # NOTE: We deliberately pass the global batch size\n # The dataloader shards the dataset across all processes\n args.batch_size,\n *image_shape,\n num_workers=8,\n prefetch_buffer_size=1,\n seed=args.seed,\n )\n initial_state = grain_dataloader._create_initial_state()\n grain_iterator = grain.DataLoaderIterator(grain_dataloader, initial_state)\n\n # --- Restore checkpoint ---\n if args.restore_ckpt:\n abstract_optimizer = nnx.eval_shape(lambda: optimizer)\n abstract_optimizer_state = nnx.state(abstract_optimizer)\n restored = checkpoint_manager.restore(\n checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore(abstract_optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointRestore(grain_iterator), # type: ignore\n ),\n )\n restored_optimizer_state = restored[""model_state""]\n nnx.update(optimizer, restored_optimizer_state)\n grain_iterator = restored[""dataloader_state""]\n step = checkpoint_manager.latest_step() or 0\n print(f""Restored dataloader and model state from step {step}"")\n\n # --- TRAIN LOOP ---\n dataloader = (\n jax.make_array_from_process_local_data(videos_sharding, elem)\n for elem in grain_iterator\n )\n print(f""Starting training from step {step}..."")\n while step < args.num_steps:\n for videos in dataloader:\n # --- Train step ---\n inputs = dict(videos=videos)\n loss, recon, metrics = train_step(tokenizer, optimizer, inputs)\n metrics[""lr""] = lr_schedule(step)\n print(f""Step {step}, loss: {loss}"")\n step += 1\n\n # --- Logging ---\n if args.log:\n if step % args.log_interval == 0 and jax.process_index() == 0:\n wandb.log(\n {\n ""loss"": loss,\n ""step"": step,\n **metrics,\n }\n )\n if step % args.log_image_interval == 0:\n gt_seq = inputs[""videos""][0].astype(jnp.float32) / 255.0\n recon_seq = recon[0].clip(0, 1)\n comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n comparison_seq = einops.rearrange(\n comparison_seq * 255, ""t h w c -> h (t w) c""\n )\n # NOTE: Process-dependent control flow deliberately happens\n # after indexing operation since it must not contain code\n # sections that lead to cross-accelerator communication.\n if jax.process_index() == 0:\n log_images = dict(\n image=wandb.Image(np.asarray(gt_seq[0])),\n recon=wandb.Image(np.asarray(recon_seq[0])),\n true_vs_recon=wandb.Image(\n np.asarray(comparison_seq.astype(np.uint8))\n ),\n )\n wandb.log(log_images)\n # --- Checkpointing ---\n if args.save_ckpt and step % args.log_checkpoint_interval == 0:\n optimizer_state = nnx.state(optimizer)\n checkpoint_manager.save(\n step,\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeSave(optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointSave( # type: ignore\n grain_iterator # type: ignore\n ),\n ),\n )\n print(f""Saved checkpoint at step {step}"")\n if step >= args.num_steps:\n break\n\n checkpoint_manager.close()\n",python,tab +6,16207,"train_dynamics.py",0,0,"from dataclasses import dataclass, field\nimport os\nfrom typing import cast\n\nimport einops\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax.checkpoint as ocp\nimport numpy as np\nimport dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\nimport grain\nimport flax.nnx as nnx\n\nfrom genie import Genie, restore_genie_components\nfrom utils.dataloader import get_dataloader\nfrom utils.lr_utils import get_lr_schedule\nfrom utils.parameter_utils import count_parameters_by_component\n\n\n@dataclass\nclass Args:\n # Experiment\n num_steps: int = 200_000\n seed: int = 0\n seq_len: int = 16\n image_channels: int = 3\n image_height: int = 90\n image_width: int = 160\n data_dir: str = """"\n save_ckpt: bool = False\n restore_ckpt: bool = False\n # Optimization\n batch_size: int = 36\n init_lr: float = 0.0\n max_lr: float = 3e-5\n decay_end: float = 0.0\n wsd_decay_steps: int = (\n 10000 # NOTE: wsd_decay_steps will only be used when using a wsd-schedule\n )\n warmup_steps: int = 5000\n lr_schedule: str = ""wsd"" # supported options: wsd, cos\n # Tokenizer\n tokenizer_dim: int = 512\n tokenizer_ffn_dim: int = 2048\n latent_patch_dim: int = 32\n num_patch_latents: int = 1024\n patch_size: int = 4\n tokenizer_num_blocks: int = 4\n tokenizer_num_heads: int = 8\n tokenizer_checkpoint: str = """"\n # LAM\n lam_dim: int = 512\n lam_ffn_dim: int = 2048\n latent_action_dim: int = 32\n num_latent_actions: int = 6\n lam_patch_size: int = 16\n lam_num_blocks: int = 4\n lam_num_heads: int = 8\n lam_checkpoint: str = """"\n # Dynamics\n dyna_dim: int = 512\n dyna_ffn_dim: int = 2048\n dyna_num_blocks: int = 6\n dyna_num_heads: int = 8\n dropout: float = 0.0\n mask_limit: float = 0.5\n param_dtype = jnp.float32\n dtype = jnp.bfloat16\n use_flash_attention: bool = True\n # Logging\n log: bool = False\n entity: str = """"\n project: str = """"\n name: str = ""train_dynamics""\n tags: list[str] = field(default_factory=lambda: [""dynamics""])\n log_interval: int = 5\n log_image_interval: int = 250\n ckpt_dir: str = """"\n log_checkpoint_interval: int = 25000\n log_checkpoint_keep_period: int = 20000\n log_gradients: bool = False\n wandb_id: str = """"\n\n\nargs = tyro.cli(Args)\n\n\ndef dynamics_loss_fn(\n model: Genie, inputs: dict\n) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n """"""Compute masked dynamics loss""""""\n gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n inputs[""videos""] = gt.astype(args.dtype)\n model.train()\n outputs = model(inputs, training=True)\n mask = outputs[""mask""]\n outputs[""token_logits""] = outputs[""token_logits""].astype(jnp.float32)\n ce_loss = optax.softmax_cross_entropy_with_integer_labels(\n outputs[""token_logits""], outputs[""video_tokens""]\n )\n ce_loss = (mask * ce_loss).sum() / mask.sum()\n acc = outputs[""token_logits""].argmax(-1) == outputs[""video_tokens""]\n acc = (mask * acc).sum() / mask.sum()\n select_probs = jax.nn.softmax(outputs[""token_logits""])\n gt = gt.clip(0, 1).reshape(-1, *gt.shape[2:])\n recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n psnr = jnp.asarray(pix.psnr(gt, recon)).mean()\n ssim = jnp.asarray(pix.ssim(gt, recon)).mean()\n _, index_counts_lam = jnp.unique_counts(\n jnp.ravel(outputs[""lam_indices""]), size=args.num_latent_actions, fill_value=0\n )\n _, index_counts_tokenizer = jnp.unique_counts(\n jnp.ravel(outputs[""video_tokens""]), size=args.num_patch_latents, fill_value=0\n )\n codebook_usage_lam = (index_counts_lam != 0).mean()\n codebook_usage_tokenizer = (index_counts_tokenizer != 0).mean()\n metrics = dict(\n cross_entropy_loss=ce_loss,\n masked_token_accuracy=acc,\n select_logit=outputs[""token_logits""].max(-1).mean(),\n select_p=select_probs.max(-1).mean(),\n entropy=jax.scipy.special.entr(select_probs).sum(-1).mean(),\n psnr=psnr,\n ssim=ssim,\n codebook_usage_lam=codebook_usage_lam,\n codebook_usage_tokenizer=codebook_usage_tokenizer,\n )\n return ce_loss, (outputs[""recon""], metrics)\n\n\n@nnx.jit\ndef train_step(\n model: Genie, optimizer: nnx.Optimizer, inputs: dict\n) -> tuple[jax.Array, jax.Array, dict]:\n """"""Update state and compute metrics""""""\n\n def loss_fn(model: Genie) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n return dynamics_loss_fn(model, inputs)\n\n (loss, (recon, metrics)), grads = nnx.value_and_grad(loss_fn, has_aux=True)(model)\n optimizer.update(grads)\n if args.log_gradients:\n metrics[""gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""dynamics""]\n )\n return loss, recon, metrics\n\n\nif __name__ == ""__main__"":\n jax.distributed.initialize()\n num_devices = jax.device_count()\n if num_devices == 0:\n raise ValueError(""No JAX devices found."")\n print(f""Running on {num_devices} devices."")\n\n if args.batch_size % num_devices != 0:\n raise ValueError(\n f""Global batch size {args.batch_size} must be divisible by ""\n f""number of devices {num_devices}.""\n )\n\n per_device_batch_size_for_init = args.batch_size // num_devices\n\n rng = jax.random.key(args.seed)\n\n # --- Initialize model ---\n rng, _rng = jax.random.split(rng)\n rngs = nnx.Rngs(_rng)\n genie = Genie(\n # Tokenizer\n in_dim=args.image_channels,\n tokenizer_dim=args.tokenizer_dim,\n tokenizer_ffn_dim=args.tokenizer_ffn_dim,\n latent_patch_dim=args.latent_patch_dim,\n num_patch_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n tokenizer_num_blocks=args.tokenizer_num_blocks,\n tokenizer_num_heads=args.tokenizer_num_heads,\n # LAM\n lam_dim=args.lam_dim,\n lam_ffn_dim=args.lam_ffn_dim,\n latent_action_dim=args.latent_action_dim,\n num_latent_actions=args.num_latent_actions,\n lam_patch_size=args.lam_patch_size,\n lam_num_blocks=args.lam_num_blocks,\n lam_num_heads=args.lam_num_heads,\n lam_co_train=not args.lam_checkpoint,\n # Dynamics\n dyna_dim=args.dyna_dim,\n dyna_ffn_dim=args.dyna_ffn_dim,\n dyna_num_blocks=args.dyna_num_blocks,\n dyna_num_heads=args.dyna_num_heads,\n dropout=args.dropout,\n mask_limit=args.mask_limit,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n )\n\n _, params, _ = nnx.split(genie, nnx.Param, ...)\n param_counts = count_parameters_by_component(params)\n\n if args.log and jax.process_index() == 0:\n wandb_init_kwargs = {\n ""entity"": args.entity,\n ""project"": args.project,\n ""name"": args.name,\n ""tags"": args.tags,\n ""group"": ""debug"",\n ""config"": args,\n }\n\n if args.wandb_id:\n wandb_init_kwargs.update(\n {\n ""id"": args.wandb_id,\n ""resume"": ""allow"",\n }\n )\n wandb.init(**wandb_init_kwargs)\n\n wandb.config.update({""model_param_count"": param_counts})\n\n print(""Parameter counts:"")\n print(param_counts)\n\n # --- Initialize optimizer ---\n lr_schedule = get_lr_schedule(\n args.lr_schedule,\n args.init_lr,\n args.max_lr,\n args.decay_end,\n args.num_steps,\n args.warmup_steps,\n args.wsd_decay_steps,\n )\n tx = optax.adamw(\n learning_rate=lr_schedule,\n b1=0.9,\n b2=0.9,\n weight_decay=1e-4,\n mu_dtype=args.dtype,\n )\n optimizer = nnx.Optimizer(genie, tx)\n del genie\n\n # FIXME: switch to create_hybrid_device_mesh for runs spanning multiple nodes\n device_mesh_arr = create_device_mesh((num_devices,))\n mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n\n replicated_sharding = NamedSharding(mesh, PartitionSpec())\n videos_sharding = NamedSharding(mesh, PartitionSpec(""data"", None, None, None, None))\n\n model_state = nnx.state(optimizer.model)\n model_sharded_state = jax.lax.with_sharding_constraint(\n model_state, replicated_sharding\n )\n nnx.update(optimizer.model, model_sharded_state)\n optimizer_state = nnx.state(optimizer, nnx.optimizer.OptState)\n optimizer_sharded_state = jax.lax.with_sharding_constraint(\n optimizer_state, replicated_sharding\n )\n nnx.update(optimizer, optimizer_sharded_state)\n\n # --- Initialize checkpoint manager ---\n step = 0\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointSave,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n handler_registry.add(\n ""dataloader_state"",\n grain.checkpoint.CheckpointRestore,\n cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n save_interval_steps=args.log_checkpoint_interval,\n max_to_keep=3,\n keep_period=args.log_checkpoint_keep_period,\n step_format_fixed_length=6,\n cleanup_tmp_directories=True,\n )\n\n checkpoint_manager = ocp.CheckpointManager(\n args.ckpt_dir,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n\n # --- Create DataLoaderIterator from dataloader ---\n image_shape = (args.image_height, args.image_width, args.image_channels)\n array_record_files = [\n os.path.join(args.data_dir, x)\n for x in os.listdir(args.data_dir)\n if x.endswith("".array_record"")\n ]\n grain_dataloader = get_dataloader(\n array_record_files,\n args.seq_len,\n # NOTE: We deliberately pass the global batch size\n # The dataloader shards the dataset across all processes\n args.batch_size,\n *image_shape,\n num_workers=8,\n prefetch_buffer_size=1,\n seed=args.seed,\n )\n initial_state = grain_dataloader._create_initial_state()\n grain_iterator = grain.DataLoaderIterator(grain_dataloader, initial_state)\n\n # --- Restore checkpoint ---\n if args.restore_ckpt:\n abstract_optimizer = nnx.eval_shape(lambda: optimizer)\n abstract_optimizer_state = nnx.state(abstract_optimizer)\n restored = checkpoint_manager.restore(\n checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore(abstract_optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointRestore(grain_iterator), # type: ignore\n ),\n )\n restored_optimizer_state = restored[""model_state""]\n nnx.update(optimizer, restored_optimizer_state)\n grain_iterator = restored[""dataloader_state""]\n step = checkpoint_manager.latest_step() or 0\n print(f""Restored dataloader and model state from step {step}"")\n else:\n # Restore from pre-trained tokenizer (and LAM)\n optimizer = restore_genie_components(optimizer, replicated_sharding, rng, args)\n # NOTE: We have to remove the (unused) tokenizer vq dropout due flax.nnx lazily initializing modules.\n # Specifically, the first dynamics model checkpoint will contain the vq dropout module,\n # but the first full restore will fail due to nnx not initializing the module when\n # dropout is set to 0.0.\n del optimizer.model.tokenizer.vq.drop\n\n # --- TRAIN LOOP ---\n dataloader = (\n jax.make_array_from_process_local_data(videos_sharding, elem)\n for elem in grain_iterator\n )\n print(f""Starting training from step {step}..."")\n while step < args.num_steps:\n for videos in dataloader:\n # --- Train step ---\n rng, _rng_mask = jax.random.split(rng, 2)\n inputs = dict(videos=videos, mask_rng=_rng_mask)\n loss, recon, metrics = train_step(optimizer.model, optimizer, inputs)\n metrics[""lr""] = lr_schedule(step)\n print(f""Step {step}, loss: {loss}"")\n step += 1\n\n # --- Logging ---\n if args.log:\n if step % args.log_interval == 0 and jax.process_index() == 0:\n wandb.log(\n {\n ""loss"": loss,\n ""step"": step,\n **metrics,\n }\n )\n if step % args.log_image_interval == 0:\n gt_seq = inputs[""videos""][0].astype(jnp.float32) / 255.0\n recon_seq = recon[0].clip(0, 1)\n comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n comparison_seq = einops.rearrange(\n comparison_seq * 255, ""t h w c -> h (t w) c""\n )\n if jax.process_index() == 0:\n log_images = dict(\n image=wandb.Image(np.asarray(gt_seq[args.seq_len - 1])),\n recon=wandb.Image(np.asarray(recon_seq[args.seq_len - 1])),\n true_vs_recon=wandb.Image(\n np.asarray(comparison_seq.astype(np.uint8))\n ),\n )\n wandb.log(log_images)\n # --- Checkpointing ---\n if args.save_ckpt and step % args.log_checkpoint_interval == 0:\n optimizer_state = nnx.state(optimizer)\n checkpoint_manager.save(\n step,\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeSave(optimizer_state), # type: ignore\n dataloader_state=grain.checkpoint.CheckpointSave( # type: ignore\n grain_iterator # type: ignore\n ),\n ),\n )\n print(f""Saved checkpoint at step {step}"")\n if step >= args.num_steps:\n break\n\n checkpoint_manager.close()\n",python,tab +7,22697,"train_dynamics.py",10557,0,"",python,selection_mouse +8,186546,"train_dynamics.py",12378,0,"",python,selection_mouse +9,190846,"train_dynamics.py",4277,0,"",python,selection_command +10,221560,"train_dynamics.py",1717,4890," dyna_type: str = ""maskgit"" # supported options: maskgit, causal\n dyna_dim: int = 512\n dyna_ffn_dim: int = 2048\n dyna_num_blocks: int = 6\n dyna_num_heads: int = 8\n dropout: float = 0.0\n mask_limit: float = 0.5\n param_dtype = jnp.float32\n dtype = jnp.bfloat16\n use_flash_attention: bool = True\n # Logging\n log: bool = False\n entity: str = """"\n project: str = """"\n name: str = ""train_dynamics""\n tags: list[str] = field(default_factory=lambda: [""dynamics""])\n log_interval: int = 5\n log_image_interval: int = 250\n ckpt_dir: str = """"\n log_checkpoint_interval: int = 25000\n log_checkpoint_keep_period: int = 20000\n log_gradients: bool = False\n wandb_id: str = """"\n\n\nargs = tyro.cli(Args)\n\n\ndef dynamics_loss_fn(\n model: Genie, inputs: dict\n) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n """"""Compute masked dynamics loss""""""\n gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n inputs[""videos""] = gt.astype(args.dtype)\n model.train()\n outputs = model(inputs, training=True)\n mask = outputs[""mask""]\n outputs[""token_logits""] = outputs[""token_logits""].astype(jnp.float32)\n ce_loss = optax.softmax_cross_entropy_with_integer_labels(\n outputs[""token_logits""], outputs[""video_tokens""]\n )\n ce_loss = (mask * ce_loss).sum() / mask.sum()\n acc = outputs[""token_logits""].argmax(-1) == outputs[""video_tokens""]\n acc = (mask * acc).sum() / mask.sum()\n select_probs = jax.nn.softmax(outputs[""token_logits""])\n gt = gt.clip(0, 1).reshape(-1, *gt.shape[2:])\n recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n psnr = jnp.asarray(pix.psnr(gt, recon)).mean()\n ssim = jnp.asarray(pix.ssim(gt, recon)).mean()\n _, index_counts_lam = jnp.unique_counts(\n jnp.ravel(outputs[""lam_indices""]), size=args.num_latent_actions, fill_value=0\n )\n _, index_counts_tokenizer = jnp.unique_counts(\n jnp.ravel(outputs[""video_tokens""]), size=args.num_patch_latents, fill_value=0\n )\n codebook_usage_lam = (index_counts_lam != 0).mean()\n codebook_usage_tokenizer = (index_counts_tokenizer != 0).mean()\n metrics = dict(\n cross_entropy_loss=ce_loss,\n masked_token_accuracy=acc,\n select_logit=outputs[""token_logits""].max(-1).mean(),\n select_p=select_probs.max(-1).mean(),\n entropy=jax.scipy.special.entr(select_probs).sum(-1).mean(),\n psnr=psnr,\n ssim=ssim,\n codebook_usage_lam=codebook_usage_lam,\n codebook_usage_tokenizer=codebook_usage_tokenizer,\n )\n return ce_loss, (outputs[""recon""], metrics)\n\n\n@nnx.jit\ndef train_step(\n model: Genie, optimizer: nnx.Optimizer, inputs: dict\n) -> tuple[jax.Array, jax.Array, dict]:\n """"""Update state and compute metrics""""""\n\n def loss_fn(model: Genie) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n return dynamics_loss_fn(model, inputs)\n\n (loss, (recon, metrics)), grads = nnx.value_and_grad(loss_fn, has_aux=True)(model)\n optimizer.update(grads)\n if args.log_gradients:\n metrics[""gradients_std/""] = jax.tree.map(\n lambda x: x.std(), grads[""params""][""dynamics""]\n )\n return loss, recon, metrics\n\n\nif __name__ == ""__main__"":\n jax.distributed.initialize()\n num_devices = jax.device_count()\n if num_devices == 0:\n raise ValueError(""No JAX devices found."")\n print(f""Running on {num_devices} devices."")\n\n if args.batch_size % num_devices != 0:\n raise ValueError(\n f""Global batch size {args.batch_size} must be divisible by ""\n f""number of devices {num_devices}.""\n )\n\n per_device_batch_size_for_init = args.batch_size // num_devices\n\n rng = jax.random.key(args.seed)\n\n # --- Initialize model ---\n rng, _rng = jax.random.split(rng)\n rngs = nnx.Rngs(_rng)\n genie = Genie(\n # Tokenizer\n in_dim=args.image_channels,\n tokenizer_dim=args.tokenizer_dim,\n tokenizer_ffn_dim=args.tokenizer_ffn_dim,\n latent_patch_dim=args.latent_patch_dim,\n num_patch_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n tokenizer_num_blocks=args.tokenizer_num_blocks,\n tokenizer_num_heads=args.tokenizer_num_heads,\n # LAM\n lam_dim=args.lam_dim,\n lam_ffn_dim=args.lam_ffn_dim,\n latent_action_dim=args.latent_action_dim,\n num_latent_actions=args.num_latent_actions,\n lam_patch_size=args.lam_patch_size,\n lam_num_blocks=args.lam_num_blocks,\n lam_num_heads=args.lam_num_heads,\n lam_co_train=not args.lam_checkpoint,\n # Dynamics\n dyna_type=args.dyna_type,\n dyna_dim=args.dyna_dim,\n dyna_ffn_dim=args.dyna_ffn_dim,\n dyna_num_blocks=args.dyna_num_blocks,\n dyna_num_heads=args.dyna_num_heads,\n dropout=args.dropout,\n mask_limit=args.mask_limit,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n decode=False,\n",python,content +11,8915782,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab +12,8917431,"train_dynamics.py",0,0,"",python,tab +13,8917453,"TERMINAL",0,0,"",,terminal_focus +14,8917701,"TERMINAL",0,0,"source /Users/franzsrambical/Documents/pdoom/jafar/.venv/bin/activate",,terminal_command +15,8917702,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +16,8918492,"TERMINAL",0,0,"cd ..",,terminal_command +17,8918492,"TERMINAL",0,0,"]633;C% \r \r",,terminal_output +18,8918908,"TERMINAL",0,0,"ls",,terminal_command +19,8918917,"TERMINAL",0,0,"]633;Cbig_vision\t\t\t\t\t\t\t\tminecraft_scraping_list-2.txt\r\nCraftax\t\t\t\t\t\t\t\t\tminecraft_scraping_list.txt\r\ncrowd-code\t\t\t\t\t\t\t\tp-board\r\ndatawhale\t\t\t\t\t\t\t\tpdoom.org\r\nexpanded-verification-signals-warrant-reward-redistribution.pages\treward-redistribution\r\nexpanded-verification-signals-warrant-reward-redistribution.pdf\t\tverification_signals_warrant_reward_redistribution.pages\r\ngsc-grant-introduction.pages\t\t\t\t\t\tverification_signals_warrant_reward_redistribution.pdf\r\ngsc-grant-project-description.pages\t\t\t\t\tverl\r\njafar\t\t\t\t\t\t\t\t\tvid-tag\r\n% \r \r",,terminal_output +20,8929853,"TERMINAL",0,0,"git clone https://github.com/tinygrad/tinygrad.git",,terminal_command +21,8929921,"TERMINAL",0,0,"]633;CCloning into 'tinygrad'...\r\n",,terminal_output +22,8931585,"TERMINAL",0,0,"remote: Enumerating objects: 81085, done.\r\nremote: Counting objects: 0% (1/175)\rremote: Counting objects: 1% (2/175)\rremote: Counting objects: 2% (4/175)\rremote: Counting objects: 3% (6/175)\rremote: Counting objects: 4% (7/175)\rremote: Counting objects: 5% (9/175)\rremote: Counting objects: 6% (11/175)\rremote: Counting objects: 7% (13/175)\rremote: Counting objects: 8% (14/175)\rremote: Counting objects: 9% (16/175)\rremote: Counting objects: 10% (18/175)\rremote: Counting objects: 11% (20/175)\rremote: Counting objects: 12% (21/175)\rremote: Counting objects: 13% (23/175)\rremote: Counting objects: 14% (25/175)\rremote: Counting objects: 15% (27/175)\rremote: Counting objects: 16% (28/175)\rremote: Counting objects: 17% (30/175)\rremote: Counting objects: 18% (32/175)\rremote: Counting objects: 19% (34/175)\rremote: Counting objects: 20% (35/175)\rremote: Counting objects: 21% (37/175)\rremote: Counting objects: 22% (39/175)\rremote: Counting objects: 23% (41/175)\rremote: Counting objects: 24% (42/175)\rremote: Counting objects: 25% (44/175)\rremote: Counting objects: 26% (46/175)\rremote: Counting objects: 27% (48/175)\rremote: Counting objects: 28% (49/175)\rremote: Counting objects: 29% (51/175)\rremote: Counting objects: 30% (53/175)\rremote: Counting objects: 31% (55/175)\rremote: Counting objects: 32% (56/175)\rremote: Counting objects: 33% (58/175)\rremote: Counting objects: 34% (60/175)\rremote: Counting objects: 35% (62/175)\rremote: Counting objects: 36% (63/175)\rremote: Counting objects: 37% (65/175)\rremote: Counting objects: 38% (67/175)\rremote: Counting objects: 39% (69/175)\rremote: Counting objects: 40% (70/175)\rremote: Counting objects: 41% (72/175)\rremote: Counting objects: 42% (74/175)\rremote: Counting objects: 43% (76/175)\rremote: Counting objects: 44% (77/175)\rremote: Counting objects: 45% (79/175)\rremote: Counting objects: 46% (81/175)\rremote: Counting objects: 47% (83/175)\rremote: Counting objects: 48% (84/175)\rremote: Counting objects: 49% (86/175)\rremote: Counting objects: 50% (88/175)\rremote: Counting objects: 51% (90/175)\rremote: Counting objects: 52% (91/175)\rremote: Counting objects: 53% (93/175)\rremote: Counting objects: 54% (95/175)\rremote: Counting objects: 55% (97/175)\rremote: Counting objects: 56% (98/175)\rremote: Counting objects: 57% (100/175)\rremote: Counting objects: 58% (102/175)\rremote: Counting objects: 59% (104/175)\rremote: Counting objects: 60% (105/175)\rremote: Counting objects: 61% (107/175)\rremote: Counting objects: 62% (109/175)\rremote: Counting objects: 63% (111/175)\rremote: Counting objects: 64% (112/175)\rremote: Counting objects: 65% (114/175)\rremote: Counting objects: 66% (116/175)\rremote: Counting objects: 67% (118/175)\rremote: Counting objects: 68% (119/175)\rremote: Counting objects: 69% (121/175)\rremote: Counting objects: 70% (123/175)\rremote: Counting objects: 71% (125/175)\rremote: Counting objects: 72% (126/175)\rremote: Counting objects: 73% (128/175)\rremote: Counting objects: 74% (130/175)\rremote: Counting objects: 75% (132/175)\rremote: Counting objects: 76% (133/175)\rremote: Counting objects: 77% (135/175)\rremote: Counting objects: 78% (137/175)\rremote: Counting objects: 79% (139/175)\rremote: Counting objects: 80% (140/175)\rremote: Counting objects: 81% (142/175)\rremote: Counting objects: 82% (144/175)\rremote: Counting objects: 83% (146/175)\rremote: Counting objects: 84% (147/175)\rremote: Counting objects: 85% (149/175)\rremote: Counting objects: 86% (151/175)\rremote: Counting objects: 87% (153/175)\rremote: Counting objects: 88% (154/175)\rremote: Counting objects: 89% (156/175)\rremote: Counting objects: 90% (158/175)\rremote: Counting objects: 91% (160/175)\rremote: Counting objects: 92% (161/175)\rremote: Counting objects: 93% (163/175)\rremote: Counting objects: 94% (165/175)\rremote: Counting objects: 95% (167/175)\rremote: Counting objects: 96% (168/175)\rremote: Counting objects: 97% (170/175)\rremote: Counting objects: 98% (172/175)\rremote: Counting objects: 99% (174/175)\rremote: Counting objects: 100% (175/175)\rremote: Counting objects: 100% (175/175), done.\r\nremote: Compressing objects: 1% (1/93)\rremote: Compressing objects: 2% (2/93)\rremote: Compressing objects: 3% (3/93)\rremote: Compressing objects: 4% (4/93)\rremote: Compressing objects: 5% (5/93)\rremote: Compressing objects: 6% (6/93)\rremote: Compressing objects: 7% (7/93)\rremote: Compressing objects: 8% (8/93)\rremote: Compressing objects: 9% (9/93)\rremote: Compressing objects: 10% (10/93)\rremote: Compressing objects: 11% (11/93)\rremote: Compressing objects: 12% (12/93)\rremote: Compressing objects: 13% (13/93)\rremote: Compressing objects: 15% (14/93)\rremote: Compressing objects: 16% (15/93)\rremote: Compressing objects: 17% (16/93)\rremote: Compressing objects: 18% (17/93)\rremote: Compressing objects: 19% (18/93)\rremote: Compressing objects: 20% (19/93)\rremote: Compressing objects: 21% (20/93)\rremote: Compressing objects: 22% (21/93)\rremote: Compressing objects: 23% (22/93)\rremote: Compressing objects: 24% (23/93)\rremote: Compressing objects: 25% (24/93)\rremote: Compressing objects: 26% (25/93)\rremote: Compressing objects: 27% (26/93)\rremote: Compressing objects: 29% (27/93)\rremote: Compressing objects: 30% (28/93)\rremote: Compressing objects: 31% (29/93)\rremote: Compressing objects: 32% (30/93)\rremote: Compressing objects: 33% (31/93)\rremote: Compressing objects: 34% (32/93)\rremote: Compressing objects: 35% (33/93)\rremote: Compressing objects: 36% (34/93)\rremote: Compressing objects: 37% (35/93)\rremote: Compressing objects: 38% (36/93)\rremote: Compressing objects: 39% (37/93)\rremote: Compressing objects: 40% (38/93)\rremote: Compressing objects: 41% (39/93)\rremote: Compressing objects: 43% (40/93)\rremote: Compressing objects: 44% (41/93)\rremote: Compressing objects: 45% (42/93)\rremote: Compressing objects: 46% (43/93)\rremote: Compressing objects: 47% (44/93)\rremote: Compressing objects: 48% (45/93)\rremote: Compressing objects: 49% (46/93)\rremote: Compressing objects: 50% (47/93)\rremote: Compressing objects: 51% (48/93)\rremote: Compressing objects: 52% (49/93)\rremote: Compressing objects: 53% (50/93)\rremote: Compressing objects: 54% (51/93)\rremote: Compressing objects: 55% (52/93)\rremote: Compressing objects: 56% (53/93)\rremote: Compressing objects: 58% (54/93)\rremote: Compressing objects: 59% (55/93)\rremote: Compressing objects: 60% (56/93)\rremote: Compressing objects: 61% (57/93)\rremote: Compressing objects: 62% (58/93)\rremote: Compressing objects: 63% (59/93)\rremote: Compressing objects: 64% (60/93)\rremote: Compressing objects: 65% (61/93)\rremote: Compressing objects: 66% (62/93)\rremote: Compressing objects: 67% (63/93)\rremote: Compressing objects: 68% (64/93)\rremote: Compressing objects: 69% (65/93)\rremote: Compressing objects: 70% (66/93)\rremote: Compressing objects: 72% (67/93)\rremote: Compressing objects: 73% (68/93)\rremote: Compressing objects: 74% (69/93)\rremote: Compressing objects: 75% (70/93)\rremote: Compressing objects: 76% (71/93)\rremote: Compressing objects: 77% (72/93)\rremote: Compressing objects: 78% (73/93)\rremote: Compressing objects: 79% (74/93)\rremote: Compressing objects: 80% (75/93)\rremote: Compressing objects: 81% (76/93)\rremote: Compressing objects: 82% (77/93)\rremote: Compressing objects: 83% (78/93)\rremote: Compressing objects: 84% (79/93)\rremote: Compressing objects: 86% (80/93)\rremote: Compressing objects: 87% (81/93)\rremote: Compressing objects: 88% (82/93)\rremote: Compressing objects: 89% (83/93)\rremote: Compressing objects: 90% (84/93)\rremote: Compressing objects: 91% (85/93)\rremote: Compressing objects: 92% (86/93)\rremote: Compressing objects: 93% (87/93)\rremote: Compressing objects: 94% (88/93)\rremote: Compressing objects: 95% (89/93)\rremote: Compressing objects: 96% (90/93)\rremote: Compressing objects: 97% (91/93)\rremote: Compressing objects: 98% (92/93)\rremote: Compressing objects: 100% (93/93)\rremote: Compressing objects: 100% (93/93), done.\r\nReceiving objects: 0% (1/81085)\r",,terminal_output +23,8932632,"TERMINAL",0,0,"Receiving objects: 0% (319/81085), 156.00 KiB | 197.00 KiB/s\r",,terminal_output +24,8933638,"TERMINAL",0,0,"Receiving objects: 0% (541/81085), 340.00 KiB | 179.00 KiB/s\r",,terminal_output +25,8934622,"TERMINAL",0,0,"Receiving objects: 0% (739/81085), 476.00 KiB | 157.00 KiB/s\r",,terminal_output +26,8935009,"TERMINAL",0,0,"Receiving objects: 1% (811/81085), 476.00 KiB | 157.00 KiB/s\r",,terminal_output +27,8935611,"TERMINAL",0,0,"Receiving objects: 1% (920/81085), 556.00 KiB | 150.00 KiB/s\r",,terminal_output +28,8936698,"TERMINAL",0,0,"Receiving objects: 1% (1108/81085), 700.00 KiB | 145.00 KiB/s\r",,terminal_output +29,8937791,"TERMINAL",0,0,"Receiving objects: 1% (1271/81085), 820.00 KiB | 130.00 KiB/s\r",,terminal_output +30,8938649,"TERMINAL",0,0,"Receiving objects: 1% (1430/81085), 932.00 KiB | 117.00 KiB/s\r",,terminal_output +31,8939612,"TERMINAL",0,0,"Receiving objects: 1% (1563/81085), 1.01 MiB | 110.00 KiB/s \r",,terminal_output +32,8939824,"TERMINAL",0,0,"Receiving objects: 2% (1622/81085), 1.01 MiB | 110.00 KiB/s\r",,terminal_output +33,8940669,"TERMINAL",0,0,"Receiving objects: 2% (1732/81085), 1.08 MiB | 111.00 KiB/s\r",,terminal_output +34,8941605,"TERMINAL",0,0,"Receiving objects: 2% (1856/81085), 1.20 MiB | 105.00 KiB/s\r",,terminal_output +35,8942596,"TERMINAL",0,0,"Receiving objects: 2% (2056/81085), 1.32 MiB | 106.00 KiB/s\r",,terminal_output +36,8943657,"TERMINAL",0,0,"Receiving objects: 2% (2299/81085), 1.43 MiB | 106.00 KiB/s\r",,terminal_output +37,8944158,"TERMINAL",0,0,"Receiving objects: 3% (2433/81085), 1.50 MiB | 108.00 KiB/s\r",,terminal_output +38,8944629,"TERMINAL",0,0,"Receiving objects: 3% (2519/81085), 1.50 MiB | 108.00 KiB/s\r",,terminal_output +39,8945667,"TERMINAL",0,0,"Receiving objects: 3% (2654/81085), 1.64 MiB | 112.00 KiB/s\r",,terminal_output +40,8946604,"TERMINAL",0,0,"Receiving objects: 3% (2793/81085), 1.75 MiB | 112.00 KiB/s\r",,terminal_output +41,8947637,"TERMINAL",0,0,"Receiving objects: 3% (2959/81085), 1.87 MiB | 110.00 KiB/s\r",,terminal_output +42,8948708,"TERMINAL",0,0,"Receiving objects: 3% (3105/81085), 1.98 MiB | 108.00 KiB/s\r",,terminal_output +43,8949532,"TERMINAL",0,0,"Receiving objects: 4% (3244/81085), 2.05 MiB | 112.00 KiB/s\r",,terminal_output +44,8949613,"TERMINAL",0,0,"Receiving objects: 4% (3257/81085), 2.05 MiB | 112.00 KiB/s\r",,terminal_output +45,8950598,"TERMINAL",0,0,"Receiving objects: 4% (3393/81085), 2.16 MiB | 106.00 KiB/s\r",,terminal_output +46,8951677,"TERMINAL",0,0,"Receiving objects: 4% (3554/81085), 2.28 MiB | 108.00 KiB/s\r",,terminal_output +47,8952594,"TERMINAL",0,0,"Receiving objects: 4% (3697/81085), 2.39 MiB | 107.00 KiB/s\r",,terminal_output +48,8953633,"TERMINAL",0,0,"Receiving objects: 4% (3811/81085), 2.49 MiB | 105.00 KiB/s\r",,terminal_output +49,8954658,"TERMINAL",0,0,"Receiving objects: 4% (4003/81085), 2.52 MiB | 96.00 KiB/s \r",,terminal_output +50,8954909,"TERMINAL",0,0,"Receiving objects: 5% (4055/81085), 2.64 MiB | 109.00 KiB/s\r",,terminal_output +51,8955584,"TERMINAL",0,0,"Receiving objects: 5% (4144/81085), 2.70 MiB | 109.00 KiB/s\r",,terminal_output +52,8956589,"TERMINAL",0,0,"Receiving objects: 5% (4277/81085), 2.81 MiB | 110.00 KiB/s\r",,terminal_output +53,8957594,"TERMINAL",0,0,"Receiving objects: 5% (4428/81085), 2.93 MiB | 109.00 KiB/s\r",,terminal_output +54,8958657,"TERMINAL",0,0,"Receiving objects: 5% (4586/81085), 3.05 MiB | 116.00 KiB/s\r",,terminal_output +55,8959620,"TERMINAL",0,0,"Receiving objects: 5% (4738/81085), 3.17 MiB | 112.00 KiB/s\r",,terminal_output +56,8960403,"TERMINAL",0,0,"Receiving objects: 6% (4866/81085), 3.23 MiB | 111.00 KiB/s\r",,terminal_output +57,8960663,"TERMINAL",0,0,"Receiving objects: 6% (4889/81085), 3.29 MiB | 111.00 KiB/s\r",,terminal_output +58,8961674,"TERMINAL",0,0,"Receiving objects: 6% (5043/81085), 3.36 MiB | 115.00 KiB/s\r",,terminal_output +59,8962614,"TERMINAL",0,0,"Receiving objects: 6% (5188/81085), 3.46 MiB | 115.00 KiB/s\r",,terminal_output +60,8964047,"TERMINAL",0,0,"Receiving objects: 6% (5356/81085), 3.61 MiB | 103.00 KiB/s\r",,terminal_output +61,8964666,"TERMINAL",0,0,"Receiving objects: 6% (5491/81085), 3.71 MiB | 111.00 KiB/s\r",,terminal_output +62,8965621,"TERMINAL",0,0,"Receiving objects: 6% (5642/81085), 3.82 MiB | 111.00 KiB/s\r",,terminal_output +63,8965792,"TERMINAL",0,0,"Receiving objects: 7% (5676/81085), 3.82 MiB | 111.00 KiB/s\r",,terminal_output +64,8966607,"TERMINAL",0,0,"Receiving objects: 7% (5754/81085), 3.87 MiB | 105.00 KiB/s\r",,terminal_output +65,8967615,"TERMINAL",0,0,"Receiving objects: 7% (5867/81085), 3.96 MiB | 102.00 KiB/s\r",,terminal_output +66,8968593,"TERMINAL",0,0,"Receiving objects: 7% (6023/81085), 4.10 MiB | 105.00 KiB/s\r",,terminal_output +67,8969611,"TERMINAL",0,0,"Receiving objects: 7% (6183/81085), 4.19 MiB | 101.00 KiB/s\r",,terminal_output +68,8970603,"TERMINAL",0,0,"Receiving objects: 7% (6323/81085), 4.33 MiB | 106.00 KiB/s\r",,terminal_output +69,8971599,"TERMINAL",0,0,"Receiving objects: 7% (6456/81085), 4.45 MiB | 110.00 KiB/s\r",,terminal_output +70,8971887,"TERMINAL",0,0,"Receiving objects: 8% (6487/81085), 4.45 MiB | 110.00 KiB/s\r",,terminal_output +71,8972634,"TERMINAL",0,0,"Receiving objects: 8% (6588/81085), 4.50 MiB | 113.00 KiB/s\r",,terminal_output +72,8973688,"TERMINAL",0,0,"Receiving objects: 8% (6732/81085), 4.68 MiB | 119.00 KiB/s\r",,terminal_output +73,8974635,"TERMINAL",0,0,"Receiving objects: 8% (6843/81085), 4.75 MiB | 118.00 KiB/s\r",,terminal_output +74,8975595,"TERMINAL",0,0,"Receiving objects: 8% (7009/81085), 4.85 MiB | 111.00 KiB/s\r",,terminal_output +75,8976603,"TERMINAL",0,0,"Receiving objects: 8% (7153/81085), 4.96 MiB | 110.00 KiB/s\r",,terminal_output +76,8977134,"TERMINAL",0,0,"Receiving objects: 9% (7298/81085), 5.03 MiB | 113.00 KiB/s\r",,terminal_output +77,8977641,"TERMINAL",0,0,"Receiving objects: 9% (7355/81085), 5.07 MiB | 110.00 KiB/s\r",,terminal_output +78,8978648,"TERMINAL",0,0,"Receiving objects: 9% (7426/81085), 5.15 MiB | 98.00 KiB/s \r",,terminal_output +79,8979653,"TERMINAL",0,0,"Receiving objects: 9% (7613/81085), 5.22 MiB | 98.00 KiB/s\r",,terminal_output +80,8980629,"TERMINAL",0,0,"Receiving objects: 9% (7732/81085), 5.37 MiB | 107.00 KiB/s\r",,terminal_output +81,8981650,"TERMINAL",0,0,"Receiving objects: 9% (7866/81085), 5.48 MiB | 105.00 KiB/s\r",,terminal_output +82,8982714,"TERMINAL",0,0,"Receiving objects: 9% (7996/81085), 5.61 MiB | 107.00 KiB/s\r",,terminal_output +83,8983413,"TERMINAL",0,0,"Receiving objects: 10% (8109/81085), 5.65 MiB | 104.00 KiB/s\r",,terminal_output +84,8983680,"TERMINAL",0,0,"Receiving objects: 10% (8171/81085), 5.70 MiB | 114.00 KiB/s\r",,terminal_output +85,8984590,"TERMINAL",0,0,"Receiving objects: 10% (8356/81085), 5.75 MiB | 100.00 KiB/s\r",,terminal_output +86,8985644,"TERMINAL",0,0,"Receiving objects: 10% (8617/81085), 5.94 MiB | 110.00 KiB/s\r",,terminal_output +87,8986616,"TERMINAL",0,0,"Receiving objects: 10% (8803/81085), 5.99 MiB | 108.00 KiB/s\r",,terminal_output +88,8987218,"TERMINAL",0,0,"Receiving objects: 11% (8920/81085), 6.06 MiB | 107.00 KiB/s\r",,terminal_output +89,8987799,"TERMINAL",0,0,"Receiving objects: 11% (9007/81085), 6.14 MiB | 111.00 KiB/s\r",,terminal_output +90,8988590,"TERMINAL",0,0,"Receiving objects: 11% (9185/81085), 6.26 MiB | 113.00 KiB/s\r",,terminal_output +91,8989619,"TERMINAL",0,0,"Receiving objects: 11% (9367/81085), 6.33 MiB | 116.00 KiB/s\r",,terminal_output +92,8990654,"TERMINAL",0,0,"Receiving objects: 11% (9575/81085), 6.44 MiB | 111.00 KiB/s\r",,terminal_output +93,8991240,"TERMINAL",0,0,"Receiving objects: 12% (9731/81085), 6.50 MiB | 110.00 KiB/s\r",,terminal_output +94,8991642,"TERMINAL",0,0,"Receiving objects: 12% (9902/81085), 6.56 MiB | 112.00 KiB/s\r",,terminal_output +95,8992600,"TERMINAL",0,0,"Receiving objects: 12% (10280/81085), 6.68 MiB | 111.00 KiB/s\r",,terminal_output +96,8992957,"TERMINAL",0,0,"Receiving objects: 13% (10542/81085), 6.74 MiB | 115.00 KiB/s\r",,terminal_output +97,8993646,"TERMINAL",0,0,"Receiving objects: 13% (10882/81085), 6.79 MiB | 110.00 KiB/s\r",,terminal_output +98,8994607,"TERMINAL",0,0,"Receiving objects: 13% (11254/81085), 6.93 MiB | 111.00 KiB/s\r",,terminal_output +99,8994789,"TERMINAL",0,0,"Receiving objects: 14% (11352/81085), 6.93 MiB | 111.00 KiB/s\r",,terminal_output +100,8995734,"TERMINAL",0,0,"Receiving objects: 14% (11478/81085), 7.05 MiB | 112.00 KiB/s\r",,terminal_output +101,8996848,"TERMINAL",0,0,"Receiving objects: 14% (11478/81085), 7.18 MiB | 113.00 KiB/s\r",,terminal_output +102,8997661,"TERMINAL",0,0,"Receiving objects: 14% (11499/81085), 7.25 MiB | 115.00 KiB/s\r",,terminal_output +103,8999162,"TERMINAL",0,0,"Receiving objects: 14% (11515/81085), 7.40 MiB | 108.00 KiB/s\r",,terminal_output +104,8999691,"TERMINAL",0,0,"Receiving objects: 14% (11515/81085), 7.45 MiB | 105.00 KiB/s\r",,terminal_output +105,9000755,"TERMINAL",0,0,"Receiving objects: 14% (11527/81085), 7.57 MiB | 105.00 KiB/s\r",,terminal_output +106,9001856,"TERMINAL",0,0,"Receiving objects: 14% (11552/81085), 7.68 MiB | 102.00 KiB/s\r",,terminal_output +107,9003065,"TERMINAL",0,0,"Receiving objects: 14% (11553/81085), 7.79 MiB | 94.00 KiB/s \r",,terminal_output +108,9004200,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 7.86 MiB | 92.00 KiB/s\r",,terminal_output +109,9004718,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 7.97 MiB | 106.00 KiB/s\r",,terminal_output +110,9005763,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.10 MiB | 108.00 KiB/s\r",,terminal_output +111,9006838,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.23 MiB | 114.00 KiB/s\r",,terminal_output +112,9007923,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.34 MiB | 117.00 KiB/s\r",,terminal_output +113,9009092,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.47 MiB | 129.00 KiB/s\r",,terminal_output +114,9009975,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.51 MiB | 105.00 KiB/s\r",,terminal_output +115,9011055,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.69 MiB | 115.00 KiB/s\r",,terminal_output +116,9011607,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.75 MiB | 112.00 KiB/s\r",,terminal_output +117,9012762,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.86 MiB | 108.00 KiB/s\r",,terminal_output +118,9013815,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 8.96 MiB | 108.00 KiB/s\r",,terminal_output +119,9014887,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.09 MiB | 120.00 KiB/s\r",,terminal_output +120,9015963,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.21 MiB | 107.00 KiB/s\r",,terminal_output +121,9017147,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.31 MiB | 106.00 KiB/s\r",,terminal_output +122,9018050,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.35 MiB | 95.00 KiB/s \r",,terminal_output +123,9018606,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.47 MiB | 109.00 KiB/s\r",,terminal_output +124,9019771,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.58 MiB | 106.00 KiB/s\r",,terminal_output +125,9020788,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.71 MiB | 105.00 KiB/s\r",,terminal_output +126,9021917,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.83 MiB | 109.00 KiB/s\r",,terminal_output +127,9023121,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.96 MiB | 123.00 KiB/s\r",,terminal_output +128,9023659,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.02 MiB | 110.00 KiB/s\r",,terminal_output +129,9024998,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.13 MiB | 107.00 KiB/s\r",,terminal_output +130,9025587,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.24 MiB | 114.00 KiB/s\r",,terminal_output +131,9027025,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.39 MiB | 115.00 KiB/s\r",,terminal_output +132,9028331,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.41 MiB | 92.00 KiB/s \r",,terminal_output +133,9028873,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.46 MiB | 91.00 KiB/s\r",,terminal_output +134,9029808,"TERMINAL",0,0,"Receiving objects: 14% (11560/81085), 10.58 MiB | 102.00 KiB/s\r",,terminal_output +135,9030595,"TERMINAL",0,0,"Receiving objects: 14% (11650/81085), 10.77 MiB | 111.00 KiB/s\r",,terminal_output +136,9031999,"TERMINAL",0,0,"Receiving objects: 14% (11670/81085), 10.94 MiB | 117.00 KiB/s\r",,terminal_output +137,9033262,"TERMINAL",0,0,"Receiving objects: 14% (11670/81085), 11.07 MiB | 137.00 KiB/s\r",,terminal_output +138,9033622,"TERMINAL",0,0,"Receiving objects: 14% (11691/81085), 11.07 MiB | 137.00 KiB/s\r",,terminal_output +139,9034887,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 11.24 MiB | 109.00 KiB/s\r",,terminal_output +140,9036198,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 11.36 MiB | 105.00 KiB/s\r",,terminal_output +141,9036754,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 11.42 MiB | 103.00 KiB/s\r",,terminal_output +142,9037790,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 11.54 MiB | 104.00 KiB/s\r",,terminal_output +143,9038857,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 11.64 MiB | 105.00 KiB/s\r",,terminal_output +144,9039941,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 11.77 MiB | 107.00 KiB/s\r",,terminal_output +145,9040974,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 11.88 MiB | 110.00 KiB/s\r",,terminal_output +146,9042114,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 12.00 MiB | 111.00 KiB/s\r",,terminal_output +147,9042632,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 12.07 MiB | 112.00 KiB/s\r",,terminal_output +148,9043811,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 12.18 MiB | 111.00 KiB/s\r",,terminal_output +149,9044964,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 12.31 MiB | 109.00 KiB/s\r",,terminal_output +150,9046364,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 12.40 MiB | 99.00 KiB/s \r",,terminal_output +151,9046880,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 12.51 MiB | 109.00 KiB/s\r",,terminal_output +152,9047895,"TERMINAL",0,0,"Receiving objects: 14% (11866/81085), 12.61 MiB | 104.00 KiB/s\r",,terminal_output +153,9048959,"TERMINAL",0,0,"Receiving objects: 14% (11876/81085), 12.73 MiB | 108.00 KiB/s\r",,terminal_output +154,9050073,"TERMINAL",0,0,"Receiving objects: 14% (11876/81085), 12.83 MiB | 105.00 KiB/s\r",,terminal_output +155,9050639,"TERMINAL",0,0,"Receiving objects: 14% (11876/81085), 12.89 MiB | 106.00 KiB/s\r",,terminal_output +156,9051684,"TERMINAL",0,0,"Receiving objects: 14% (11876/81085), 13.00 MiB | 105.00 KiB/s\r",,terminal_output +157,9052820,"TERMINAL",0,0,"Receiving objects: 14% (11877/81085), 13.02 MiB | 96.00 KiB/s \r",,terminal_output +158,9053938,"TERMINAL",0,0,"Receiving objects: 14% (11909/81085), 13.24 MiB | 104.00 KiB/s\r",,terminal_output +159,9054623,"TERMINAL",0,0,"Receiving objects: 14% (11910/81085), 13.29 MiB | 107.00 KiB/s\r",,terminal_output +160,9055916,"TERMINAL",0,0,"Receiving objects: 14% (11911/81085), 13.40 MiB | 108.00 KiB/s\r",,terminal_output +161,9056664,"TERMINAL",0,0,"Receiving objects: 14% (11929/81085), 13.54 MiB | 110.00 KiB/s\r",,terminal_output +162,9057692,"TERMINAL",0,0,"Receiving objects: 14% (11993/81085), 13.66 MiB | 113.00 KiB/s\r",,terminal_output +163,9059102,"TERMINAL",0,0,"Receiving objects: 14% (12074/81085), 13.78 MiB | 106.00 KiB/s\r",,terminal_output +164,9059615,"TERMINAL",0,0,"Receiving objects: 14% (12074/81085), 13.88 MiB | 117.00 KiB/s\r",,terminal_output +165,9060633,"TERMINAL",0,0,"Receiving objects: 14% (12074/81085), 13.99 MiB | 116.00 KiB/s\r",,terminal_output +166,9061733,"TERMINAL",0,0,"Receiving objects: 14% (12110/81085), 14.09 MiB | 110.00 KiB/s\r",,terminal_output +167,9062384,"TERMINAL",0,0,"Receiving objects: 15% (12163/81085), 14.15 MiB | 109.00 KiB/s\r",,terminal_output +168,9062829,"TERMINAL",0,0,"Receiving objects: 15% (12207/81085), 14.21 MiB | 110.00 KiB/s\r",,terminal_output +169,9063951,"TERMINAL",0,0,"Receiving objects: 15% (12208/81085), 14.32 MiB | 115.00 KiB/s\r",,terminal_output +170,9064740,"TERMINAL",0,0,"Receiving objects: 15% (12217/81085), 14.36 MiB | 99.00 KiB/s \r",,terminal_output +171,9065839,"TERMINAL",0,0,"Receiving objects: 15% (12220/81085), 14.43 MiB | 93.00 KiB/s\r",,terminal_output +172,9066606,"TERMINAL",0,0,"Receiving objects: 15% (12250/81085), 14.61 MiB | 108.00 KiB/s\r",,terminal_output +173,9067598,"TERMINAL",0,0,"Receiving objects: 15% (12295/81085), 14.68 MiB | 109.00 KiB/s\r",,terminal_output +174,9068593,"TERMINAL",0,0,"Receiving objects: 15% (12328/81085), 14.79 MiB | 107.00 KiB/s\r",,terminal_output +175,9069626,"TERMINAL",0,0,"Receiving objects: 15% (12397/81085), 14.91 MiB | 110.00 KiB/s\r",,terminal_output +176,9070628,"TERMINAL",0,0,"Receiving objects: 15% (12525/81085), 15.04 MiB | 123.00 KiB/s\r",,terminal_output +177,9071615,"TERMINAL",0,0,"Receiving objects: 15% (12620/81085), 15.15 MiB | 110.00 KiB/s\r",,terminal_output +178,9072696,"TERMINAL",0,0,"Receiving objects: 15% (12621/81085), 15.26 MiB | 111.00 KiB/s\r",,terminal_output +179,9073608,"TERMINAL",0,0,"Receiving objects: 15% (12644/81085), 15.33 MiB | 110.00 KiB/s\r",,terminal_output +180,9074658,"TERMINAL",0,0,"Receiving objects: 15% (12704/81085), 15.44 MiB | 104.00 KiB/s\r",,terminal_output +181,9075622,"TERMINAL",0,0,"Receiving objects: 15% (12771/81085), 15.57 MiB | 107.00 KiB/s\r",,terminal_output +182,9077068,"TERMINAL",0,0,"Receiving objects: 15% (12870/81085), 15.68 MiB | 99.00 KiB/s \r",,terminal_output +183,9077640,"TERMINAL",0,0,"Receiving objects: 15% (12962/81085), 15.80 MiB | 109.00 KiB/s\r",,terminal_output +184,9077702,"TERMINAL",0,0,"Receiving objects: 16% (12974/81085), 15.80 MiB | 109.00 KiB/s\r",,terminal_output +185,9078764,"TERMINAL",0,0,"Receiving objects: 16% (12996/81085), 15.93 MiB | 112.00 KiB/s\r",,terminal_output +186,9079803,"TERMINAL",0,0,"Receiving objects: 16% (13000/81085), 16.03 MiB | 114.00 KiB/s\r",,terminal_output +187,9080834,"TERMINAL",0,0,"Receiving objects: 16% (13000/81085), 16.15 MiB | 113.00 KiB/s\r",,terminal_output +188,9082025,"TERMINAL",0,0,"Receiving objects: 16% (13001/81085), 16.27 MiB | 123.00 KiB/s\r",,terminal_output +189,9083028,"TERMINAL",0,0,"Receiving objects: 16% (13002/81085), 16.32 MiB | 107.00 KiB/s\r",,terminal_output +190,9084025,"TERMINAL",0,0,"Receiving objects: 16% (13008/81085), 16.47 MiB | 112.00 KiB/s\r",,terminal_output +191,9085028,"TERMINAL",0,0,"Receiving objects: 16% (13009/81085), 16.61 MiB | 118.00 KiB/s\r",,terminal_output +192,9086026,"TERMINAL",0,0,"Receiving objects: 16% (13010/81085), 16.67 MiB | 115.00 KiB/s\r",,terminal_output +193,9088027,"TERMINAL",0,0,"Receiving objects: 16% (13010/81085), 16.81 MiB | 107.00 KiB/s\rReceiving objects: 16% (13011/81085), 16.86 MiB | 109.00 KiB/s\r",,terminal_output +194,9089026,"TERMINAL",0,0,"Receiving objects: 16% (13011/81085), 17.00 MiB | 105.00 KiB/s\r",,terminal_output +195,9091027,"TERMINAL",0,0,"Receiving objects: 16% (13011/81085), 17.11 MiB | 94.00 KiB/s \rReceiving objects: 16% (13011/81085), 17.23 MiB | 104.00 KiB/s\r",,terminal_output +196,9092025,"TERMINAL",0,0,"Receiving objects: 16% (13012/81085), 17.28 MiB | 105.00 KiB/s\r",,terminal_output +197,9093029,"TERMINAL",0,0,"Receiving objects: 16% (13019/81085), 17.32 MiB | 103.00 KiB/s\r",,terminal_output +198,9094027,"TERMINAL",0,0,"Receiving objects: 16% (13028/81085), 17.46 MiB | 109.00 KiB/s\r",,terminal_output +199,9095025,"TERMINAL",0,0,"Receiving objects: 16% (13034/81085), 17.64 MiB | 109.00 KiB/s\r",,terminal_output +200,9096024,"TERMINAL",0,0,"Receiving objects: 16% (13034/81085), 17.76 MiB | 109.00 KiB/s\r",,terminal_output +201,9097026,"TERMINAL",0,0,"Receiving objects: 16% (13035/81085), 17.89 MiB | 120.00 KiB/s\r",,terminal_output +202,9098027,"TERMINAL",0,0,"Receiving objects: 16% (13043/81085), 17.94 MiB | 117.00 KiB/s\r",,terminal_output +203,9099025,"TERMINAL",0,0,"Receiving objects: 16% (13083/81085), 18.06 MiB | 118.00 KiB/s\r",,terminal_output +204,9100025,"TERMINAL",0,0,"Receiving objects: 16% (13153/81085), 18.18 MiB | 118.00 KiB/s\r",,terminal_output +205,9101025,"TERMINAL",0,0,"Receiving objects: 16% (13200/81085), 18.30 MiB | 115.00 KiB/s\r",,terminal_output +206,9102025,"TERMINAL",0,0,"Receiving objects: 16% (13205/81085), 18.40 MiB | 111.00 KiB/s\r",,terminal_output +207,9103027,"TERMINAL",0,0,"Receiving objects: 16% (13389/81085), 18.47 MiB | 113.00 KiB/s\r",,terminal_output +208,9105034,"TERMINAL",0,0,"Receiving objects: 16% (13623/81085), 18.59 MiB | 109.00 KiB/s\rReceiving objects: 17% (13785/81085), 18.65 MiB | 109.00 KiB/s\rReceiving objects: 17% (13789/81085), 18.67 MiB | 97.00 KiB/s \r",,terminal_output +209,9106025,"TERMINAL",0,0,"Receiving objects: 17% (14222/81085), 18.78 MiB | 107.00 KiB/s\r",,terminal_output +210,9107025,"TERMINAL",0,0,"Receiving objects: 17% (14530/81085), 18.89 MiB | 110.00 KiB/s\r",,terminal_output +211,9108026,"TERMINAL",0,0,"Receiving objects: 17% (14595/81085), 19.00 MiB | 107.00 KiB/s\rReceiving objects: 18% (14596/81085), 19.00 MiB | 107.00 KiB/s\r",,terminal_output +212,9109029,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.18 MiB | 112.00 KiB/s\r",,terminal_output +213,9110022,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.29 MiB | 110.00 KiB/s\r",,terminal_output +214,9111029,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.40 MiB | 107.00 KiB/s\r",,terminal_output +215,9112025,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.43 MiB | 100.00 KiB/s\r",,terminal_output +216,9113027,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.53 MiB | 97.00 KiB/s \r",,terminal_output +217,9114025,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.65 MiB | 96.00 KiB/s\r",,terminal_output +218,9115028,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.79 MiB | 100.00 KiB/s\r",,terminal_output +219,9116025,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.91 MiB | 104.00 KiB/s\r",,terminal_output +220,9117025,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 19.97 MiB | 110.00 KiB/s\r",,terminal_output +221,9118028,"TERMINAL",0,0,"Receiving objects: 18% (14651/81085), 20.04 MiB | 107.00 KiB/s\r",,terminal_output +222,9119035,"TERMINAL",0,0,"Receiving objects: 18% (14702/81085), 20.17 MiB | 115.00 KiB/s\r",,terminal_output +223,9120026,"TERMINAL",0,0,"Receiving objects: 18% (14764/81085), 20.30 MiB | 117.00 KiB/s\r",,terminal_output +224,9121023,"TERMINAL",0,0,"Receiving objects: 18% (14892/81085), 20.42 MiB | 114.00 KiB/s\r",,terminal_output +225,9122025,"TERMINAL",0,0,"Receiving objects: 18% (15214/81085), 20.48 MiB | 114.00 KiB/s\r",,terminal_output +226,9123027,"TERMINAL",0,0,"Receiving objects: 18% (15293/81085), 20.61 MiB | 115.00 KiB/s\r",,terminal_output +227,9124027,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 20.79 MiB | 111.00 KiB/s\r",,terminal_output +228,9125026,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 20.82 MiB | 100.00 KiB/s\r",,terminal_output +229,9126024,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 20.96 MiB | 105.00 KiB/s\r",,terminal_output +230,9127027,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 21.14 MiB | 116.00 KiB/s\r",,terminal_output +231,9129025,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 21.24 MiB | 112.00 KiB/s\rReceiving objects: 18% (15359/81085), 21.29 MiB | 110.00 KiB/s\r",,terminal_output +232,9130027,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 21.42 MiB | 120.00 KiB/s\r",,terminal_output +233,9131027,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 21.55 MiB | 120.00 KiB/s\r",,terminal_output +234,9132025,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 21.66 MiB | 107.00 KiB/s\r",,terminal_output +235,9133028,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 21.71 MiB | 102.00 KiB/s\r",,terminal_output +236,9134027,"TERMINAL",0,0,"Receiving objects: 18% (15359/81085), 21.90 MiB | 117.00 KiB/s\r",,terminal_output +237,9135027,"TERMINAL",0,0,"Receiving objects: 19% (15407/81085), 21.96 MiB | 116.00 KiB/s\rReceiving objects: 19% (15412/81085), 21.96 MiB | 116.00 KiB/s\r",,terminal_output +238,9137025,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 22.11 MiB | 109.00 KiB/s\rReceiving objects: 19% (15479/81085), 22.16 MiB | 101.00 KiB/s\r",,terminal_output +239,9138024,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 22.29 MiB | 115.00 KiB/s\r",,terminal_output +240,9140025,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 22.40 MiB | 100.00 KiB/s\rReceiving objects: 19% (15479/81085), 22.45 MiB | 98.00 KiB/s \r",,terminal_output +241,9141028,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 22.61 MiB | 104.00 KiB/s\r",,terminal_output +242,9142025,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 22.67 MiB | 104.00 KiB/s\r",,terminal_output +243,9143023,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 22.78 MiB | 106.00 KiB/s\r",,terminal_output +244,9144023,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 22.90 MiB | 107.00 KiB/s\r",,terminal_output +245,9145024,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 23.02 MiB | 112.00 KiB/s\r",,terminal_output +246,9146025,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 23.12 MiB | 106.00 KiB/s\r",,terminal_output +247,9148024,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 23.21 MiB | 92.00 KiB/s \r",,terminal_output +248,9149025,"TERMINAL",0,0,"Receiving objects: 19% (15479/81085), 23.23 MiB | 79.00 KiB/s\r",,terminal_output +249,9150024,"TERMINAL",0,0,"Receiving objects: 19% (15486/81085), 23.47 MiB | 98.00 KiB/s\r",,terminal_output +250,9151024,"TERMINAL",0,0,"Receiving objects: 19% (15486/81085), 23.60 MiB | 98.00 KiB/s\r",,terminal_output +251,9152024,"TERMINAL",0,0,"Receiving objects: 19% (15486/81085), 23.73 MiB | 104.00 KiB/s\r",,terminal_output +252,9154024,"TERMINAL",0,0,"Receiving objects: 19% (15486/81085), 23.86 MiB | 121.00 KiB/s\r",,terminal_output +253,9155023,"TERMINAL",0,0,"Receiving objects: 19% (15486/81085), 23.93 MiB | 128.00 KiB/s\r",,terminal_output +254,9156023,"TERMINAL",0,0,"Receiving objects: 19% (15486/81085), 23.98 MiB | 96.00 KiB/s \rReceiving objects: 19% (15486/81085), 24.10 MiB | 103.00 KiB/s\r",,terminal_output +255,9157024,"TERMINAL",0,0,"Receiving objects: 19% (15486/81085), 24.22 MiB | 106.00 KiB/s\r",,terminal_output +256,9158024,"TERMINAL",0,0,"Receiving objects: 19% (15487/81085), 24.29 MiB | 106.00 KiB/s\r",,terminal_output +257,9159024,"TERMINAL",0,0,"Receiving objects: 19% (15567/81085), 24.46 MiB | 104.00 KiB/s\r",,terminal_output +258,9160024,"TERMINAL",0,0,"Receiving objects: 19% (15749/81085), 24.53 MiB | 111.00 KiB/s\r",,terminal_output +259,9161024,"TERMINAL",0,0,"Receiving objects: 19% (16097/81085), 24.65 MiB | 117.00 KiB/s\r",,terminal_output +260,9162024,"TERMINAL",0,0,"Receiving objects: 19% (16202/81085), 24.77 MiB | 114.00 KiB/s\rReceiving objects: 20% (16217/81085), 24.77 MiB | 114.00 KiB/s\r",,terminal_output +261,9163024,"TERMINAL",0,0,"Receiving objects: 20% (16314/81085), 24.83 MiB | 115.00 KiB/s\r",,terminal_output +262,9164023,"TERMINAL",0,0,"Receiving objects: 20% (16386/81085), 24.96 MiB | 110.00 KiB/s\r",,terminal_output +263,9165024,"TERMINAL",0,0,"Receiving objects: 20% (16436/81085), 25.08 MiB | 113.00 KiB/s\r",,terminal_output +264,9166024,"TERMINAL",0,0,"Receiving objects: 20% (16534/81085), 25.21 MiB | 113.00 KiB/s\r",,terminal_output +265,9167023,"TERMINAL",0,0,"Receiving objects: 20% (16578/81085), 25.29 MiB | 100.00 KiB/s\r",,terminal_output +266,9169022,"TERMINAL",0,0,"Receiving objects: 20% (16621/81085), 25.47 MiB | 113.00 KiB/s\r",,terminal_output +267,9170024,"TERMINAL",0,0,"Receiving objects: 20% (16621/81085), 25.60 MiB | 115.00 KiB/s\rReceiving objects: 20% (16621/81085), 25.64 MiB | 111.00 KiB/s\r",,terminal_output +268,9171024,"TERMINAL",0,0,"Receiving objects: 20% (16621/81085), 25.75 MiB | 107.00 KiB/s\r",,terminal_output +269,9172023,"TERMINAL",0,0,"Receiving objects: 20% (16622/81085), 25.79 MiB | 107.00 KiB/s\r",,terminal_output +270,9174022,"TERMINAL",0,0,"Receiving objects: 20% (16622/81085), 25.99 MiB | 105.00 KiB/s\r",,terminal_output +271,9175022,"TERMINAL",0,0,"Receiving objects: 20% (16622/81085), 26.07 MiB | 97.00 KiB/s \rReceiving objects: 20% (16622/81085), 26.17 MiB | 105.00 KiB/s\r",,terminal_output +272,9176023,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 26.34 MiB | 115.00 KiB/s\r",,terminal_output +273,9177075,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 26.43 MiB | 112.00 KiB/s\r",,terminal_output +274,9178023,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 26.49 MiB | 106.00 KiB/s\r",,terminal_output +275,9179023,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 26.64 MiB | 119.00 KiB/s\r",,terminal_output +276,9181024,"TERMINAL",0,0,"Receiving objects: 20% (16624/81085), 26.79 MiB | 121.00 KiB/s\rReceiving objects: 20% (16624/81085), 26.82 MiB | 109.00 KiB/s\r",,terminal_output +277,9181981,"TERMINAL",0,0,"Receiving objects: 20% (16624/81085), 26.96 MiB | 106.00 KiB/s\r",,terminal_output +278,9182792,"TERMINAL",0,0,"Receiving objects: 20% (16625/81085), 27.01 MiB | 110.00 KiB/s\r",,terminal_output +279,9183646,"TERMINAL",0,0,"Receiving objects: 20% (16628/81085), 27.14 MiB | 109.00 KiB/s\r",,terminal_output +280,9184752,"TERMINAL",0,0,"Receiving objects: 20% (16629/81085), 27.20 MiB | 100.00 KiB/s\r",,terminal_output +281,9185955,"TERMINAL",0,0,"Receiving objects: 20% (16633/81085), 27.41 MiB | 107.00 KiB/s\r",,terminal_output +282,9187055,"TERMINAL",0,0,"Receiving objects: 20% (16633/81085), 27.50 MiB | 102.00 KiB/s\r",,terminal_output +283,9187993,"TERMINAL",0,0,"Receiving objects: 20% (16633/81085), 27.54 MiB | 98.00 KiB/s \r",,terminal_output +284,9189038,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 27.76 MiB | 123.00 KiB/s\r",,terminal_output +285,9190138,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 27.89 MiB | 130.00 KiB/s\r",,terminal_output +286,9190686,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 27.95 MiB | 128.00 KiB/s\r",,terminal_output +287,9192079,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 28.08 MiB | 117.00 KiB/s\r",,terminal_output +288,9192602,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 28.15 MiB | 121.00 KiB/s\r",,terminal_output +289,9193692,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 28.29 MiB | 123.00 KiB/s\r",,terminal_output +290,9194811,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 28.43 MiB | 122.00 KiB/s\r",,terminal_output +291,9196021,"TERMINAL",0,0,"Receiving objects: 20% (16634/81085), 28.55 MiB | 115.00 KiB/s\r",,terminal_output +292,9197077,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 28.67 MiB | 120.00 KiB/s\r",,terminal_output +293,9198109,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 28.79 MiB | 116.00 KiB/s\r",,terminal_output +294,9198610,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 28.86 MiB | 117.00 KiB/s\r",,terminal_output +295,9199715,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 28.98 MiB | 114.00 KiB/s\r",,terminal_output +296,9200759,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 29.09 MiB | 116.00 KiB/s\r",,terminal_output +297,9202024,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 29.22 MiB | 114.00 KiB/s\r",,terminal_output +298,9203080,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 29.34 MiB | 112.00 KiB/s\r",,terminal_output +299,9203676,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 29.41 MiB | 112.00 KiB/s\r",,terminal_output +300,9204841,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 29.54 MiB | 110.00 KiB/s\r",,terminal_output +301,9205998,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 29.64 MiB | 108.00 KiB/s\r",,terminal_output +302,9207572,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 29.77 MiB | 101.00 KiB/s\r",,terminal_output +303,9208100,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 29.89 MiB | 112.00 KiB/s\r",,terminal_output +304,9208694,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 29.96 MiB | 112.00 KiB/s\r",,terminal_output +305,9209766,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 30.09 MiB | 117.00 KiB/s\r",,terminal_output +306,9210909,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 30.21 MiB | 119.00 KiB/s\r",,terminal_output +307,9211901,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 30.27 MiB | 116.00 KiB/s\r",,terminal_output +308,9213192,"TERMINAL",0,0,"Receiving objects: 20% (16645/81085), 30.43 MiB | 108.00 KiB/s\r",,terminal_output +309,9213730,"TERMINAL",0,0,"Receiving objects: 20% (16645/81085), 30.48 MiB | 106.00 KiB/s\r",,terminal_output +310,9214712,"TERMINAL",0,0,"Receiving objects: 20% (16646/81085), 30.55 MiB | 109.00 KiB/s\r",,terminal_output +311,9215623,"TERMINAL",0,0,"Receiving objects: 20% (16700/81085), 30.68 MiB | 108.00 KiB/s\r",,terminal_output +312,9216684,"TERMINAL",0,0,"Receiving objects: 20% (16821/81085), 30.80 MiB | 105.00 KiB/s\r",,terminal_output +313,9217694,"TERMINAL",0,0,"Receiving objects: 20% (16838/81085), 30.93 MiB | 105.00 KiB/s\r",,terminal_output +314,9218742,"TERMINAL",0,0,"Receiving objects: 20% (16838/81085), 31.02 MiB | 110.00 KiB/s\r",,terminal_output +315,9219839,"TERMINAL",0,0,"Receiving objects: 20% (16838/81085), 31.16 MiB | 117.00 KiB/s\r",,terminal_output +316,9220921,"TERMINAL",0,0,"Receiving objects: 20% (16838/81085), 31.29 MiB | 120.00 KiB/s\r",,terminal_output +317,9222039,"TERMINAL",0,0,"Receiving objects: 20% (16838/81085), 31.40 MiB | 112.00 KiB/s\r",,terminal_output +318,9222590,"TERMINAL",0,0,"Receiving objects: 20% (16839/81085), 31.48 MiB | 116.00 KiB/s\r",,terminal_output +319,9223690,"TERMINAL",0,0,"Receiving objects: 20% (16839/81085), 31.60 MiB | 119.00 KiB/s\r",,terminal_output +320,9224759,"TERMINAL",0,0,"Receiving objects: 20% (16839/81085), 31.72 MiB | 117.00 KiB/s\r",,terminal_output +321,9225605,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.78 MiB | 111.00 KiB/s\r",,terminal_output +322,9226674,"TERMINAL",0,0,"Receiving objects: 20% (16845/81085), 31.87 MiB | 107.00 KiB/s\r",,terminal_output +323,9227647,"TERMINAL",0,0,"Receiving objects: 20% (16849/81085), 32.01 MiB | 110.00 KiB/s\r",,terminal_output +324,9228734,"TERMINAL",0,0,"Receiving objects: 20% (16917/81085), 32.10 MiB | 101.00 KiB/s\r",,terminal_output +325,9229460,"TERMINAL",0,0,"Receiving objects: 21% (17028/81085), 32.20 MiB | 106.00 KiB/s\r",,terminal_output +326,9229827,"TERMINAL",0,0,"Receiving objects: 21% (17036/81085), 32.27 MiB | 110.00 KiB/s\r",,terminal_output +327,9230990,"TERMINAL",0,0,"Receiving objects: 21% (17036/81085), 32.35 MiB | 102.00 KiB/s\r",,terminal_output +328,9231868,"TERMINAL",0,0,"Receiving objects: 21% (17037/81085), 32.43 MiB | 113.00 KiB/s\r",,terminal_output +329,9232640,"TERMINAL",0,0,"Receiving objects: 21% (17086/81085), 32.57 MiB | 113.00 KiB/s\r",,terminal_output +330,9233621,"TERMINAL",0,0,"Receiving objects: 21% (17237/81085), 32.64 MiB | 114.00 KiB/s\r",,terminal_output +331,9234667,"TERMINAL",0,0,"Receiving objects: 21% (17326/81085), 32.77 MiB | 115.00 KiB/s\r",,terminal_output +332,9235765,"TERMINAL",0,0,"Receiving objects: 21% (17424/81085), 32.87 MiB | 103.00 KiB/s\r",,terminal_output +333,9236855,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.00 MiB | 107.00 KiB/s\r",,terminal_output +334,9237861,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.14 MiB | 110.00 KiB/s\r",,terminal_output +335,9238948,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.25 MiB | 109.00 KiB/s\r",,terminal_output +336,9239956,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.37 MiB | 109.00 KiB/s\r",,terminal_output +337,9241065,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.49 MiB | 112.00 KiB/s\r",,terminal_output +338,9242175,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.58 MiB | 107.00 KiB/s\r",,terminal_output +339,9242675,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.61 MiB | 101.00 KiB/s\r",,terminal_output +340,9243715,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.65 MiB | 85.00 KiB/s \r",,terminal_output +341,9244759,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.82 MiB | 95.00 KiB/s\r",,terminal_output +342,9245815,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 33.92 MiB | 92.00 KiB/s\r",,terminal_output +343,9246908,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 34.05 MiB | 101.00 KiB/s\r",,terminal_output +344,9248030,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 34.17 MiB | 110.00 KiB/s\r",,terminal_output +345,9248592,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 34.24 MiB | 123.00 KiB/s\r",,terminal_output +346,9249664,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 34.35 MiB | 111.00 KiB/s\r",,terminal_output +347,9250866,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 34.43 MiB | 104.00 KiB/s\r",,terminal_output +348,9251932,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 34.57 MiB | 105.00 KiB/s\r",,terminal_output +349,9252796,"TERMINAL",0,0,"Receiving objects: 21% (17581/81085), 34.61 MiB | 94.00 KiB/s \r",,terminal_output +350,9253641,"TERMINAL",0,0,"Receiving objects: 21% (17837/81085), 34.70 MiB | 102.00 KiB/s\rReceiving objects: 22% (17839/81085), 34.70 MiB | 102.00 KiB/s\r",,terminal_output +351,9254589,"TERMINAL",0,0,"Receiving objects: 22% (18195/81085), 34.84 MiB | 106.00 KiB/s\r",,terminal_output +352,9255590,"TERMINAL",0,0,"Receiving objects: 22% (18329/81085), 34.96 MiB | 111.00 KiB/s\r",,terminal_output +353,9256612,"TERMINAL",0,0,"Receiving objects: 22% (18492/81085), 35.08 MiB | 117.00 KiB/s\r",,terminal_output +354,9257439,"TERMINAL",0,0,"Receiving objects: 23% (18650/81085), 35.14 MiB | 114.00 KiB/s\r",,terminal_output +355,9257623,"TERMINAL",0,0,"Receiving objects: 23% (18681/81085), 35.21 MiB | 127.00 KiB/s\r",,terminal_output +356,9258637,"TERMINAL",0,0,"Receiving objects: 23% (18987/81085), 35.26 MiB | 118.00 KiB/s\r",,terminal_output +357,9259814,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 35.44 MiB | 112.00 KiB/s\r",,terminal_output +358,9261085,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 35.57 MiB | 108.00 KiB/s\r",,terminal_output +359,9261638,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 35.64 MiB | 111.00 KiB/s\r",,terminal_output +360,9262685,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 35.75 MiB | 109.00 KiB/s\r",,terminal_output +361,9263743,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 35.84 MiB | 104.00 KiB/s\r",,terminal_output +362,9264800,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 35.93 MiB | 99.00 KiB/s \r",,terminal_output +363,9265901,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 36.03 MiB | 98.00 KiB/s\r",,terminal_output +364,9266736,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 36.04 MiB | 80.00 KiB/s\r",,terminal_output +365,9269082,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 36.21 MiB | 75.00 KiB/s\r",,terminal_output +366,9269603,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 36.25 MiB | 73.00 KiB/s\r",,terminal_output +367,9270647,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 36.49 MiB | 96.00 KiB/s\r",,terminal_output +368,9271784,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 36.61 MiB | 100.00 KiB/s\r",,terminal_output +369,9272880,"TERMINAL",0,0,"Receiving objects: 23% (19074/81085), 36.73 MiB | 116.00 KiB/s\r",,terminal_output +370,9273969,"TERMINAL",0,0,"Receiving objects: 23% (19085/81085), 36.86 MiB | 134.00 KiB/s\r",,terminal_output +371,9275071,"TERMINAL",0,0,"Receiving objects: 23% (19085/81085), 36.99 MiB | 114.00 KiB/s\r",,terminal_output +372,9275632,"TERMINAL",0,0,"Receiving objects: 23% (19085/81085), 37.05 MiB | 115.00 KiB/s\r",,terminal_output +373,9276680,"TERMINAL",0,0,"Receiving objects: 23% (19085/81085), 37.11 MiB | 107.00 KiB/s\r",,terminal_output +374,9277608,"TERMINAL",0,0,"Receiving objects: 23% (19116/81085), 37.23 MiB | 115.00 KiB/s\r",,terminal_output +375,9279078,"TERMINAL",0,0,"Receiving objects: 23% (19142/81085), 37.41 MiB | 111.00 KiB/s\r",,terminal_output +376,9279663,"TERMINAL",0,0,"Receiving objects: 23% (19142/81085), 37.47 MiB | 111.00 KiB/s\r",,terminal_output +377,9281199,"TERMINAL",0,0,"Receiving objects: 23% (19142/81085), 37.61 MiB | 103.00 KiB/s\r",,terminal_output +378,9281905,"TERMINAL",0,0,"Receiving objects: 23% (19142/81085), 37.67 MiB | 100.00 KiB/s\r",,terminal_output +379,9282647,"TERMINAL",0,0,"Receiving objects: 23% (19143/81085), 37.75 MiB | 110.00 KiB/s\r",,terminal_output +380,9283705,"TERMINAL",0,0,"Receiving objects: 23% (19226/81085), 37.83 MiB | 94.00 KiB/s \r",,terminal_output +381,9284589,"TERMINAL",0,0,"Receiving objects: 23% (19358/81085), 37.94 MiB | 103.00 KiB/s\r",,terminal_output +382,9285591,"TERMINAL",0,0,"Receiving objects: 23% (19446/81085), 38.07 MiB | 105.00 KiB/s\r",,terminal_output +383,9285662,"TERMINAL",0,0,"Receiving objects: 24% (19461/81085), 38.07 MiB | 105.00 KiB/s\r",,terminal_output +384,9286612,"TERMINAL",0,0,"Receiving objects: 24% (19533/81085), 38.18 MiB | 109.00 KiB/s\r",,terminal_output +385,9287720,"TERMINAL",0,0,"Receiving objects: 24% (19575/81085), 38.25 MiB | 99.00 KiB/s \r",,terminal_output +386,9288811,"TERMINAL",0,0,"Receiving objects: 24% (19591/81085), 38.36 MiB | 105.00 KiB/s\r",,terminal_output +387,9289878,"TERMINAL",0,0,"Receiving objects: 24% (19592/81085), 38.48 MiB | 94.00 KiB/s \r",,terminal_output +388,9290993,"TERMINAL",0,0,"Receiving objects: 24% (19592/81085), 38.59 MiB | 93.00 KiB/s\r",,terminal_output +389,9291863,"TERMINAL",0,0,"Receiving objects: 24% (19592/81085), 38.61 MiB | 82.00 KiB/s\r",,terminal_output +390,9292715,"TERMINAL",0,0,"Receiving objects: 24% (19597/81085), 38.71 MiB | 96.00 KiB/s\r",,terminal_output +391,9293645,"TERMINAL",0,0,"Receiving objects: 24% (19643/81085), 38.82 MiB | 99.00 KiB/s\r",,terminal_output +392,9294630,"TERMINAL",0,0,"Receiving objects: 24% (19890/81085), 38.87 MiB | 97.00 KiB/s\r",,terminal_output +393,9296070,"TERMINAL",0,0,"Receiving objects: 24% (19989/81085), 39.01 MiB | 86.00 KiB/s\r",,terminal_output +394,9296635,"TERMINAL",0,0,"Receiving objects: 24% (19989/81085), 39.07 MiB | 86.00 KiB/s\r",,terminal_output +395,9297773,"TERMINAL",0,0,"Receiving objects: 24% (19989/81085), 39.19 MiB | 92.00 KiB/s\r",,terminal_output +396,9298874,"TERMINAL",0,0,"Receiving objects: 24% (19989/81085), 39.29 MiB | 91.00 KiB/s\r",,terminal_output +397,9300022,"TERMINAL",0,0,"Receiving objects: 24% (19989/81085), 39.39 MiB | 95.00 KiB/s\r",,terminal_output +398,9300658,"TERMINAL",0,0,"Receiving objects: 24% (20037/81085), 39.45 MiB | 101.00 KiB/s\r",,terminal_output +399,9301674,"TERMINAL",0,0,"Receiving objects: 24% (20095/81085), 39.54 MiB | 95.00 KiB/s \r",,terminal_output +400,9302657,"TERMINAL",0,0,"Receiving objects: 24% (20159/81085), 39.59 MiB | 94.00 KiB/s\r",,terminal_output +401,9303616,"TERMINAL",0,0,"Receiving objects: 24% (20197/81085), 39.65 MiB | 82.00 KiB/s\r",,terminal_output +402,9303708,"TERMINAL",0,0,"Receiving objects: 25% (20272/81085), 39.65 MiB | 82.00 KiB/s\r",,terminal_output +403,9304692,"TERMINAL",0,0,"Receiving objects: 25% (20629/81085), 39.75 MiB | 84.00 KiB/s\r",,terminal_output +404,9305659,"TERMINAL",0,0,"Receiving objects: 25% (20747/81085), 39.86 MiB | 85.00 KiB/s\r",,terminal_output +405,9306607,"TERMINAL",0,0,"Receiving objects: 25% (20794/81085), 39.99 MiB | 94.00 KiB/s\r",,terminal_output +406,9307685,"TERMINAL",0,0,"Receiving objects: 25% (20794/81085), 40.10 MiB | 100.00 KiB/s\r",,terminal_output +407,9308822,"TERMINAL",0,0,"Receiving objects: 25% (20794/81085), 40.22 MiB | 110.00 KiB/s\r",,terminal_output +408,9309927,"TERMINAL",0,0,"Receiving objects: 25% (20794/81085), 40.33 MiB | 109.00 KiB/s\r",,terminal_output +409,9310652,"TERMINAL",0,0,"Receiving objects: 25% (20797/81085), 40.39 MiB | 110.00 KiB/s\r",,terminal_output +410,9312034,"TERMINAL",0,0,"Receiving objects: 25% (20834/81085), 40.47 MiB | 91.00 KiB/s \r",,terminal_output +411,9312656,"TERMINAL",0,0,"Receiving objects: 25% (20847/81085), 40.57 MiB | 98.00 KiB/s\r",,terminal_output +412,9313710,"TERMINAL",0,0,"Receiving objects: 25% (20847/81085), 40.68 MiB | 96.00 KiB/s\r",,terminal_output +413,9314650,"TERMINAL",0,0,"Receiving objects: 25% (20884/81085), 40.74 MiB | 96.00 KiB/s\r",,terminal_output +414,9314883,"TERMINAL",0,0,"Receiving objects: 26% (21083/81085), 40.81 MiB | 97.00 KiB/s\r",,terminal_output +415,9315678,"TERMINAL",0,0,"Receiving objects: 26% (21269/81085), 40.86 MiB | 99.00 KiB/s\r",,terminal_output +416,9316614,"TERMINAL",0,0,"Receiving objects: 26% (21368/81085), 40.95 MiB | 96.00 KiB/s\r",,terminal_output +417,9318114,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.12 MiB | 101.00 KiB/s\r",,terminal_output +418,9318648,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.18 MiB | 103.00 KiB/s\r",,terminal_output +419,9319798,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.29 MiB | 99.00 KiB/s \r",,terminal_output +420,9320925,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.40 MiB | 101.00 KiB/s\r",,terminal_output +421,9322082,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.51 MiB | 102.00 KiB/s\r",,terminal_output +422,9322660,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.54 MiB | 94.00 KiB/s \r",,terminal_output +423,9323700,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.61 MiB | 88.00 KiB/s\r",,terminal_output +424,9324864,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.73 MiB | 90.00 KiB/s\r",,terminal_output +425,9325936,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.82 MiB | 86.00 KiB/s\r",,terminal_output +426,9326991,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.90 MiB | 81.00 KiB/s\r",,terminal_output +427,9328106,"TERMINAL",0,0,"Receiving objects: 26% (21499/81085), 41.99 MiB | 86.00 KiB/s\r",,terminal_output +428,9328595,"TERMINAL",0,0,"Receiving objects: 26% (21541/81085), 41.99 MiB | 86.00 KiB/s\r",,terminal_output +429,9329588,"TERMINAL",0,0,"Receiving objects: 26% (21631/81085), 42.11 MiB | 99.00 KiB/s\r",,terminal_output +430,9330857,"TERMINAL",0,0,"Receiving objects: 26% (21733/81085), 42.25 MiB | 87.00 KiB/s\r",,terminal_output +431,9331571,"TERMINAL",0,0,"Receiving objects: 26% (21854/81085), 42.36 MiB | 101.00 KiB/s\r",,terminal_output +432,9331858,"TERMINAL",0,0,"Receiving objects: 27% (21893/81085), 42.36 MiB | 101.00 KiB/s\r",,terminal_output +433,9332576,"TERMINAL",0,0,"Receiving objects: 27% (21956/81085), 42.46 MiB | 105.00 KiB/s\r",,terminal_output +434,9334167,"TERMINAL",0,0,"Receiving objects: 27% (21969/81085), 42.64 MiB | 109.00 KiB/s\r",,terminal_output +435,9334677,"TERMINAL",0,0,"Receiving objects: 27% (21969/81085), 42.70 MiB | 110.00 KiB/s\r",,terminal_output +436,9335775,"TERMINAL",0,0,"Receiving objects: 27% (21969/81085), 42.79 MiB | 113.00 KiB/s\r",,terminal_output +437,9336971,"TERMINAL",0,0,"Receiving objects: 27% (21969/81085), 42.89 MiB | 98.00 KiB/s \r",,terminal_output +438,9337609,"TERMINAL",0,0,"Receiving objects: 27% (21969/81085), 42.94 MiB | 96.00 KiB/s\r",,terminal_output +439,9338569,"TERMINAL",0,0,"Receiving objects: 27% (22048/81085), 43.02 MiB | 98.00 KiB/s\r",,terminal_output +440,9339862,"TERMINAL",0,0,"Receiving objects: 27% (22175/81085), 43.14 MiB | 89.00 KiB/s\r",,terminal_output +441,9340624,"TERMINAL",0,0,"Receiving objects: 27% (22297/81085), 43.26 MiB | 100.00 KiB/s\r",,terminal_output +442,9341624,"TERMINAL",0,0,"Receiving objects: 27% (22419/81085), 43.39 MiB | 106.00 KiB/s\r",,terminal_output +443,9342788,"TERMINAL",0,0,"Receiving objects: 27% (22551/81085), 43.50 MiB | 107.00 KiB/s\r",,terminal_output +444,9343666,"TERMINAL",0,0,"Receiving objects: 27% (22642/81085), 43.58 MiB | 115.00 KiB/s\r",,terminal_output +445,9343940,"TERMINAL",0,0,"Receiving objects: 28% (22704/81085), 43.64 MiB | 112.00 KiB/s\r",,terminal_output +446,9344626,"TERMINAL",0,0,"Receiving objects: 28% (22740/81085), 43.70 MiB | 111.00 KiB/s\r",,terminal_output +447,9345971,"TERMINAL",0,0,"Receiving objects: 28% (22829/81085), 43.86 MiB | 108.00 KiB/s\r",,terminal_output +448,9347023,"TERMINAL",0,0,"Receiving objects: 28% (22829/81085), 43.98 MiB | 108.00 KiB/s\r",,terminal_output +449,9348098,"TERMINAL",0,0,"Receiving objects: 28% (22829/81085), 44.10 MiB | 110.00 KiB/s\r",,terminal_output +450,9348616,"TERMINAL",0,0,"Receiving objects: 28% (22829/81085), 44.16 MiB | 110.00 KiB/s\r",,terminal_output +451,9349704,"TERMINAL",0,0,"Receiving objects: 28% (22829/81085), 44.28 MiB | 112.00 KiB/s\r",,terminal_output +452,9350615,"TERMINAL",0,0,"Receiving objects: 28% (22906/81085), 44.34 MiB | 115.00 KiB/s\r",,terminal_output +453,9351577,"TERMINAL",0,0,"Receiving objects: 28% (22996/81085), 44.45 MiB | 113.00 KiB/s\r",,terminal_output +454,9352906,"TERMINAL",0,0,"Receiving objects: 28% (23277/81085), 44.61 MiB | 109.00 KiB/s\r",,terminal_output +455,9354013,"TERMINAL",0,0,"Receiving objects: 28% (23277/81085), 44.73 MiB | 106.00 KiB/s\r",,terminal_output +456,9354597,"TERMINAL",0,0,"Receiving objects: 28% (23277/81085), 44.79 MiB | 107.00 KiB/s\r",,terminal_output +457,9355662,"TERMINAL",0,0,"Receiving objects: 28% (23278/81085), 44.82 MiB | 93.00 KiB/s \r",,terminal_output +458,9356604,"TERMINAL",0,0,"Receiving objects: 28% (23279/81085), 45.00 MiB | 105.00 KiB/s\r",,terminal_output +459,9357602,"TERMINAL",0,0,"Receiving objects: 28% (23281/81085), 45.07 MiB | 111.00 KiB/s\r",,terminal_output +460,9358787,"TERMINAL",0,0,"Receiving objects: 28% (23281/81085), 45.25 MiB | 108.00 KiB/s\r",,terminal_output +461,9359859,"TERMINAL",0,0,"Receiving objects: 28% (23281/81085), 45.36 MiB | 111.00 KiB/s\r",,terminal_output +462,9360990,"TERMINAL",0,0,"Receiving objects: 28% (23282/81085), 45.48 MiB | 111.00 KiB/s\r",,terminal_output +463,9362087,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 45.61 MiB | 113.00 KiB/s\r",,terminal_output +464,9362687,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 45.67 MiB | 112.00 KiB/s\r",,terminal_output +465,9363930,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 45.71 MiB | 91.00 KiB/s \r",,terminal_output +466,9365034,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 45.88 MiB | 102.00 KiB/s\r",,terminal_output +467,9365601,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 45.95 MiB | 105.00 KiB/s\r",,terminal_output +468,9366637,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 46.07 MiB | 104.00 KiB/s\r",,terminal_output +469,9367665,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 46.19 MiB | 107.00 KiB/s\r",,terminal_output +470,9369175,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 46.31 MiB | 117.00 KiB/s\r",,terminal_output +471,9369700,"TERMINAL",0,0,"Receiving objects: 28% (23283/81085), 46.37 MiB | 108.00 KiB/s\r",,terminal_output +472,9370740,"TERMINAL",0,0,"Receiving objects: 28% (23284/81085), 46.50 MiB | 110.00 KiB/s\r",,terminal_output +473,9371793,"TERMINAL",0,0,"Receiving objects: 28% (23284/81085), 46.64 MiB | 113.00 KiB/s\r",,terminal_output +474,9372741,"TERMINAL",0,0,"Receiving objects: 28% (23289/81085), 46.70 MiB | 111.00 KiB/s\r",,terminal_output +475,9373934,"TERMINAL",0,0,"Receiving objects: 28% (23296/81085), 46.83 MiB | 110.00 KiB/s\r",,terminal_output +476,9374925,"TERMINAL",0,0,"Receiving objects: 28% (23296/81085), 46.92 MiB | 107.00 KiB/s\r",,terminal_output +477,9375888,"TERMINAL",0,0,"Receiving objects: 28% (23297/81085), 47.02 MiB | 100.00 KiB/s\r",,terminal_output +478,9377058,"TERMINAL",0,0,"Receiving objects: 28% (23297/81085), 47.20 MiB | 109.00 KiB/s\r",,terminal_output +479,9378076,"TERMINAL",0,0,"Receiving objects: 28% (23299/81085), 47.32 MiB | 112.00 KiB/s\r",,terminal_output +480,9378585,"TERMINAL",0,0,"Receiving objects: 28% (23299/81085), 47.39 MiB | 112.00 KiB/s\r",,terminal_output +481,9379714,"TERMINAL",0,0,"Receiving objects: 28% (23299/81085), 47.52 MiB | 116.00 KiB/s\r",,terminal_output +482,9380761,"TERMINAL",0,0,"Receiving objects: 28% (23300/81085), 47.63 MiB | 128.00 KiB/s\r",,terminal_output +483,9381844,"TERMINAL",0,0,"Receiving objects: 28% (23301/81085), 47.75 MiB | 118.00 KiB/s\r",,terminal_output +484,9383417,"TERMINAL",0,0,"Receiving objects: 28% (23307/81085), 47.86 MiB | 103.00 KiB/s\r",,terminal_output +485,9383619,"TERMINAL",0,0,"Receiving objects: 28% (23312/81085), 47.86 MiB | 103.00 KiB/s\r",,terminal_output +486,9384590,"TERMINAL",0,0,"Receiving objects: 28% (23430/81085), 48.02 MiB | 110.00 KiB/s\r",,terminal_output +487,9385258,"TERMINAL",0,0,"Receiving objects: 29% (23515/81085), 48.07 MiB | 108.00 KiB/s\r",,terminal_output +488,9385622,"TERMINAL",0,0,"Receiving objects: 29% (23556/81085), 48.13 MiB | 105.00 KiB/s\r",,terminal_output +489,9386663,"TERMINAL",0,0,"Receiving objects: 29% (23566/81085), 48.21 MiB | 99.00 KiB/s \r",,terminal_output +490,9388310,"TERMINAL",0,0,"Receiving objects: 29% (23566/81085), 48.25 MiB | 74.00 KiB/s\r",,terminal_output +491,9389174,"TERMINAL",0,0,"Receiving objects: 29% (23566/81085), 48.28 MiB | 73.00 KiB/s\r",,terminal_output +492,9389697,"TERMINAL",0,0,"Receiving objects: 29% (23566/81085), 48.45 MiB | 85.00 KiB/s\r",,terminal_output +493,9390638,"TERMINAL",0,0,"Receiving objects: 29% (23634/81085), 48.54 MiB | 92.00 KiB/s\r",,terminal_output +494,9391568,"TERMINAL",0,0,"Receiving objects: 29% (23845/81085), 48.68 MiB | 96.00 KiB/s\r",,terminal_output +495,9392603,"TERMINAL",0,0,"Receiving objects: 29% (23962/81085), 48.79 MiB | 102.00 KiB/s\r",,terminal_output +496,9393618,"TERMINAL",0,0,"Receiving objects: 29% (24268/81085), 48.92 MiB | 130.00 KiB/s\r",,terminal_output +497,9393922,"TERMINAL",0,0,"Receiving objects: 30% (24326/81085), 48.92 MiB | 130.00 KiB/s\r",,terminal_output +498,9394631,"TERMINAL",0,0,"Receiving objects: 30% (24395/81085), 49.04 MiB | 121.00 KiB/s\r",,terminal_output +499,9395722,"TERMINAL",0,0,"Receiving objects: 30% (24485/81085), 49.14 MiB | 111.00 KiB/s\r",,terminal_output +500,9396862,"TERMINAL",0,0,"Receiving objects: 30% (24529/81085), 49.25 MiB | 106.00 KiB/s\r",,terminal_output +501,9397924,"TERMINAL",0,0,"Receiving objects: 30% (24529/81085), 49.38 MiB | 108.00 KiB/s\r",,terminal_output +502,9399304,"TERMINAL",0,0,"Receiving objects: 30% (24529/81085), 49.47 MiB | 98.00 KiB/s \r",,terminal_output +503,9399823,"TERMINAL",0,0,"Receiving objects: 30% (24529/81085), 49.58 MiB | 107.00 KiB/s\r",,terminal_output +504,9400604,"TERMINAL",0,0,"Receiving objects: 30% (24598/81085), 49.64 MiB | 107.00 KiB/s\r",,terminal_output +505,9401866,"TERMINAL",0,0,"Receiving objects: 30% (24754/81085), 49.83 MiB | 118.00 KiB/s\r",,terminal_output +506,9402942,"TERMINAL",0,0,"Receiving objects: 30% (24754/81085), 49.95 MiB | 116.00 KiB/s\r",,terminal_output +507,9404015,"TERMINAL",0,0,"Receiving objects: 30% (24754/81085), 50.05 MiB | 125.00 KiB/s\r",,terminal_output +508,9405075,"TERMINAL",0,0,"Receiving objects: 30% (24754/81085), 50.16 MiB | 111.00 KiB/s\r",,terminal_output +509,9405585,"TERMINAL",0,0,"Receiving objects: 30% (24755/81085), 50.22 MiB | 113.00 KiB/s\r",,terminal_output +510,9406671,"TERMINAL",0,0,"Receiving objects: 30% (24768/81085), 50.25 MiB | 98.00 KiB/s \r",,terminal_output +511,9407573,"TERMINAL",0,0,"Receiving objects: 30% (24938/81085), 50.41 MiB | 106.00 KiB/s\r",,terminal_output +512,9409054,"TERMINAL",0,0,"Receiving objects: 30% (25067/81085), 50.60 MiB | 111.00 KiB/s\r",,terminal_output +513,9409591,"TERMINAL",0,0,"Receiving objects: 30% (25085/81085), 50.65 MiB | 111.00 KiB/s\r",,terminal_output +514,9409955,"TERMINAL",0,0,"Receiving objects: 31% (25137/81085), 50.65 MiB | 111.00 KiB/s\r",,terminal_output +515,9410622,"TERMINAL",0,0,"Receiving objects: 31% (25224/81085), 50.77 MiB | 111.00 KiB/s\r",,terminal_output +516,9411699,"TERMINAL",0,0,"Receiving objects: 31% (25224/81085), 50.88 MiB | 114.00 KiB/s\r",,terminal_output +517,9413098,"TERMINAL",0,0,"Receiving objects: 31% (25224/81085), 51.02 MiB | 108.00 KiB/s\r",,terminal_output +518,9413670,"TERMINAL",0,0,"Receiving objects: 31% (25224/81085), 51.09 MiB | 109.00 KiB/s\r",,terminal_output +519,9414604,"TERMINAL",0,0,"Receiving objects: 31% (25303/81085), 51.14 MiB | 108.00 KiB/s\r",,terminal_output +520,9415608,"TERMINAL",0,0,"Receiving objects: 31% (25403/81085), 51.23 MiB | 103.00 KiB/s\r",,terminal_output +521,9416971,"TERMINAL",0,0,"Receiving objects: 31% (25552/81085), 51.40 MiB | 100.00 KiB/s\r",,terminal_output +522,9418002,"TERMINAL",0,0,"Receiving objects: 31% (25552/81085), 51.50 MiB | 99.00 KiB/s \r",,terminal_output +523,9419048,"TERMINAL",0,0,"Receiving objects: 31% (25552/81085), 51.58 MiB | 92.00 KiB/s\r",,terminal_output +524,9419620,"TERMINAL",0,0,"Receiving objects: 31% (25552/81085), 51.64 MiB | 93.00 KiB/s\r",,terminal_output +525,9420946,"TERMINAL",0,0,"Receiving objects: 31% (25552/81085), 51.74 MiB | 92.00 KiB/s\r",,terminal_output +526,9421564,"TERMINAL",0,0,"Receiving objects: 31% (25552/81085), 51.86 MiB | 102.00 KiB/s\r",,terminal_output +527,9422585,"TERMINAL",0,0,"Receiving objects: 31% (25723/81085), 51.93 MiB | 105.00 KiB/s\r",,terminal_output +528,9423655,"TERMINAL",0,0,"Receiving objects: 31% (25802/81085), 52.04 MiB | 103.00 KiB/s\r",,terminal_output +529,9425051,"TERMINAL",0,0,"Receiving objects: 31% (25905/81085), 52.20 MiB | 106.00 KiB/s\r",,terminal_output +530,9425957,"TERMINAL",0,0,"Receiving objects: 31% (25906/81085), 52.25 MiB | 103.00 KiB/s\r",,terminal_output +531,9426536,"TERMINAL",0,0,"Receiving objects: 32% (25948/81085), 52.31 MiB | 113.00 KiB/s\r",,terminal_output +532,9426657,"TERMINAL",0,0,"Receiving objects: 32% (25963/81085), 52.37 MiB | 103.00 KiB/s\r",,terminal_output +533,9428141,"TERMINAL",0,0,"Receiving objects: 32% (26086/81085), 52.46 MiB | 89.00 KiB/s \r",,terminal_output +534,9428647,"TERMINAL",0,0,"Receiving objects: 32% (26095/81085), 52.55 MiB | 99.00 KiB/s\r",,terminal_output +535,9429769,"TERMINAL",0,0,"Receiving objects: 32% (26095/81085), 52.71 MiB | 103.00 KiB/s\r",,terminal_output +536,9430814,"TERMINAL",0,0,"Receiving objects: 32% (26095/81085), 52.82 MiB | 111.00 KiB/s\r",,terminal_output +537,9431905,"TERMINAL",0,0,"Receiving objects: 32% (26095/81085), 52.91 MiB | 105.00 KiB/s\r",,terminal_output +538,9432937,"TERMINAL",0,0,"Receiving objects: 32% (26095/81085), 53.04 MiB | 124.00 KiB/s\r",,terminal_output +539,9434211,"TERMINAL",0,0,"Receiving objects: 32% (26095/81085), 53.16 MiB | 105.00 KiB/s\r",,terminal_output +540,9435023,"TERMINAL",0,0,"Receiving objects: 32% (26095/81085), 53.21 MiB | 98.00 KiB/s \r",,terminal_output +541,9435818,"TERMINAL",0,0,"Receiving objects: 32% (26096/81085), 53.30 MiB | 103.00 KiB/s\r",,terminal_output +542,9436661,"TERMINAL",0,0,"Receiving objects: 32% (26193/81085), 53.42 MiB | 106.00 KiB/s\r",,terminal_output +543,9437621,"TERMINAL",0,0,"Receiving objects: 32% (26373/81085), 53.46 MiB | 107.00 KiB/s\r",,terminal_output +544,9438805,"TERMINAL",0,0,"Receiving objects: 32% (26477/81085), 53.54 MiB | 86.00 KiB/s \rReceiving objects: 33% (26759/81085), 53.54 MiB | 86.00 KiB/s\r",,terminal_output +545,9439330,"TERMINAL",0,0,"Receiving objects: 34% (27569/81085), 53.67 MiB | 99.00 KiB/s\r",,terminal_output +546,9439628,"TERMINAL",0,0,"Receiving objects: 34% (28175/81085), 53.67 MiB | 99.00 KiB/s\r",,terminal_output +547,9439968,"TERMINAL",0,0,"Receiving objects: 35% (28380/81085), 53.77 MiB | 109.00 KiB/s\r",,terminal_output +548,9441017,"TERMINAL",0,0,"Receiving objects: 35% (28497/81085), 53.91 MiB | 114.00 KiB/s\r",,terminal_output +549,9441650,"TERMINAL",0,0,"Receiving objects: 35% (28498/81085), 53.97 MiB | 115.00 KiB/s\r",,terminal_output +550,9442624,"TERMINAL",0,0,"Receiving objects: 35% (28598/81085), 54.07 MiB | 114.00 KiB/s\r",,terminal_output +551,9444117,"TERMINAL",0,0,"Receiving objects: 35% (29051/81085), 54.19 MiB | 126.00 KiB/s\rReceiving objects: 36% (29191/81085), 54.19 MiB | 126.00 KiB/s\r",,terminal_output +552,9444568,"TERMINAL",0,0,"Receiving objects: 36% (29394/81085), 54.19 MiB | 126.00 KiB/s\r",,terminal_output +553,9445652,"TERMINAL",0,0,"Receiving objects: 36% (29481/81085), 54.40 MiB | 112.00 KiB/s\r",,terminal_output +554,9446577,"TERMINAL",0,0,"Receiving objects: 36% (29589/81085), 54.45 MiB | 106.00 KiB/s\r",,terminal_output +555,9447012,"TERMINAL",0,0,"Receiving objects: 37% (30002/81085), 54.51 MiB | 106.00 KiB/s\r",,terminal_output +556,9447579,"TERMINAL",0,0,"Receiving objects: 37% (30509/81085), 54.54 MiB | 103.00 KiB/s\r",,terminal_output +557,9447785,"TERMINAL",0,0,"Receiving objects: 38% (30813/81085), 54.62 MiB | 108.00 KiB/s\r",,terminal_output +558,9448637,"TERMINAL",0,0,"Receiving objects: 38% (31360/81085), 54.69 MiB | 110.00 KiB/s\r",,terminal_output +559,9449568,"TERMINAL",0,0,"Receiving objects: 38% (31539/81085), 54.80 MiB | 107.00 KiB/s\r",,terminal_output +560,9449931,"TERMINAL",0,0,"Receiving objects: 39% (31624/81085), 54.86 MiB | 112.00 KiB/s\r",,terminal_output +561,9450607,"TERMINAL",0,0,"Receiving objects: 39% (31902/81085), 54.93 MiB | 113.00 KiB/s\r",,terminal_output +562,9451822,"TERMINAL",0,0,"Receiving objects: 39% (32030/81085), 55.04 MiB | 111.00 KiB/s\r",,terminal_output +563,9452588,"TERMINAL",0,0,"Receiving objects: 39% (32212/81085), 55.13 MiB | 108.00 KiB/s\r",,terminal_output +564,9453611,"TERMINAL",0,0,"Receiving objects: 39% (32388/81085), 55.19 MiB | 106.00 KiB/s\r",,terminal_output +565,9453797,"TERMINAL",0,0,"Receiving objects: 40% (32434/81085), 55.25 MiB | 105.00 KiB/s\r",,terminal_output +566,9454614,"TERMINAL",0,0,"Receiving objects: 40% (32639/81085), 55.30 MiB | 104.00 KiB/s\r",,terminal_output +567,9455601,"TERMINAL",0,0,"Receiving objects: 40% (33233/81085), 55.41 MiB | 98.00 KiB/s \r",,terminal_output +568,9455670,"TERMINAL",0,0,"Receiving objects: 41% (33245/81085), 55.41 MiB | 98.00 KiB/s\r",,terminal_output +569,9456595,"TERMINAL",0,0,"Receiving objects: 41% (33405/81085), 55.54 MiB | 102.00 KiB/s\r",,terminal_output +570,9457703,"TERMINAL",0,0,"Receiving objects: 41% (33547/81085), 55.62 MiB | 98.00 KiB/s \r",,terminal_output +571,9458825,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 55.79 MiB | 105.00 KiB/s\r",,terminal_output +572,9459963,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 55.93 MiB | 114.00 KiB/s\r",,terminal_output +573,9461006,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.04 MiB | 114.00 KiB/s\r",,terminal_output +574,9461626,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.11 MiB | 114.00 KiB/s\r",,terminal_output +575,9462777,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.24 MiB | 124.00 KiB/s\r",,terminal_output +576,9463968,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.38 MiB | 118.00 KiB/s\r",,terminal_output +577,9464664,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.40 MiB | 106.00 KiB/s\r",,terminal_output +578,9465811,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.57 MiB | 112.00 KiB/s\r",,terminal_output +579,9466929,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.68 MiB | 110.00 KiB/s\r",,terminal_output +580,9468030,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.79 MiB | 106.00 KiB/s\r",,terminal_output +581,9468618,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.85 MiB | 105.00 KiB/s\r",,terminal_output +582,9469745,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 56.96 MiB | 113.00 KiB/s\r",,terminal_output +583,9471144,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.05 MiB | 93.00 KiB/s \r",,terminal_output +584,9471909,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.14 MiB | 97.00 KiB/s\r",,terminal_output +585,9472932,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.28 MiB | 101.00 KiB/s\r",,terminal_output +586,9474046,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 57.39 MiB | 101.00 KiB/s\r",,terminal_output +587,9474610,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 57.46 MiB | 106.00 KiB/s\r",,terminal_output +588,9475863,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 57.59 MiB | 105.00 KiB/s\r",,terminal_output +589,9477004,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 57.71 MiB | 114.00 KiB/s\r",,terminal_output +590,9478071,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 57.84 MiB | 112.00 KiB/s\r",,terminal_output +591,9478621,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 57.90 MiB | 113.00 KiB/s\r",,terminal_output +592,9479686,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.01 MiB | 110.00 KiB/s\r",,terminal_output +593,9480852,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.14 MiB | 112.00 KiB/s\r",,terminal_output +594,9482018,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.25 MiB | 108.00 KiB/s\r",,terminal_output +595,9483126,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.37 MiB | 107.00 KiB/s\r",,terminal_output +596,9483682,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.43 MiB | 107.00 KiB/s\r",,terminal_output +597,9484757,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.55 MiB | 108.00 KiB/s\r",,terminal_output +598,9485795,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.61 MiB | 99.00 KiB/s \r",,terminal_output +599,9486804,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.77 MiB | 109.00 KiB/s\r",,terminal_output +600,9487944,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 58.89 MiB | 111.00 KiB/s\r",,terminal_output +601,9489078,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.00 MiB | 108.00 KiB/s\r",,terminal_output +602,9489667,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.07 MiB | 108.00 KiB/s\r",,terminal_output +603,9490712,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.19 MiB | 122.00 KiB/s\r",,terminal_output +604,9491738,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.30 MiB | 110.00 KiB/s\r",,terminal_output +605,9492827,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.43 MiB | 111.00 KiB/s\r",,terminal_output +606,9493922,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.56 MiB | 117.00 KiB/s\r",,terminal_output +607,9494983,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.67 MiB | 114.00 KiB/s\r",,terminal_output +608,9496178,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.75 MiB | 106.00 KiB/s\r",,terminal_output +609,9496729,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.86 MiB | 113.00 KiB/s\r",,terminal_output +610,9497787,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 59.96 MiB | 109.00 KiB/s\r",,terminal_output +611,9498900,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.09 MiB | 109.00 KiB/s\r",,terminal_output +612,9500034,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.20 MiB | 107.00 KiB/s\r",,terminal_output +613,9500577,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.27 MiB | 109.00 KiB/s\r",,terminal_output +614,9501644,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.39 MiB | 112.00 KiB/s\r",,terminal_output +615,9502616,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.46 MiB | 107.00 KiB/s\r",,terminal_output +616,9503771,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.64 MiB | 116.00 KiB/s\r",,terminal_output +617,9504867,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.76 MiB | 120.00 KiB/s\r",,terminal_output +618,9506047,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.88 MiB | 114.00 KiB/s\r",,terminal_output +619,9506604,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 60.93 MiB | 110.00 KiB/s\r",,terminal_output +620,9507752,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.04 MiB | 116.00 KiB/s\r",,terminal_output +621,9508953,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.05 MiB | 85.00 KiB/s \r",,terminal_output +622,9510075,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.30 MiB | 106.00 KiB/s\r",,terminal_output +623,9510957,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.36 MiB | 99.00 KiB/s \r",,terminal_output +624,9512019,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.54 MiB | 112.00 KiB/s\r",,terminal_output +625,9513110,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.66 MiB | 115.00 KiB/s\r",,terminal_output +626,9513641,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.72 MiB | 118.00 KiB/s\r",,terminal_output +627,9514713,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.83 MiB | 118.00 KiB/s\r",,terminal_output +628,9515795,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 61.96 MiB | 127.00 KiB/s\r",,terminal_output +629,9516866,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.01 MiB | 100.00 KiB/s\r",,terminal_output +630,9517963,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.20 MiB | 113.00 KiB/s\r",,terminal_output +631,9519033,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.32 MiB | 115.00 KiB/s\r",,terminal_output +632,9519606,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.39 MiB | 116.00 KiB/s\r",,terminal_output +633,9520694,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.52 MiB | 117.00 KiB/s\r",,terminal_output +634,9521790,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.64 MiB | 131.00 KiB/s\r",,terminal_output +635,9522845,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.76 MiB | 118.00 KiB/s\r",,terminal_output +636,9523933,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.87 MiB | 114.00 KiB/s\r",,terminal_output +637,9524991,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 62.99 MiB | 113.00 KiB/s\r",,terminal_output +638,9526109,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.11 MiB | 110.00 KiB/s\r",,terminal_output +639,9526638,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.15 MiB | 107.00 KiB/s\r",,terminal_output +640,9527700,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.27 MiB | 107.00 KiB/s\r",,terminal_output +641,9529312,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.38 MiB | 96.00 KiB/s \r",,terminal_output +642,9529814,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.48 MiB | 106.00 KiB/s\r",,terminal_output +643,9531037,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.61 MiB | 105.00 KiB/s\r",,terminal_output +644,9531601,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.66 MiB | 103.00 KiB/s\r",,terminal_output +645,9532662,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.77 MiB | 103.00 KiB/s\r",,terminal_output +646,9533920,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 63.91 MiB | 106.00 KiB/s\r",,terminal_output +647,9534992,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.03 MiB | 108.00 KiB/s\r",,terminal_output +648,9536108,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.16 MiB | 112.00 KiB/s\r",,terminal_output +649,9536653,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.21 MiB | 112.00 KiB/s\r",,terminal_output +650,9537689,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.34 MiB | 116.00 KiB/s\r",,terminal_output +651,9539259,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.46 MiB | 106.00 KiB/s\r",,terminal_output +652,9539822,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.57 MiB | 114.00 KiB/s\r",,terminal_output +653,9540921,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.69 MiB | 113.00 KiB/s\r",,terminal_output +654,9542093,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.80 MiB | 110.00 KiB/s\r",,terminal_output +655,9542653,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.86 MiB | 107.00 KiB/s\r",,terminal_output +656,9543895,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 64.99 MiB | 107.00 KiB/s\r",,terminal_output +657,9545030,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.10 MiB | 103.00 KiB/s\r",,terminal_output +658,9546096,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.20 MiB | 100.00 KiB/s\r",,terminal_output +659,9547000,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.23 MiB | 91.00 KiB/s \r",,terminal_output +660,9548072,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.40 MiB | 103.00 KiB/s\r",,terminal_output +661,9548620,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.45 MiB | 100.00 KiB/s\r",,terminal_output +662,9549702,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.57 MiB | 107.00 KiB/s\r",,terminal_output +663,9550834,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.69 MiB | 108.00 KiB/s\r",,terminal_output +664,9551862,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.82 MiB | 123.00 KiB/s\r",,terminal_output +665,9553306,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 65.89 MiB | 94.00 KiB/s \r",,terminal_output +666,9553866,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 66.00 MiB | 108.00 KiB/s\r",,terminal_output +667,9555137,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 66.14 MiB | 107.00 KiB/s\r",,terminal_output +668,9555693,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 66.21 MiB | 109.00 KiB/s\r",,terminal_output +669,9556070,"train_dynamics.py",4881,0,"",python,selection_keyboard +670,9556506,"train_dynamics.py",4891,0,"",python,selection_command +671,9556670,"train_dynamics.py",4915,0,"",python,selection_command +672,9556749,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 66.32 MiB | 107.00 KiB/s\r",,terminal_output +673,9557914,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 66.43 MiB | 106.00 KiB/s\r",,terminal_output +674,9558938,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 66.54 MiB | 108.00 KiB/s\r",,terminal_output +675,9560478,"TERMINAL",0,0,"Receiving objects: 41% (33905/81085), 66.67 MiB | 100.00 KiB/s\r",,terminal_output +676,9560935,"TERMINAL",0,0,"Receiving objects: 41% (33906/81085), 66.67 MiB | 100.00 KiB/s\r",,terminal_output +677,9561602,"TERMINAL",0,0,"Receiving objects: 41% (34029/81085), 66.86 MiB | 117.00 KiB/s\r",,terminal_output +678,9561670,"TERMINAL",0,0,"Receiving objects: 42% (34056/81085), 66.86 MiB | 117.00 KiB/s\r",,terminal_output +679,9562569,"TERMINAL",0,0,"Receiving objects: 42% (34244/81085), 66.93 MiB | 114.00 KiB/s\r",,terminal_output +680,9563586,"TERMINAL",0,0,"Receiving objects: 42% (34467/81085), 67.05 MiB | 119.00 KiB/s\r",,terminal_output +681,9564752,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.22 MiB | 117.00 KiB/s\r",,terminal_output +682,9565770,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.33 MiB | 114.00 KiB/s\r",,terminal_output +683,9566859,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.46 MiB | 114.00 KiB/s\r",,terminal_output +684,9568092,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.58 MiB | 110.00 KiB/s\r",,terminal_output +685,9568503,"train_dynamics.py",4881,0,"",python,selection_keyboard +686,9568664,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.65 MiB | 111.00 KiB/s\r",,terminal_output +687,9569184,"train_dynamics.py",4891,0,"",python,selection_command +688,9569512,"train_dynamics.py",4915,0,"",python,selection_command +689,9569809,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.75 MiB | 106.00 KiB/s\r",,terminal_output +690,9570950,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.87 MiB | 106.00 KiB/s\r",,terminal_output +691,9572061,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.97 MiB | 101.00 KiB/s\r",,terminal_output +692,9572611,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.03 MiB | 101.00 KiB/s\r",,terminal_output +693,9573871,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.16 MiB | 99.00 KiB/s \r",,terminal_output +694,9575053,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.30 MiB | 108.00 KiB/s\r",,terminal_output +695,9575577,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.36 MiB | 109.00 KiB/s\r",,terminal_output +696,9576602,"TERMINAL",0,0,"Receiving objects: 42% (34543/81085), 68.43 MiB | 108.00 KiB/s\r",,terminal_output +697,9577812,"TERMINAL",0,0,"Receiving objects: 42% (34778/81085), 68.61 MiB | 113.00 KiB/s\r",,terminal_output +698,9578891,"TERMINAL",0,0,"Receiving objects: 42% (34778/81085), 68.71 MiB | 113.00 KiB/s\r",,terminal_output +699,9580003,"TERMINAL",0,0,"Receiving objects: 42% (34778/81085), 68.85 MiB | 113.00 KiB/s\r",,terminal_output +700,9581085,"TERMINAL",0,0,"Receiving objects: 42% (34778/81085), 68.97 MiB | 114.00 KiB/s\r",,terminal_output +701,9581598,"TERMINAL",0,0,"Receiving objects: 42% (34778/81085), 69.02 MiB | 111.00 KiB/s\r",,terminal_output +702,9582225,"TERMINAL",0,0,"Receiving objects: 43% (34867/81085), 69.08 MiB | 112.00 KiB/s\r",,terminal_output +703,9582995,"TERMINAL",0,0,"Receiving objects: 43% (34909/81085), 69.11 MiB | 100.00 KiB/s\r",,terminal_output +704,9583651,"TERMINAL",0,0,"Receiving objects: 43% (35198/81085), 69.21 MiB | 109.00 KiB/s\r",,terminal_output +705,9584565,"TERMINAL",0,0,"Receiving objects: 43% (35440/81085), 69.29 MiB | 111.00 KiB/s\r",,terminal_output +706,9585686,"TERMINAL",0,0,"Receiving objects: 43% (35610/81085), 69.46 MiB | 111.00 KiB/s\r",,terminal_output +707,9586219,"TERMINAL",0,0,"Receiving objects: 44% (35678/81085), 69.53 MiB | 110.00 KiB/s\r",,terminal_output +708,9586671,"TERMINAL",0,0,"Receiving objects: 44% (35706/81085), 69.53 MiB | 110.00 KiB/s\r",,terminal_output +709,9587653,"TERMINAL",0,0,"Receiving objects: 44% (35859/81085), 69.63 MiB | 108.00 KiB/s\r",,terminal_output +710,9588603,"TERMINAL",0,0,"Receiving objects: 44% (36077/81085), 69.71 MiB | 107.00 KiB/s\r",,terminal_output +711,9590045,"TERMINAL",0,0,"Receiving objects: 44% (36098/81085), 69.90 MiB | 102.00 KiB/s\r",,terminal_output +712,9590569,"TERMINAL",0,0,"Receiving objects: 44% (36167/81085), 69.96 MiB | 104.00 KiB/s\r",,terminal_output +713,9591620,"TERMINAL",0,0,"Receiving objects: 44% (36316/81085), 70.07 MiB | 103.00 KiB/s\r",,terminal_output +714,9592746,"TERMINAL",0,0,"Receiving objects: 44% (36386/81085), 70.13 MiB | 93.00 KiB/s \rReceiving objects: 45% (36489/81085), 70.13 MiB | 93.00 KiB/s\r",,terminal_output +715,9593628,"TERMINAL",0,0,"Receiving objects: 45% (36621/81085), 70.24 MiB | 105.00 KiB/s\r",,terminal_output +716,9594613,"TERMINAL",0,0,"Receiving objects: 45% (37183/81085), 70.36 MiB | 106.00 KiB/s\rReceiving objects: 46% (37300/81085), 70.36 MiB | 106.00 KiB/s\r",,terminal_output +717,9595201,"TERMINAL",0,0,"Receiving objects: 47% (38110/81085), 70.42 MiB | 109.00 KiB/s\r",,terminal_output +718,9595585,"TERMINAL",0,0,"Receiving objects: 47% (38317/81085), 70.47 MiB | 110.00 KiB/s\r",,terminal_output +719,9596668,"TERMINAL",0,0,"Receiving objects: 47% (38857/81085), 70.59 MiB | 110.00 KiB/s\r",,terminal_output +720,9597016,"TERMINAL",0,0,"Receiving objects: 48% (38921/81085), 70.65 MiB | 111.00 KiB/s\r",,terminal_output +721,9597616,"TERMINAL",0,0,"Receiving objects: 48% (39015/81085), 70.71 MiB | 125.00 KiB/s\r",,terminal_output +722,9598577,"TERMINAL",0,0,"Receiving objects: 48% (39629/81085), 70.82 MiB | 113.00 KiB/s\r",,terminal_output +723,9598707,"TERMINAL",0,0,"Receiving objects: 49% (39732/81085), 70.82 MiB | 113.00 KiB/s\r",,terminal_output +724,9599604,"TERMINAL",0,0,"Receiving objects: 49% (40125/81085), 70.87 MiB | 110.00 KiB/s\r",,terminal_output +725,9600124,"TERMINAL",0,0,"Receiving objects: 50% (40543/81085), 70.92 MiB | 103.00 KiB/s\r",,terminal_output +726,9600591,"TERMINAL",0,0,"Receiving objects: 50% (40719/81085), 70.98 MiB | 104.00 KiB/s\r",,terminal_output +727,9602208,"TERMINAL",0,0,"Receiving objects: 50% (40870/81085), 71.08 MiB | 86.00 KiB/s \r",,terminal_output +728,9602335,"TERMINAL",0,0,"Receiving objects: 51% (41354/81085), 71.08 MiB | 86.00 KiB/s\r",,terminal_output +729,9602577,"TERMINAL",0,0,"Receiving objects: 51% (41502/81085), 71.08 MiB | 86.00 KiB/s\r",,terminal_output +730,9603808,"TERMINAL",0,0,"Receiving objects: 51% (41580/81085), 71.32 MiB | 95.00 KiB/s\r",,terminal_output +731,9604892,"TERMINAL",0,0,"Receiving objects: 51% (41741/81085), 71.42 MiB | 96.00 KiB/s\r",,terminal_output +732,9605991,"TERMINAL",0,0,"Receiving objects: 51% (41741/81085), 71.55 MiB | 103.00 KiB/s\r",,terminal_output +733,9607084,"TERMINAL",0,0,"Receiving objects: 51% (41741/81085), 71.68 MiB | 124.00 KiB/s\r",,terminal_output +734,9607638,"TERMINAL",0,0,"Receiving objects: 51% (41741/81085), 71.72 MiB | 111.00 KiB/s\r",,terminal_output +735,9608809,"TERMINAL",0,0,"Receiving objects: 51% (41741/81085), 71.86 MiB | 110.00 KiB/s\r",,terminal_output +736,9609577,"TERMINAL",0,0,"Receiving objects: 51% (41788/81085), 71.92 MiB | 113.00 KiB/s\r",,terminal_output +737,9610642,"TERMINAL",0,0,"Receiving objects: 51% (41918/81085), 72.04 MiB | 114.00 KiB/s\r",,terminal_output +738,9611026,"TERMINAL",0,0,"Receiving objects: 52% (42165/81085), 72.08 MiB | 109.00 KiB/s\r",,terminal_output +739,9611668,"TERMINAL",0,0,"Receiving objects: 52% (42656/81085), 72.14 MiB | 109.00 KiB/s\r",,terminal_output +740,9611891,"TERMINAL",0,0,"Receiving objects: 53% (42976/81085), 72.14 MiB | 109.00 KiB/s\r",,terminal_output +741,9612565,"TERMINAL",0,0,"Receiving objects: 53% (43351/81085), 72.26 MiB | 112.00 KiB/s\r",,terminal_output +742,9612865,"TERMINAL",0,0,"Receiving objects: 54% (43786/81085), 72.26 MiB | 112.00 KiB/s\r",,terminal_output +743,9613605,"TERMINAL",0,0,"Receiving objects: 54% (44435/81085), 72.38 MiB | 111.00 KiB/s\r",,terminal_output +744,9613952,"TERMINAL",0,0,"Receiving objects: 55% (44597/81085), 72.38 MiB | 111.00 KiB/s\r",,terminal_output +745,9614716,"TERMINAL",0,0,"Receiving objects: 55% (44799/81085), 72.46 MiB | 100.00 KiB/s\r",,terminal_output +746,9614888,"TERMINAL",0,0,"Receiving objects: 56% (45408/81085), 72.46 MiB | 100.00 KiB/s\r",,terminal_output +747,9615566,"TERMINAL",0,0,"Receiving objects: 56% (45701/81085), 72.57 MiB | 113.00 KiB/s\r",,terminal_output +748,9616579,"TERMINAL",0,0,"Receiving objects: 56% (45787/81085), 72.68 MiB | 114.00 KiB/s\r",,terminal_output +749,9617618,"TERMINAL",0,0,"Receiving objects: 56% (45926/81085), 72.80 MiB | 114.00 KiB/s\r",,terminal_output +750,9618609,"TERMINAL",0,0,"Receiving objects: 56% (46036/81085), 72.91 MiB | 111.00 KiB/s\r",,terminal_output +751,9619384,"TERMINAL",0,0,"Receiving objects: 57% (46219/81085), 72.97 MiB | 110.00 KiB/s\r",,terminal_output +752,9619576,"TERMINAL",0,0,"Receiving objects: 57% (46244/81085), 73.03 MiB | 120.00 KiB/s\r",,terminal_output +753,9620572,"TERMINAL",0,0,"Receiving objects: 57% (46443/81085), 73.09 MiB | 107.00 KiB/s\r",,terminal_output +754,9621609,"TERMINAL",0,0,"Receiving objects: 57% (46554/81085), 73.19 MiB | 105.00 KiB/s\r",,terminal_output +755,9622310,"TERMINAL",0,0,"Receiving objects: 58% (47030/81085), 73.29 MiB | 101.00 KiB/s\r",,terminal_output +756,9622946,"TERMINAL",0,0,"Receiving objects: 58% (47097/81085), 73.29 MiB | 90.00 KiB/s \r",,terminal_output +757,9624039,"TERMINAL",0,0,"Receiving objects: 58% (47257/81085), 73.46 MiB | 100.00 KiB/s\r",,terminal_output +758,9624576,"TERMINAL",0,0,"Receiving objects: 58% (47257/81085), 73.52 MiB | 100.00 KiB/s\r",,terminal_output +759,9625623,"TERMINAL",0,0,"Receiving objects: 58% (47257/81085), 73.64 MiB | 104.00 KiB/s\r",,terminal_output +760,9626655,"TERMINAL",0,0,"Receiving objects: 58% (47305/81085), 73.69 MiB | 103.00 KiB/s\r",,terminal_output +761,9628141,"TERMINAL",0,0,"Receiving objects: 58% (47788/81085), 73.80 MiB | 100.00 KiB/s\r",,terminal_output +762,9629011,"TERMINAL",0,0,"Receiving objects: 58% (47802/81085), 73.81 MiB | 75.00 KiB/s \r",,terminal_output +763,9629145,"TERMINAL",0,0,"Receiving objects: 59% (47841/81085), 73.81 MiB | 75.00 KiB/s\r",,terminal_output +764,9629575,"TERMINAL",0,0,"Receiving objects: 59% (47879/81085), 73.85 MiB | 71.00 KiB/s\r",,terminal_output +765,9630438,"TERMINAL",0,0,"Receiving objects: 60% (48651/81085), 73.94 MiB | 78.00 KiB/s\r",,terminal_output +766,9630570,"TERMINAL",0,0,"Receiving objects: 60% (48898/81085), 74.10 MiB | 96.00 KiB/s\r",,terminal_output +767,9631601,"TERMINAL",0,0,"Receiving objects: 60% (49195/81085), 74.16 MiB | 97.00 KiB/s\r",,terminal_output +768,9632809,"TERMINAL",0,0,"Receiving objects: 60% (49268/81085), 74.36 MiB | 106.00 KiB/s\r",,terminal_output +769,9633855,"TERMINAL",0,0,"Receiving objects: 60% (49268/81085), 74.46 MiB | 138.00 KiB/s\r",,terminal_output +770,9634944,"TERMINAL",0,0,"Receiving objects: 60% (49268/81085), 74.59 MiB | 136.00 KiB/s\r",,terminal_output +771,9635582,"TERMINAL",0,0,"Receiving objects: 60% (49268/81085), 74.66 MiB | 115.00 KiB/s\r",,terminal_output +772,9636686,"TERMINAL",0,0,"Receiving objects: 60% (49393/81085), 74.79 MiB | 118.00 KiB/s\r",,terminal_output +773,9637163,"TERMINAL",0,0,"Receiving objects: 61% (49462/81085), 74.79 MiB | 118.00 KiB/s\r",,terminal_output +774,9637765,"TERMINAL",0,0,"Receiving objects: 61% (49686/81085), 74.92 MiB | 116.00 KiB/s\r",,terminal_output +775,9638825,"TERMINAL",0,0,"Receiving objects: 61% (49686/81085), 75.03 MiB | 115.00 KiB/s\r",,terminal_output +776,9639885,"TERMINAL",0,0,"Receiving objects: 61% (49686/81085), 75.14 MiB | 113.00 KiB/s\r",,terminal_output +777,9641060,"TERMINAL",0,0,"Receiving objects: 61% (49686/81085), 75.28 MiB | 112.00 KiB/s\r",,terminal_output +778,9641569,"TERMINAL",0,0,"Receiving objects: 61% (49686/81085), 75.34 MiB | 114.00 KiB/s\r",,terminal_output +779,9642654,"TERMINAL",0,0,"Receiving objects: 61% (49866/81085), 75.46 MiB | 112.00 KiB/s\r",,terminal_output +780,9643124,"TERMINAL",0,0,"Receiving objects: 62% (50273/81085), 75.46 MiB | 112.00 KiB/s\r",,terminal_output +781,9643629,"TERMINAL",0,0,"Receiving objects: 62% (50689/81085), 75.52 MiB | 112.00 KiB/s\r",,terminal_output +782,9644534,"TERMINAL",0,0,"Receiving objects: 63% (51084/81085), 75.64 MiB | 113.00 KiB/s\r",,terminal_output +783,9644642,"TERMINAL",0,0,"Receiving objects: 63% (51194/81085), 75.64 MiB | 113.00 KiB/s\r",,terminal_output +784,9645785,"TERMINAL",0,0,"Receiving objects: 63% (51345/81085), 75.74 MiB | 109.00 KiB/s\r",,terminal_output +785,9646640,"TERMINAL",0,0,"Receiving objects: 63% (51537/81085), 75.84 MiB | 101.00 KiB/s\r",,terminal_output +786,9647720,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 75.96 MiB | 101.00 KiB/s\r",,terminal_output +787,9648879,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.07 MiB | 99.00 KiB/s \r",,terminal_output +788,9649998,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.19 MiB | 100.00 KiB/s\r",,terminal_output +789,9650728,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.25 MiB | 99.00 KiB/s \r",,terminal_output +790,9651957,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.39 MiB | 105.00 KiB/s\r",,terminal_output +791,9653292,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.46 MiB | 93.00 KiB/s \r",,terminal_output +792,9653845,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.59 MiB | 105.00 KiB/s\r",,terminal_output +793,9654932,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.69 MiB | 103.00 KiB/s\r",,terminal_output +794,9655984,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.81 MiB | 108.00 KiB/s\r",,terminal_output +795,9657008,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.92 MiB | 107.00 KiB/s\r",,terminal_output +796,9657616,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 76.97 MiB | 108.00 KiB/s\r",,terminal_output +797,9658752,"TERMINAL",0,0,"Receiving objects: 63% (51812/81085), 77.10 MiB | 106.00 KiB/s\r",,terminal_output +798,9659586,"TERMINAL",0,0,"Receiving objects: 63% (51813/81085), 77.16 MiB | 109.00 KiB/s\r",,terminal_output +799,9659978,"TERMINAL",0,0,"Receiving objects: 64% (51895/81085), 77.20 MiB | 107.00 KiB/s\r",,terminal_output +800,9661053,"TERMINAL",0,0,"Receiving objects: 64% (52033/81085), 77.35 MiB | 108.00 KiB/s\r",,terminal_output +801,9661589,"TERMINAL",0,0,"Receiving objects: 64% (52033/81085), 77.40 MiB | 107.00 KiB/s\r",,terminal_output +802,9662808,"TERMINAL",0,0,"Receiving objects: 64% (52033/81085), 77.52 MiB | 107.00 KiB/s\r",,terminal_output +803,9663864,"TERMINAL",0,0,"Receiving objects: 64% (52033/81085), 77.65 MiB | 111.00 KiB/s\r",,terminal_output +804,9664794,"TERMINAL",0,0,"Receiving objects: 64% (52034/81085), 77.71 MiB | 111.00 KiB/s\r",,terminal_output +805,9665970,"TERMINAL",0,0,"Receiving objects: 64% (52081/81085), 77.89 MiB | 112.00 KiB/s\r",,terminal_output +806,9667076,"TERMINAL",0,0,"Receiving objects: 64% (52081/81085), 78.00 MiB | 112.00 KiB/s\r",,terminal_output +807,9667717,"TERMINAL",0,0,"Receiving objects: 64% (52081/81085), 78.07 MiB | 114.00 KiB/s\r",,terminal_output +808,9668917,"TERMINAL",0,0,"Receiving objects: 64% (52081/81085), 78.19 MiB | 109.00 KiB/s\r",,terminal_output +809,9669681,"TERMINAL",0,0,"Receiving objects: 64% (52082/81085), 78.25 MiB | 108.00 KiB/s\r",,terminal_output +810,9670653,"TERMINAL",0,0,"Receiving objects: 64% (52188/81085), 78.33 MiB | 102.00 KiB/s\r",,terminal_output +811,9671707,"TERMINAL",0,0,"Receiving objects: 64% (52338/81085), 78.46 MiB | 103.00 KiB/s\r",,terminal_output +812,9672824,"TERMINAL",0,0,"Receiving objects: 64% (52338/81085), 78.58 MiB | 103.00 KiB/s\r",,terminal_output +813,9673869,"TERMINAL",0,0,"Receiving objects: 64% (52338/81085), 78.66 MiB | 96.00 KiB/s \r",,terminal_output +814,9674976,"TERMINAL",0,0,"Receiving objects: 64% (52338/81085), 78.80 MiB | 104.00 KiB/s\r",,terminal_output +815,9676060,"TERMINAL",0,0,"Receiving objects: 64% (52338/81085), 78.90 MiB | 105.00 KiB/s\r",,terminal_output +816,9676627,"TERMINAL",0,0,"Receiving objects: 64% (52415/81085), 78.98 MiB | 107.00 KiB/s\r",,terminal_output +817,9677870,"TERMINAL",0,0,"Receiving objects: 64% (52625/81085), 79.08 MiB | 101.00 KiB/s\rReceiving objects: 65% (52706/81085), 79.08 MiB | 101.00 KiB/s\r",,terminal_output +818,9678656,"TERMINAL",0,0,"Receiving objects: 65% (53089/81085), 79.19 MiB | 113.00 KiB/s\r",,terminal_output +819,9679625,"TERMINAL",0,0,"Receiving objects: 65% (53288/81085), 79.31 MiB | 109.00 KiB/s\r",,terminal_output +820,9680001,"TERMINAL",0,0,"Receiving objects: 66% (53517/81085), 79.31 MiB | 109.00 KiB/s\r",,terminal_output +821,9680571,"TERMINAL",0,0,"Receiving objects: 66% (53828/81085), 79.36 MiB | 112.00 KiB/s\r",,terminal_output +822,9681390,"TERMINAL",0,0,"Receiving objects: 67% (54327/81085), 79.48 MiB | 115.00 KiB/s\r",,terminal_output +823,9681584,"TERMINAL",0,0,"Receiving objects: 67% (54485/81085), 79.48 MiB | 115.00 KiB/s\r",,terminal_output +824,9682433,"TERMINAL",0,0,"Receiving objects: 68% (55138/81085), 79.61 MiB | 112.00 KiB/s\r",,terminal_output +825,9682597,"TERMINAL",0,0,"Receiving objects: 68% (55220/81085), 79.61 MiB | 112.00 KiB/s\r",,terminal_output +826,9683231,"TERMINAL",0,0,"Receiving objects: 69% (55949/81085), 79.67 MiB | 122.00 KiB/s\r",,terminal_output +827,9683668,"TERMINAL",0,0,"Receiving objects: 69% (56339/81085), 79.73 MiB | 113.00 KiB/s\r",,terminal_output +828,9684668,"TERMINAL",0,0,"Receiving objects: 69% (56611/81085), 79.85 MiB | 114.00 KiB/s\r",,terminal_output +829,9685064,"TERMINAL",0,0,"Receiving objects: 70% (56760/81085), 79.90 MiB | 112.00 KiB/s\r",,terminal_output +830,9685908,"TERMINAL",0,0,"Receiving objects: 70% (57286/81085), 79.96 MiB | 113.00 KiB/s\r",,terminal_output +831,9686036,"TERMINAL",0,0,"Receiving objects: 71% (57571/81085), 80.00 MiB | 109.00 KiB/s\r",,terminal_output +832,9686782,"TERMINAL",0,0,"Receiving objects: 71% (57997/81085), 80.05 MiB | 102.00 KiB/s\r",,terminal_output +833,9687140,"TERMINAL",0,0,"Receiving objects: 72% (58382/81085), 80.05 MiB | 102.00 KiB/s\r",,terminal_output +834,9687599,"TERMINAL",0,0,"Receiving objects: 72% (59061/81085), 80.11 MiB | 102.00 KiB/s\r",,terminal_output +835,9687841,"TERMINAL",0,0,"Receiving objects: 73% (59193/81085), 80.21 MiB | 110.00 KiB/s\r",,terminal_output +836,9688596,"TERMINAL",0,0,"Receiving objects: 73% (59474/81085), 80.29 MiB | 112.00 KiB/s\r",,terminal_output +837,9689636,"TERMINAL",0,0,"Receiving objects: 73% (59727/81085), 80.37 MiB | 105.00 KiB/s\r",,terminal_output +838,9689935,"TERMINAL",0,0,"Receiving objects: 74% (60003/81085), 80.42 MiB | 107.00 KiB/s\r",,terminal_output +839,9690686,"TERMINAL",0,0,"Receiving objects: 74% (60509/81085), 80.48 MiB | 108.00 KiB/s\r",,terminal_output +840,9690985,"TERMINAL",0,0,"Receiving objects: 75% (60814/81085), 80.55 MiB | 114.00 KiB/s\r",,terminal_output +841,9691583,"TERMINAL",0,0,"Receiving objects: 75% (61030/81085), 80.61 MiB | 119.00 KiB/s\r",,terminal_output +842,9692565,"TERMINAL",0,0,"Receiving objects: 75% (61408/81085), 80.72 MiB | 110.00 KiB/s\r",,terminal_output +843,9692944,"TERMINAL",0,0,"Receiving objects: 76% (61625/81085), 80.72 MiB | 110.00 KiB/s\r",,terminal_output +844,9693844,"TERMINAL",0,0,"Receiving objects: 76% (61817/81085), 80.82 MiB | 99.00 KiB/s \r",,terminal_output +845,9694590,"TERMINAL",0,0,"Receiving objects: 76% (62321/81085), 80.89 MiB | 102.00 KiB/s\rReceiving objects: 77% (62436/81085), 80.89 MiB | 102.00 KiB/s\r",,terminal_output +846,9695124,"TERMINAL",0,0,"Receiving objects: 78% (63247/81085), 81.00 MiB | 113.00 KiB/s\r",,terminal_output +847,9695709,"TERMINAL",0,0,"Receiving objects: 78% (63652/81085), 81.05 MiB | 111.00 KiB/s\r",,terminal_output +848,9697000,"TERMINAL",0,0,"Receiving objects: 78% (63666/81085), 81.11 MiB | 93.00 KiB/s \r",,terminal_output +849,9697198,"TERMINAL",0,0,"Receiving objects: 79% (64058/81085), 81.11 MiB | 93.00 KiB/s\r",,terminal_output +850,9697562,"TERMINAL",0,0,"Receiving objects: 79% (64627/81085), 81.22 MiB | 104.00 KiB/s\r",,terminal_output +851,9697685,"TERMINAL",0,0,"Receiving objects: 80% (64868/81085), 81.22 MiB | 104.00 KiB/s\r",,terminal_output +852,9698642,"TERMINAL",0,0,"Receiving objects: 80% (65277/81085), 81.34 MiB | 102.00 KiB/s\r",,terminal_output +853,9699541,"TERMINAL",0,0,"Receiving objects: 81% (65679/81085), 81.39 MiB | 109.00 KiB/s\r",,terminal_output +854,9699611,"TERMINAL",0,0,"Receiving objects: 81% (65743/81085), 81.39 MiB | 109.00 KiB/s\r",,terminal_output +855,9700767,"TERMINAL",0,0,"Receiving objects: 81% (66342/81085), 81.54 MiB | 98.00 KiB/s \r",,terminal_output +856,9701575,"TERMINAL",0,0,"Receiving objects: 81% (66429/81085), 81.60 MiB | 100.00 KiB/s\r",,terminal_output +857,9701647,"TERMINAL",0,0,"Receiving objects: 82% (66490/81085), 81.60 MiB | 100.00 KiB/s\r",,terminal_output +858,9702234,"TERMINAL",0,0,"Receiving objects: 83% (67301/81085), 81.68 MiB | 117.00 KiB/s\r",,terminal_output +859,9702607,"TERMINAL",0,0,"Receiving objects: 83% (67729/81085), 81.74 MiB | 106.00 KiB/s\r",,terminal_output +860,9703134,"TERMINAL",0,0,"Receiving objects: 84% (68112/81085), 81.79 MiB | 104.00 KiB/s\r",,terminal_output +861,9703561,"TERMINAL",0,0,"Receiving objects: 84% (68601/81085), 81.85 MiB | 105.00 KiB/s\r",,terminal_output +862,9703900,"TERMINAL",0,0,"Receiving objects: 85% (68923/81085), 81.85 MiB | 105.00 KiB/s\r",,terminal_output +863,9704574,"TERMINAL",0,0,"Receiving objects: 85% (69501/81085), 81.91 MiB | 109.00 KiB/s\r",,terminal_output +864,9704697,"TERMINAL",0,0,"Receiving objects: 86% (69734/81085), 81.97 MiB | 112.00 KiB/s\r",,terminal_output +865,9705426,"TERMINAL",0,0,"Receiving objects: 87% (70544/81085), 82.04 MiB | 115.00 KiB/s\r",,terminal_output +866,9705620,"TERMINAL",0,0,"Receiving objects: 87% (70623/81085), 82.04 MiB | 115.00 KiB/s\r",,terminal_output +867,9706333,"TERMINAL",0,0,"Receiving objects: 88% (71355/81085), 82.10 MiB | 113.00 KiB/s\r",,terminal_output +868,9706584,"TERMINAL",0,0,"Receiving objects: 88% (71471/81085), 82.17 MiB | 117.00 KiB/s\r",,terminal_output +869,9707617,"TERMINAL",0,0,"Receiving objects: 88% (71714/81085), 82.29 MiB | 111.00 KiB/s\r",,terminal_output +870,9708075,"TERMINAL",0,0,"Receiving objects: 89% (72166/81085), 82.33 MiB | 112.00 KiB/s\r",,terminal_output +871,9708990,"TERMINAL",0,0,"Receiving objects: 89% (72304/81085), 82.37 MiB | 98.00 KiB/s \r",,terminal_output +872,9709709,"TERMINAL",0,0,"Receiving objects: 89% (72938/81085), 82.48 MiB | 107.00 KiB/s\rReceiving objects: 90% (72977/81085), 82.48 MiB | 107.00 KiB/s\r",,terminal_output +873,9710593,"TERMINAL",0,0,"Receiving objects: 90% (73429/81085), 82.58 MiB | 103.00 KiB/s\r",,terminal_output +874,9710934,"TERMINAL",0,0,"Receiving objects: 91% (73788/81085), 82.58 MiB | 103.00 KiB/s\r",,terminal_output +875,9711586,"TERMINAL",0,0,"Receiving objects: 91% (74199/81085), 82.64 MiB | 103.00 KiB/s\r",,terminal_output +876,9712066,"TERMINAL",0,0,"Receiving objects: 92% (74599/81085), 82.70 MiB | 102.00 KiB/s\r",,terminal_output +877,9712638,"TERMINAL",0,0,"Receiving objects: 92% (74976/81085), 82.75 MiB | 104.00 KiB/s\r",,terminal_output +878,9713115,"TERMINAL",0,0,"Receiving objects: 93% (75410/81085), 82.82 MiB | 103.00 KiB/s\r",,terminal_output +879,9713592,"TERMINAL",0,0,"Receiving objects: 93% (75758/81085), 82.88 MiB | 105.00 KiB/s\r",,terminal_output +880,9714313,"TERMINAL",0,0,"Receiving objects: 94% (76220/81085), 82.93 MiB | 117.00 KiB/s\r",,terminal_output +881,9714625,"TERMINAL",0,0,"Receiving objects: 94% (76430/81085), 83.00 MiB | 109.00 KiB/s\r",,terminal_output +882,9715354,"TERMINAL",0,0,"error: RPC failed; curl 18 Transferred a partial file\r\nerror: 4510 bytes of body are still expected\r\nfetch-pack: unexpected disconnect while reading sideband packet\r\nfatal: early EOF\r\nfatal: fetch-pack: invalid index-pack output\r\n% \r \r",,terminal_output +883,9978854,"TERMINAL",0,0,"git clone https://github.com/tinygrad/tinygrad.git",,terminal_command +884,9978924,"TERMINAL",0,0,"]633;CCloning into 'tinygrad'...\r\n",,terminal_output +885,9981360,"TERMINAL",0,0,"remote: Enumerating objects: 81085, done.\r\nremote: Counting objects: 0% (1/175)\rremote: Counting objects: 1% (2/175)\rremote: Counting objects: 2% (4/175)\rremote: Counting objects: 3% (6/175)\rremote: Counting objects: 4% (7/175)\rremote: Counting objects: 5% (9/175)\rremote: Counting objects: 6% (11/175)\rremote: Counting objects: 7% (13/175)\rremote: Counting objects: 8% (14/175)\rremote: Counting objects: 9% (16/175)\rremote: Counting objects: 10% (18/175)\rremote: Counting objects: 11% (20/175)\rremote: Counting objects: 12% (21/175)\rremote: Counting objects: 13% (23/175)\rremote: Counting objects: 14% (25/175)\rremote: Counting objects: 15% (27/175)\rremote: Counting objects: 16% (28/175)\rremote: Counting objects: 17% (30/175)\rremote: Counting objects: 18% (32/175)\rremote: Counting objects: 19% (34/175)\rremote: Counting objects: 20% (35/175)\rremote: Counting objects: 21% (37/175)\rremote: Counting objects: 22% (39/175)\rremote: Counting objects: 23% (41/175)\rremote: Counting objects: 24% (42/175)\rremote: Counting objects: 25% (44/175)\rremote: Counting objects: 26% (46/175)\rremote: Counting objects: 27% (48/175)\rremote: Counting objects: 28% (49/175)\rremote: Counting objects: 29% (51/175)\rremote: Counting objects: 30% (53/175)\rremote: Counting objects: 31% (55/175)\rremote: Counting objects: 32% (56/175)\rremote: Counting objects: 33% (58/175)\rremote: Counting objects: 34% (60/175)\rremote: Counting objects: 35% (62/175)\rremote: Counting objects: 36% (63/175)\rremote: Counting objects: 37% (65/175)\rremote: Counting objects: 38% (67/175)\rremote: Counting objects: 39% (69/175)\rremote: Counting objects: 40% (70/175)\rremote: Counting objects: 41% (72/175)\rremote: Counting objects: 42% (74/175)\rremote: Counting objects: 43% (76/175)\rremote: Counting objects: 44% (77/175)\rremote: Counting objects: 45% (79/175)\rremote: Counting objects: 46% (81/175)\rremote: Counting objects: 47% (83/175)\rremote: Counting objects: 48% (84/175)\rremote: Counting objects: 49% (86/175)\rremote: Counting objects: 50% (88/175)\rremote: Counting objects: 51% (90/175)\rremote: Counting objects: 52% (91/175)\rremote: Counting objects: 53% (93/175)\rremote: Counting objects: 54% (95/175)\rremote: Counting objects: 55% (97/175)\rremote: Counting objects: 56% (98/175)\rremote: Counting objects: 57% (100/175)\rremote: Counting objects: 58% (102/175)\rremote: Counting objects: 59% (104/175)\rremote: Counting objects: 60% (105/175)\rremote: Counting objects: 61% (107/175)\rremote: Counting objects: 62% (109/175)\rremote: Counting objects: 63% (111/175)\rremote: Counting objects: 64% (112/175)\rremote: Counting objects: 65% (114/175)\rremote: Counting objects: 66% (116/175)\rremote: Counting objects: 67% (118/175)\rremote: Counting objects: 68% (119/175)\rremote: Counting objects: 69% (121/175)\rremote: Counting objects: 70% (123/175)\rremote: Counting objects: 71% (125/175)\rremote: Counting objects: 72% (126/175)\rremote: Counting objects: 73% (128/175)\rremote: Counting objects: 74% (130/175)\rremote: Counting objects: 75% (132/175)\rremote: Counting objects: 76% (133/175)\rremote: Counting objects: 77% (135/175)\rremote: Counting objects: 78% (137/175)\rremote: Counting objects: 79% (139/175)\rremote: Counting objects: 80% (140/175)\rremote: Counting objects: 81% (142/175)\rremote: Counting objects: 82% (144/175)\rremote: Counting objects: 83% (146/175)\rremote: Counting objects: 84% (147/175)\rremote: Counting objects: 85% (149/175)\rremote: Counting objects: 86% (151/175)\rremote: Counting objects: 87% (153/175)\rremote: Counting objects: 88% (154/175)\rremote: Counting objects: 89% (156/175)\rremote: Counting objects: 90% (158/175)\rremote: Counting objects: 91% (160/175)\rremote: Counting objects: 92% (161/175)\rremote: Counting objects: 93% (163/175)\rremote: Counting objects: 94% (165/175)\rremote: Counting objects: 95% (167/175)\rremote: Counting objects: 96% (168/175)\rremote: Counting objects: 97% (170/175)\rremote: Counting objects: 98% (172/175)\rremote: Counting objects: 99% (174/175)\rremote: Counting objects: 100% (175/175)\rremote: Counting objects: 100% (175/175), done.\r\nremote: Compressing objects: 1% (1/93)\rremote: Compressing objects: 2% (2/93)\rremote: Compressing objects: 3% (3/93)\rremote: Compressing objects: 4% (4/93)\rremote: Compressing objects: 5% (5/93)\rremote: Compressing objects: 6% (6/93)\rremote: Compressing objects: 7% (7/93)\rremote: Compressing objects: 8% (8/93)\rremote: Compressing objects: 9% (9/93)\rremote: Compressing objects: 10% (10/93)\rremote: Compressing objects: 11% (11/93)\rremote: Compressing objects: 12% (12/93)\rremote: Compressing objects: 13% (13/93)\rremote: Compressing objects: 15% (14/93)\rremote: Compressing objects: 16% (15/93)\rremote: Compressing objects: 17% (16/93)\rremote: Compressing objects: 18% (17/93)\rremote: Compressing objects: 19% (18/93)\rremote: Compressing objects: 20% (19/93)\rremote: Compressing objects: 21% (20/93)\rremote: Compressing objects: 22% (21/93)\rremote: Compressing objects: 23% (22/93)\rremote: Compressing objects: 24% (23/93)\rremote: Compressing objects: 25% (24/93)\rremote: Compressing objects: 26% (25/93)\rremote: Compressing objects: 27% (26/93)\rremote: Compressing objects: 29% (27/93)\rremote: Compressing objects: 30% (28/93)\rremote: Compressing objects: 31% (29/93)\rremote: Compressing objects: 32% (30/93)\rremote: Compressing objects: 33% (31/93)\rremote: Compressing objects: 34% (32/93)\rremote: Compressing objects: 35% (33/93)\rremote: Compressing objects: 36% (34/93)\rremote: Compressing objects: 37% (35/93)\rremote: Compressing objects: 38% (36/93)\rremote: Compressing objects: 39% (37/93)\rremote: Compressing objects: 40% (38/93)\rremote: Compressing objects: 41% (39/93)\rremote: Compressing objects: 43% (40/93)\rremote: Compressing objects: 44% (41/93)\rremote: Compressing objects: 45% (42/93)\rremote: Compressing objects: 46% (43/93)\rremote: Compressing objects: 47% (44/93)\rremote: Compressing objects: 48% (45/93)\rremote: Compressing objects: 49% (46/93)\rremote: Compressing objects: 50% (47/93)\rremote: Compressing objects: 51% (48/93)\rremote: Compressing objects: 52% (49/93)\rremote: Compressing objects: 53% (50/93)\rremote: Compressing objects: 54% (51/93)\rremote: Compressing objects: 55% (52/93)\rremote: Compressing objects: 56% (53/93)\rremote: Compressing objects: 58% (54/93)\rremote: Compressing objects: 59% (55/93)\rremote: Compressing objects: 60% (56/93)\rremote: Compressing objects: 61% (57/93)\rremote: Compressing objects: 62% (58/93)\rremote: Compressing objects: 63% (59/93)\rremote: Compressing objects: 64% (60/93)\rremote: Compressing objects: 65% (61/93)\rremote: Compressing objects: 66% (62/93)\rremote: Compressing objects: 67% (63/93)\rremote: Compressing objects: 68% (64/93)\rremote: Compressing objects: 69% (65/93)\rremote: Compressing objects: 70% (66/93)\rremote: Compressing objects: 72% (67/93)\rremote: Compressing objects: 73% (68/93)\rremote: Compressing objects: 74% (69/93)\rremote: Compressing objects: 75% (70/93)\rremote: Compressing objects: 76% (71/93)\rremote: Compressing objects: 77% (72/93)\rremote: Compressing objects: 78% (73/93)\rremote: Compressing objects: 79% (74/93)\rremote: Compressing objects: 80% (75/93)\rremote: Compressing objects: 81% (76/93)\rremote: Compressing objects: 82% (77/93)\rremote: Compressing objects: 83% (78/93)\rremote: Compressing objects: 84% (79/93)\rremote: Compressing objects: 86% (80/93)\rremote: Compressing objects: 87% (81/93)\rremote: Compressing objects: 88% (82/93)\rremote: Compressing objects: 89% (83/93)\rremote: Compressing objects: 90% (84/93)\rremote: Compressing objects: 91% (85/93)\rremote: Compressing objects: 92% (86/93)\rremote: Compressing objects: 93% (87/93)\rremote: Compressing objects: 94% (88/93)\rremote: Compressing objects: 95% (89/93)\rremote: Compressing objects: 96% (90/93)\rremote: Compressing objects: 97% (91/93)\rremote: Compressing objects: 98% (92/93)\rremote: Compressing objects: 100% (93/93)\rremote: Compressing objects: 100% (93/93), done.\r\nReceiving objects: 0% (1/81085)\r",,terminal_output +886,9982390,"TERMINAL",0,0,"Receiving objects: 0% (352/81085), 172.00 KiB | 268.00 KiB/s\r",,terminal_output +887,9983429,"TERMINAL",0,0,"Receiving objects: 0% (573/81085), 316.00 KiB | 184.00 KiB/s\r",,terminal_output +888,9984456,"TERMINAL",0,0,"Receiving objects: 0% (739/81085), 460.00 KiB | 167.00 KiB/s\r",,terminal_output +889,9984896,"TERMINAL",0,0,"Receiving objects: 1% (811/81085), 508.00 KiB | 155.00 KiB/s\r",,terminal_output +890,9985412,"TERMINAL",0,0,"Receiving objects: 1% (896/81085), 564.00 KiB | 149.00 KiB/s\r",,terminal_output +891,9986360,"TERMINAL",0,0,"Receiving objects: 1% (1049/81085), 684.00 KiB | 138.00 KiB/s\r",,terminal_output +892,9987378,"TERMINAL",0,0,"Receiving objects: 1% (1228/81085), 812.00 KiB | 121.00 KiB/s\r",,terminal_output +893,9989596,"TERMINAL",0,0,"^Cfetch-pack: unexpected disconnect while reading sideband packet\r\n\r\n% \r \r",,terminal_output +894,9989769,"TERMINAL",0,0,"",,terminal_command +895,9989769,"TERMINAL",0,0,"]633;C",,terminal_output +896,11856716,"TERMINAL",0,0,"git clone https://github.com/tinygrad/tinygrad.git",,terminal_command +897,11856778,"TERMINAL",0,0,"]633;CCloning into 'tinygrad'...\r\n",,terminal_output +898,11860866,"TERMINAL",0,0,"remote: Enumerating objects: 81085, done.\r\nremote: Counting objects: 0% (1/175)\rremote: Counting objects: 1% (2/175)\rremote: Counting objects: 2% (4/175)\rremote: Counting objects: 3% (6/175)\rremote: Counting objects: 4% (7/175)\rremote: Counting objects: 5% (9/175)\rremote: Counting objects: 6% (11/175)\rremote: Counting objects: 7% (13/175)\rremote: Counting objects: 8% (14/175)\rremote: Counting objects: 9% (16/175)\rremote: Counting objects: 10% (18/175)\rremote: Counting objects: 11% (20/175)\rremote: Counting objects: 12% (21/175)\rremote: Counting objects: 13% (23/175)\rremote: Counting objects: 14% (25/175)\rremote: Counting objects: 15% (27/175)\rremote: Counting objects: 16% (28/175)\rremote: Counting objects: 17% (30/175)\rremote: Counting objects: 18% (32/175)\rremote: Counting objects: 19% (34/175)\rremote: Counting objects: 20% (35/175)\rremote: Counting objects: 21% (37/175)\rremote: Counting objects: 22% (39/175)\rremote: Counting objects: 23% (41/175)\rremote: Counting objects: 24% (42/175)\rremote: Counting objects: 25% (44/175)\rremote: Counting objects: 26% (46/175)\rremote: Counting objects: 27% (48/175)\rremote: Counting objects: 28% (49/175)\rremote: Counting objects: 29% (51/175)\rremote: Counting objects: 30% (53/175)\rremote: Counting objects: 31% (55/175)\rremote: Counting objects: 32% (56/175)\rremote: Counting objects: 33% (58/175)\rremote: Counting objects: 34% (60/175)\rremote: Counting objects: 35% (62/175)\rremote: Counting objects: 36% (63/175)\rremote: Counting objects: 37% (65/175)\rremote: Counting objects: 38% (67/175)\rremote: Counting objects: 39% (69/175)\rremote: Counting objects: 40% (70/175)\rremote: Counting objects: 41% (72/175)\rremote: Counting objects: 42% (74/175)\rremote: Counting objects: 43% (76/175)\rremote: Counting objects: 44% (77/175)\rremote: Counting objects: 45% (79/175)\rremote: Counting objects: 46% (81/175)\rremote: Counting objects: 47% (83/175)\rremote: Counting objects: 48% (84/175)\rremote: Counting objects: 49% (86/175)\rremote: Counting objects: 50% (88/175)\rremote: Counting objects: 51% (90/175)\rremote: Counting objects: 52% (91/175)\rremote: Counting objects: 53% (93/175)\rremote: Counting objects: 54% (95/175)\rremote: Counting objects: 55% (97/175)\rremote: Counting objects: 56% (98/175)\rremote: Counting objects: 57% (100/175)\rremote: Counting objects: 58% (102/175)\rremote: Counting objects: 59% (104/175)\rremote: Counting objects: 60% (105/175)\rremote: Counting objects: 61% (107/175)\rremote: Counting objects: 62% (109/175)\rremote: Counting objects: 63% (111/175)\rremote: Counting objects: 64% (112/175)\rremote: Counting objects: 65% (114/175)\rremote: Counting objects: 66% (116/175)\rremote: Counting objects: 67% (118/175)\rremote: Counting objects: 68% (119/175)\rremote: Counting objects: 69% (121/175)\rremote: Counting objects: 70% (123/175)\rremote: Counting objects: 71% (125/175)\rremote: Counting objects: 72% (126/175)\rremote: Counting objects: 73% (128/175)\rremote: Counting objects: 74% (130/175)\rremote: Counting objects: 75% (132/175)\rremote: Counting objects: 76% (133/175)\rremote: Counting objects: 77% (135/175)\rremote: Counting objects: 78% (137/175)\rremote: Counting objects: 79% (139/175)\rremote: Counting objects: 80% (140/175)\rremote: Counting objects: 81% (142/175)\rremote: Counting objects: 82% (144/175)\rremote: Counting objects: 83% (146/175)\rremote: Counting objects: 84% (147/175)\rremote: Counting objects: 85% (149/175)\rremote: Counting objects: 86% (151/175)\rremote: Counting objects: 87% (153/175)\rremote: Counting objects: 88% (154/175)\rremote: Counting objects: 89% (156/175)\rremote: Counting objects: 90% (158/175)\rremote: Counting objects: 91% (160/175)\rremote: Counting objects: 92% (161/175)\rremote: Counting objects: 93% (163/175)\rremote: Counting objects: 94% (165/175)\rremote: Counting objects: 95% (167/175)\rremote: Counting objects: 96% (168/175)\rremote: Counting objects: 97% (170/175)\rremote: Counting objects: 98% (172/175)\rremote: Counting objects: 99% (174/175)\rremote: Counting objects: 100% (175/175)\rremote: Counting objects: 100% (175/175), done.\r\nremote: Compressing objects: 1% (1/92)\rremote: Compressing objects: 2% (2/92)\rremote: Compressing objects: 3% (3/92)\rremote: Compressing objects: 4% (4/92)\rremote: Compressing objects: 5% (5/92)\rremote: Compressing objects: 6% (6/92)\rremote: Compressing objects: 7% (7/92)\rremote: Compressing objects: 8% (8/92)\rremote: Compressing objects: 9% (9/92)\rremote: Compressing objects: 10% (10/92)\rremote: Compressing objects: 11% (11/92)\rremote: Compressing objects: 13% (12/92)\rremote: Compressing objects: 14% (13/92)\rremote: Compressing objects: 15% (14/92)\rremote: Compressing objects: 16% (15/92)\rremote: Compressing objects: 17% (16/92)\rremote: Compressing objects: 18% (17/92)\rremote: Compressing objects: 19% (18/92)\rremote: Compressing objects: 20% (19/92)\rremote: Compressing objects: 21% (20/92)\rremote: Compressing objects: 22% (21/92)\rremote: Compressing objects: 23% (22/92)\rremote: Compressing objects: 25% (23/92)\rremote: Compressing objects: 26% (24/92)\rremote: Compressing objects: 27% (25/92)\rremote: Compressing objects: 28% (26/92)\rremote: Compressing objects: 29% (27/92)\rremote: Compressing objects: 30% (28/92)\rremote: Compressing objects: 31% (29/92)\rremote: Compressing objects: 32% (30/92)\rremote: Compressing objects: 33% (31/92)\rremote: Compressing objects: 34% (32/92)\rremote: Compressing objects: 35% (33/92)\rremote: Compressing objects: 36% (34/92)\rremote: Compressing objects: 38% (35/92)\rremote: Compressing objects: 39% (36/92)\rremote: Compressing objects: 40% (37/92)\rremote: Compressing objects: 41% (38/92)\rremote: Compressing objects: 42% (39/92)\rremote: Compressing objects: 43% (40/92)\rremote: Compressing objects: 44% (41/92)\rremote: Compressing objects: 45% (42/92)\rremote: Compressing objects: 46% (43/92)\rremote: Compressing objects: 47% (44/92)\rremote: Compressing objects: 48% (45/92)\rremote: Compressing objects: 50% (46/92)\rremote: Compressing objects: 51% (47/92)\rremote: Compressing objects: 52% (48/92)\rremote: Compressing objects: 53% (49/92)\rremote: Compressing objects: 54% (50/92)\rremote: Compressing objects: 55% (51/92)\rremote: Compressing objects: 56% (52/92)\rremote: Compressing objects: 57% (53/92)\rremote: Compressing objects: 58% (54/92)\rremote: Compressing objects: 59% (55/92)\rremote: Compressing objects: 60% (56/92)\rremote: Compressing objects: 61% (57/92)\rremote: Compressing objects: 63% (58/92)\rremote: Compressing objects: 64% (59/92)\rremote: Compressing objects: 65% (60/92)\rremote: Compressing objects: 66% (61/92)\rremote: Compressing objects: 67% (62/92)\rremote: Compressing objects: 68% (63/92)\rremote: Compressing objects: 69% (64/92)\rremote: Compressing objects: 70% (65/92)\rremote: Compressing objects: 71% (66/92)\rremote: Compressing objects: 72% (67/92)\rremote: Compressing objects: 73% (68/92)\rremote: Compressing objects: 75% (69/92)\rremote: Compressing objects: 76% (70/92)\rremote: Compressing objects: 77% (71/92)\rremote: Compressing objects: 78% (72/92)\rremote: Compressing objects: 79% (73/92)\rremote: Compressing objects: 80% (74/92)\rremote: Compressing objects: 81% (75/92)\rremote: Compressing objects: 82% (76/92)\rremote: Compressing objects: 83% (77/92)\rremote: Compressing objects: 84% (78/92)\rremote: Compressing objects: 85% (79/92)\rremote: Compressing objects: 86% (80/92)\rremote: Compressing objects: 88% (81/92)\rremote: Compressing objects: 89% (82/92)\rremote: Compressing objects: 90% (83/92)\rremote: Compressing objects: 91% (84/92)\rremote: Compressing objects: 92% (85/92)\rremote: Compressing objects: 93% (86/92)\rremote: Compressing objects: 94% (87/92)\rremote: Compressing objects: 95% (88/92)\rremote: Compressing objects: 96% (89/92)\rremote: Compressing objects: 97% (90/92)\rremote: Compressing objects: 98% (91/92)\rremote: Compressing objects: 100% (92/92)\rremote: Compressing objects: 100% (92/92), done.\r\n",,terminal_output +899,11860919,"TERMINAL",0,0,"Receiving objects: 0% (1/81085)\r",,terminal_output +900,11861940,"TERMINAL",0,0,"Receiving objects: 0% (246/81085), 84.00 KiB | 138.00 KiB/s\r",,terminal_output +901,11862990,"TERMINAL",0,0,"Receiving objects: 0% (446/81085), 236.00 KiB | 136.00 KiB/s\r",,terminal_output +902,11863950,"TERMINAL",0,0,"Receiving objects: 0% (591/81085), 356.00 KiB | 126.00 KiB/s\r",,terminal_output +903,11865001,"TERMINAL",0,0,"Receiving objects: 0% (728/81085), 468.00 KiB | 118.00 KiB/s\r",,terminal_output +904,11865612,"TERMINAL",0,0,"Receiving objects: 1% (811/81085), 508.00 KiB | 109.00 KiB/s\r",,terminal_output +905,11866004,"TERMINAL",0,0,"Receiving objects: 1% (848/81085), 572.00 KiB | 111.00 KiB/s\r",,terminal_output +906,11866926,"TERMINAL",0,0,"Receiving objects: 1% (960/81085), 612.00 KiB | 104.00 KiB/s\r",,terminal_output +907,11867964,"TERMINAL",0,0,"Receiving objects: 1% (1228/81085), 740.00 KiB | 102.00 KiB/s\r",,terminal_output +908,11868965,"TERMINAL",0,0,"Receiving objects: 1% (1430/81085), 892.00 KiB | 108.00 KiB/s\r",,terminal_output +909,11870009,"TERMINAL",0,0,"Receiving objects: 1% (1584/81085), 1012.00 KiB | 113.00 KiB/s\r",,terminal_output +910,11870162,"TERMINAL",0,0,"Receiving objects: 2% (1622/81085), 1.04 MiB | 119.00 KiB/s \r",,terminal_output +911,11870949,"TERMINAL",0,0,"Receiving objects: 2% (1732/81085), 1.10 MiB | 117.00 KiB/s\r",,terminal_output +912,11871962,"TERMINAL",0,0,"Receiving objects: 2% (1888/81085), 1.24 MiB | 128.00 KiB/s\r",,terminal_output +913,11872971,"TERMINAL",0,0,"Receiving objects: 2% (2126/81085), 1.35 MiB | 115.00 KiB/s\r",,terminal_output +914,11873930,"TERMINAL",0,0,"Receiving objects: 2% (2372/81085), 1.47 MiB | 114.00 KiB/s\r",,terminal_output +915,11874308,"TERMINAL",0,0,"Receiving objects: 3% (2433/81085), 1.47 MiB | 114.00 KiB/s\r",,terminal_output +916,11875093,"TERMINAL",0,0,"Receiving objects: 3% (2538/81085), 1.58 MiB | 112.00 KiB/s\r",,terminal_output +917,11875971,"TERMINAL",0,0,"Receiving objects: 3% (2664/81085), 1.64 MiB | 113.00 KiB/s\r",,terminal_output +918,11876921,"TERMINAL",0,0,"Receiving objects: 3% (2831/81085), 1.78 MiB | 111.00 KiB/s\r",,terminal_output +919,11878036,"TERMINAL",0,0,"Receiving objects: 3% (3012/81085), 1.89 MiB | 111.00 KiB/s\r",,terminal_output +920,11878985,"TERMINAL",0,0,"Receiving objects: 3% (3162/81085), 2.01 MiB | 109.00 KiB/s\r",,terminal_output +921,11879543,"TERMINAL",0,0,"Receiving objects: 4% (3244/81085), 2.07 MiB | 109.00 KiB/s\r",,terminal_output +922,11879934,"TERMINAL",0,0,"Receiving objects: 4% (3295/81085), 2.07 MiB | 109.00 KiB/s\r",,terminal_output +923,11881035,"TERMINAL",0,0,"Receiving objects: 4% (3461/81085), 2.20 MiB | 113.00 KiB/s\r",,terminal_output +924,11881964,"TERMINAL",0,0,"Receiving objects: 4% (3554/81085), 2.28 MiB | 103.00 KiB/s\r",,terminal_output +925,11882967,"TERMINAL",0,0,"Receiving objects: 4% (3707/81085), 2.39 MiB | 102.00 KiB/s\r",,terminal_output +926,11883972,"TERMINAL",0,0,"Receiving objects: 4% (3867/81085), 2.53 MiB | 105.00 KiB/s\r",,terminal_output +927,11884923,"TERMINAL",0,0,"Receiving objects: 4% (4012/81085), 2.59 MiB | 106.00 KiB/s\r",,terminal_output +928,11885217,"TERMINAL",0,0,"Receiving objects: 5% (4055/81085), 2.64 MiB | 104.00 KiB/s\r",,terminal_output +929,11885978,"TERMINAL",0,0,"Receiving objects: 5% (4135/81085), 2.69 MiB | 102.00 KiB/s\r",,terminal_output +930,11886931,"TERMINAL",0,0,"Receiving objects: 5% (4301/81085), 2.82 MiB | 111.00 KiB/s\r",,terminal_output +931,11888001,"TERMINAL",0,0,"Receiving objects: 5% (4471/81085), 2.95 MiB | 115.00 KiB/s\r",,terminal_output +932,11888994,"TERMINAL",0,0,"Receiving objects: 5% (4627/81085), 3.07 MiB | 114.00 KiB/s\r",,terminal_output +933,11889943,"TERMINAL",0,0,"Receiving objects: 5% (4756/81085), 3.18 MiB | 113.00 KiB/s\r",,terminal_output +934,11890596,"TERMINAL",0,0,"Receiving objects: 6% (4866/81085), 3.25 MiB | 116.00 KiB/s\r",,terminal_output +935,11891408,"TERMINAL",0,0,"Receiving objects: 6% (4889/81085), 3.29 MiB | 106.00 KiB/s\r",,terminal_output +936,11891988,"TERMINAL",0,0,"Receiving objects: 6% (5024/81085), 3.38 MiB | 109.00 KiB/s\r",,terminal_output +937,11892975,"TERMINAL",0,0,"Receiving objects: 6% (5166/81085), 3.49 MiB | 106.00 KiB/s\r",,terminal_output +938,11894011,"TERMINAL",0,0,"Receiving objects: 6% (5340/81085), 3.55 MiB | 105.00 KiB/s\r",,terminal_output +939,11894935,"TERMINAL",0,0,"Receiving objects: 6% (5482/81085), 3.68 MiB | 102.00 KiB/s\r",,terminal_output +940,11895925,"TERMINAL",0,0,"Receiving objects: 6% (5615/81085), 3.81 MiB | 105.00 KiB/s\r",,terminal_output +941,11896242,"TERMINAL",0,0,"Receiving objects: 7% (5676/81085), 3.81 MiB | 105.00 KiB/s\r",,terminal_output +942,11896953,"TERMINAL",0,0,"Receiving objects: 7% (5754/81085), 3.91 MiB | 108.00 KiB/s\r",,terminal_output +943,11897934,"TERMINAL",0,0,"Receiving objects: 7% (5901/81085), 3.94 MiB | 97.00 KiB/s \r",,terminal_output +944,11898992,"TERMINAL",0,0,"Receiving objects: 7% (6062/81085), 4.11 MiB | 111.00 KiB/s\r",,terminal_output +945,11899927,"TERMINAL",0,0,"Receiving objects: 7% (6221/81085), 4.25 MiB | 117.00 KiB/s\r",,terminal_output +946,11900949,"TERMINAL",0,0,"Receiving objects: 7% (6323/81085), 4.34 MiB | 108.00 KiB/s\r",,terminal_output +947,11901775,"TERMINAL",0,0,"Receiving objects: 8% (6487/81085), 4.41 MiB | 111.00 KiB/s\r",,terminal_output +948,11901947,"TERMINAL",0,0,"Receiving objects: 8% (6513/81085), 4.41 MiB | 111.00 KiB/s\r",,terminal_output +949,11902927,"TERMINAL",0,0,"Receiving objects: 8% (6657/81085), 4.57 MiB | 133.00 KiB/s\r",,terminal_output +950,11903962,"TERMINAL",0,0,"Receiving objects: 8% (6798/81085), 4.69 MiB | 123.00 KiB/s\r",,terminal_output +951,11904950,"TERMINAL",0,0,"Receiving objects: 8% (6950/81085), 4.81 MiB | 119.00 KiB/s\r",,terminal_output +952,11906404,"TERMINAL",0,0,"Receiving objects: 8% (7102/81085), 4.96 MiB | 113.00 KiB/s\r",,terminal_output +953,11906792,"TERMINAL",0,0,"Receiving objects: 9% (7298/81085), 4.96 MiB | 113.00 KiB/s\r",,terminal_output +954,11906997,"TERMINAL",0,0,"Receiving objects: 9% (7322/81085), 5.07 MiB | 117.00 KiB/s\r",,terminal_output +955,11907923,"TERMINAL",0,0,"Receiving objects: 9% (7458/81085), 5.13 MiB | 117.00 KiB/s\r",,terminal_output +956,11909301,"TERMINAL",0,0,"Receiving objects: 9% (7478/81085), 5.19 MiB | 89.00 KiB/s \r",,terminal_output +957,11909942,"TERMINAL",0,0,"Receiving objects: 9% (7712/81085), 5.37 MiB | 111.00 KiB/s\r",,terminal_output +958,11910983,"TERMINAL",0,0,"Receiving objects: 9% (7866/81085), 5.43 MiB | 104.00 KiB/s\r",,terminal_output +959,11911929,"TERMINAL",0,0,"Receiving objects: 9% (7995/81085), 5.62 MiB | 114.00 KiB/s\r",,terminal_output +960,11912474,"TERMINAL",0,0,"Receiving objects: 10% (8109/81085), 5.68 MiB | 122.00 KiB/s\r",,terminal_output +961,11912943,"TERMINAL",0,0,"Receiving objects: 10% (8213/81085), 5.68 MiB | 122.00 KiB/s\r",,terminal_output +962,11914012,"TERMINAL",0,0,"Receiving objects: 10% (8417/81085), 5.81 MiB | 114.00 KiB/s\r",,terminal_output +963,11914996,"TERMINAL",0,0,"Receiving objects: 10% (8661/81085), 5.93 MiB | 143.00 KiB/s\r",,terminal_output +964,11915984,"TERMINAL",0,0,"Receiving objects: 10% (8846/81085), 6.05 MiB | 129.00 KiB/s\r",,terminal_output +965,11916340,"TERMINAL",0,0,"Receiving objects: 11% (8920/81085), 6.11 MiB | 116.00 KiB/s\r",,terminal_output +966,11916929,"TERMINAL",0,0,"Receiving objects: 11% (9033/81085), 6.17 MiB | 114.00 KiB/s\r",,terminal_output +967,11917989,"TERMINAL",0,0,"Receiving objects: 11% (9206/81085), 6.29 MiB | 110.00 KiB/s\r",,terminal_output +968,11918975,"TERMINAL",0,0,"Receiving objects: 11% (9409/81085), 6.32 MiB | 101.00 KiB/s\r",,terminal_output +969,11919944,"TERMINAL",0,0,"Receiving objects: 11% (9634/81085), 6.50 MiB | 112.00 KiB/s\r",,terminal_output +970,11920216,"TERMINAL",0,0,"Receiving objects: 12% (9731/81085), 6.50 MiB | 112.00 KiB/s\r",,terminal_output +971,11920969,"TERMINAL",0,0,"Receiving objects: 12% (10078/81085), 6.63 MiB | 115.00 KiB/s\r",,terminal_output +972,11921865,"TERMINAL",0,0,"Receiving objects: 13% (10542/81085), 6.69 MiB | 117.00 KiB/s\r",,terminal_output +973,11921990,"TERMINAL",0,0,"Receiving objects: 13% (10602/81085), 6.75 MiB | 117.00 KiB/s\r",,terminal_output +974,11922973,"TERMINAL",0,0,"Receiving objects: 13% (11050/81085), 6.86 MiB | 118.00 KiB/s\r",,terminal_output +975,11923719,"TERMINAL",0,0,"Receiving objects: 14% (11352/81085), 6.93 MiB | 130.00 KiB/s\r",,terminal_output +976,11923961,"TERMINAL",0,0,"Receiving objects: 14% (11461/81085), 6.93 MiB | 130.00 KiB/s\r",,terminal_output +977,11925204,"TERMINAL",0,0,"Receiving objects: 14% (11478/81085), 7.11 MiB | 113.00 KiB/s\r",,terminal_output +978,11926299,"TERMINAL",0,0,"Receiving objects: 14% (11478/81085), 7.20 MiB | 106.00 KiB/s\r",,terminal_output +979,11927126,"TERMINAL",0,0,"Receiving objects: 14% (11479/81085), 7.22 MiB | 92.00 KiB/s \r",,terminal_output +980,11928451,"TERMINAL",0,0,"Receiving objects: 14% (11515/81085), 7.39 MiB | 99.00 KiB/s\r",,terminal_output +981,11929045,"TERMINAL",0,0,"Receiving objects: 14% (11517/81085), 7.46 MiB | 98.00 KiB/s\r",,terminal_output +982,11929965,"TERMINAL",0,0,"Receiving objects: 14% (11528/81085), 7.54 MiB | 100.00 KiB/s\r",,terminal_output +983,11931159,"TERMINAL",0,0,"Receiving objects: 14% (11552/81085), 7.69 MiB | 100.00 KiB/s\r",,terminal_output +984,11932241,"TERMINAL",0,0,"Receiving objects: 14% (11553/81085), 7.79 MiB | 114.00 KiB/s\r",,terminal_output +985,11933505,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 7.91 MiB | 104.00 KiB/s\r",,terminal_output +986,11934058,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 7.98 MiB | 105.00 KiB/s\r",,terminal_output +987,11935111,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.07 MiB | 99.00 KiB/s \r",,terminal_output +988,11936125,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.18 MiB | 100.00 KiB/s\r",,terminal_output +989,11937233,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.28 MiB | 99.00 KiB/s \r",,terminal_output +990,11938381,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.39 MiB | 100.00 KiB/s\r",,terminal_output +991,11939000,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.45 MiB | 97.00 KiB/s \r",,terminal_output +992,11940141,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.58 MiB | 104.00 KiB/s\r",,terminal_output +993,11941209,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.70 MiB | 105.00 KiB/s\r",,terminal_output +994,11942298,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.79 MiB | 104.00 KiB/s\r",,terminal_output +995,11943342,"TERMINAL",0,0,"Receiving objects: 14% (11554/81085), 8.91 MiB | 108.00 KiB/s\r",,terminal_output +996,11944393,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.04 MiB | 109.00 KiB/s\r",,terminal_output +997,11945466,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.15 MiB | 108.00 KiB/s\r",,terminal_output +998,11945977,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.21 MiB | 109.00 KiB/s\r",,terminal_output +999,11947352,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.36 MiB | 115.00 KiB/s\r",,terminal_output +1000,11948526,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.49 MiB | 114.00 KiB/s\r",,terminal_output +1001,11949114,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.57 MiB | 116.00 KiB/s\r",,terminal_output +1002,11950151,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.68 MiB | 114.00 KiB/s\r",,terminal_output +1003,11951418,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.75 MiB | 101.00 KiB/s\r",,terminal_output +1004,11952025,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.84 MiB | 115.00 KiB/s\r",,terminal_output +1005,11953049,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 9.98 MiB | 110.00 KiB/s\r",,terminal_output +1006,11954520,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.09 MiB | 99.00 KiB/s\r",,terminal_output +1007,11955082,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.21 MiB | 108.00 KiB/s\r",,terminal_output +1008,11956152,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.33 MiB | 118.00 KiB/s\r",,terminal_output +1009,11957248,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.46 MiB | 121.00 KiB/s\r",,terminal_output +1010,11958313,"TERMINAL",0,0,"Receiving objects: 14% (11558/81085), 10.57 MiB | 115.00 KiB/s\r",,terminal_output +1011,11959282,"TERMINAL",0,0,"Receiving objects: 14% (11560/81085), 10.64 MiB | 115.00 KiB/s\r",,terminal_output +1012,11960035,"TERMINAL",0,0,"Receiving objects: 14% (11642/81085), 10.76 MiB | 117.00 KiB/s\r",,terminal_output +1013,11961066,"TERMINAL",0,0,"Receiving objects: 14% (11671/81085), 10.88 MiB | 114.00 KiB/s\r",,terminal_output +1014,11962362,"TERMINAL",0,0,"Receiving objects: 14% (11671/81085), 10.98 MiB | 104.00 KiB/s\r",,terminal_output +1015,11962960,"TERMINAL",0,0,"Receiving objects: 14% (11679/81085), 11.10 MiB | 116.00 KiB/s\r",,terminal_output +1016,11964036,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 11.21 MiB | 114.00 KiB/s\r",,terminal_output +1017,11965104,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 11.34 MiB | 114.00 KiB/s\r",,terminal_output +1018,11966175,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 11.46 MiB | 115.00 KiB/s\r",,terminal_output +1019,11967274,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 11.58 MiB | 125.00 KiB/s\r",,terminal_output +1020,11968346,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 11.71 MiB | 114.00 KiB/s\r",,terminal_output +1021,11969882,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 11.82 MiB | 103.00 KiB/s\r",,terminal_output +1022,11970399,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 11.94 MiB | 116.00 KiB/s\r",,terminal_output +1023,11970986,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 12.00 MiB | 115.00 KiB/s\r",,terminal_output +1024,11972084,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 12.12 MiB | 115.00 KiB/s\r",,terminal_output +1025,11973124,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 12.25 MiB | 116.00 KiB/s\r",,terminal_output +1026,11974305,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 12.38 MiB | 114.00 KiB/s\r",,terminal_output +1027,11975423,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 12.50 MiB | 113.00 KiB/s\r",,terminal_output +1028,11975965,"TERMINAL",0,0,"Receiving objects: 14% (11867/81085), 12.57 MiB | 117.00 KiB/s\r",,terminal_output +1029,11977001,"TERMINAL",0,0,"Receiving objects: 14% (11877/81085), 12.69 MiB | 118.00 KiB/s\r",,terminal_output +1030,11978150,"TERMINAL",0,0,"Receiving objects: 14% (11877/81085), 12.82 MiB | 116.00 KiB/s\r",,terminal_output +1031,11979221,"TERMINAL",0,0,"Receiving objects: 14% (11877/81085), 12.89 MiB | 107.00 KiB/s\r",,terminal_output +1032,11980284,"TERMINAL",0,0,"Receiving objects: 14% (11877/81085), 13.07 MiB | 121.00 KiB/s\r",,terminal_output +1033,11980942,"TERMINAL",0,0,"Receiving objects: 14% (11909/81085), 13.14 MiB | 119.00 KiB/s\r",,terminal_output +1034,11982400,"TERMINAL",0,0,"Receiving objects: 14% (11910/81085), 13.32 MiB | 119.00 KiB/s\r",,terminal_output +1035,11982944,"TERMINAL",0,0,"Receiving objects: 14% (11911/81085), 13.38 MiB | 118.00 KiB/s\r",,terminal_output +1036,11983990,"TERMINAL",0,0,"Receiving objects: 14% (11915/81085), 13.50 MiB | 129.00 KiB/s\r",,terminal_output +1037,11984922,"TERMINAL",0,0,"Receiving objects: 14% (11953/81085), 13.56 MiB | 116.00 KiB/s\r",,terminal_output +1038,11985957,"TERMINAL",0,0,"Receiving objects: 14% (12007/81085), 13.68 MiB | 113.00 KiB/s\r",,terminal_output +1039,11987282,"TERMINAL",0,0,"Receiving objects: 14% (12075/81085), 13.86 MiB | 113.00 KiB/s\r",,terminal_output +1040,11988789,"TERMINAL",0,0,"Receiving objects: 14% (12075/81085), 13.97 MiB | 102.00 KiB/s\r",,terminal_output +1041,11988962,"TERMINAL",0,0,"Receiving objects: 14% (12100/81085), 13.97 MiB | 102.00 KiB/s\r",,terminal_output +1042,11989938,"TERMINAL",0,0,"Receiving objects: 14% (12161/81085), 14.15 MiB | 115.00 KiB/s\rReceiving objects: 15% (12163/81085), 14.15 MiB | 115.00 KiB/s\r",,terminal_output +1043,11991441,"TERMINAL",0,0,"Receiving objects: 15% (12209/81085), 14.32 MiB | 114.00 KiB/s\r",,terminal_output +1044,11991956,"TERMINAL",0,0,"Receiving objects: 15% (12217/81085), 14.38 MiB | 113.00 KiB/s\r",,terminal_output +1045,11993213,"TERMINAL",0,0,"Receiving objects: 15% (12220/81085), 14.49 MiB | 107.00 KiB/s\r",,terminal_output +1046,11994094,"TERMINAL",0,0,"Receiving objects: 15% (12228/81085), 14.57 MiB | 114.00 KiB/s\r",,terminal_output +1047,11994953,"TERMINAL",0,0,"Receiving objects: 15% (12281/81085), 14.67 MiB | 110.00 KiB/s\r",,terminal_output +1048,11995965,"TERMINAL",0,0,"Receiving objects: 15% (12329/81085), 14.79 MiB | 112.00 KiB/s\r",,terminal_output +1049,11996958,"TERMINAL",0,0,"Receiving objects: 15% (12385/81085), 14.91 MiB | 112.00 KiB/s\r",,terminal_output +1050,11998001,"TERMINAL",0,0,"Receiving objects: 15% (12513/81085), 15.03 MiB | 114.00 KiB/s\r",,terminal_output +1051,11998975,"TERMINAL",0,0,"Receiving objects: 15% (12615/81085), 15.14 MiB | 122.00 KiB/s\r",,terminal_output +1052,11999937,"TERMINAL",0,0,"Receiving objects: 15% (12622/81085), 15.19 MiB | 111.00 KiB/s\r",,terminal_output +1053,12000966,"TERMINAL",0,0,"Receiving objects: 15% (12645/81085), 15.29 MiB | 100.00 KiB/s\r",,terminal_output +1054,12001990,"TERMINAL",0,0,"Receiving objects: 15% (12707/81085), 15.46 MiB | 109.00 KiB/s\r",,terminal_output +1055,12003018,"TERMINAL",0,0,"Receiving objects: 15% (12782/81085), 15.59 MiB | 110.00 KiB/s\r",,terminal_output +1056,12003923,"TERMINAL",0,0,"Receiving objects: 15% (12871/81085), 15.63 MiB | 111.00 KiB/s\r",,terminal_output +1057,12004980,"TERMINAL",0,0,"Receiving objects: 15% (12955/81085), 15.75 MiB | 110.00 KiB/s\r",,terminal_output +1058,12005250,"TERMINAL",0,0,"Receiving objects: 16% (12974/81085), 15.82 MiB | 107.00 KiB/s\r",,terminal_output +1059,12006265,"TERMINAL",0,0,"Receiving objects: 16% (12999/81085), 15.93 MiB | 109.00 KiB/s\r",,terminal_output +1060,12007377,"TERMINAL",0,0,"Receiving objects: 16% (13003/81085), 16.04 MiB | 108.00 KiB/s\r",,terminal_output +1061,12008868,"TERMINAL",0,0,"Receiving objects: 16% (13003/81085), 16.18 MiB | 104.00 KiB/s\r",,terminal_output +1062,12009385,"TERMINAL",0,0,"Receiving objects: 16% (13004/81085), 16.29 MiB | 112.00 KiB/s\r",,terminal_output +1063,12009926,"TERMINAL",0,0,"Receiving objects: 16% (13004/81085), 16.35 MiB | 117.00 KiB/s\r",,terminal_output +1064,12010938,"TERMINAL",0,0,"Receiving objects: 16% (13011/81085), 16.39 MiB | 113.00 KiB/s\r",,terminal_output +1065,12012101,"TERMINAL",0,0,"Receiving objects: 16% (13012/81085), 16.61 MiB | 117.00 KiB/s\r",,terminal_output +1066,12013203,"TERMINAL",0,0,"Receiving objects: 16% (13013/81085), 16.72 MiB | 117.00 KiB/s\r",,terminal_output +1067,12014077,"TERMINAL",0,0,"Receiving objects: 16% (13014/81085), 16.79 MiB | 129.00 KiB/s\r",,terminal_output +1068,12015400,"TERMINAL",0,0,"Receiving objects: 16% (13014/81085), 16.96 MiB | 117.00 KiB/s\r",,terminal_output +1069,12016453,"TERMINAL",0,0,"Receiving objects: 16% (13014/81085), 17.07 MiB | 112.00 KiB/s\r",,terminal_output +1070,12017111,"TERMINAL",0,0,"Receiving objects: 16% (13014/81085), 17.13 MiB | 107.00 KiB/s\r",,terminal_output +1071,12018229,"TERMINAL",0,0,"Receiving objects: 16% (13014/81085), 17.25 MiB | 108.00 KiB/s\r",,terminal_output +1072,12018997,"TERMINAL",0,0,"Receiving objects: 16% (13020/81085), 17.31 MiB | 106.00 KiB/s\r",,terminal_output +1073,12019954,"TERMINAL",0,0,"Receiving objects: 16% (13024/81085), 17.40 MiB | 106.00 KiB/s\r",,terminal_output +1074,12020965,"TERMINAL",0,0,"Receiving objects: 16% (13029/81085), 17.50 MiB | 101.00 KiB/s\r",,terminal_output +1075,12022152,"TERMINAL",0,0,"Receiving objects: 16% (13037/81085), 17.63 MiB | 101.00 KiB/s\r",,terminal_output +1076,12023223,"TERMINAL",0,0,"Receiving objects: 16% (13037/81085), 17.78 MiB | 107.00 KiB/s\r",,terminal_output +1077,12024006,"TERMINAL",0,0,"Receiving objects: 16% (13038/81085), 17.79 MiB | 97.00 KiB/s \r",,terminal_output +1078,12024919,"TERMINAL",0,0,"Receiving objects: 16% (13039/81085), 17.91 MiB | 100.00 KiB/s\r",,terminal_output +1079,12026004,"TERMINAL",0,0,"Receiving objects: 16% (13091/81085), 18.08 MiB | 119.00 KiB/s\r",,terminal_output +1080,12027055,"TERMINAL",0,0,"Receiving objects: 16% (13156/81085), 18.22 MiB | 124.00 KiB/s\r",,terminal_output +1081,12028560,"TERMINAL",0,0,"Receiving objects: 16% (13171/81085), 18.25 MiB | 91.00 KiB/s \r",,terminal_output +1082,12029043,"TERMINAL",0,0,"Receiving objects: 16% (13180/81085), 18.25 MiB | 91.00 KiB/s\r",,terminal_output +1083,12030071,"TERMINAL",0,0,"Receiving objects: 16% (13206/81085), 18.32 MiB | 91.00 KiB/s\r",,terminal_output +1084,12031045,"TERMINAL",0,0,"Receiving objects: 16% (13511/81085), 18.56 MiB | 100.00 KiB/s\r",,terminal_output +1085,12032171,"TERMINAL",0,0,"Receiving objects: 16% (13694/81085), 18.64 MiB | 87.00 KiB/s \r",,terminal_output +1086,12032333,"TERMINAL",0,0,"Receiving objects: 17% (13785/81085), 18.64 MiB | 87.00 KiB/s\r",,terminal_output +1087,12033279,"TERMINAL",0,0,"Receiving objects: 17% (14016/81085), 18.77 MiB | 96.00 KiB/s\r",,terminal_output +1088,12033920,"TERMINAL",0,0,"Receiving objects: 17% (14047/81085), 18.79 MiB | 101.00 KiB/s\r",,terminal_output +1089,12035445,"TERMINAL",0,0,"Receiving objects: 17% (14099/81085), 18.80 MiB | 85.00 KiB/s \r",,terminal_output +1090,12036139,"TERMINAL",0,0,"Receiving objects: 17% (14172/81085), 18.82 MiB | 79.00 KiB/s\r",,terminal_output +1091,12037404,"TERMINAL",0,0,"Receiving objects: 17% (14239/81085), 18.84 MiB | 60.00 KiB/s\r",,terminal_output +1092,12037921,"TERMINAL",0,0,"Receiving objects: 17% (14302/81085), 18.86 MiB | 43.00 KiB/s\r",,terminal_output +1093,12039477,"TERMINAL",0,0,"Receiving objects: 17% (14452/81085), 18.89 MiB | 35.00 KiB/s\r",,terminal_output +1094,12039962,"TERMINAL",0,0,"Receiving objects: 17% (14504/81085), 18.89 MiB | 35.00 KiB/s\r",,terminal_output +1095,12041070,"TERMINAL",0,0,"Receiving objects: 17% (14524/81085), 18.94 MiB | 22.00 KiB/s\r",,terminal_output +1096,12042094,"TERMINAL",0,0,"Receiving objects: 17% (14564/81085), 18.95 MiB | 20.00 KiB/s\r",,terminal_output +1097,12043244,"TERMINAL",0,0,"Receiving objects: 17% (14587/81085), 18.99 MiB | 23.00 KiB/s\r",,terminal_output +1098,12061609,"TERMINAL",0,0,"Receiving objects: 17% (14587/81085), 19.04 MiB | 8.00 KiB/s \rReceiving objects: 18% (14596/81085), 19.04 MiB | 8.00 KiB/s\r",,terminal_output +1099,12062126,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 19.18 MiB | 13.00 KiB/s\r",,terminal_output +1100,12063159,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 19.37 MiB | 19.00 KiB/s\r",,terminal_output +1101,12064212,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 19.51 MiB | 26.00 KiB/s\r",,terminal_output +1102,12065346,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 19.63 MiB | 29.00 KiB/s\r",,terminal_output +1103,12066471,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 19.76 MiB | 151.00 KiB/s\r",,terminal_output +1104,12066975,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 19.82 MiB | 137.00 KiB/s\r",,terminal_output +1105,12068273,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 19.89 MiB | 104.00 KiB/s\r",,terminal_output +1106,12069343,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 20.07 MiB | 110.00 KiB/s\r",,terminal_output +1107,12069957,"TERMINAL",0,0,"Receiving objects: 18% (14646/81085), 20.13 MiB | 114.00 KiB/s\r",,terminal_output +1108,12070979,"TERMINAL",0,0,"Receiving objects: 18% (14720/81085), 20.19 MiB | 111.00 KiB/s\r",,terminal_output +1109,12071953,"TERMINAL",0,0,"Receiving objects: 18% (14772/81085), 20.32 MiB | 110.00 KiB/s\r",,terminal_output +1110,12072929,"TERMINAL",0,0,"Receiving objects: 18% (14966/81085), 20.43 MiB | 109.00 KiB/s\r",,terminal_output +1111,12073971,"TERMINAL",0,0,"Receiving objects: 18% (15225/81085), 20.55 MiB | 115.00 KiB/s\r",,terminal_output +1112,12075436,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 20.74 MiB | 113.00 KiB/s\r",,terminal_output +1113,12075942,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 20.79 MiB | 114.00 KiB/s\r",,terminal_output +1114,12076991,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 20.89 MiB | 107.00 KiB/s\r",,terminal_output +1115,12078112,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.01 MiB | 108.00 KiB/s\r",,terminal_output +1116,12079202,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.14 MiB | 108.00 KiB/s\r",,terminal_output +1117,12080305,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.27 MiB | 111.00 KiB/s\r",,terminal_output +1118,12081398,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.39 MiB | 115.00 KiB/s\r",,terminal_output +1119,12081919,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.44 MiB | 115.00 KiB/s\r",,terminal_output +1120,12083048,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.57 MiB | 115.00 KiB/s\r",,terminal_output +1121,12084081,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.69 MiB | 116.00 KiB/s\r",,terminal_output +1122,12085139,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.79 MiB | 110.00 KiB/s\r",,terminal_output +1123,12086286,"TERMINAL",0,0,"Receiving objects: 18% (15362/81085), 21.91 MiB | 109.00 KiB/s\r",,terminal_output +1124,12086662,"TERMINAL",0,0,"Receiving objects: 19% (15407/81085), 21.91 MiB | 109.00 KiB/s\r",,terminal_output +1125,12087638,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.03 MiB | 103.00 KiB/s\r",,terminal_output +1126,12088186,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.13 MiB | 112.00 KiB/s\r",,terminal_output +1127,12089209,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.27 MiB | 114.00 KiB/s\r",,terminal_output +1128,12090399,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.36 MiB | 111.00 KiB/s\r",,terminal_output +1129,12090935,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.46 MiB | 117.00 KiB/s\r",,terminal_output +1130,12092103,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.58 MiB | 116.00 KiB/s\r",,terminal_output +1131,12093406,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.66 MiB | 104.00 KiB/s\r",,terminal_output +1132,12093946,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.76 MiB | 108.00 KiB/s\r",,terminal_output +1133,12095272,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.88 MiB | 104.00 KiB/s\r",,terminal_output +1134,12096402,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 22.98 MiB | 96.00 KiB/s \r",,terminal_output +1135,12096934,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 23.10 MiB | 111.00 KiB/s\r",,terminal_output +1136,12098008,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 23.22 MiB | 114.00 KiB/s\r",,terminal_output +1137,12099067,"TERMINAL",0,0,"Receiving objects: 19% (15483/81085), 23.35 MiB | 117.00 KiB/s\r",,terminal_output +1138,12100133,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 23.48 MiB | 126.00 KiB/s\r",,terminal_output +1139,12101272,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 23.61 MiB | 131.00 KiB/s\r",,terminal_output +1140,12102425,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 23.72 MiB | 118.00 KiB/s\r",,terminal_output +1141,12103109,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 23.77 MiB | 109.00 KiB/s\r",,terminal_output +1142,12104217,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 23.94 MiB | 118.00 KiB/s\r",,terminal_output +1143,12105257,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 24.06 MiB | 115.00 KiB/s\r",,terminal_output +1144,12106285,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 24.18 MiB | 118.00 KiB/s\r",,terminal_output +1145,12107377,"TERMINAL",0,0,"Receiving objects: 19% (15490/81085), 24.30 MiB | 119.00 KiB/s\r",,terminal_output +1146,12107948,"TERMINAL",0,0,"Receiving objects: 19% (15516/81085), 24.36 MiB | 126.00 KiB/s\r",,terminal_output +1147,12108959,"TERMINAL",0,0,"Receiving objects: 19% (15572/81085), 24.43 MiB | 115.00 KiB/s\r",,terminal_output +1148,12110590,"TERMINAL",0,0,"Receiving objects: 19% (16061/81085), 24.60 MiB | 103.00 KiB/s\r",,terminal_output +1149,12111143,"TERMINAL",0,0,"Receiving objects: 19% (16149/81085), 24.69 MiB | 107.00 KiB/s\r",,terminal_output +1150,12111723,"TERMINAL",0,0,"Receiving objects: 20% (16217/81085), 24.78 MiB | 111.00 KiB/s\r",,terminal_output +1151,12111938,"TERMINAL",0,0,"Receiving objects: 20% (16221/81085), 24.78 MiB | 111.00 KiB/s\r",,terminal_output +1152,12112962,"TERMINAL",0,0,"Receiving objects: 20% (16368/81085), 24.89 MiB | 113.00 KiB/s\r",,terminal_output +1153,12114043,"TERMINAL",0,0,"Receiving objects: 20% (16399/81085), 25.03 MiB | 108.00 KiB/s\r",,terminal_output +1154,12114920,"TERMINAL",0,0,"Receiving objects: 20% (16483/81085), 25.10 MiB | 111.00 KiB/s\r",,terminal_output +1155,12115967,"TERMINAL",0,0,"Receiving objects: 20% (16568/81085), 25.23 MiB | 127.00 KiB/s\r",,terminal_output +1156,12117304,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 25.41 MiB | 115.00 KiB/s\r",,terminal_output +1157,12118359,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 25.53 MiB | 113.00 KiB/s\r",,terminal_output +1158,12119339,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 25.55 MiB | 101.00 KiB/s\r",,terminal_output +1159,12120394,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 25.71 MiB | 104.00 KiB/s\r",,terminal_output +1160,12121134,"TERMINAL",0,0,"Receiving objects: 20% (16623/81085), 25.76 MiB | 99.00 KiB/s \r",,terminal_output +1161,12122217,"TERMINAL",0,0,"Receiving objects: 20% (16624/81085), 25.81 MiB | 92.00 KiB/s\r",,terminal_output +1162,12123369,"TERMINAL",0,0,"Receiving objects: 20% (16624/81085), 25.96 MiB | 91.00 KiB/s\r",,terminal_output +1163,12124413,"TERMINAL",0,0,"Receiving objects: 20% (16624/81085), 26.04 MiB | 99.00 KiB/s\r",,terminal_output +1164,12124929,"TERMINAL",0,0,"Receiving objects: 20% (16624/81085), 26.11 MiB | 88.00 KiB/s\r",,terminal_output +1165,12126023,"TERMINAL",0,0,"Receiving objects: 20% (16624/81085), 26.23 MiB | 98.00 KiB/s\r",,terminal_output +1166,12127461,"TERMINAL",0,0,"Receiving objects: 20% (16625/81085), 26.32 MiB | 92.00 KiB/s\r",,terminal_output +1167,12128141,"TERMINAL",0,0,"Receiving objects: 20% (16625/81085), 26.36 MiB | 87.00 KiB/s\r",,terminal_output +1168,12129290,"TERMINAL",0,0,"Receiving objects: 20% (16625/81085), 26.52 MiB | 99.00 KiB/s\r",,terminal_output +1169,12130227,"TERMINAL",0,0,"Receiving objects: 20% (16626/81085), 26.61 MiB | 107.00 KiB/s\r",,terminal_output +1170,12130944,"TERMINAL",0,0,"Receiving objects: 20% (16626/81085), 26.77 MiB | 113.00 KiB/s\r",,terminal_output +1171,12132067,"TERMINAL",0,0,"Receiving objects: 20% (16626/81085), 26.93 MiB | 129.00 KiB/s\r",,terminal_output +1172,12132967,"TERMINAL",0,0,"Receiving objects: 20% (16627/81085), 27.00 MiB | 135.00 KiB/s\r",,terminal_output +1173,12134401,"TERMINAL",0,0,"Receiving objects: 20% (16630/81085), 27.19 MiB | 134.00 KiB/s\r",,terminal_output +1174,12134931,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 27.24 MiB | 126.00 KiB/s\r",,terminal_output +1175,12136097,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 27.32 MiB | 108.00 KiB/s\r",,terminal_output +1176,12137157,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 27.45 MiB | 103.00 KiB/s\r",,terminal_output +1177,12138428,"TERMINAL",0,0,"Receiving objects: 20% (16635/81085), 27.56 MiB | 98.00 KiB/s \r",,terminal_output +1178,12139101,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 27.61 MiB | 90.00 KiB/s\r",,terminal_output +1179,12140159,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 27.69 MiB | 88.00 KiB/s\r",,terminal_output +1180,12141301,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 27.82 MiB | 99.00 KiB/s\r",,terminal_output +1181,12141991,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 27.91 MiB | 95.00 KiB/s\r",,terminal_output +1182,12143109,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 28.03 MiB | 110.00 KiB/s\r",,terminal_output +1183,12144266,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 28.16 MiB | 110.00 KiB/s\r",,terminal_output +1184,12145326,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 28.28 MiB | 116.00 KiB/s\r",,terminal_output +1185,12145930,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 28.33 MiB | 107.00 KiB/s\r",,terminal_output +1186,12147092,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 28.46 MiB | 111.00 KiB/s\r",,terminal_output +1187,12149065,"TERMINAL",0,0,"Receiving objects: 20% (16636/81085), 28.47 MiB | 79.00 KiB/s \r",,terminal_output +1188,12150161,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 28.75 MiB | 103.00 KiB/s\r",,terminal_output +1189,12151271,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 28.86 MiB | 101.00 KiB/s\r",,terminal_output +1190,12152324,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 28.97 MiB | 102.00 KiB/s\r",,terminal_output +1191,12153448,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 29.07 MiB | 96.00 KiB/s \r",,terminal_output +1192,12153993,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 29.12 MiB | 134.00 KiB/s\r",,terminal_output +1193,12155145,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 29.18 MiB | 90.00 KiB/s \r",,terminal_output +1194,12156037,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 29.23 MiB | 84.00 KiB/s\r",,terminal_output +1195,12157252,"TERMINAL",0,0,"Receiving objects: 20% (16637/81085), 29.25 MiB | 68.00 KiB/s\r",,terminal_output +1196,12158377,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.42 MiB | 75.00 KiB/s\r",,terminal_output +1197,12159520,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.50 MiB | 73.00 KiB/s\r",,terminal_output +1198,12160157,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.55 MiB | 71.00 KiB/s\r",,terminal_output +1199,12161397,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.61 MiB | 70.00 KiB/s\r",,terminal_output +1200,12162021,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.64 MiB | 69.00 KiB/s\r",,terminal_output +1201,12163448,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.66 MiB | 52.00 KiB/s\r",,terminal_output +1202,12164287,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.68 MiB | 46.00 KiB/s\r",,terminal_output +1203,12165190,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.69 MiB | 39.00 KiB/s\r",,terminal_output +1204,12166006,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.72 MiB | 34.00 KiB/s\r",,terminal_output +1205,12166978,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.73 MiB | 27.00 KiB/s\r",,terminal_output +1206,12169133,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.79 MiB | 22.00 KiB/s\r",,terminal_output +1207,12170386,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.80 MiB | 20.00 KiB/s\r",,terminal_output +1208,12171172,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.81 MiB | 19.00 KiB/s\r",,terminal_output +1209,12172355,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.84 MiB | 19.00 KiB/s\r",,terminal_output +1210,12173120,"TERMINAL",0,0,"Receiving objects: 20% (16638/81085), 29.86 MiB | 21.00 KiB/s\r",,terminal_output +1211,12174132,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 29.88 MiB | 20.00 KiB/s\r",,terminal_output +1212,12175188,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 29.95 MiB | 25.00 KiB/s\r",,terminal_output +1213,12175921,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 29.98 MiB | 29.00 KiB/s\r",,terminal_output +1214,12177133,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 30.08 MiB | 42.00 KiB/s\r",,terminal_output +1215,12178178,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 30.19 MiB | 61.00 KiB/s\r",,terminal_output +1216,12179345,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 30.24 MiB | 66.00 KiB/s\r",,terminal_output +1217,12179927,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 30.25 MiB | 64.00 KiB/s\r",,terminal_output +1218,12181347,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 30.31 MiB | 61.00 KiB/s\r",,terminal_output +1219,12182375,"TERMINAL",0,0,"Receiving objects: 20% (16639/81085), 30.32 MiB | 46.00 KiB/s\r",,terminal_output +1220,12183235,"TERMINAL",0,0,"Receiving objects: 20% (16640/81085), 30.33 MiB | 41.00 KiB/s\r",,terminal_output +1221,12183999,"TERMINAL",0,0,"Receiving objects: 20% (16642/81085), 30.37 MiB | 36.00 KiB/s\r",,terminal_output +1222,12185143,"TERMINAL",0,0,"Receiving objects: 20% (16646/81085), 30.39 MiB | 27.00 KiB/s\r",,terminal_output +1223,12186412,"TERMINAL",0,0,"Receiving objects: 20% (16646/81085), 30.43 MiB | 28.00 KiB/s\r",,terminal_output +1224,12187117,"TERMINAL",0,0,"Receiving objects: 20% (16646/81085), 30.48 MiB | 31.00 KiB/s\r",,terminal_output +1225,12188171,"TERMINAL",0,0,"Receiving objects: 20% (16646/81085), 30.54 MiB | 40.00 KiB/s\r",,terminal_output +1226,12189002,"TERMINAL",0,0,"Receiving objects: 20% (16658/81085), 30.59 MiB | 46.00 KiB/s\r",,terminal_output +1227,12190231,"TERMINAL",0,0,"Receiving objects: 20% (16673/81085), 30.64 MiB | 47.00 KiB/s\r",,terminal_output +1228,12191051,"TERMINAL",0,0,"Receiving objects: 20% (16706/81085), 30.68 MiB | 53.00 KiB/s\r",,terminal_output +1229,12192142,"TERMINAL",0,0,"Receiving objects: 20% (16738/81085), 30.75 MiB | 55.00 KiB/s\r",,terminal_output +1230,12192945,"TERMINAL",0,0,"Receiving objects: 20% (16782/81085), 30.77 MiB | 50.00 KiB/s\r",,terminal_output +1231,12194190,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 30.82 MiB | 47.00 KiB/s\r",,terminal_output +1232,12195104,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 30.84 MiB | 41.00 KiB/s\r",,terminal_output +1233,12196260,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 30.88 MiB | 41.00 KiB/s\r",,terminal_output +1234,12196948,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 30.90 MiB | 36.00 KiB/s\r",,terminal_output +1235,12198168,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 30.97 MiB | 38.00 KiB/s\r",,terminal_output +1236,12199600,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.04 MiB | 40.00 KiB/s\r",,terminal_output +1237,12200104,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.07 MiB | 41.00 KiB/s\r",,terminal_output +1238,12201179,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.11 MiB | 47.00 KiB/s\r",,terminal_output +1239,12202323,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.15 MiB | 47.00 KiB/s\r",,terminal_output +1240,12203373,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.22 MiB | 49.00 KiB/s\r",,terminal_output +1241,12203941,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.25 MiB | 49.00 KiB/s\r",,terminal_output +1242,12205662,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.29 MiB | 40.00 KiB/s\r",,terminal_output +1243,12206223,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.36 MiB | 48.00 KiB/s\r",,terminal_output +1244,12207401,"TERMINAL",0,0,"Receiving objects: 20% (16840/81085), 31.46 MiB | 58.00 KiB/s\r",,terminal_output +1245,12207954,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.50 MiB | 62.00 KiB/s\r",,terminal_output +1246,12209056,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.55 MiB | 59.00 KiB/s\r",,terminal_output +1247,12210395,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.57 MiB | 52.00 KiB/s\r",,terminal_output +1248,12211031,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.60 MiB | 51.00 KiB/s\r",,terminal_output +1249,12212056,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.65 MiB | 50.00 KiB/s\r",,terminal_output +1250,12214158,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.70 MiB | 36.00 KiB/s\r",,terminal_output +1251,12215241,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.72 MiB | 29.00 KiB/s\r",,terminal_output +1252,12216184,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.76 MiB | 30.00 KiB/s\r",,terminal_output +1253,12217182,"TERMINAL",0,0,"Receiving objects: 20% (16841/81085), 31.78 MiB | 30.00 KiB/s\r",,terminal_output +1254,12218518,"TERMINAL",0,0,"Receiving objects: 20% (16845/81085), 31.81 MiB | 26.00 KiB/s\r",,terminal_output +1255,12219062,"TERMINAL",0,0,"Receiving objects: 20% (16846/81085), 31.84 MiB | 27.00 KiB/s\r",,terminal_output +1256,12220176,"TERMINAL",0,0,"Receiving objects: 20% (16846/81085), 31.87 MiB | 29.00 KiB/s\r",,terminal_output +1257,12221047,"TERMINAL",0,0,"Receiving objects: 20% (16847/81085), 31.89 MiB | 30.00 KiB/s\r",,terminal_output +1258,12221972,"TERMINAL",0,0,"Receiving objects: 20% (16850/81085), 31.96 MiB | 37.00 KiB/s\r",,terminal_output +1259,12223069,"TERMINAL",0,0,"Receiving objects: 20% (16854/81085), 32.03 MiB | 45.00 KiB/s\r",,terminal_output +1260,12224281,"TERMINAL",0,0,"Receiving objects: 20% (16878/81085), 32.07 MiB | 47.00 KiB/s\r",,terminal_output +1261,12225013,"TERMINAL",0,0,"Receiving objects: 20% (16941/81085), 32.11 MiB | 48.00 KiB/s\r",,terminal_output +1262,12226205,"TERMINAL",0,0,"Receiving objects: 20% (16985/81085), 32.19 MiB | 54.00 KiB/s\r",,terminal_output +1263,12227055,"TERMINAL",0,0,"Receiving objects: 20% (17010/81085), 32.21 MiB | 52.00 KiB/s\r",,terminal_output +1264,12227257,"TERMINAL",0,0,"Receiving objects: 21% (17028/81085), 32.21 MiB | 52.00 KiB/s\r",,terminal_output +1265,12228479,"TERMINAL",0,0,"Receiving objects: 21% (17036/81085), 32.30 MiB | 51.00 KiB/s\r",,terminal_output +1266,12229047,"TERMINAL",0,0,"Receiving objects: 21% (17036/81085), 32.40 MiB | 59.00 KiB/s\r",,terminal_output +1267,12229922,"TERMINAL",0,0,"Receiving objects: 21% (17152/81085), 32.48 MiB | 69.00 KiB/s\r",,terminal_output +1268,12230928,"TERMINAL",0,0,"Receiving objects: 21% (17263/81085), 32.71 MiB | 103.00 KiB/s\r",,terminal_output +1269,12231966,"TERMINAL",0,0,"Receiving objects: 21% (17419/81085), 32.85 MiB | 117.00 KiB/s\r",,terminal_output +1270,12232968,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 32.96 MiB | 132.00 KiB/s\r",,terminal_output +1271,12234015,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.07 MiB | 138.00 KiB/s\r",,terminal_output +1272,12235037,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.18 MiB | 109.00 KiB/s\r",,terminal_output +1273,12236153,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.31 MiB | 121.00 KiB/s\r",,terminal_output +1274,12237250,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.43 MiB | 109.00 KiB/s\r",,terminal_output +1275,12238432,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.49 MiB | 95.00 KiB/s \r",,terminal_output +1276,12239022,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.62 MiB | 111.00 KiB/s\r",,terminal_output +1277,12240118,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.73 MiB | 111.00 KiB/s\r",,terminal_output +1278,12241288,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.86 MiB | 109.00 KiB/s\r",,terminal_output +1279,12242334,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 33.93 MiB | 100.00 KiB/s\r",,terminal_output +1280,12242953,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 34.02 MiB | 107.00 KiB/s\r",,terminal_output +1281,12244042,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 34.14 MiB | 106.00 KiB/s\r",,terminal_output +1282,12245097,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 34.25 MiB | 107.00 KiB/s\r",,terminal_output +1283,12245939,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 34.33 MiB | 105.00 KiB/s\r",,terminal_output +1284,12247049,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 34.48 MiB | 114.00 KiB/s\r",,terminal_output +1285,12248093,"TERMINAL",0,0,"Receiving objects: 21% (17586/81085), 34.60 MiB | 114.00 KiB/s\r",,terminal_output +1286,12248956,"TERMINAL",0,0,"Receiving objects: 21% (17729/81085), 34.66 MiB | 113.00 KiB/s\r",,terminal_output +1287,12249305,"TERMINAL",0,0,"Receiving objects: 22% (17839/81085), 34.72 MiB | 114.00 KiB/s\r",,terminal_output +1288,12249971,"TERMINAL",0,0,"Receiving objects: 22% (18137/81085), 34.79 MiB | 114.00 KiB/s\r",,terminal_output +1289,12251280,"TERMINAL",0,0,"Receiving objects: 22% (18274/81085), 34.90 MiB | 120.00 KiB/s\r",,terminal_output +1290,12251922,"TERMINAL",0,0,"Receiving objects: 22% (18370/81085), 34.95 MiB | 107.00 KiB/s\r",,terminal_output +1291,12252959,"TERMINAL",0,0,"Receiving objects: 22% (18538/81085), 35.07 MiB | 108.00 KiB/s\r",,terminal_output +1292,12253758,"TERMINAL",0,0,"Receiving objects: 23% (18650/81085), 35.17 MiB | 105.00 KiB/s\r",,terminal_output +1293,12253947,"TERMINAL",0,0,"Receiving objects: 23% (18692/81085), 35.17 MiB | 105.00 KiB/s\r",,terminal_output +1294,12254973,"TERMINAL",0,0,"Receiving objects: 23% (19002/81085), 35.29 MiB | 103.00 KiB/s\r",,terminal_output +1295,12256327,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 35.40 MiB | 92.00 KiB/s \r",,terminal_output +1296,12256954,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 35.52 MiB | 106.00 KiB/s\r",,terminal_output +1297,12258044,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 35.64 MiB | 107.00 KiB/s\r",,terminal_output +1298,12259104,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 35.77 MiB | 112.00 KiB/s\r",,terminal_output +1299,12260236,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 35.87 MiB | 108.00 KiB/s\r",,terminal_output +1300,12261416,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.00 MiB | 119.00 KiB/s\r",,terminal_output +1301,12261951,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.07 MiB | 112.00 KiB/s\r",,terminal_output +1302,12263553,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.18 MiB | 98.00 KiB/s \r",,terminal_output +1303,12264134,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.29 MiB | 104.00 KiB/s\r",,terminal_output +1304,12265187,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.39 MiB | 104.00 KiB/s\r",,terminal_output +1305,12266221,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.50 MiB | 107.00 KiB/s\r",,terminal_output +1306,12267454,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.62 MiB | 103.00 KiB/s\r",,terminal_output +1307,12268367,"TERMINAL",0,0,"Receiving objects: 23% (19089/81085), 36.63 MiB | 89.00 KiB/s \r",,terminal_output +1308,12268960,"TERMINAL",0,0,"Receiving objects: 23% (19090/81085), 36.71 MiB | 103.00 KiB/s\r",,terminal_output +1309,12269928,"TERMINAL",0,0,"Receiving objects: 23% (19100/81085), 36.91 MiB | 110.00 KiB/s\r",,terminal_output +1310,12271053,"TERMINAL",0,0,"Receiving objects: 23% (19100/81085), 37.02 MiB | 109.00 KiB/s\r",,terminal_output +1311,12272084,"TERMINAL",0,0,"Receiving objects: 23% (19100/81085), 37.13 MiB | 107.00 KiB/s\r",,terminal_output +1312,12272934,"TERMINAL",0,0,"Receiving objects: 23% (19111/81085), 37.14 MiB | 101.00 KiB/s\r",,terminal_output +1313,12273967,"TERMINAL",0,0,"Receiving objects: 23% (19154/81085), 37.30 MiB | 118.00 KiB/s\r",,terminal_output +1314,12274977,"TERMINAL",0,0,"Receiving objects: 23% (19154/81085), 37.44 MiB | 107.00 KiB/s\r",,terminal_output +1315,12276092,"TERMINAL",0,0,"Receiving objects: 23% (19154/81085), 37.57 MiB | 111.00 KiB/s\r",,terminal_output +1316,12277110,"TERMINAL",0,0,"Receiving objects: 23% (19154/81085), 37.67 MiB | 109.00 KiB/s\r",,terminal_output +1317,12278178,"TERMINAL",0,0,"Receiving objects: 23% (19154/81085), 37.77 MiB | 107.00 KiB/s\r",,terminal_output +1318,12278925,"TERMINAL",0,0,"Receiving objects: 23% (19246/81085), 37.82 MiB | 113.00 KiB/s\r",,terminal_output +1319,12279983,"TERMINAL",0,0,"Receiving objects: 23% (19359/81085), 37.96 MiB | 108.00 KiB/s\r",,terminal_output +1320,12280973,"TERMINAL",0,0,"Receiving objects: 23% (19456/81085), 38.09 MiB | 109.00 KiB/s\rReceiving objects: 24% (19461/81085), 38.09 MiB | 109.00 KiB/s\r",,terminal_output +1321,12281959,"TERMINAL",0,0,"Receiving objects: 24% (19558/81085), 38.15 MiB | 110.00 KiB/s\r",,terminal_output +1322,12283060,"TERMINAL",0,0,"Receiving objects: 24% (19601/81085), 38.27 MiB | 103.00 KiB/s\r",,terminal_output +1323,12284097,"TERMINAL",0,0,"Receiving objects: 24% (19602/81085), 38.44 MiB | 117.00 KiB/s\r",,terminal_output +1324,12285248,"TERMINAL",0,0,"Receiving objects: 24% (19602/81085), 38.57 MiB | 115.00 KiB/s\r",,terminal_output +1325,12286311,"TERMINAL",0,0,"Receiving objects: 24% (19602/81085), 38.68 MiB | 114.00 KiB/s\r",,terminal_output +1326,12287081,"TERMINAL",0,0,"Receiving objects: 24% (19609/81085), 38.74 MiB | 112.00 KiB/s\r",,terminal_output +1327,12287969,"TERMINAL",0,0,"Receiving objects: 24% (19659/81085), 38.85 MiB | 122.00 KiB/s\r",,terminal_output +1328,12288968,"TERMINAL",0,0,"Receiving objects: 24% (19994/81085), 38.96 MiB | 108.00 KiB/s\r",,terminal_output +1329,12290038,"TERMINAL",0,0,"Receiving objects: 24% (19994/81085), 39.07 MiB | 106.00 KiB/s\r",,terminal_output +1330,12291127,"TERMINAL",0,0,"Receiving objects: 24% (19994/81085), 39.18 MiB | 104.00 KiB/s\r",,terminal_output +1331,12292742,"TERMINAL",0,0,"Receiving objects: 24% (19994/81085), 39.29 MiB | 94.00 KiB/s \r",,terminal_output +1332,12293253,"TERMINAL",0,0,"Receiving objects: 24% (19994/81085), 39.40 MiB | 106.00 KiB/s\r",,terminal_output +1333,12294023,"TERMINAL",0,0,"Receiving objects: 24% (20064/81085), 39.46 MiB | 105.00 KiB/s\r",,terminal_output +1334,12294935,"TERMINAL",0,0,"Receiving objects: 24% (20140/81085), 39.57 MiB | 108.00 KiB/s\r",,terminal_output +1335,12295760,"TERMINAL",0,0,"Receiving objects: 25% (20272/81085), 39.64 MiB | 109.00 KiB/s\r",,terminal_output +1336,12295970,"TERMINAL",0,0,"Receiving objects: 25% (20380/81085), 39.70 MiB | 111.00 KiB/s\r",,terminal_output +1337,12296948,"TERMINAL",0,0,"Receiving objects: 25% (20704/81085), 39.77 MiB | 113.00 KiB/s\r",,terminal_output +1338,12298089,"TERMINAL",0,0,"Receiving objects: 25% (20810/81085), 39.95 MiB | 115.00 KiB/s\r",,terminal_output +1339,12299106,"TERMINAL",0,0,"Receiving objects: 25% (20810/81085), 40.06 MiB | 115.00 KiB/s\r",,terminal_output +1340,12300304,"TERMINAL",0,0,"Receiving objects: 25% (20810/81085), 40.17 MiB | 110.00 KiB/s\r",,terminal_output +1341,12301479,"TERMINAL",0,0,"Receiving objects: 25% (20810/81085), 40.27 MiB | 102.00 KiB/s\r",,terminal_output +1342,12302217,"TERMINAL",0,0,"Receiving objects: 25% (20810/81085), 40.34 MiB | 100.00 KiB/s\r",,terminal_output +1343,12302982,"TERMINAL",0,0,"Receiving objects: 25% (20832/81085), 40.41 MiB | 101.00 KiB/s\r",,terminal_output +1344,12304133,"TERMINAL",0,0,"Receiving objects: 25% (20862/81085), 40.52 MiB | 99.00 KiB/s \r",,terminal_output +1345,12304975,"TERMINAL",0,0,"Receiving objects: 25% (20866/81085), 40.63 MiB | 101.00 KiB/s\r",,terminal_output +1346,12306039,"TERMINAL",0,0,"Receiving objects: 25% (20888/81085), 40.75 MiB | 105.00 KiB/s\r",,terminal_output +1347,12306472,"TERMINAL",0,0,"Receiving objects: 26% (21083/81085), 40.75 MiB | 105.00 KiB/s\r",,terminal_output +1348,12307031,"TERMINAL",0,0,"Receiving objects: 26% (21261/81085), 40.81 MiB | 108.00 KiB/s\r",,terminal_output +1349,12307944,"TERMINAL",0,0,"Receiving objects: 26% (21399/81085), 40.93 MiB | 109.00 KiB/s\r",,terminal_output +1350,12309218,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.11 MiB | 114.00 KiB/s\r",,terminal_output +1351,12310391,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.19 MiB | 104.00 KiB/s\r",,terminal_output +1352,12310922,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.30 MiB | 114.00 KiB/s\r",,terminal_output +1353,12311979,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.43 MiB | 118.00 KiB/s\r",,terminal_output +1354,12312999,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.54 MiB | 115.00 KiB/s\r",,terminal_output +1355,12314053,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.64 MiB | 110.00 KiB/s\r",,terminal_output +1356,12315220,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.75 MiB | 119.00 KiB/s\r",,terminal_output +1357,12316325,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.86 MiB | 105.00 KiB/s\r",,terminal_output +1358,12317428,"TERMINAL",0,0,"Receiving objects: 26% (21505/81085), 41.99 MiB | 105.00 KiB/s\r",,terminal_output +1359,12317915,"TERMINAL",0,0,"Receiving objects: 26% (21561/81085), 41.99 MiB | 105.00 KiB/s\r",,terminal_output +1360,12319064,"TERMINAL",0,0,"Receiving objects: 26% (21672/81085), 42.18 MiB | 110.00 KiB/s\r",,terminal_output +1361,12320467,"TERMINAL",0,0,"Receiving objects: 26% (21744/81085), 42.27 MiB | 100.00 KiB/s\r",,terminal_output +1362,12320955,"TERMINAL",0,0,"Receiving objects: 26% (21887/81085), 42.27 MiB | 100.00 KiB/s\rReceiving objects: 27% (21893/81085), 42.27 MiB | 100.00 KiB/s\r",,terminal_output +1363,12322119,"TERMINAL",0,0,"Receiving objects: 27% (21968/81085), 42.52 MiB | 116.00 KiB/s\r",,terminal_output +1364,12323181,"TERMINAL",0,0,"Receiving objects: 27% (21968/81085), 42.64 MiB | 115.00 KiB/s\r",,terminal_output +1365,12324241,"TERMINAL",0,0,"Receiving objects: 27% (21968/81085), 42.75 MiB | 114.00 KiB/s\r",,terminal_output +1366,12325284,"TERMINAL",0,0,"Receiving objects: 27% (21968/81085), 42.89 MiB | 131.00 KiB/s\r",,terminal_output +1367,12326152,"TERMINAL",0,0,"Receiving objects: 27% (21969/81085), 42.95 MiB | 120.00 KiB/s\r",,terminal_output +1368,12326926,"TERMINAL",0,0,"Receiving objects: 27% (22051/81085), 43.07 MiB | 118.00 KiB/s\r",,terminal_output +1369,12327971,"TERMINAL",0,0,"Receiving objects: 27% (22220/81085), 43.11 MiB | 106.00 KiB/s\r",,terminal_output +1370,12328942,"TERMINAL",0,0,"Receiving objects: 27% (22307/81085), 43.27 MiB | 113.00 KiB/s\r",,terminal_output +1371,12329951,"TERMINAL",0,0,"Receiving objects: 27% (22439/81085), 43.40 MiB | 113.00 KiB/s\r",,terminal_output +1372,12330941,"TERMINAL",0,0,"Receiving objects: 27% (22573/81085), 43.46 MiB | 114.00 KiB/s\r",,terminal_output +1373,12331946,"TERMINAL",0,0,"Receiving objects: 27% (22689/81085), 43.59 MiB | 115.00 KiB/s\r",,terminal_output +1374,12332008,"TERMINAL",0,0,"Receiving objects: 28% (22704/81085), 43.59 MiB | 115.00 KiB/s\r",,terminal_output +1375,12332939,"TERMINAL",0,0,"Receiving objects: 28% (22800/81085), 43.71 MiB | 127.00 KiB/s\r",,terminal_output +1376,12334520,"TERMINAL",0,0,"Receiving objects: 28% (22828/81085), 43.89 MiB | 110.00 KiB/s\r",,terminal_output +1377,12335025,"TERMINAL",0,0,"Receiving objects: 28% (22828/81085), 43.96 MiB | 113.00 KiB/s\r",,terminal_output +1378,12336158,"TERMINAL",0,0,"Receiving objects: 28% (22828/81085), 44.09 MiB | 111.00 KiB/s\r",,terminal_output +1379,12337326,"TERMINAL",0,0,"Receiving objects: 28% (22828/81085), 44.21 MiB | 110.00 KiB/s\r",,terminal_output +1380,12337950,"TERMINAL",0,0,"Receiving objects: 28% (22828/81085), 44.24 MiB | 100.00 KiB/s\r",,terminal_output +1381,12339094,"TERMINAL",0,0,"Receiving objects: 28% (22931/81085), 44.40 MiB | 108.00 KiB/s\r",,terminal_output +1382,12339961,"TERMINAL",0,0,"Receiving objects: 28% (23257/81085), 44.45 MiB | 113.00 KiB/s\r",,terminal_output +1383,12341238,"TERMINAL",0,0,"Receiving objects: 28% (23286/81085), 44.64 MiB | 111.00 KiB/s\r",,terminal_output +1384,12342287,"TERMINAL",0,0,"Receiving objects: 28% (23286/81085), 44.77 MiB | 114.00 KiB/s\r",,terminal_output +1385,12343370,"TERMINAL",0,0,"Receiving objects: 28% (23286/81085), 44.88 MiB | 113.00 KiB/s\r",,terminal_output +1386,12343957,"TERMINAL",0,0,"Receiving objects: 28% (23287/81085), 44.93 MiB | 111.00 KiB/s\r",,terminal_output +1387,12345002,"TERMINAL",0,0,"Receiving objects: 28% (23289/81085), 45.05 MiB | 112.00 KiB/s\r",,terminal_output +1388,12346083,"TERMINAL",0,0,"Receiving objects: 28% (23290/81085), 45.16 MiB | 109.00 KiB/s\r",,terminal_output +1389,12346962,"TERMINAL",0,0,"Receiving objects: 28% (23290/81085), 45.19 MiB | 96.00 KiB/s \r",,terminal_output +1390,12348160,"TERMINAL",0,0,"Receiving objects: 28% (23290/81085), 45.36 MiB | 102.00 KiB/s\r",,terminal_output +1391,12349200,"TERMINAL",0,0,"Receiving objects: 28% (23291/81085), 45.49 MiB | 108.00 KiB/s\r",,terminal_output +1392,12350407,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 45.58 MiB | 100.00 KiB/s\r",,terminal_output +1393,12350939,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 45.67 MiB | 106.00 KiB/s\r",,terminal_output +1394,12352099,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 45.80 MiB | 121.00 KiB/s\r",,terminal_output +1395,12353264,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 45.91 MiB | 109.00 KiB/s\r",,terminal_output +1396,12354402,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 46.04 MiB | 109.00 KiB/s\r",,terminal_output +1397,12354936,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 46.10 MiB | 112.00 KiB/s\r",,terminal_output +1398,12356193,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 46.21 MiB | 106.00 KiB/s\r",,terminal_output +1399,12357293,"TERMINAL",0,0,"Receiving objects: 28% (23292/81085), 46.35 MiB | 107.00 KiB/s\r",,terminal_output +1400,12357987,"TERMINAL",0,0,"Receiving objects: 28% (23293/81085), 46.40 MiB | 111.00 KiB/s\r",,terminal_output +1401,12358928,"TERMINAL",0,0,"Receiving objects: 28% (23293/81085), 46.52 MiB | 110.00 KiB/s\r",,terminal_output +1402,12360007,"TERMINAL",0,0,"Receiving objects: 28% (23293/81085), 46.62 MiB | 105.00 KiB/s\r",,terminal_output +1403,12361132,"TERMINAL",0,0,"Receiving objects: 28% (23297/81085), 46.74 MiB | 108.00 KiB/s\r",,terminal_output +1404,12363708,"TERMINAL",0,0,"Receiving objects: 28% (23304/81085), 46.79 MiB | 81.00 KiB/s \r",,terminal_output +1405,12364031,"TERMINAL",0,0,"Receiving objects: 28% (23306/81085), 46.79 MiB | 81.00 KiB/s\r",,terminal_output +1406,12365302,"TERMINAL",0,0,"Receiving objects: 28% (23306/81085), 47.11 MiB | 94.00 KiB/s\r",,terminal_output +1407,12366141,"TERMINAL",0,0,"Receiving objects: 28% (23307/81085), 47.17 MiB | 96.00 KiB/s\r",,terminal_output +1408,12367480,"TERMINAL",0,0,"Receiving objects: 28% (23308/81085), 47.32 MiB | 95.00 KiB/s\r",,terminal_output +1409,12368003,"TERMINAL",0,0,"Receiving objects: 28% (23308/81085), 47.40 MiB | 99.00 KiB/s\r",,terminal_output +1410,12369091,"TERMINAL",0,0,"Receiving objects: 28% (23308/81085), 47.51 MiB | 138.00 KiB/s\r",,terminal_output +1411,12370148,"TERMINAL",0,0,"Receiving objects: 28% (23309/81085), 47.63 MiB | 112.00 KiB/s\r",,terminal_output +1412,12371308,"TERMINAL",0,0,"Receiving objects: 28% (23310/81085), 47.76 MiB | 110.00 KiB/s\r",,terminal_output +1413,12372463,"TERMINAL",0,0,"Receiving objects: 28% (23316/81085), 47.88 MiB | 115.00 KiB/s\r",,terminal_output +1414,12372920,"TERMINAL",0,0,"Receiving objects: 28% (23328/81085), 47.88 MiB | 115.00 KiB/s\r",,terminal_output +1415,12373959,"TERMINAL",0,0,"Receiving objects: 28% (23453/81085), 47.99 MiB | 109.00 KiB/s\r",,terminal_output +1416,12374497,"TERMINAL",0,0,"Receiving objects: 29% (23515/81085), 48.06 MiB | 111.00 KiB/s\r",,terminal_output +1417,12375139,"TERMINAL",0,0,"Receiving objects: 29% (23581/81085), 48.17 MiB | 110.00 KiB/s\r",,terminal_output +1418,12376193,"TERMINAL",0,0,"Receiving objects: 29% (23581/81085), 48.27 MiB | 106.00 KiB/s\r",,terminal_output +1419,12377276,"TERMINAL",0,0,"Receiving objects: 29% (23581/81085), 48.38 MiB | 106.00 KiB/s\r",,terminal_output +1420,12378419,"TERMINAL",0,0,"Receiving objects: 29% (23581/81085), 48.50 MiB | 107.00 KiB/s\r",,terminal_output +1421,12378998,"TERMINAL",0,0,"Receiving objects: 29% (23623/81085), 48.57 MiB | 107.00 KiB/s\r",,terminal_output +1422,12380019,"TERMINAL",0,0,"Receiving objects: 29% (23827/81085), 48.60 MiB | 99.00 KiB/s \r",,terminal_output +1423,12380924,"TERMINAL",0,0,"Receiving objects: 29% (23954/81085), 48.77 MiB | 112.00 KiB/s\r",,terminal_output +1424,12381926,"TERMINAL",0,0,"Receiving objects: 29% (24184/81085), 48.90 MiB | 115.00 KiB/s\r",,terminal_output +1425,12382246,"TERMINAL",0,0,"Receiving objects: 30% (24326/81085), 48.90 MiB | 115.00 KiB/s\r",,terminal_output +1426,12382936,"TERMINAL",0,0,"Receiving objects: 30% (24403/81085), 49.02 MiB | 116.00 KiB/s\r",,terminal_output +1427,12384021,"TERMINAL",0,0,"Receiving objects: 30% (24500/81085), 49.14 MiB | 115.00 KiB/s\r",,terminal_output +1428,12385156,"TERMINAL",0,0,"Receiving objects: 30% (24549/81085), 49.25 MiB | 111.00 KiB/s\r",,terminal_output +1429,12386282,"TERMINAL",0,0,"Receiving objects: 30% (24549/81085), 49.36 MiB | 110.00 KiB/s\r",,terminal_output +1430,12387076,"TERMINAL",0,0,"Receiving objects: 30% (24549/81085), 49.38 MiB | 94.00 KiB/s \r",,terminal_output +1431,12388916,"TERMINAL",0,0,"Receiving objects: 30% (24549/81085), 49.44 MiB | 75.00 KiB/s\r",,terminal_output +1432,12389949,"TERMINAL",0,0,"Receiving objects: 30% (24642/81085), 49.64 MiB | 98.00 KiB/s\r",,terminal_output +1433,12391001,"TERMINAL",0,0,"Receiving objects: 30% (24782/81085), 49.82 MiB | 101.00 KiB/s\r",,terminal_output +1434,12392054,"TERMINAL",0,0,"Receiving objects: 30% (24782/81085), 49.94 MiB | 102.00 KiB/s\r",,terminal_output +1435,12393093,"TERMINAL",0,0,"Receiving objects: 30% (24782/81085), 50.04 MiB | 113.00 KiB/s\r",,terminal_output +1436,12394144,"TERMINAL",0,0,"Receiving objects: 30% (24782/81085), 50.16 MiB | 112.00 KiB/s\r",,terminal_output +1437,12394934,"TERMINAL",0,0,"Receiving objects: 30% (24783/81085), 50.21 MiB | 110.00 KiB/s\r",,terminal_output +1438,12396267,"TERMINAL",0,0,"Receiving objects: 30% (24797/81085), 50.32 MiB | 105.00 KiB/s\r",,terminal_output +1439,12396935,"TERMINAL",0,0,"Receiving objects: 30% (24911/81085), 50.40 MiB | 96.00 KiB/s \r",,terminal_output +1440,12398452,"TERMINAL",0,0,"Receiving objects: 30% (25022/81085), 50.50 MiB | 88.00 KiB/s\r",,terminal_output +1441,12398976,"TERMINAL",0,0,"Receiving objects: 30% (25075/81085), 50.61 MiB | 97.00 KiB/s\r",,terminal_output +1442,12399618,"TERMINAL",0,0,"Receiving objects: 31% (25137/81085), 50.68 MiB | 99.00 KiB/s\r",,terminal_output +1443,12400010,"TERMINAL",0,0,"Receiving objects: 31% (25208/81085), 50.68 MiB | 99.00 KiB/s\r",,terminal_output +1444,12401131,"TERMINAL",0,0,"Receiving objects: 31% (25221/81085), 50.82 MiB | 97.00 KiB/s\r",,terminal_output +1445,12402226,"TERMINAL",0,0,"Receiving objects: 31% (25221/81085), 50.93 MiB | 102.00 KiB/s\r",,terminal_output +1446,12403378,"TERMINAL",0,0,"Receiving objects: 31% (25221/81085), 51.00 MiB | 103.00 KiB/s\r",,terminal_output +1447,12403918,"TERMINAL",0,0,"Receiving objects: 31% (25221/81085), 51.07 MiB | 95.00 KiB/s \r",,terminal_output +1448,12404968,"TERMINAL",0,0,"Receiving objects: 31% (25309/81085), 51.14 MiB | 94.00 KiB/s\r",,terminal_output +1449,12406020,"TERMINAL",0,0,"Receiving objects: 31% (25409/81085), 51.24 MiB | 92.00 KiB/s\r",,terminal_output +1450,12407388,"TERMINAL",0,0,"Receiving objects: 31% (25563/81085), 51.39 MiB | 91.00 KiB/s\r",,terminal_output +1451,12407950,"TERMINAL",0,0,"Receiving objects: 31% (25563/81085), 51.44 MiB | 92.00 KiB/s\r",,terminal_output +1452,12409258,"TERMINAL",0,0,"Receiving objects: 31% (25563/81085), 51.52 MiB | 85.00 KiB/s\r",,terminal_output +1453,12410339,"TERMINAL",0,0,"Receiving objects: 31% (25563/81085), 51.68 MiB | 92.00 KiB/s\r",,terminal_output +1454,12411391,"TERMINAL",0,0,"Receiving objects: 31% (25563/81085), 51.79 MiB | 98.00 KiB/s\r",,terminal_output +1455,12411953,"TERMINAL",0,0,"Receiving objects: 31% (25563/81085), 51.85 MiB | 101.00 KiB/s\r",,terminal_output +1456,12412996,"TERMINAL",0,0,"Receiving objects: 31% (25666/81085), 51.90 MiB | 101.00 KiB/s\r",,terminal_output +1457,12414160,"TERMINAL",0,0,"Receiving objects: 31% (25788/81085), 52.07 MiB | 114.00 KiB/s\r",,terminal_output +1458,12414951,"TERMINAL",0,0,"Receiving objects: 31% (25881/81085), 52.13 MiB | 107.00 KiB/s\r",,terminal_output +1459,12415938,"TERMINAL",0,0,"Receiving objects: 31% (25909/81085), 52.26 MiB | 107.00 KiB/s\r",,terminal_output +1460,12416791,"TERMINAL",0,0,"Receiving objects: 32% (25948/81085), 52.32 MiB | 109.00 KiB/s\r",,terminal_output +1461,12417043,"TERMINAL",0,0,"Receiving objects: 32% (25972/81085), 52.38 MiB | 107.00 KiB/s\r",,terminal_output +1462,12418078,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 52.51 MiB | 111.00 KiB/s\r",,terminal_output +1463,12419136,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 52.61 MiB | 112.00 KiB/s\r",,terminal_output +1464,12420213,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 52.72 MiB | 114.00 KiB/s\r",,terminal_output +1465,12421269,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 52.83 MiB | 108.00 KiB/s\r",,terminal_output +1466,12422345,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 52.89 MiB | 95.00 KiB/s \r",,terminal_output +1467,12422961,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 53.00 MiB | 103.00 KiB/s\r",,terminal_output +1468,12424006,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 53.11 MiB | 105.00 KiB/s\r",,terminal_output +1469,12425076,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 53.22 MiB | 105.00 KiB/s\r",,terminal_output +1470,12426173,"TERMINAL",0,0,"Receiving objects: 32% (26108/81085), 53.34 MiB | 106.00 KiB/s\r",,terminal_output +1471,12427001,"TERMINAL",0,0,"Receiving objects: 32% (26207/81085), 53.40 MiB | 107.00 KiB/s\r",,terminal_output +1472,12427925,"TERMINAL",0,0,"Receiving objects: 32% (26427/81085), 53.51 MiB | 108.00 KiB/s\r",,terminal_output +1473,12428184,"TERMINAL",0,0,"Receiving objects: 33% (26759/81085), 53.51 MiB | 108.00 KiB/s\r",,terminal_output +1474,12429122,"TERMINAL",0,0,"Receiving objects: 33% (27319/81085), 53.63 MiB | 111.00 KiB/s\r",,terminal_output +1475,12429317,"TERMINAL",0,0,"Receiving objects: 34% (27569/81085), 53.69 MiB | 110.00 KiB/s\r",,terminal_output +1476,12429951,"TERMINAL",0,0,"Receiving objects: 34% (28271/81085), 53.75 MiB | 114.00 KiB/s\r",,terminal_output +1477,12430054,"TERMINAL",0,0,"Receiving objects: 35% (28380/81085), 53.75 MiB | 114.00 KiB/s\r",,terminal_output +1478,12431137,"TERMINAL",0,0,"Receiving objects: 35% (28517/81085), 53.85 MiB | 104.00 KiB/s\r",,terminal_output +1479,12431936,"TERMINAL",0,0,"Receiving objects: 35% (28518/81085), 53.96 MiB | 115.00 KiB/s\r",,terminal_output +1480,12433030,"TERMINAL",0,0,"Receiving objects: 35% (28654/81085), 54.09 MiB | 119.00 KiB/s\r",,terminal_output +1481,12433929,"TERMINAL",0,0,"Receiving objects: 35% (29075/81085), 54.18 MiB | 113.00 KiB/s\r",,terminal_output +1482,12434003,"TERMINAL",0,0,"Receiving objects: 36% (29191/81085), 54.18 MiB | 113.00 KiB/s\r",,terminal_output +1483,12434962,"TERMINAL",0,0,"Receiving objects: 36% (29390/81085), 54.28 MiB | 106.00 KiB/s\r",,terminal_output +1484,12435946,"TERMINAL",0,0,"Receiving objects: 36% (29468/81085), 54.39 MiB | 115.00 KiB/s\r",,terminal_output +1485,12436966,"TERMINAL",0,0,"Receiving objects: 36% (29606/81085), 54.45 MiB | 103.00 KiB/s\r",,terminal_output +1486,12437349,"TERMINAL",0,0,"Receiving objects: 37% (30002/81085), 54.51 MiB | 100.00 KiB/s\r",,terminal_output +1487,12437968,"TERMINAL",0,0,"Receiving objects: 37% (30526/81085), 54.57 MiB | 100.00 KiB/s\r",,terminal_output +1488,12438224,"TERMINAL",0,0,"Receiving objects: 38% (30813/81085), 54.63 MiB | 103.00 KiB/s\r",,terminal_output +1489,12439004,"TERMINAL",0,0,"Receiving objects: 38% (31356/81085), 54.70 MiB | 107.00 KiB/s\r",,terminal_output +1490,12439987,"TERMINAL",0,0,"Receiving objects: 38% (31510/81085), 54.75 MiB | 97.00 KiB/s \r",,terminal_output +1491,12440275,"TERMINAL",0,0,"Receiving objects: 39% (31624/81085), 54.86 MiB | 109.00 KiB/s\r",,terminal_output +1492,12440952,"TERMINAL",0,0,"Receiving objects: 39% (31903/81085), 54.91 MiB | 108.00 KiB/s\r",,terminal_output +1493,12441936,"TERMINAL",0,0,"Receiving objects: 39% (32010/81085), 55.00 MiB | 104.00 KiB/s\r",,terminal_output +1494,12442917,"TERMINAL",0,0,"Receiving objects: 39% (32186/81085), 55.06 MiB | 101.00 KiB/s\r",,terminal_output +1495,12443926,"TERMINAL",0,0,"Receiving objects: 39% (32360/81085), 55.18 MiB | 101.00 KiB/s\r",,terminal_output +1496,12444252,"TERMINAL",0,0,"Receiving objects: 40% (32434/81085), 55.23 MiB | 98.00 KiB/s \r",,terminal_output +1497,12444950,"TERMINAL",0,0,"Receiving objects: 40% (32574/81085), 55.29 MiB | 108.00 KiB/s\r",,terminal_output +1498,12445988,"TERMINAL",0,0,"Receiving objects: 40% (33240/81085), 55.40 MiB | 100.00 KiB/s\rReceiving objects: 41% (33245/81085), 55.40 MiB | 100.00 KiB/s\r",,terminal_output +1499,12446942,"TERMINAL",0,0,"Receiving objects: 41% (33409/81085), 55.46 MiB | 102.00 KiB/s\r",,terminal_output +1500,12447917,"TERMINAL",0,0,"Receiving objects: 41% (33700/81085), 55.59 MiB | 109.00 KiB/s\r",,terminal_output +1501,12449222,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 55.77 MiB | 111.00 KiB/s\r",,terminal_output +1502,12450282,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 55.89 MiB | 114.00 KiB/s\r",,terminal_output +1503,12451310,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.01 MiB | 116.00 KiB/s\r",,terminal_output +1504,12452405,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.14 MiB | 116.00 KiB/s\r",,terminal_output +1505,12452990,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.18 MiB | 111.00 KiB/s\r",,terminal_output +1506,12453969,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.24 MiB | 100.00 KiB/s\r",,terminal_output +1507,12455060,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.39 MiB | 106.00 KiB/s\r",,terminal_output +1508,12456153,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.49 MiB | 102.00 KiB/s\r",,terminal_output +1509,12457190,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.59 MiB | 99.00 KiB/s \r",,terminal_output +1510,12458253,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.70 MiB | 100.00 KiB/s\r",,terminal_output +1511,12459328,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.82 MiB | 102.00 KiB/s\r",,terminal_output +1512,12460447,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 56.95 MiB | 107.00 KiB/s\r",,terminal_output +1513,12460981,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 57.00 MiB | 109.00 KiB/s\r",,terminal_output +1514,12462061,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 57.11 MiB | 110.00 KiB/s\r",,terminal_output +1515,12463155,"TERMINAL",0,0,"Receiving objects: 41% (33901/81085), 57.21 MiB | 107.00 KiB/s\r",,terminal_output +1516,12464273,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.36 MiB | 110.00 KiB/s\r",,terminal_output +1517,12464919,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.42 MiB | 108.00 KiB/s\r",,terminal_output +1518,12466010,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.52 MiB | 105.00 KiB/s\r",,terminal_output +1519,12467073,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.64 MiB | 106.00 KiB/s\r",,terminal_output +1520,12468225,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.75 MiB | 109.00 KiB/s\r",,terminal_output +1521,12469120,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.79 MiB | 96.00 KiB/s \r",,terminal_output +1522,12470233,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 57.94 MiB | 100.00 KiB/s\r",,terminal_output +1523,12471356,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.04 MiB | 100.00 KiB/s\r",,terminal_output +1524,12471982,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.11 MiB | 100.00 KiB/s\r",,terminal_output +1525,12473052,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.23 MiB | 100.00 KiB/s\r",,terminal_output +1526,12474141,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.36 MiB | 114.00 KiB/s\r",,terminal_output +1527,12475279,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.48 MiB | 109.00 KiB/s\r",,terminal_output +1528,12476337,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.57 MiB | 107.00 KiB/s\r",,terminal_output +1529,12477412,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.69 MiB | 109.00 KiB/s\r",,terminal_output +1530,12478291,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.75 MiB | 100.00 KiB/s\r",,terminal_output +1531,12479428,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.93 MiB | 110.00 KiB/s\r",,terminal_output +1532,12480000,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 58.97 MiB | 106.00 KiB/s\r",,terminal_output +1533,12481056,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.11 MiB | 111.00 KiB/s\r",,terminal_output +1534,12482151,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.23 MiB | 116.00 KiB/s\r",,terminal_output +1535,12483271,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.34 MiB | 122.00 KiB/s\r",,terminal_output +1536,12484358,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.46 MiB | 112.00 KiB/s\r",,terminal_output +1537,12485449,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.59 MiB | 113.00 KiB/s\r",,terminal_output +1538,12486369,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.64 MiB | 102.00 KiB/s\r",,terminal_output +1539,12486967,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.73 MiB | 106.00 KiB/s\r",,terminal_output +1540,12487999,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.86 MiB | 109.00 KiB/s\r",,terminal_output +1541,12489091,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 59.96 MiB | 108.00 KiB/s\r",,terminal_output +1542,12490199,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.07 MiB | 106.00 KiB/s\r",,terminal_output +1543,12491234,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.18 MiB | 115.00 KiB/s\r",,terminal_output +1544,12492307,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.30 MiB | 106.00 KiB/s\r",,terminal_output +1545,12493316,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.41 MiB | 108.00 KiB/s\r",,terminal_output +1546,12494436,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.54 MiB | 111.00 KiB/s\r",,terminal_output +1547,12494969,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.60 MiB | 112.00 KiB/s\r",,terminal_output +1548,12496508,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.71 MiB | 103.00 KiB/s\r",,terminal_output +1549,12497173,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.83 MiB | 110.00 KiB/s\r",,terminal_output +1550,12498230,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 60.94 MiB | 112.00 KiB/s\r",,terminal_output +1551,12499370,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.07 MiB | 111.00 KiB/s\r",,terminal_output +1552,12499970,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.13 MiB | 109.00 KiB/s\r",,terminal_output +1553,12501115,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.25 MiB | 108.00 KiB/s\r",,terminal_output +1554,12502328,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.36 MiB | 105.00 KiB/s\r",,terminal_output +1555,12503531,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.47 MiB | 102.00 KiB/s\r",,terminal_output +1556,12504065,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.54 MiB | 105.00 KiB/s\r",,terminal_output +1557,12505172,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.66 MiB | 104.00 KiB/s\r",,terminal_output +1558,12506304,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.77 MiB | 103.00 KiB/s\r",,terminal_output +1559,12507403,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.90 MiB | 108.00 KiB/s\r",,terminal_output +1560,12508011,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.96 MiB | 106.00 KiB/s\r",,terminal_output +1561,12509363,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 61.97 MiB | 87.00 KiB/s \r",,terminal_output +1562,12510431,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.25 MiB | 113.00 KiB/s\r",,terminal_output +1563,12510950,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.31 MiB | 114.00 KiB/s\r",,terminal_output +1564,12512065,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.43 MiB | 118.00 KiB/s\r",,terminal_output +1565,12513147,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.56 MiB | 117.00 KiB/s\r",,terminal_output +1566,12514187,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.68 MiB | 149.00 KiB/s\r",,terminal_output +1567,12515392,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.75 MiB | 103.00 KiB/s\r",,terminal_output +1568,12515929,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.86 MiB | 114.00 KiB/s\r",,terminal_output +1569,12516989,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 62.99 MiB | 115.00 KiB/s\r",,terminal_output +1570,12518160,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.11 MiB | 113.00 KiB/s\r",,terminal_output +1571,12519302,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.24 MiB | 112.00 KiB/s\r",,terminal_output +1572,12520386,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.37 MiB | 128.00 KiB/s\r",,terminal_output +1573,12520932,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.43 MiB | 116.00 KiB/s\r",,terminal_output +1574,12521973,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.55 MiB | 115.00 KiB/s\r",,terminal_output +1575,12523082,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.68 MiB | 118.00 KiB/s\r",,terminal_output +1576,12524100,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.81 MiB | 121.00 KiB/s\r",,terminal_output +1577,12525182,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 63.93 MiB | 120.00 KiB/s\r",,terminal_output +1578,12526266,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.07 MiB | 122.00 KiB/s\r",,terminal_output +1579,12527288,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.18 MiB | 119.00 KiB/s\r",,terminal_output +1580,12528329,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.29 MiB | 118.00 KiB/s\r",,terminal_output +1581,12529468,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.41 MiB | 125.00 KiB/s\r",,terminal_output +1582,12529998,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.46 MiB | 113.00 KiB/s\r",,terminal_output +1583,12531062,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.58 MiB | 110.00 KiB/s\r",,terminal_output +1584,12532158,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.65 MiB | 100.00 KiB/s\r",,terminal_output +1585,12533198,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.83 MiB | 113.00 KiB/s\r",,terminal_output +1586,12534261,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 64.96 MiB | 117.00 KiB/s\r",,terminal_output +1587,12535272,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.07 MiB | 118.00 KiB/s\r",,terminal_output +1588,12536353,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.21 MiB | 121.00 KiB/s\r",,terminal_output +1589,12537532,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.33 MiB | 118.00 KiB/s\r",,terminal_output +1590,12538051,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.39 MiB | 117.00 KiB/s\r",,terminal_output +1591,12539136,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.50 MiB | 114.00 KiB/s\r",,terminal_output +1592,12540192,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.63 MiB | 115.00 KiB/s\r",,terminal_output +1593,12541250,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.75 MiB | 112.00 KiB/s\r",,terminal_output +1594,12542289,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 65.87 MiB | 116.00 KiB/s\r",,terminal_output +1595,12543390,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.00 MiB | 116.00 KiB/s\r",,terminal_output +1596,12543944,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.07 MiB | 119.00 KiB/s\r",,terminal_output +1597,12545157,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.21 MiB | 119.00 KiB/s\r",,terminal_output +1598,12546256,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.33 MiB | 119.00 KiB/s\r",,terminal_output +1599,12547315,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.46 MiB | 119.00 KiB/s\r",,terminal_output +1600,12548381,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.57 MiB | 117.00 KiB/s\r",,terminal_output +1601,12548929,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.64 MiB | 117.00 KiB/s\r",,terminal_output +1602,12549952,"TERMINAL",0,0,"Receiving objects: 41% (33903/81085), 66.76 MiB | 118.00 KiB/s\r",,terminal_output +1603,12550773,"TERMINAL",0,0,"Receiving objects: 42% (34056/81085), 66.82 MiB | 120.00 KiB/s\r",,terminal_output +1604,12550966,"TERMINAL",0,0,"Receiving objects: 42% (34099/81085), 66.82 MiB | 120.00 KiB/s\r",,terminal_output +1605,12551962,"TERMINAL",0,0,"Receiving objects: 42% (34300/81085), 66.95 MiB | 119.00 KiB/s\r",,terminal_output +1606,12552937,"TERMINAL",0,0,"Receiving objects: 42% (34483/81085), 67.04 MiB | 109.00 KiB/s\r",,terminal_output +1607,12554424,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.29 MiB | 120.00 KiB/s\r",,terminal_output +1608,12554983,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.35 MiB | 119.00 KiB/s\r",,terminal_output +1609,12556056,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.47 MiB | 119.00 KiB/s\r",,terminal_output +1610,12557089,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.58 MiB | 115.00 KiB/s\r",,terminal_output +1611,12558144,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.70 MiB | 114.00 KiB/s\r",,terminal_output +1612,12559219,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.81 MiB | 111.00 KiB/s\r",,terminal_output +1613,12560336,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 67.93 MiB | 111.00 KiB/s\r",,terminal_output +1614,12561487,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.07 MiB | 112.00 KiB/s\r",,terminal_output +1615,12562064,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.14 MiB | 114.00 KiB/s\r",,terminal_output +1616,12563154,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.27 MiB | 116.00 KiB/s\r",,terminal_output +1617,12564223,"TERMINAL",0,0,"Receiving objects: 42% (34542/81085), 68.39 MiB | 118.00 KiB/s\r",,terminal_output +1618,12564948,"TERMINAL",0,0,"Receiving objects: 42% (34586/81085), 68.45 MiB | 118.00 KiB/s\r",,terminal_output +1619,12566324,"TERMINAL",0,0,"Receiving objects: 42% (34771/81085), 68.63 MiB | 119.00 KiB/s\r",,terminal_output +1620,12566930,"TERMINAL",0,0,"Receiving objects: 42% (34771/81085), 68.68 MiB | 113.00 KiB/s\r",,terminal_output +1621,12568012,"TERMINAL",0,0,"Receiving objects: 42% (34771/81085), 68.80 MiB | 112.00 KiB/s\r",,terminal_output +1622,12569053,"TERMINAL",0,0,"Receiving objects: 42% (34771/81085), 68.91 MiB | 111.00 KiB/s\r",,terminal_output +1623,12570099,"TERMINAL",0,0,"Receiving objects: 42% (34771/81085), 69.04 MiB | 111.00 KiB/s\r",,terminal_output +1624,12570883,"TERMINAL",0,0,"Receiving objects: 43% (34867/81085), 69.07 MiB | 101.00 KiB/s\r",,terminal_output +1625,12570936,"TERMINAL",0,0,"Receiving objects: 43% (35039/81085), 69.07 MiB | 101.00 KiB/s\r",,terminal_output +1626,12572018,"TERMINAL",0,0,"Receiving objects: 43% (35241/81085), 69.23 MiB | 111.00 KiB/s\r",,terminal_output +1627,12573078,"TERMINAL",0,0,"Receiving objects: 43% (35424/81085), 69.35 MiB | 110.00 KiB/s\r",,terminal_output +1628,12573929,"TERMINAL",0,0,"Receiving objects: 43% (35564/81085), 69.41 MiB | 110.00 KiB/s\r",,terminal_output +1629,12574883,"TERMINAL",0,0,"Receiving objects: 44% (35678/81085), 69.54 MiB | 114.00 KiB/s\r",,terminal_output +1630,12574967,"TERMINAL",0,0,"Receiving objects: 44% (35685/81085), 69.54 MiB | 114.00 KiB/s\r",,terminal_output +1631,12575959,"TERMINAL",0,0,"Receiving objects: 44% (36007/81085), 69.66 MiB | 124.00 KiB/s\r",,terminal_output +1632,12577423,"TERMINAL",0,0,"Receiving objects: 44% (36030/81085), 69.82 MiB | 112.00 KiB/s\r",,terminal_output +1633,12578107,"TERMINAL",0,0,"Receiving objects: 44% (36030/81085), 69.85 MiB | 101.00 KiB/s\r",,terminal_output +1634,12579182,"TERMINAL",0,0,"Receiving objects: 44% (36211/81085), 70.00 MiB | 107.00 KiB/s\r",,terminal_output +1635,12579951,"TERMINAL",0,0,"Receiving objects: 44% (36335/81085), 70.07 MiB | 106.00 KiB/s\r",,terminal_output +1636,12580432,"TERMINAL",0,0,"Receiving objects: 45% (36489/81085), 70.13 MiB | 107.00 KiB/s\r",,terminal_output +1637,12580935,"TERMINAL",0,0,"Receiving objects: 45% (36584/81085), 70.19 MiB | 108.00 KiB/s\r",,terminal_output +1638,12581952,"TERMINAL",0,0,"Receiving objects: 45% (37298/81085), 70.32 MiB | 112.00 KiB/s\rReceiving objects: 46% (37300/81085), 70.32 MiB | 112.00 KiB/s\r",,terminal_output +1639,12582476,"TERMINAL",0,0,"Receiving objects: 47% (38110/81085), 70.38 MiB | 115.00 KiB/s\r",,terminal_output +1640,12582931,"TERMINAL",0,0,"Receiving objects: 47% (38278/81085), 70.44 MiB | 126.00 KiB/s\r",,terminal_output +1641,12584015,"TERMINAL",0,0,"Receiving objects: 47% (38792/81085), 70.56 MiB | 117.00 KiB/s\r",,terminal_output +1642,12584576,"TERMINAL",0,0,"Receiving objects: 48% (38921/81085), 70.61 MiB | 115.00 KiB/s\r",,terminal_output +1643,12584976,"TERMINAL",0,0,"Receiving objects: 48% (38952/81085), 70.61 MiB | 115.00 KiB/s\r",,terminal_output +1644,12585917,"TERMINAL",0,0,"Receiving objects: 48% (39688/81085), 70.75 MiB | 114.00 KiB/s\r",,terminal_output +1645,12586031,"TERMINAL",0,0,"Receiving objects: 49% (39732/81085), 70.75 MiB | 114.00 KiB/s\r",,terminal_output +1646,12586934,"TERMINAL",0,0,"Receiving objects: 49% (40376/81085), 70.86 MiB | 113.00 KiB/s\r",,terminal_output +1647,12587126,"TERMINAL",0,0,"Receiving objects: 50% (40543/81085), 70.86 MiB | 113.00 KiB/s\r",,terminal_output +1648,12587949,"TERMINAL",0,0,"Receiving objects: 50% (40738/81085), 70.98 MiB | 114.00 KiB/s\r",,terminal_output +1649,12588736,"TERMINAL",0,0,"Receiving objects: 51% (41354/81085), 71.04 MiB | 116.00 KiB/s\r",,terminal_output +1650,12588944,"TERMINAL",0,0,"Receiving objects: 51% (41464/81085), 71.10 MiB | 115.00 KiB/s\r",,terminal_output +1651,12590409,"TERMINAL",0,0,"Receiving objects: 51% (41517/81085), 71.27 MiB | 113.00 KiB/s\r",,terminal_output +1652,12590933,"TERMINAL",0,0,"Receiving objects: 51% (41674/81085), 71.33 MiB | 113.00 KiB/s\r",,terminal_output +1653,12592012,"TERMINAL",0,0,"Receiving objects: 51% (41674/81085), 71.45 MiB | 114.00 KiB/s\r",,terminal_output +1654,12593078,"TERMINAL",0,0,"Receiving objects: 51% (41674/81085), 71.57 MiB | 112.00 KiB/s\r",,terminal_output +1655,12594130,"TERMINAL",0,0,"Receiving objects: 51% (41674/81085), 71.70 MiB | 116.00 KiB/s\r",,terminal_output +1656,12595222,"TERMINAL",0,0,"Receiving objects: 51% (41674/81085), 71.81 MiB | 114.00 KiB/s\r",,terminal_output +1657,12596143,"TERMINAL",0,0,"Receiving objects: 51% (41717/81085), 71.88 MiB | 107.00 KiB/s\r",,terminal_output +1658,12596919,"TERMINAL",0,0,"Receiving objects: 51% (42036/81085), 71.99 MiB | 117.00 KiB/s\r",,terminal_output +1659,12597048,"TERMINAL",0,0,"Receiving objects: 52% (42165/81085), 71.99 MiB | 117.00 KiB/s\r",,terminal_output +1660,12597961,"TERMINAL",0,0,"Receiving objects: 52% (42803/81085), 72.11 MiB | 116.00 KiB/s\rReceiving objects: 53% (42976/81085), 72.11 MiB | 116.00 KiB/s\r",,terminal_output +1661,12598973,"TERMINAL",0,0,"Receiving objects: 53% (43747/81085), 72.21 MiB | 113.00 KiB/s\r",,terminal_output +1662,12599161,"TERMINAL",0,0,"Receiving objects: 54% (43786/81085), 72.21 MiB | 113.00 KiB/s\r",,terminal_output +1663,12599923,"TERMINAL",0,0,"Receiving objects: 54% (44417/81085), 72.32 MiB | 111.00 KiB/s\r",,terminal_output +1664,12600290,"TERMINAL",0,0,"Receiving objects: 55% (44597/81085), 72.32 MiB | 111.00 KiB/s\r",,terminal_output +1665,12600916,"TERMINAL",0,0,"Receiving objects: 55% (45076/81085), 72.38 MiB | 111.00 KiB/s\r",,terminal_output +1666,12601126,"TERMINAL",0,0,"Receiving objects: 56% (45408/81085), 72.44 MiB | 119.00 KiB/s\r",,terminal_output +1667,12601968,"TERMINAL",0,0,"Receiving objects: 56% (45662/81085), 72.50 MiB | 107.00 KiB/s\r",,terminal_output +1668,12602938,"TERMINAL",0,0,"Receiving objects: 56% (45741/81085), 72.61 MiB | 107.00 KiB/s\r",,terminal_output +1669,12603957,"TERMINAL",0,0,"Receiving objects: 56% (45908/81085), 72.74 MiB | 110.00 KiB/s\r",,terminal_output +1670,12604927,"TERMINAL",0,0,"Receiving objects: 56% (46090/81085), 72.86 MiB | 111.00 KiB/s\r",,terminal_output +1671,12605513,"TERMINAL",0,0,"Receiving objects: 57% (46219/81085), 72.93 MiB | 116.00 KiB/s\r",,terminal_output +1672,12606016,"TERMINAL",0,0,"Receiving objects: 57% (46226/81085), 72.96 MiB | 104.00 KiB/s\r",,terminal_output +1673,12606968,"TERMINAL",0,0,"Receiving objects: 57% (46480/81085), 73.06 MiB | 115.00 KiB/s\r",,terminal_output +1674,12607960,"TERMINAL",0,0,"Receiving objects: 58% (47030/81085), 73.17 MiB | 115.00 KiB/s\rReceiving objects: 58% (47051/81085), 73.17 MiB | 115.00 KiB/s\r",,terminal_output +1675,12609140,"TERMINAL",0,0,"Receiving objects: 58% (47230/81085), 73.35 MiB | 113.00 KiB/s\r",,terminal_output +1676,12610306,"TERMINAL",0,0,"Receiving objects: 58% (47230/81085), 73.47 MiB | 111.00 KiB/s\r",,terminal_output +1677,12611351,"TERMINAL",0,0,"Receiving objects: 58% (47230/81085), 73.58 MiB | 111.00 KiB/s\r",,terminal_output +1678,12611929,"TERMINAL",0,0,"Receiving objects: 58% (47336/81085), 73.64 MiB | 112.00 KiB/s\r",,terminal_output +1679,12612960,"TERMINAL",0,0,"Receiving objects: 58% (47833/81085), 73.75 MiB | 111.00 KiB/s\r",,terminal_output +1680,12613023,"TERMINAL",0,0,"Receiving objects: 59% (47841/81085), 73.75 MiB | 111.00 KiB/s\r",,terminal_output +1681,12613948,"TERMINAL",0,0,"Receiving objects: 59% (48043/81085), 73.82 MiB | 112.00 KiB/s\r",,terminal_output +1682,12614776,"TERMINAL",0,0,"Receiving objects: 60% (48651/81085), 73.94 MiB | 111.00 KiB/s\r",,terminal_output +1683,12614961,"TERMINAL",0,0,"Receiving objects: 60% (48786/81085), 73.94 MiB | 111.00 KiB/s\r",,terminal_output +1684,12615961,"TERMINAL",0,0,"Receiving objects: 60% (49011/81085), 74.07 MiB | 116.00 KiB/s\r",,terminal_output +1685,12617400,"TERMINAL",0,0,"Receiving objects: 60% (49260/81085), 74.21 MiB | 104.00 KiB/s\r",,terminal_output +1686,12617985,"TERMINAL",0,0,"Receiving objects: 60% (49260/81085), 74.32 MiB | 113.00 KiB/s\r",,terminal_output +1687,12619139,"TERMINAL",0,0,"Receiving objects: 60% (49260/81085), 74.46 MiB | 116.00 KiB/s\r",,terminal_output +1688,12620316,"TERMINAL",0,0,"Receiving objects: 60% (49260/81085), 74.60 MiB | 114.00 KiB/s\r",,terminal_output +1689,12620932,"TERMINAL",0,0,"Receiving objects: 60% (49260/81085), 74.66 MiB | 113.00 KiB/s\r",,terminal_output +1690,12622012,"TERMINAL",0,0,"Receiving objects: 60% (49417/81085), 74.71 MiB | 111.00 KiB/s\r",,terminal_output +1691,12622151,"TERMINAL",0,0,"Receiving objects: 61% (49462/81085), 74.79 MiB | 112.00 KiB/s\r",,terminal_output +1692,12623173,"TERMINAL",0,0,"Receiving objects: 61% (49679/81085), 74.91 MiB | 117.00 KiB/s\r",,terminal_output +1693,12624240,"TERMINAL",0,0,"Receiving objects: 61% (49679/81085), 75.04 MiB | 116.00 KiB/s\r",,terminal_output +1694,12625323,"TERMINAL",0,0,"Receiving objects: 61% (49679/81085), 75.17 MiB | 116.00 KiB/s\r",,terminal_output +1695,12626378,"TERMINAL",0,0,"Receiving objects: 61% (49679/81085), 75.29 MiB | 121.00 KiB/s\r",,terminal_output +1696,12626975,"TERMINAL",0,0,"Receiving objects: 61% (49774/81085), 75.35 MiB | 119.00 KiB/s\r",,terminal_output +1697,12629141,"TERMINAL",0,0,"Receiving objects: 61% (49868/81085), 75.43 MiB | 88.00 KiB/s \r",,terminal_output +1698,12629587,"TERMINAL",0,0,"Receiving objects: 62% (50273/81085), 75.43 MiB | 88.00 KiB/s\r",,terminal_output +1699,12629943,"TERMINAL",0,0,"Receiving objects: 62% (51051/81085), 75.52 MiB | 93.00 KiB/s\rReceiving objects: 63% (51084/81085), 75.52 MiB | 93.00 KiB/s\r",,terminal_output +1700,12630976,"TERMINAL",0,0,"Receiving objects: 63% (51388/81085), 75.72 MiB | 107.00 KiB/s\r",,terminal_output +1701,12631996,"TERMINAL",0,0,"Receiving objects: 63% (51631/81085), 75.82 MiB | 101.00 KiB/s\r",,terminal_output +1702,12633464,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.00 MiB | 100.00 KiB/s\r",,terminal_output +1703,12634084,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.07 MiB | 134.00 KiB/s\r",,terminal_output +1704,12635552,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.20 MiB | 103.00 KiB/s\r",,terminal_output +1705,12636120,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.32 MiB | 114.00 KiB/s\r",,terminal_output +1706,12637229,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.46 MiB | 120.00 KiB/s\r",,terminal_output +1707,12638317,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.58 MiB | 120.00 KiB/s\r",,terminal_output +1708,12639335,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.68 MiB | 118.00 KiB/s\r",,terminal_output +1709,12640438,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.82 MiB | 129.00 KiB/s\r",,terminal_output +1710,12640979,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.86 MiB | 112.00 KiB/s\r",,terminal_output +1711,12642032,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 76.96 MiB | 108.00 KiB/s\r",,terminal_output +1712,12643109,"TERMINAL",0,0,"Receiving objects: 63% (51785/81085), 77.09 MiB | 108.00 KiB/s\r",,terminal_output +1713,12643866,"TERMINAL",0,0,"Receiving objects: 64% (51895/81085), 77.16 MiB | 113.00 KiB/s\r",,terminal_output +1714,12643918,"TERMINAL",0,0,"Receiving objects: 64% (51906/81085), 77.16 MiB | 113.00 KiB/s\r",,terminal_output +1715,12645560,"TERMINAL",0,0,"Receiving objects: 64% (52005/81085), 77.31 MiB | 98.00 KiB/s \r",,terminal_output +1716,12646146,"TERMINAL",0,0,"Receiving objects: 64% (52005/81085), 77.43 MiB | 113.00 KiB/s\r",,terminal_output +1717,12647151,"TERMINAL",0,0,"Receiving objects: 64% (52005/81085), 77.53 MiB | 112.00 KiB/s\r",,terminal_output +1718,12648193,"TERMINAL",0,0,"Receiving objects: 64% (52005/81085), 77.62 MiB | 107.00 KiB/s\r",,terminal_output +1719,12649026,"TERMINAL",0,0,"Receiving objects: 64% (52006/81085), 77.69 MiB | 107.00 KiB/s\r",,terminal_output +1720,12650479,"TERMINAL",0,0,"Receiving objects: 64% (52062/81085), 77.86 MiB | 114.00 KiB/s\r",,terminal_output +1721,12651042,"TERMINAL",0,0,"Receiving objects: 64% (52062/81085), 77.92 MiB | 103.00 KiB/s\r",,terminal_output +1722,12652237,"TERMINAL",0,0,"Receiving objects: 64% (52062/81085), 78.04 MiB | 102.00 KiB/s\r",,terminal_output +1723,12653270,"TERMINAL",0,0,"Receiving objects: 64% (52062/81085), 78.09 MiB | 94.00 KiB/s \r",,terminal_output +1724,12653934,"TERMINAL",0,0,"Receiving objects: 64% (52062/81085), 78.20 MiB | 103.00 KiB/s\r",,terminal_output +1725,12654929,"TERMINAL",0,0,"Receiving objects: 64% (52261/81085), 78.28 MiB | 105.00 KiB/s\r",,terminal_output +1726,12656070,"TERMINAL",0,0,"Receiving objects: 64% (52319/81085), 78.46 MiB | 110.00 KiB/s\r",,terminal_output +1727,12657105,"TERMINAL",0,0,"Receiving objects: 64% (52319/81085), 78.57 MiB | 109.00 KiB/s\r",,terminal_output +1728,12658231,"TERMINAL",0,0,"Receiving objects: 64% (52319/81085), 78.70 MiB | 125.00 KiB/s\r",,terminal_output +1729,12659326,"TERMINAL",0,0,"Receiving objects: 64% (52319/81085), 78.80 MiB | 110.00 KiB/s\r",,terminal_output +1730,12660235,"TERMINAL",0,0,"Receiving objects: 64% (52320/81085), 78.86 MiB | 109.00 KiB/s\r",,terminal_output +1731,12660928,"TERMINAL",0,0,"Receiving objects: 64% (52501/81085), 78.93 MiB | 108.00 KiB/s\r",,terminal_output +1732,12661821,"TERMINAL",0,0,"Receiving objects: 65% (52706/81085), 78.98 MiB | 99.00 KiB/s \r",,terminal_output +1733,12661963,"TERMINAL",0,0,"Receiving objects: 65% (52813/81085), 78.98 MiB | 99.00 KiB/s\r",,terminal_output +1734,12662991,"TERMINAL",0,0,"Receiving objects: 65% (53093/81085), 79.15 MiB | 108.00 KiB/s\r",,terminal_output +1735,12663990,"TERMINAL",0,0,"Receiving objects: 65% (53425/81085), 79.26 MiB | 105.00 KiB/s\r",,terminal_output +1736,12664048,"TERMINAL",0,0,"Receiving objects: 66% (53517/81085), 79.26 MiB | 105.00 KiB/s\r",,terminal_output +1737,12665024,"TERMINAL",0,0,"Receiving objects: 66% (54048/81085), 79.39 MiB | 110.00 KiB/s\r",,terminal_output +1738,12665339,"TERMINAL",0,0,"Receiving objects: 67% (54327/81085), 79.44 MiB | 110.00 KiB/s\r",,terminal_output +1739,12665980,"TERMINAL",0,0,"Receiving objects: 67% (54644/81085), 79.50 MiB | 110.00 KiB/s\r",,terminal_output +1740,12666546,"TERMINAL",0,0,"Receiving objects: 68% (55138/81085), 79.57 MiB | 123.00 KiB/s\r",,terminal_output +1741,12666941,"TERMINAL",0,0,"Receiving objects: 68% (55492/81085), 79.62 MiB | 112.00 KiB/s\r",,terminal_output +1742,12667362,"TERMINAL",0,0,"Receiving objects: 69% (55949/81085), 79.62 MiB | 112.00 KiB/s\r",,terminal_output +1743,12667971,"TERMINAL",0,0,"Receiving objects: 69% (56450/81085), 79.75 MiB | 117.00 KiB/s\r",,terminal_output +1744,12668927,"TERMINAL",0,0,"Receiving objects: 69% (56698/81085), 79.80 MiB | 115.00 KiB/s\r",,terminal_output +1745,12669058,"TERMINAL",0,0,"Receiving objects: 70% (56760/81085), 79.87 MiB | 116.00 KiB/s\r",,terminal_output +1746,12670305,"TERMINAL",0,0,"Receiving objects: 70% (57431/81085), 79.94 MiB | 101.00 KiB/s\rReceiving objects: 71% (57571/81085), 79.94 MiB | 101.00 KiB/s\r",,terminal_output +1747,12670643,"TERMINAL",0,0,"Receiving objects: 72% (58382/81085), 79.94 MiB | 101.00 KiB/s\r",,terminal_output +1748,12670942,"TERMINAL",0,0,"Receiving objects: 72% (58686/81085), 80.06 MiB | 113.00 KiB/s\r",,terminal_output +1749,12671939,"TERMINAL",0,0,"Receiving objects: 72% (59067/81085), 80.16 MiB | 108.00 KiB/s\r",,terminal_output +1750,12672138,"TERMINAL",0,0,"Receiving objects: 73% (59193/81085), 80.16 MiB | 108.00 KiB/s\r",,terminal_output +1751,12673003,"TERMINAL",0,0,"Receiving objects: 73% (59447/81085), 80.26 MiB | 105.00 KiB/s\r",,terminal_output +1752,12673949,"TERMINAL",0,0,"Receiving objects: 73% (59789/81085), 80.32 MiB | 103.00 KiB/s\r",,terminal_output +1753,12674123,"TERMINAL",0,0,"Receiving objects: 74% (60003/81085), 80.39 MiB | 104.00 KiB/s\r",,terminal_output +1754,12674973,"TERMINAL",0,0,"Receiving objects: 74% (60530/81085), 80.45 MiB | 106.00 KiB/s\r",,terminal_output +1755,12675333,"TERMINAL",0,0,"Receiving objects: 75% (60814/81085), 80.50 MiB | 117.00 KiB/s\r",,terminal_output +1756,12675946,"TERMINAL",0,0,"Receiving objects: 75% (61093/81085), 80.57 MiB | 106.00 KiB/s\r",,terminal_output +1757,12676997,"TERMINAL",0,0,"Receiving objects: 75% (61444/81085), 80.69 MiB | 107.00 KiB/s\r",,terminal_output +1758,12677298,"TERMINAL",0,0,"Receiving objects: 76% (61625/81085), 80.69 MiB | 107.00 KiB/s\r",,terminal_output +1759,12677935,"TERMINAL",0,0,"Receiving objects: 76% (61988/81085), 80.75 MiB | 107.00 KiB/s\r",,terminal_output +1760,12678437,"TERMINAL",0,0,"Receiving objects: 77% (62436/81085), 80.82 MiB | 112.00 KiB/s\r",,terminal_output +1761,12678999,"TERMINAL",0,0,"Receiving objects: 77% (62893/81085), 80.87 MiB | 113.00 KiB/s\r",,terminal_output +1762,12679354,"TERMINAL",0,0,"Receiving objects: 78% (63247/81085), 80.93 MiB | 110.00 KiB/s\r",,terminal_output +1763,12679938,"TERMINAL",0,0,"Receiving objects: 78% (63671/81085), 80.99 MiB | 111.00 KiB/s\r",,terminal_output +1764,12681084,"TERMINAL",0,0,"Receiving objects: 78% (63685/81085), 81.07 MiB | 97.00 KiB/s \rReceiving objects: 79% (64058/81085), 81.07 MiB | 97.00 KiB/s\r",,terminal_output +1765,12681650,"TERMINAL",0,0,"Receiving objects: 80% (64868/81085), 81.19 MiB | 109.00 KiB/s\r",,terminal_output +1766,12681978,"TERMINAL",0,0,"Receiving objects: 80% (65121/81085), 81.19 MiB | 109.00 KiB/s\r",,terminal_output +1767,12682977,"TERMINAL",0,0,"Receiving objects: 80% (65575/81085), 81.32 MiB | 111.00 KiB/s\r",,terminal_output +1768,12683153,"TERMINAL",0,0,"Receiving objects: 81% (65679/81085), 81.32 MiB | 111.00 KiB/s\r",,terminal_output +1769,12683948,"TERMINAL",0,0,"Receiving objects: 81% (66296/81085), 81.43 MiB | 109.00 KiB/s\r",,terminal_output +1770,12685010,"TERMINAL",0,0,"Receiving objects: 81% (66411/81085), 81.55 MiB | 110.00 KiB/s\r",,terminal_output +1771,12685190,"TERMINAL",0,0,"Receiving objects: 82% (66490/81085), 81.55 MiB | 110.00 KiB/s\r",,terminal_output +1772,12685860,"TERMINAL",0,0,"Receiving objects: 83% (67301/81085), 81.61 MiB | 113.00 KiB/s\r",,terminal_output +1773,12685953,"TERMINAL",0,0,"Receiving objects: 83% (67324/81085), 81.68 MiB | 126.00 KiB/s\r",,terminal_output +1774,12686734,"TERMINAL",0,0,"Receiving objects: 84% (68112/81085), 81.72 MiB | 112.00 KiB/s\r",,terminal_output +1775,12686971,"TERMINAL",0,0,"Receiving objects: 84% (68326/81085), 81.78 MiB | 112.00 KiB/s\r",,terminal_output +1776,12687590,"TERMINAL",0,0,"Receiving objects: 85% (68923/81085), 81.83 MiB | 110.00 KiB/s\r",,terminal_output +1777,12687924,"TERMINAL",0,0,"Receiving objects: 85% (69160/81085), 81.83 MiB | 110.00 KiB/s\r",,terminal_output +1778,12688419,"TERMINAL",0,0,"Receiving objects: 86% (69734/81085), 81.89 MiB | 111.00 KiB/s\r",,terminal_output +1779,12688961,"TERMINAL",0,0,"Receiving objects: 86% (70260/81085), 81.96 MiB | 113.00 KiB/s\r",,terminal_output +1780,12689609,"TERMINAL",0,0,"Receiving objects: 87% (70544/81085), 82.03 MiB | 104.00 KiB/s\r",,terminal_output +1781,12689932,"TERMINAL",0,0,"Receiving objects: 87% (71328/81085), 82.03 MiB | 104.00 KiB/s\r",,terminal_output +1782,12689985,"TERMINAL",0,0,"Receiving objects: 88% (71355/81085), 82.03 MiB | 104.00 KiB/s\r",,terminal_output +1783,12691034,"TERMINAL",0,0,"Receiving objects: 88% (71605/81085), 82.21 MiB | 115.00 KiB/s\r",,terminal_output +1784,12691693,"TERMINAL",0,0,"Receiving objects: 89% (72166/81085), 82.25 MiB | 113.00 KiB/s\r",,terminal_output +1785,12692000,"TERMINAL",0,0,"Receiving objects: 89% (72299/81085), 82.32 MiB | 113.00 KiB/s\r",,terminal_output +1786,12692934,"TERMINAL",0,0,"Receiving objects: 89% (72897/81085), 82.45 MiB | 116.00 KiB/s\r",,terminal_output +1787,12693196,"TERMINAL",0,0,"Receiving objects: 90% (72977/81085), 82.45 MiB | 116.00 KiB/s\r",,terminal_output +1788,12693963,"TERMINAL",0,0,"Receiving objects: 90% (73559/81085), 82.51 MiB | 116.00 KiB/s\r",,terminal_output +1789,12694236,"TERMINAL",0,0,"Receiving objects: 91% (73788/81085), 82.57 MiB | 113.00 KiB/s\r",,terminal_output +1790,12694968,"TERMINAL",0,0,"Receiving objects: 91% (74266/81085), 82.63 MiB | 122.00 KiB/s\r",,terminal_output +1791,12695384,"TERMINAL",0,0,"Receiving objects: 92% (74599/81085), 82.68 MiB | 110.00 KiB/s\r",,terminal_output +1792,12695936,"TERMINAL",0,0,"Receiving objects: 92% (75016/81085), 82.74 MiB | 108.00 KiB/s\r",,terminal_output +1793,12696313,"TERMINAL",0,0,"Receiving objects: 93% (75410/81085), 82.80 MiB | 112.00 KiB/s\r",,terminal_output +1794,12696921,"TERMINAL",0,0,"Receiving objects: 93% (75819/81085), 82.86 MiB | 115.00 KiB/s\r",,terminal_output +1795,12697461,"TERMINAL",0,0,"Receiving objects: 94% (76220/81085), 82.93 MiB | 115.00 KiB/s\r",,terminal_output +1796,12697946,"TERMINAL",0,0,"Receiving objects: 94% (76515/81085), 82.99 MiB | 114.00 KiB/s\r",,terminal_output +1797,12698515,"TERMINAL",0,0,"Receiving objects: 95% (77031/81085), 83.06 MiB | 116.00 KiB/s\r",,terminal_output +1798,12698956,"TERMINAL",0,0,"Receiving objects: 95% (77279/81085), 83.11 MiB | 115.00 KiB/s\r",,terminal_output +1799,12699555,"TERMINAL",0,0,"Receiving objects: 96% (77842/81085), 83.17 MiB | 114.00 KiB/s\r",,terminal_output +1800,12699936,"TERMINAL",0,0,"Receiving objects: 96% (78255/81085), 83.17 MiB | 114.00 KiB/s\r",,terminal_output +1801,12700380,"TERMINAL",0,0,"Receiving objects: 97% (78653/81085), 83.25 MiB | 118.00 KiB/s\r",,terminal_output +1802,12700937,"TERMINAL",0,0,"Receiving objects: 97% (79076/81085), 83.31 MiB | 120.00 KiB/s\r",,terminal_output +1803,12701346,"TERMINAL",0,0,"Receiving objects: 98% (79464/81085), 83.36 MiB | 118.00 KiB/s\r",,terminal_output +1804,12701957,"TERMINAL",0,0,"Receiving objects: 98% (79864/81085), 83.43 MiB | 119.00 KiB/s\r",,terminal_output +1805,12702378,"TERMINAL",0,0,"Receiving objects: 99% (80275/81085), 83.50 MiB | 119.00 KiB/s\r",,terminal_output +1806,12703114,"TERMINAL",0,0,"Receiving objects: 99% (80740/81085), 83.56 MiB | 120.00 KiB/s\r",,terminal_output +1807,12704006,"TERMINAL",0,0,"Receiving objects: 99% (80797/81085), 83.66 MiB | 116.00 KiB/s\r",,terminal_output +1808,12704935,"TERMINAL",0,0,"Receiving objects: 99% (80918/81085), 83.77 MiB | 112.00 KiB/s\r",,terminal_output +1809,12706302,"TERMINAL",0,0,"Receiving objects: 99% (81026/81085), 83.89 MiB | 109.00 KiB/s\r",,terminal_output +1810,12706941,"TERMINAL",0,0,"Receiving objects: 99% (81029/81085), 84.00 MiB | 105.00 KiB/s\r",,terminal_output +1811,12707934,"TERMINAL",0,0,"Receiving objects: 99% (81046/81085), 84.06 MiB | 106.00 KiB/s\r",,terminal_output +1812,12708770,"TERMINAL",0,0,"remote: Total 81085 (delta 131), reused 83 (delta 83), pack-reused 80910 (from 3)\r\nReceiving objects: 100% (81085/81085), 84.13 MiB | 108.00 KiB/s\rReceiving objects: 100% (81085/81085), 84.21 MiB | 101.00 KiB/s, done.\r\nResolving deltas: 0% (0/61058)\rResolving deltas: 1% (613/61058)\rResolving deltas: 2% (1222/61058)\rResolving deltas: 3% (1832/61058)\rResolving deltas: 4% (2443/61058)\rResolving deltas: 5% (3054/61058)\rResolving deltas: 6% (3664/61058)\rResolving deltas: 7% (4277/61058)\rResolving deltas: 8% (4885/61058)\rResolving deltas: 9% (5496/61058)\rResolving deltas: 10% (6106/61058)\rResolving deltas: 11% (6717/61058)\rResolving deltas: 12% (7327/61058)\r",,terminal_output +1813,12708954,"TERMINAL",0,0,"Resolving deltas: 13% (7938/61058)\rResolving deltas: 14% (8549/61058)\rResolving deltas: 15% (9159/61058)\rResolving deltas: 16% (9770/61058)\rResolving deltas: 17% (10381/61058)\rResolving deltas: 18% (10991/61058)\rResolving deltas: 19% (11602/61058)\rResolving deltas: 20% (12212/61058)\rResolving deltas: 21% (12823/61058)\r",,terminal_output +1814,12709523,"TERMINAL",0,0,"Resolving deltas: 22% (13433/61058)\rResolving deltas: 23% (14044/61058)\rResolving deltas: 24% (14654/61058)\rResolving deltas: 25% (15265/61058)\rResolving deltas: 26% (15876/61058)\rResolving deltas: 27% (16487/61058)\rResolving deltas: 28% (17097/61058)\rResolving deltas: 29% (17708/61058)\rResolving deltas: 30% (18318/61058)\rResolving deltas: 31% (18930/61058)\rResolving deltas: 32% (19539/61058)\rResolving deltas: 33% (20150/61058)\rResolving deltas: 34% (20761/61058)\rResolving deltas: 35% (21371/61058)\rResolving deltas: 36% (21981/61058)\rResolving deltas: 37% (22593/61058)\rResolving deltas: 38% (23203/61058)\rResolving deltas: 39% (23813/61058)\rResolving deltas: 40% (24424/61058)\rResolving deltas: 41% (25034/61058)\rResolving deltas: 42% (25645/61058)\rResolving deltas: 43% (26257/61058)\rResolving deltas: 44% (26867/61058)\rResolving deltas: 45% (27477/61058)\rResolving deltas: 46% (28087/61058)\rResolving deltas: 47% (28700/61058)\rResolving deltas: 48% (29308/61058)\rResolving deltas: 49% (29919/61058)\rResolving deltas: 50% (30529/61058)\rResolving deltas: 51% (31140/61058)\rResolving deltas: 52% (31751/61058)\rResolving deltas: 53% (32361/61058)\rResolving deltas: 54% (32974/61058)\rResolving deltas: 55% (33582/61058)\rResolving deltas: 56% (34194/61058)\rResolving deltas: 57% (34804/61058)\rResolving deltas: 58% (35414/61058)\rResolving deltas: 59% (36025/61058)\rResolving deltas: 60% (36635/61058)\rResolving deltas: 61% (37246/61058)\rResolving deltas: 62% (37858/61058)\rResolving deltas: 63% (38467/61058)\rResolving deltas: 64% (39078/61058)\rResolving deltas: 65% (39689/61058)\rResolving deltas: 66% (40300/61058)\rResolving deltas: 67% (40909/61058)\rResolving deltas: 68% (41521/61058)\rResolving deltas: 69% (42132/61058)\rResolving deltas: 70% (42741/61058)\rResolving deltas: 71% (43352/61058)\rResolving deltas: 72% (43962/61058)\rResolving deltas: 73% (44574/61058)\rResolving deltas: 74% (45183/61058)\rResolving deltas: 75% (45795/61058)\rResolving deltas: 76% (46405/61058)\rResolving deltas: 77% (47015/61058)\rResolving deltas: 78% (47626/61058)\rResolving deltas: 79% (48237/61058)\rResolving deltas: 80% (48847/61058)\rResolving deltas: 81% (49457/61058)\rResolving deltas: 82% (50068/61058)\rResolving deltas: 83% (50679/61058)\rResolving deltas: 84% (51289/61058)\rResolving deltas: 85% (51901/61058)\rResolving deltas: 86% (52511/61058)\rResolving deltas: 87% (53121/61058)\rResolving deltas: 88% (53733/61058)\rResolving deltas: 89% (54343/61058)\rResolving deltas: 90% (54953/61058)\rResolving deltas: 91% (55564/61058)\rResolving deltas: 92% (56175/61058)\rResolving deltas: 93% (56784/61058)\rResolving deltas: 94% (57395/61058)\rResolving deltas: 95% (58006/61058)\rResolving deltas: 96% (58616/61058)\rResolving deltas: 97% (59227/61058)\rResolving deltas: 98% (59837/61058)\rResolving deltas: 99% (60449/61058)\rResolving deltas: 100% (61058/61058)\rResolving deltas: 100% (61058/61058), done.\r\n",,terminal_output +1815,12709782,"TERMINAL",0,0,"% \r \r",,terminal_output +1816,13304422,"train_dynamics.py",0,0,"",python,selection_command +1817,13305180,"train_dynamics.py",373,0,"",python,selection_command +1818,13306075,"train_dynamics.py",377,0,"",python,selection_command +1819,13306237,"train_dynamics.py",378,0,"",python,selection_command +1820,13306994,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",0,0,"# Copyright 2024 The Flax Authors.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\nfrom flax.linen.pooling import avg_pool as avg_pool\nfrom flax.linen.pooling import max_pool as max_pool\nfrom flax.linen.pooling import min_pool as min_pool\nfrom flax.linen.pooling import pool as pool\nfrom flax.typing import Initializer as Initializer\n\nfrom .bridge import wrappers as wrappers\nfrom .filterlib import WithTag as WithTag\nfrom .filterlib import PathContains as PathContains\nfrom .filterlib import OfType as OfType\nfrom .filterlib import Any as Any\nfrom .filterlib import All as All\nfrom .filterlib import Not as Not\nfrom .filterlib import Everything as Everything\nfrom .filterlib import Nothing as Nothing\nfrom .graph import GraphDef as GraphDef\nfrom .graph import GraphState as GraphState\nfrom .graph import PureState as PureState\nfrom .object import Object as Object\nfrom .helpers import Dict as Dict\nfrom .helpers import Sequential as Sequential\nfrom .helpers import TrainState as TrainState\nfrom .module import M as M\nfrom .module import Module as Module\nfrom .graph import merge as merge\nfrom .graph import UpdateContext as UpdateContext\nfrom .graph import update_context as update_context\nfrom .graph import current_update_context as current_update_context\nfrom .graph import split as split\nfrom .graph import update as update\nfrom .graph import clone as clone\nfrom .graph import pop as pop\nfrom .graph import state as state\nfrom .graph import graphdef as graphdef\nfrom .graph import iter_graph as iter_graph\nfrom .graph import call as call\nfrom .graph import SplitContext as SplitContext\nfrom .graph import split_context as split_context\nfrom .graph import MergeContext as MergeContext\nfrom .graph import merge_context as merge_context\nfrom .graph import variables as variables\nfrom .graph import cached_partial as cached_partial\nfrom .nn import initializers as initializers\nfrom .nn.activations import celu as celu\nfrom .nn.activations import elu as elu\nfrom .nn.activations import gelu as gelu\nfrom .nn.activations import glu as glu\nfrom .nn.activations import hard_sigmoid as hard_sigmoid\nfrom .nn.activations import hard_silu as hard_silu\nfrom .nn.activations import hard_swish as hard_swish\nfrom .nn.activations import hard_tanh as hard_tanh\nfrom .nn.activations import leaky_relu as leaky_relu\nfrom .nn.activations import log_sigmoid as log_sigmoid\nfrom .nn.activations import log_softmax as log_softmax\nfrom .nn.activations import logsumexp as logsumexp\nfrom .nn.activations import one_hot as one_hot\nfrom .nn.activations import relu as relu\nfrom .nn.activations import relu6 as relu6\nfrom .nn.activations import selu as selu\nfrom .nn.activations import sigmoid as sigmoid\nfrom .nn.activations import silu as silu\nfrom .nn.activations import soft_sign as soft_sign\nfrom .nn.activations import softmax as softmax\nfrom .nn.activations import softplus as softplus\nfrom .nn.activations import standardize as standardize\nfrom .nn.activations import swish as swish\nfrom .nn.activations import tanh as tanh\nfrom .nn.attention import MultiHeadAttention as MultiHeadAttention\nfrom .nn.attention import combine_masks as combine_masks\nfrom .nn.attention import dot_product_attention as dot_product_attention\nfrom .nn.attention import make_attention_mask as make_attention_mask\nfrom .nn.attention import make_causal_mask as make_causal_mask\nfrom .nn.recurrent import RNNCellBase as RNNCellBase\nfrom .nn.recurrent import LSTMCell as LSTMCell\nfrom .nn.recurrent import GRUCell as GRUCell\nfrom .nn.recurrent import OptimizedLSTMCell as OptimizedLSTMCell\nfrom .nn.recurrent import SimpleCell as SimpleCell\nfrom .nn.recurrent import RNN as RNN\nfrom .nn.recurrent import Bidirectional as Bidirectional\nfrom .nn.linear import Conv as Conv\nfrom .nn.linear import ConvTranspose as ConvTranspose\nfrom .nn.linear import Embed as Embed\nfrom .nn.linear import Linear as Linear\nfrom .nn.linear import LinearGeneral as LinearGeneral\nfrom .nn.linear import Einsum as Einsum\nfrom .nn.lora import LoRA as LoRA\nfrom .nn.lora import LoRALinear as LoRALinear\nfrom .nn.lora import LoRAParam as LoRAParam\nfrom .nn.normalization import BatchNorm as BatchNorm\nfrom .nn.normalization import LayerNorm as LayerNorm\nfrom .nn.normalization import RMSNorm as RMSNorm\nfrom .nn.normalization import GroupNorm as GroupNorm\nfrom .nn.stochastic import Dropout as Dropout\nfrom .rnglib import Rngs as Rngs\nfrom .rnglib import RngStream as RngStream\nfrom .rnglib import RngState as RngState\nfrom .rnglib import RngKey as RngKey\nfrom .rnglib import RngCount as RngCount\nfrom .rnglib import ForkStates as ForkStates\nfrom .rnglib import fork as fork\nfrom .rnglib import reseed as reseed\nfrom .rnglib import split_rngs as split_rngs\nfrom .rnglib import restore_rngs as restore_rngs\nfrom .spmd import PARTITION_NAME as PARTITION_NAME\nfrom .spmd import get_partition_spec as get_partition_spec\nfrom .spmd import get_named_sharding as get_named_sharding\nfrom .spmd import with_partitioning as with_partitioning\nfrom .spmd import with_sharding_constraint as with_sharding_constraint\nfrom .statelib import State as State\nfrom .statelib import to_flat_state as to_flat_state\nfrom .statelib import from_flat_state as from_flat_state\nfrom .statelib import to_pure_dict as to_pure_dict\nfrom .statelib import replace_by_pure_dict as replace_by_pure_dict\nfrom .statelib import filter_state as filter_state\nfrom .statelib import merge_state as merge_state\nfrom .statelib import split_state as split_state\nfrom .statelib import map_state as map_state\nfrom .training import metrics as metrics\nfrom .variablelib import Param as Param\n# this needs to be imported before optimizer to prevent circular import\nfrom .training import optimizer as optimizer\nfrom .training.metrics import Metric as Metric\nfrom .training.metrics import MultiMetric as MultiMetric\nfrom .training.optimizer import Optimizer as Optimizer\nfrom .transforms.autodiff import DiffState as DiffState\nfrom .transforms.autodiff import grad as grad\nfrom .transforms.autodiff import value_and_grad as value_and_grad\nfrom .transforms.autodiff import custom_vjp as custom_vjp\nfrom .transforms.autodiff import remat as remat\nfrom .transforms.compilation import jit as jit\nfrom .transforms.compilation import shard_map as shard_map\nfrom .transforms.compilation import StateSharding as StateSharding\nfrom .transforms.iteration import Carry as Carry\nfrom .transforms.iteration import scan as scan\nfrom .transforms.iteration import vmap as vmap\nfrom .transforms.iteration import pmap as pmap\nfrom .transforms.transforms import eval_shape as eval_shape\nfrom .transforms.transforms import cond as cond\nfrom .transforms.transforms import switch as switch\nfrom .transforms.transforms import checkify as checkify\nfrom .transforms.iteration import while_loop as while_loop\nfrom .transforms.iteration import fori_loop as fori_loop\nfrom .transforms.iteration import StateAxes as StateAxes\nfrom .variablelib import A as A\nfrom .variablelib import BatchStat as BatchStat\nfrom .variablelib import Cache as Cache\nfrom .variablelib import Intermediate as Intermediate\nfrom .variablelib import Perturbation as Perturbation\nfrom .variablelib import Variable as Variable\nfrom .variablelib import VariableState as VariableState\nfrom .variablelib import VariableMetadata as VariableMetadata\nfrom .variablelib import with_metadata as with_metadata\nfrom .variablelib import variable_type_from_name as variable_type_from_name\nfrom .variablelib import variable_name_from_type as variable_name_from_type\nfrom .variablelib import register_variable_name as register_variable_name\nfrom .visualization import display as display\nfrom .extract import to_tree as to_tree\nfrom .extract import from_tree as from_tree\nfrom .extract import NodeStates as NodeStates\nfrom .summary import tabulate as tabulate\nfrom . import traversals as traversals\n",python,tab +1821,13314151,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",2299,0,"",python,selection_command +1822,13315357,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",2317,0,"",python,selection_command +1823,13315605,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",7481,0,"",python,selection_command +1824,13317181,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",7487,0,"",python,selection_command +1825,13317305,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",7490,0,"",python,selection_command +1826,13317754,".venv/lib/python3.10/site-packages/flax/nnx/variablelib.py",0,0,"# Copyright 2024 The Flax Authors.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# pytype: skip-file\nfrom __future__ import annotations\n\nimport dataclasses\nimport functools\nfrom functools import partial\nimport typing as tp\nfrom typing import Any\n\nimport jax\nimport treescope # type: ignore[import-untyped]\n\nfrom flax import errors\nfrom flax.nnx import filterlib, reprlib, tracers, visualization\nfrom flax.typing import Missing, PathParts, SizeBytes\nimport jax.tree_util as jtu\n\nA = tp.TypeVar('A')\nB = tp.TypeVar('B')\nF = tp.TypeVar('F', bound=tp.Callable[..., tp.Any])\nV = tp.TypeVar('V', bound='Variable[Any]')\nGetValueHook = tp.Callable[['Variable[A]', A], A]\nSetValueHook = tp.Callable[['Variable[A]', A], A]\nCreateValueHook = tp.Callable[['Variable[A]', A], A]\nAxisName = str\nAxisIndex = int\nAddAxisHook = tp.Callable[[V, AxisIndex, AxisName | None], None]\nRemoveAxisHook = tp.Callable[[V, AxisIndex, AxisName | None], None]\n\n\n@dataclasses.dataclass\nclass VariableMetadata(tp.Generic[A]):\n raw_value: A\n set_value_hooks: tuple[SetValueHook[A], ...] = ()\n get_value_hooks: tuple[GetValueHook[A], ...] = ()\n create_value_hooks: tuple[CreateValueHook[A], ...] = ()\n add_axis_hooks: tuple[AddAxisHook[Variable[A]], ...] = ()\n remove_axis_hooks: tuple[RemoveAxisHook[Variable[A]], ...] = ()\n metadata: tp.Mapping[str, tp.Any] = dataclasses.field(default_factory=dict)\n\n\nclass Variable(tp.Generic[A], reprlib.Representable):\n """"""The base class for all ``Variable`` types. Create custom ``Variable``\n types by subclassing this class. Numerous NNX graph functions can filter\n for specific ``Variable`` types, for example, :func:`split`, :func:`state`,\n :func:`pop`, and :func:`State.filter`.\n\n Example usage::\n\n >>> from flax import nnx\n >>> import jax, jax.numpy as jnp\n\n >>> class CustomVariable(nnx.Variable):\n ... pass\n\n >>> class Model(nnx.Module):\n ... def __init__(self, rngs):\n ... self.linear = nnx.Linear(2, 3, rngs=rngs)\n ... self.custom_variable = CustomVariable(jnp.ones((1, 3)))\n ... def __call__(self, x):\n ... return self.linear(x) + self.custom_variable\n >>> model = Model(rngs=nnx.Rngs(0))\n\n >>> linear_variables = nnx.state(model, nnx.Param)\n >>> jax.tree.map(jnp.shape, linear_variables)\n State({\n 'linear': {\n 'bias': VariableState(\n type=Param,\n value=(3,)\n ),\n 'kernel': VariableState(\n type=Param,\n value=(2, 3)\n )\n }\n })\n\n >>> custom_variable = nnx.state(model, CustomVariable)\n >>> jax.tree.map(jnp.shape, custom_variable)\n State({\n 'custom_variable': VariableState(\n type=CustomVariable,\n value=(1, 3)\n )\n })\n\n >>> variables = nnx.state(model)\n >>> jax.tree.map(jnp.shape, variables)\n State({\n 'custom_variable': VariableState(\n type=CustomVariable,\n value=(1, 3)\n ),\n 'linear': {\n 'bias': VariableState(\n type=Param,\n value=(3,)\n ),\n 'kernel': VariableState(\n type=Param,\n value=(2, 3)\n )\n }\n })\n """"""\n\n __slots__ = ('raw_value', '_trace_state', '_var_metadata')\n\n raw_value: A\n _trace_state: tracers.TraceState\n _var_metadata: dict[str, tp.Any]\n\n def __init__(\n self,\n value: tp.Union[A, VariableMetadata[A]],\n **metadata: tp.Any,\n ):\n var_t = type(self)\n object.__setattr__(self, '_trace_state', tracers.TraceState())\n\n if isinstance(value, VariableMetadata):\n metadata.update(value.metadata)\n value = tp.cast(A, value.raw_value)\n\n object.__setattr__(self, 'raw_value', value)\n\n if hasattr(var_t, 'on_get_value') and 'on_get_value' not in metadata:\n metadata['on_get_value'] = var_t.on_get_value\n\n if hasattr(var_t, 'on_set_value') and 'on_set_value' not in metadata:\n metadata['on_set_value'] = var_t.on_set_value\n\n if hasattr(var_t, 'on_create_value') and 'on_create_value' not in metadata:\n metadata['on_create_value'] = var_t.on_create_value\n\n if hasattr(var_t, 'on_add_axis') and 'on_add_axis' not in metadata:\n metadata['on_add_axis'] = var_t.on_add_axis\n\n if hasattr(var_t, 'on_remove_axis') and 'on_remove_axis' not in metadata:\n metadata['on_remove_axis'] = var_t.on_remove_axis\n\n object.__setattr__(self, '_var_metadata', metadata)\n # run create_value hooks\n object.__setattr__(self, 'raw_value', self.create_value(self.raw_value))\n\n def __getattr__(self, name: str) -> tp.Any:\n if name in object.__getattribute__(self, '_var_metadata'):\n return self._var_metadata[name]\n return getattr(self.value, name)\n\n def __setattr__(self, name: str, value: tp.Any):\n if not self._trace_state.is_valid():\n raise errors.TraceContextError(\n f'Cannot mutate {type(self).__name__} from a different trace level'\n )\n\n if (\n name == 'value'\n or name == 'raw_value'\n or name == '_var_metadata'\n or name == '_trace_state'\n ):\n object.__setattr__(self, name, value)\n else:\n self._var_metadata[name] = value\n\n def __delattr__(self, name: str):\n if not self._trace_state.is_valid():\n raise errors.TraceContextError(\n f'Cannot mutate {type(self).__name__} from a different trace level'\n )\n\n if (\n name == 'value'\n or name == 'raw_value'\n or name == '_var_metadata'\n or name == '_trace_state'\n ):\n object.__delattr__(self, name)\n else:\n del self._var_metadata[name]\n\n @classmethod\n def state(cls, value: A, **metadata) -> VariableState[A]:\n return cls(value, **metadata).to_state()\n\n def get_metadata(self):\n return self._var_metadata\n\n def copy_from(self, other: Variable[A]) -> None:\n if type(self) is not type(other):\n raise ValueError(\n f'Cannot copy from incompatible container, '\n f'expected {type(self).__name__}, got {type(other).__name__}'\n )\n if self is other:\n return\n self.raw_value = other.raw_value\n self._var_metadata.clear()\n self._var_metadata.update(other.get_metadata())\n\n def update_from_state(self, variable_state: VariableState[A]):\n object.__setattr__(self, 'raw_value', variable_state.value)\n object.__setattr__(\n self, '_var_metadata', variable_state._var_metadata.copy()\n )\n\n @property\n def value(self) -> A:\n value = self.raw_value\n if 'on_get_value' in self._var_metadata:\n value = self._var_metadata['on_get_value'](self, value)\n return value\n\n @value.setter\n def value(self, value: A):\n if isinstance(value, Variable):\n raise ValueError(\n 'Cannot set value to a Variable, ' 'use `copy_from` method instead'\n )\n if 'on_set_value' in self._var_metadata:\n value = self._var_metadata['on_set_value'](self, value)\n object.__setattr__(self, 'raw_value', value)\n\n def create_value(self, value: A):\n if 'on_create_value' in self._var_metadata:\n value = self._var_metadata['on_create_value'](self, value)\n return value\n\n def add_axis(self, axis_index: AxisIndex, axis_name: AxisName | None):\n if 'on_add_axis' in self._var_metadata:\n self._var_metadata['on_add_axis'](self, axis_index, axis_name)\n\n def remove_axis(self, axis_index: AxisIndex, axis_name: AxisName | None):\n if 'on_remove_axis' in self._var_metadata:\n self._var_metadata['on_remove_axis'](self, axis_index, axis_name)\n\n @tp.overload\n def replace(self, value: B, **kwargs) -> Variable[B]:\n ...\n\n @tp.overload\n def replace(self, **kwargs) -> Variable[A]:\n ...\n\n def replace(self, value: tp.Any = Missing, **kwargs) -> Variable[tp.Any]:\n if value is not Missing:\n kwargs['raw_value'] = value\n\n # rename `value` to `raw_value`\n if 'value' in kwargs:\n kwargs['raw_value'] = kwargs.pop('value')\n\n # return `value` if it is a Variable\n if 'raw_value' in kwargs and isinstance(\n value := kwargs['raw_value'], Variable\n ):\n # remove value from kwargs\n kwargs.pop('raw_value')\n if type(self) is not type(value):\n raise ValueError(\n 'Cannot replace value from incompatible container, '\n f'expected {type(self).__name__}, got {type(value).__name__}'\n )\n # if kwargs aren't empty, recursively call replace\n # else return variable value\n if kwargs:\n return value.replace(**kwargs)\n else:\n return value\n\n # get and update attributes\n # return new instance with updated attributes\n obj = object.__new__(type(self))\n object.__setattr__(obj, '_trace_state', self._trace_state)\n object.__setattr__(obj, 'raw_value', kwargs.pop('raw_value'))\n object.__setattr__(obj, '_var_metadata', self.get_metadata() | kwargs)\n return obj\n\n @classmethod\n def from_metadata(cls, value: A, attributes: dict[str, tp.Any]):\n obj = object.__new__(cls)\n object.__setattr__(obj, '_trace_state', tracers.TraceState())\n object.__setattr__(obj, 'raw_value', value)\n object.__setattr__(obj, '_var_metadata', attributes)\n return obj\n\n def copy(self: Variable[A]) -> Variable[A]:\n obj = object.__new__(type(self))\n object.__setattr__(obj, '_trace_state', self._trace_state)\n object.__setattr__(obj, 'raw_value', self.raw_value)\n object.__setattr__(obj, '_var_metadata', self.get_metadata().copy())\n return obj\n\n def to_state(self: Variable[A]) -> VariableState[A]:\n return VariableState(type(self), self.raw_value, **self._var_metadata)\n\n def __nnx_repr__(self):\n stats = SizeBytes.from_any(self.value)\n if stats:\n comment = f' # {stats}'\n else:\n comment = ''\n\n yield reprlib.Object(type=type(self).__name__, comment=comment)\n yield reprlib.Attr('value', self.raw_value)\n for name, value in self._var_metadata.items():\n yield reprlib.Attr(name, repr(value))\n\n def __treescope_repr__(self, path, subtree_renderer):\n size_bytes = SizeBytes.from_any(self.value)\n if size_bytes:\n stats_repr = f' # {size_bytes}'\n first_line_annotation = treescope.rendering_parts.comment_color(\n treescope.rendering_parts.text(f'{stats_repr}')\n )\n else:\n first_line_annotation = None\n\n children = {'value': self.raw_value, **self._var_metadata}\n return visualization.render_object_constructor(\n object_type=type(self),\n attributes=children,\n path=path,\n subtree_renderer=subtree_renderer,\n first_line_annotation=first_line_annotation,\n )\n\n # hooks API\n if tp.TYPE_CHECKING:\n\n def on_get_value(self, value: A) -> A: ...\n\n def on_set_value(self, value: A) -> A: ...\n\n def on_create_value(self, value: A) -> A: ...\n\n def on_add_axis(\n self: V, axis_index: AxisIndex, axis_name: AxisName | None\n ) -> V: ...\n\n def on_remove_axis(\n self: V, axis_index: AxisIndex, axis_name: AxisName | None\n ) -> V: ...\n\n def __jax_array__(self):\n return self.value\n\n # pickle support\n def __getstate__(self):\n return {\n 'raw_value': self.raw_value,\n '_trace_state': self._trace_state,\n '_var_metadata': self._var_metadata,\n }\n\n def __setstate__(self, state):\n object.__setattr__(self, 'raw_value', state['raw_value'])\n object.__setattr__(self, '_trace_state', state['_trace_state'])\n object.__setattr__(self, '_var_metadata', state['_var_metadata'])\n\n # --------------------------------------------\n # proxy methods\n # --------------------------------------------\n\n def __getitem__(self, key) -> tp.Any:\n return self.value[key] # type: ignore\n\n def __setitem__(self, key, value) -> None:\n self.value[key] = value # type: ignore\n\n def __call__(self, *args, **kwargs) -> tp.Any:\n return self.value(*args, **kwargs) # type: ignore\n\n def __len__(self) -> int:\n return len(self.value) # type: ignore\n\n def __iter__(self) -> tp.Iterator:\n return iter(self.value) # type: ignore\n\n def __contains__(self, item) -> bool:\n return item in self.value # type: ignore\n\n def __add__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__add__(other) # type: ignore\n\n def __sub__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__sub__(other) # type: ignore\n\n def __mul__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__mul__(other) # type: ignore\n\n def __matmul__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__matmul__(other) # type: ignore\n\n def __truediv__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__truediv__(other) # type: ignore\n\n def __floordiv__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__floordiv__(other) # type: ignore\n\n def __mod__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__mod__(other) # type: ignore\n\n def __divmod__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__divmod__(other) # type: ignore\n\n def __pow__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__pow__(other) # type: ignore\n\n def __lshift__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__lshift__(other) # type: ignore\n\n def __rshift__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rshift__(other) # type: ignore\n\n def __and__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__and__(other) # type: ignore\n\n def __xor__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__xor__(other) # type: ignore\n\n def __or__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__or__(other) # type: ignore\n\n def __radd__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__radd__(other) # type: ignore\n\n def __rsub__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rsub__(other) # type: ignore\n\n def __rmul__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rmul__(other) # type: ignore\n\n def __rmatmul__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rmatmul__(other) # type: ignore\n\n def __rtruediv__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rtruediv__(other) # type: ignore\n\n def __rfloordiv__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rfloordiv__(other) # type: ignore\n\n def __rmod__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rmod__(other) # type: ignore\n\n def __rdivmod__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rdivmod__(other) # type: ignore\n\n def __rpow__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rpow__(other) # type: ignore\n\n def __rlshift__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rlshift__(other) # type: ignore\n\n def __rrshift__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rrshift__(other) # type: ignore\n\n def __rand__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rand__(other) # type: ignore\n\n def __rxor__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__rxor__(other) # type: ignore\n\n def __ror__(self, other) -> A:\n if isinstance(other, Variable):\n other = other.value\n return self.value.__ror__(other) # type: ignore\n\n def __iadd__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__iadd__'):\n value.__iadd__(other)\n else:\n self.value = value.__add__(other)\n return self\n\n def __isub__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__isub__'):\n value.__isub__(other)\n else:\n self.value = value.__sub__(other)\n return self\n\n def __imul__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__imul__'):\n value.__imul__(other)\n else:\n self.value = value.__mul__(other)\n return self\n\n def __imatmul__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__imatmul__'):\n value.__imatmul__(other)\n else:\n self.value = value.__matmul__(other)\n return self\n\n def __itruediv__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__itruediv__'):\n value.__itruediv__(other)\n else:\n self.value = value.__truediv__(other)\n return self\n\n def __ifloordiv__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__ifloordiv__'):\n value.__ifloordiv__(other)\n else:\n self.value = value.__floordiv__(other)\n return self\n\n def __imod__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__imod__'):\n value.__imod__(other)\n else:\n self.value = value.__mod__(other)\n return self\n\n def __ipow__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__ipow__'):\n value.__ipow__(other)\n else:\n self.value = value.__pow__(other)\n return self\n\n def __ilshift__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__ilshift__'):\n value.__ilshift__(other)\n else:\n self.value = value.__lshift__(other)\n return self\n\n def __irshift__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__irshift__'):\n value.__irshift__(other)\n else:\n self.value = value.__rshift__(other)\n return self\n\n def __iand__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__iand__'):\n value.__iand__(other)\n else:\n self.value = value.__and__(other)\n return self\n\n def __ixor__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__ixor__'):\n value.__ixor__(other)\n else:\n self.value = value.__xor__(other)\n return self\n\n def __ior__(self: V, other) -> V:\n if isinstance(other, Variable):\n other = other.value\n value = self.value\n if hasattr(value, '__ior__'):\n value.__ior__(other)\n else:\n self.value = value.__or__(other)\n return self\n\n def __neg__(self) -> A:\n return self.value.__neg__() # type: ignore\n\n def __pos__(self) -> A:\n return self.value.__pos__() # type: ignore\n\n def __abs__(self) -> A:\n return self.value.__abs__() # type: ignore\n\n def __invert__(self) -> A:\n return self.value.__invert__() # type: ignore\n\n def __complex__(self) -> A:\n return self.value.__complex__() # type: ignore\n\n def __int__(self) -> A:\n return self.value.__int__() # type: ignore\n\n def __float__(self) -> A:\n return self.value.__float__() # type: ignore\n\n def __index__(self) -> A:\n return self.value.__index__() # type: ignore\n\n def __round__(self, ndigits: int) -> A:\n return self.value.__round__(ndigits) # type: ignore\n\n def __trunc__(self) -> A:\n return self.value.__trunc__() # type: ignore\n\n def __floor__(self) -> A:\n return self.value.__floor__() # type: ignore\n\n def __ceil__(self) -> A:\n return self.value.__ceil__() # type: ignore\n\n\nclass Param(Variable[A]):\n """"""The canonical learnable parameter. All learnable parameters\n in NNX layer modules will have the ``Param`` :class:`Variable`\n type::\n\n >>> from flax import nnx\n >>> import jax, jax.numpy as jnp\n\n >>> layer = nnx.Linear(2, 3, rngs=nnx.Rngs(0))\n >>> jax.tree.map(jnp.shape, nnx.state(layer))\n State({\n 'bias': VariableState(\n type=Param,\n value=(3,)\n ),\n 'kernel': VariableState(\n type=Param,\n value=(2, 3)\n )\n })\n """"""\n\n pass\n\n\nclass BatchStat(Variable[A]):\n """"""The mean and variance batch statistics stored in\n the :class:`BatchNorm` layer. Note, these are not the\n learnable scale and bias parameters, but rather the\n running average statistics that are typically used\n during post-training inference::\n\n >>> from flax import nnx\n >>> import jax, jax.numpy as jnp\n\n >>> layer = nnx.BatchNorm(3, rngs=nnx.Rngs(0))\n >>> jax.tree.map(jnp.shape, nnx.state(layer))\n State({\n 'bias': VariableState(\n type=Param,\n value=(3,)\n ),\n 'mean': VariableState(\n type=BatchStat,\n value=(3,)\n ),\n 'scale': VariableState(\n type=Param,\n value=(3,)\n ),\n 'var': VariableState(\n type=BatchStat,\n value=(3,)\n )\n })\n """"""\n\n pass\n\n\nclass Cache(Variable[A]):\n """"""Autoregressive cache in :class:`MultiHeadAttention`::\n\n >>> from flax import nnx\n >>> import jax, jax.numpy as jnp\n\n >>> layer = nnx.MultiHeadAttention(\n ... num_heads=2,\n ... in_features=3,\n ... qkv_features=6,\n ... out_features=6,\n ... decode=True,\n ... rngs=nnx.Rngs(0),\n ... )\n\n >>> layer.init_cache((1, 3))\n >>> jax.tree.map(jnp.shape, nnx.state(layer, nnx.Cache))\n State({\n 'cache_index': VariableState(\n type=Cache,\n value=()\n ),\n 'cached_key': VariableState(\n type=Cache,\n value=(1, 2, 3)\n ),\n 'cached_value': VariableState(\n type=Cache,\n value=(1, 2, 3)\n )\n })\n """"""\n\n pass\n\n\nclass Intermediate(Variable[A]):\n """""":class:`Variable` type that is typically used for\n :func:`Module.sow`::\n\n >>> from flax import nnx\n >>> import jax, jax.numpy as jnp\n\n >>> class Model(nnx.Module):\n ... def __init__(self, rngs):\n ... self.linear1 = nnx.Linear(2, 3, rngs=rngs)\n ... self.linear2 = nnx.Linear(3, 4, rngs=rngs)\n ... def __call__(self, x):\n ... x = self.linear1(x)\n ... self.sow(nnx.Intermediate, 'i', x)\n ... x = self.linear2(x)\n ... return x\n >>> model = Model(rngs=nnx.Rngs(0))\n\n >>> x = jnp.ones((1, 2))\n >>> y = model(x)\n >>> jax.tree.map(jnp.shape, nnx.state(model, nnx.Intermediate))\n State({\n 'i': VariableState(\n type=Intermediate,\n value=((1, 3),)\n )\n })\n """"""\n\n pass\n\n\nclass Perturbation(Intermediate[A]):\n """""":class:`Variable` type that is typically used for\n :func:`Module.perturb`::\n\n >>> from flax import nnx\n >>> import jax, jax.numpy as jnp\n\n >>> class Model(nnx.Module):\n ... def __init__(self, rngs):\n ... self.linear1 = nnx.Linear(2, 3, rngs=rngs)\n ... self.linear2 = nnx.Linear(3, 4, rngs=rngs)\n ... def __call__(self, x):\n ... x = self.linear1(x)\n ... x = self.perturb('i', x)\n ... x = self.linear2(x)\n ... return x\n >>> model = Model(rngs=nnx.Rngs(0))\n\n >>> x = jnp.ones((1, 2))\n >>> y = model(x)\n >>> jax.tree.map(jnp.shape, nnx.state(model, nnx.Perturbation))\n State({\n 'i': VariableState(\n type=Perturbation,\n value=(1, 3)\n )\n })\n """"""\n\n pass\n\n\nclass VariableState(tp.Generic[A], reprlib.Representable):\n __slots__ = ('type', 'value', '_var_metadata')\n type: type[Variable[A]]\n value: A\n _var_metadata: dict[str, tp.Any]\n\n def __init__(\n self,\n type: type[Variable[A]], # type: ignore [valid-type]\n value: A,\n **metadata,\n ):\n object.__setattr__(self, 'type', type)\n object.__setattr__(self, 'value', value)\n object.__setattr__(self, '_var_metadata', metadata)\n\n def __getattr__(self, name: str) -> None:\n var_metadata = object.__getattribute__(self, '_var_metadata')\n if name not in var_metadata:\n raise AttributeError(f""'VariableState' object has no attribute '{name}'"")\n return var_metadata[name]\n\n def __setattr__(self, name: str, value: Any) -> None:\n if name == 'type' or name == 'value' or name == '_var_metadata':\n object.__setattr__(self, name, value)\n else:\n self._var_metadata[name] = value\n\n def __delattr__(self, name: str) -> None:\n if name == 'type' or name == 'value' or name == '_var_metadata':\n object.__delattr__(self, name)\n else:\n del self._var_metadata[name]\n\n def __nnx_repr__(self):\n stats = SizeBytes.from_any(self.value)\n if stats:\n comment = f' # {stats}'\n else:\n comment = ''\n\n yield reprlib.Object(type=type(self), comment=comment)\n yield reprlib.Attr('type', self.type)\n yield reprlib.Attr('value', self.value)\n\n for name, value in self._var_metadata.items():\n yield reprlib.Attr(name, value)\n\n def __treescope_repr__(self, path, subtree_renderer):\n size_bytes = SizeBytes.from_any(self.value)\n if size_bytes:\n stats_repr = f' # {size_bytes}'\n first_line_annotation = treescope.rendering_parts.comment_color(\n treescope.rendering_parts.text(f'{stats_repr}')\n )\n else:\n first_line_annotation = None\n children = {'type': self.type, 'value': self.value, **self._var_metadata}\n return visualization.render_object_constructor(\n object_type=type(self),\n attributes=children,\n path=path,\n subtree_renderer=subtree_renderer,\n first_line_annotation=first_line_annotation,\n )\n\n def replace(self, value: B) -> VariableState[B]:\n return VariableState(self.type, value, **self.get_metadata())\n\n def to_variable(self) -> Variable[A]:\n # we use object.__new__ to avoid calling __init__ and bypass the\n # __init__ logic which should not be called twice\n variable = object.__new__(self.type)\n object.__setattr__(variable, '_trace_state', tracers.TraceState())\n object.__setattr__(variable, 'raw_value', self.value)\n object.__setattr__(variable, '_var_metadata', self.get_metadata().copy())\n return variable\n\n def copy(self: VariableState[A]) -> VariableState[A]:\n return jax.tree.map(lambda x: x, self)\n\n def get_metadata(self) -> dict[str, tp.Any]:\n return self._var_metadata\n\n def add_axis(self, axis_index: AxisIndex, axis_name: AxisName | None):\n if 'on_add_axis' in self._var_metadata:\n self._var_metadata['on_add_axis'](self, axis_index, axis_name)\n\n def remove_axis(self, axis_index: AxisIndex, axis_name: AxisName | None):\n if 'on_remove_axis' in self._var_metadata:\n self._var_metadata['on_remove_axis'](self, axis_index, axis_name)\n\nGraphVariableState = VariableState[VariableState[tp.Any]]\n\ndef _variable_state_flatten(x: VariableState[tp.Any], *, with_keys: bool):\n metadata = tuple(x.get_metadata().items())\n if with_keys:\n node = (jtu.GetAttrKey('value'), x.value)\n else:\n node = x.value\n\n return (node,), (x.type, metadata)\n\n\ndef _variable_state_unflatten(\n static: tuple[type[Variable[A]], tuple[tuple[str, tp.Any], ...]],\n children: tuple[A],\n) -> VariableState[A]:\n return VariableState(\n type=static[0],\n value=children[0],\n **dict(static[1]),\n )\n\n\njtu.register_pytree_with_keys(\n VariableState,\n partial(_variable_state_flatten, with_keys=True), # type: ignore\n _variable_state_unflatten, # type: ignore\n flatten_func=partial(_variable_state_flatten, with_keys=False), # type: ignore\n)\n\n\ndef with_metadata(\n initializer: F,\n set_value_hooks: tp.Union[SetValueHook[A], tp.Sequence[SetValueHook[A]]] = (),\n get_value_hooks: tp.Union[SetValueHook[A], tp.Sequence[SetValueHook[A]]] = (),\n create_value_hooks: tp.Union[\n CreateValueHook[A], tp.Sequence[CreateValueHook[A]]\n ] = (),\n add_axis_hooks: tp.Union[\n AddAxisHook[Variable[A]], tp.Sequence[AddAxisHook[Variable[A]]]\n ] = (),\n remove_axis_hooks: tp.Union[\n RemoveAxisHook[Variable[A]],\n tp.Sequence[RemoveAxisHook[Variable[A]]],\n ] = (),\n **metadata: tp.Any,\n) -> F:\n if set_value_hooks:\n if callable(set_value_hooks):\n set_value_hooks = (set_value_hooks,)\n else:\n set_value_hooks = tuple(set_value_hooks)\n else:\n set_value_hooks = ()\n\n if get_value_hooks:\n if callable(get_value_hooks):\n get_value_hooks = (get_value_hooks,)\n else:\n get_value_hooks = tuple(get_value_hooks)\n else:\n get_value_hooks = ()\n\n if create_value_hooks:\n if callable(create_value_hooks):\n create_value_hooks = (create_value_hooks,)\n else:\n create_value_hooks = tuple(create_value_hooks)\n else:\n create_value_hooks = ()\n\n if add_axis_hooks:\n if callable(add_axis_hooks):\n add_axis_hooks = (add_axis_hooks,)\n else:\n add_axis_hooks = tuple(add_axis_hooks)\n else:\n add_axis_hooks = ()\n\n if remove_axis_hooks:\n if callable(remove_axis_hooks):\n remove_axis_hooks = (remove_axis_hooks,)\n else:\n remove_axis_hooks = tuple(remove_axis_hooks)\n else:\n remove_axis_hooks = ()\n\n @functools.wraps(initializer)\n def wrapper(*args):\n return VariableMetadata(\n initializer(*args),\n set_value_hooks=set_value_hooks,\n get_value_hooks=get_value_hooks,\n create_value_hooks=create_value_hooks,\n add_axis_hooks=add_axis_hooks,\n remove_axis_hooks=remove_axis_hooks,\n metadata=metadata,\n )\n\n return wrapper # type: ignore\n\n\ndef split_flat_state(\n flat_state: tp.Iterable[tuple[PathParts, Variable | VariableState]],\n filters: tuple[filterlib.Filter, ...],\n) -> tuple[list[tuple[PathParts, Variable | VariableState]], ...]:\n predicates = filterlib.filters_to_predicates(filters)\n # we have n + 1 states, where n is the number of predicates\n # the last state is for values that don't match any predicate\n flat_states: tuple[list[tuple[PathParts, Variable | VariableState]], ...] = (\n tuple([] for _ in predicates)\n )\n\n for path, value in flat_state:\n for i, predicate in enumerate(predicates):\n if predicate(path, value):\n flat_states[i].append((path, value))\n break\n else:\n raise ValueError(\n 'Non-exhaustive filters, got a non-empty remainder: '\n f'{path} -> {value}.'\n '\nUse `...` to match all remaining elements.'\n )\n\n return flat_states\n\n\n###################################################\n### Variable type/class <-> string name mapping ###\n###################################################\n# Assumption: the mapping is 1-1 and unique.\n\nVariableTypeCache: dict[str, tp.Type[Variable[tp.Any]]] = {}\n\n\ndef variable_type_from_name(\n name: str,\n /,\n *,\n base: type[Variable[tp.Any]] = Variable,\n allow_register: bool = False,\n) -> tp.Type[Variable[tp.Any]]:\n """"""Given a Linen-style collection name, get or create its NNX Variable class.""""""\n if name not in VariableTypeCache:\n if not allow_register:\n raise ValueError(\n f'Name {name} is not registered in the registry. '\n 'To register a new name, use register_variable_name() '\n 'or set allow_register=True.'\n )\n VariableTypeCache[name] = type(name, (base,), {})\n return VariableTypeCache[name]\n\n\ndef variable_name_from_type(\n typ: tp.Type[Variable[tp.Any]], /, *, allow_register: bool = False\n) -> str:\n """"""Given an NNX Variable type, get its Linen-style collection name.\n\n Should output the exact inversed result of `variable_type_from_name()`.""""""\n for name, t in VariableTypeCache.items():\n if typ == t:\n return name\n\n if not allow_register:\n raise ValueError(\n f'Type {typ} is not registered in the registry. '\n 'To register a new type, use register_variable_name() '\n 'or set allow_register=True.'\n )\n name = typ.__name__\n if name in VariableTypeCache:\n raise ValueError(\n 'Name {name} is already registered in the registry as {VariableTypeCache[name]}. '\n 'It cannot be linked with this type {typ}.'\n )\n register_variable_name(name, typ)\n return name\n\n\nclass _Missing:\n pass\n\n\n_MISSING = _Missing()\n\n\n@tp.overload\ndef register_variable_name(\n name: str,\n typ: type[Variable[tp.Any]],\n *,\n overwrite: bool = False,\n) -> type[Variable[tp.Any]]:\n ...\n\n\n@tp.overload\ndef register_variable_name(\n name: str,\n *,\n overwrite: bool = False,\n) -> tp.Callable[[type[Variable[tp.Any]]], type[Variable[tp.Any]]]:\n ...\n\n\ndef register_variable_name(\n name: str,\n typ: type[Variable[A]] | _Missing = _MISSING,\n *,\n overwrite=False,\n) -> type[Variable[A]] | tp.Callable[[type[Variable[A]]], type[Variable[A]]]:\n """"""Register a pair of Linen collection name and its NNX type.""""""\n if typ is _MISSING:\n return partial(register_variable_name, name, overwrite=overwrite)\n typ = tp.cast(type[Variable[A]], typ)\n if not overwrite and name in VariableTypeCache:\n raise ValueError(f'Name {name} already mapped to type {VariableTypeCache[name]}. '\n 'To overwrite, call set_variable_name() with `overwrite=True`.')\n VariableTypeCache[name] = typ\n return typ\n\n\n# add known variable type names\nregister_variable_name('params', Param)\nregister_variable_name('batch_stats', BatchStat)\nregister_variable_name('cache', Cache)\nregister_variable_name('intermediates', Intermediate)\nregister_variable_name('perturbations', Perturbation)\n",python,tab +1827,13317763,".venv/lib/python3.10/site-packages/flax/nnx/variablelib.py",22343,0,"",python,selection_command +1828,13325899,".venv/lib/python3.10/site-packages/flax/nnx/nn/attention.py",0,0,"# Copyright 2024 The Flax Authors.\n#\n# Licensed under the Apache License, Version 2.0 (the ""License"");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an ""AS IS"" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n""""""Attention core modules for Flax.""""""\n\nfrom __future__ import annotations\n\nimport functools\nfrom typing import Any\nfrom collections.abc import Callable\nimport math\n\nimport jax\nimport jax.numpy as jnp\nfrom jax import lax, random\n\nfrom flax import nnx\nfrom flax.nnx import rnglib\nfrom flax.nnx.module import Module, first_from\nfrom flax.nnx.nn import initializers\nfrom flax.nnx.nn import dtypes\nfrom flax.nnx.nn.linear import (\n LinearGeneral,\n default_kernel_init,\n)\nfrom flax.nnx.nn.normalization import LayerNorm\nfrom flax.typing import (\n Dtype,\n PromoteDtypeFn,\n Shape,\n Initializer,\n PrecisionLike,\n DotGeneralT,\n)\n\nArray = jax.Array\n\n\ndef dot_product_attention_weights(\n query: Array,\n key: Array,\n bias: Array | None = None,\n mask: Array | None = None,\n broadcast_dropout: bool = True,\n dropout_rng: Array | None = None,\n dropout_rate: float = 0.0,\n deterministic: bool = False,\n dtype: Dtype | None = None,\n precision: PrecisionLike = None,\n module: Module | None = None,\n promote_dtype: PromoteDtypeFn = dtypes.promote_dtype,\n):\n """"""Computes dot-product attention weights given query and key.\n\n Used by :func:`dot_product_attention`, which is what you'll most likely use.\n But if you want access to the attention weights for introspection, then\n you can directly call this function and call einsum yourself.\n\n Args:\n query: queries for calculating attention with shape of `[batch..., q_length,\n num_heads, qk_depth_per_head]`.\n key: keys for calculating attention with shape of `[batch..., kv_length,\n num_heads, qk_depth_per_head]`.\n bias: bias for the attention weights. This should be broadcastable to the\n shape `[batch..., num_heads, q_length, kv_length]`. This can be used for\n incorporating causal masks, padding masks, proximity bias, etc.\n mask: mask for the attention weights. This should be broadcastable to the\n shape `[batch..., num_heads, q_length, kv_length]`. This can be used for\n incorporating causal masks. Attention weights are masked out if their\n corresponding mask value is `False`.\n broadcast_dropout: bool: use a broadcasted dropout along batch dims.\n dropout_rng: JAX PRNGKey: to be used for dropout\n dropout_rate: dropout rate\n deterministic: bool, deterministic or not (to apply dropout)\n dtype: the dtype of the computation (default: infer from inputs and params)\n precision: numerical precision of the computation see `jax.lax.Precision`\n for details.\n module: the Module that will sow the attention weights into the\n ``nnx.Intermediate`` collection. If ``module`` is None, the attention\n weights will not be sowed.\n promote_dtype: function to promote the dtype of the arrays to the desired\n dtype. The function should accept a tuple of ``(query, key)`` and a ``dtype``\n keyword argument, and return a tuple of arrays with the promoted dtype.\n\n Returns:\n Output of shape `[batch..., num_heads, q_length, kv_length]`.\n """"""\n query, key = promote_dtype((query, key), dtype=dtype) # type: ignore[bad-unpacking]\n dtype = query.dtype\n\n assert query.ndim == key.ndim, 'q, k must have same rank.'\n assert query.shape[:-3] == key.shape[:-3], 'q, k batch dims must match.'\n assert query.shape[-2] == key.shape[-2], 'q, k num_heads must match.'\n assert query.shape[-1] == key.shape[-1], 'q, k depths must match.'\n\n # calculate attention matrix\n depth = query.shape[-1]\n query = query / jnp.sqrt(depth).astype(dtype)\n # attn weight shape is (batch..., num_heads, q_length, kv_length)\n attn_weights = jnp.einsum(\n '...qhd,...khd->...hqk', query, key, precision=precision\n )\n\n # apply attention bias: masking, dropout, proximity bias, etc.\n if bias is not None:\n attn_weights = attn_weights + bias\n # apply attention mask\n if mask is not None:\n big_neg = jnp.finfo(dtype).min\n attn_weights = jnp.where(mask, attn_weights, big_neg)\n\n # normalize the attention weights\n attn_weights = jax.nn.softmax(attn_weights).astype(dtype)\n\n if module:\n module.sow(nnx.Intermediate, 'attention_weights', attn_weights)\n\n # apply attention dropout\n if not deterministic and dropout_rate > 0.0:\n keep_prob = 1.0 - dropout_rate\n if broadcast_dropout:\n # dropout is broadcast across the batch + head dimensions\n dropout_shape = tuple([1] * (key.ndim - 2)) + attn_weights.shape[-2:]\n keep = random.bernoulli(dropout_rng, keep_prob, dropout_shape) # type: ignore\n else:\n keep = random.bernoulli(dropout_rng, keep_prob, attn_weights.shape) # type: ignore\n multiplier = keep.astype(dtype) / jnp.asarray(keep_prob, dtype=dtype)\n attn_weights = attn_weights * multiplier\n\n return attn_weights\n\n\ndef dot_product_attention(\n query: Array,\n key: Array,\n value: Array,\n bias: Array | None = None,\n mask: Array | None = None,\n broadcast_dropout: bool = True,\n dropout_rng: Array | None = None,\n dropout_rate: float = 0.0,\n deterministic: bool = False,\n dtype: Dtype | None = None,\n precision: PrecisionLike = None,\n module: Module | None = None,\n promote_dtype: PromoteDtypeFn = dtypes.promote_dtype,\n):\n """"""Computes dot-product attention given query, key, and value.\n\n This is the core function for applying attention based on\n https://arxiv.org/abs/1706.03762. It calculates the attention weights given\n query and key and combines the values using the attention weights.\n\n Will use the more optimized `jax.nn.dot_product_attention` if dropout is\n not activated and `module=None`.\n\n .. note::\n ``query``, ``key``, ``value`` needn't have any batch dimensions.\n\n Args:\n query: queries for calculating attention with shape of ``[batch..., q_length,\n num_heads, qk_depth_per_head]``.\n key: keys for calculating attention with shape of ``[batch..., kv_length,\n num_heads, qk_depth_per_head]``.\n value: values to be used in attention with shape of ``[batch..., kv_length,\n num_heads, v_depth_per_head]``.\n bias: bias for the attention weights. This should be broadcastable to the\n shape `[batch..., num_heads, q_length, kv_length]`. This can be used for\n incorporating causal masks, padding masks, proximity bias, etc.\n mask: mask for the attention weights. This should be broadcastable to the\n shape `[batch..., num_heads, q_length, kv_length]`. This can be used for\n incorporating causal masks. Attention weights are masked out if their\n corresponding mask value is `False`.\n broadcast_dropout: bool: use a broadcasted dropout along batch dims.\n dropout_rng: JAX PRNGKey: to be used for dropout\n dropout_rate: dropout rate\n deterministic: bool, deterministic or not (to apply dropout)\n dtype: the dtype of the computation (default: infer from inputs)\n precision: numerical precision of the computation see `jax.lax.Precision`\n for details.\n module: the Module that will sow the attention weights into the\n ``nnx.Intermediate`` collection. If ``module`` is None, the attention\n weights will not be sowed.\n promote_dtype: function to promote the dtype of the arrays to the desired\n dtype. The function should accept a tuple of ``(query, key, value)`` and a\n ``dtype`` keyword argument, and return a tuple of arrays with the promoted\n dtype.\n\n Returns:\n Output of shape `[batch..., q_length, num_heads, v_depth_per_head]`.\n """"""\n query, key, value = promote_dtype((query, key, value), dtype=dtype) # type: ignore[bad-unpacking]\n dtype = query.dtype\n assert key.ndim == query.ndim == value.ndim, 'q, k, v must have same rank.'\n assert (\n query.shape[:-3] == key.shape[:-3] == value.shape[:-3]\n ), 'q, k, v batch dims must match.'\n assert (\n query.shape[-2] == key.shape[-2] == value.shape[-2]\n ), 'q, k, v num_heads must match.'\n assert key.shape[-3] == value.shape[-3], 'k, v lengths must match.'\n\n # Criteria that invoke the more optimized dot product attention\n if dropout_rate == 0.0 and module == None:\n # make sure qkv batch are compressed to one dim\n query_shape = query.shape\n if len(query_shape) > 4:\n def reshape_4d(x):\n return jnp.reshape(x, (math.prod(x.shape[:-3]), *x.shape[-3:]))\n query, key, value, bias, mask = jax.tree.map(\n reshape_4d, (query, key, value, bias, mask))\n if mask is not None:\n mask = mask.astype(jnp.bool)\n out = jax.nn.dot_product_attention(query, key, value, bias, mask)\n if len(query_shape) > 4:\n out = jnp.reshape(out, query_shape)\n return out\n\n # compute attention weights\n attn_weights = dot_product_attention_weights(\n query,\n key,\n bias,\n mask,\n broadcast_dropout,\n dropout_rng,\n dropout_rate,\n deterministic,\n dtype,\n precision,\n module,\n )\n\n # return weighted sum over values for each query position\n return jnp.einsum(\n '...hqk,...khd->...qhd', attn_weights, value, precision=precision\n )\n\n\nclass MultiHeadAttention(Module):\n """"""Multi-head attention.\n\n Example usage::\n\n >>> from flax import nnx\n >>> import jax\n\n >>> layer = nnx.MultiHeadAttention(num_heads=8, in_features=5, qkv_features=16,\n ... decode=False, rngs=nnx.Rngs(0))\n >>> key1, key2, key3 = jax.random.split(jax.random.key(0), 3)\n >>> shape = (4, 3, 2, 5)\n >>> q, k, v = (\n ... jax.random.uniform(key1, shape),\n ... jax.random.uniform(key2, shape),\n ... jax.random.uniform(key3, shape),\n ... )\n\n >>> # different inputs for inputs_q, inputs_k and inputs_v\n >>> out = layer(q, k, v)\n >>> # equivalent output when inferring v\n >>> assert (layer(q, k) == layer(q, k, k)).all()\n >>> # equivalent output when inferring k and v\n >>> assert (layer(q) == layer(q, q)).all()\n >>> assert (layer(q) == layer(q, q, q)).all()\n\n Args:\n num_heads: number of attention heads. Features (i.e. inputs_q.shape[-1])\n should be divisible by the number of heads.\n in_features: int or tuple with number of input features.\n qkv_features: dimension of the key, query, and value.\n out_features: dimension of the last projection\n dtype: the dtype of the computation (default: infer from inputs and params)\n param_dtype: the dtype passed to parameter initializers (default: float32)\n broadcast_dropout: bool: use a broadcasted dropout along batch dims.\n dropout_rate: dropout rate\n deterministic: if false, the attention weight is masked randomly using\n dropout, whereas if true, the attention weights are deterministic.\n precision: numerical precision of the computation see `jax.lax.Precision`\n for details.\n kernel_init: initializer for the kernel of the Dense layers.\n out_kernel_init: optional initializer for the kernel of the output Dense layer,\n if None, the kernel_init is used.\n bias_init: initializer for the bias of the Dense layers.\n out_bias_init: optional initializer for the bias of the output Dense layer,\n if None, the bias_init is used.\n use_bias: bool: whether pointwise QKVO dense transforms use bias.\n attention_fn: dot_product_attention or compatible function. Accepts query,\n key, value, and returns output of shape `[bs, dim1, dim2, ..., dimN,,\n num_heads, value_channels]``\n decode: whether to prepare and use an autoregressive cache.\n normalize_qk: should QK normalization be applied (arxiv.org/abs/2302.05442).\n rngs: rng key.\n """"""\n\n def __init__(\n self,\n num_heads: int,\n in_features: int,\n qkv_features: int | None = None,\n out_features: int | None = None,\n *,\n dtype: Dtype | None = None,\n param_dtype: Dtype = jnp.float32,\n broadcast_dropout: bool = True,\n dropout_rate: float = 0.0,\n deterministic: bool | None = None,\n precision: PrecisionLike = None,\n kernel_init: Initializer = default_kernel_init,\n out_kernel_init: Initializer | None = None,\n bias_init: Initializer = initializers.zeros_init(),\n out_bias_init: Initializer | None = None,\n use_bias: bool = True,\n attention_fn: Callable[..., Array] = dot_product_attention,\n decode: bool | None = None,\n normalize_qk: bool = False,\n # Deprecated, will be removed.\n qkv_dot_general: DotGeneralT | None = None,\n out_dot_general: DotGeneralT | None = None,\n qkv_dot_general_cls: Any = None,\n out_dot_general_cls: Any = None,\n rngs: rnglib.Rngs,\n ):\n self.num_heads = num_heads\n self.in_features = in_features\n self.qkv_features = (\n qkv_features if qkv_features is not None else in_features\n )\n self.out_features = (\n out_features if out_features is not None else in_features\n )\n self.dtype = dtype\n self.param_dtype = param_dtype\n self.broadcast_dropout = broadcast_dropout\n self.dropout_rate = dropout_rate\n self.deterministic = deterministic\n self.precision = precision\n self.kernel_init = kernel_init\n self.out_kernel_init = out_kernel_init\n self.bias_init = bias_init\n self.out_bias_init = out_bias_init\n self.use_bias = use_bias\n self.attention_fn = attention_fn\n self.decode = decode\n self.normalize_qk = normalize_qk\n self.qkv_dot_general = qkv_dot_general\n self.out_dot_general = out_dot_general\n self.qkv_dot_general_cls = qkv_dot_general_cls\n self.out_dot_general_cls = out_dot_general_cls\n\n if self.qkv_features % self.num_heads != 0:\n raise ValueError(\n f'Memory dimension ({self.qkv_features}) must be divisible by '\n f""'num_heads' heads ({self.num_heads}).""\n )\n\n self.head_dim = self.qkv_features // self.num_heads\n\n linear_general = functools.partial(\n LinearGeneral,\n in_features=self.in_features,\n out_features=(self.num_heads, self.head_dim),\n dtype=self.dtype,\n param_dtype=self.param_dtype,\n kernel_init=self.kernel_init,\n bias_init=self.bias_init,\n use_bias=self.use_bias,\n precision=self.precision,\n dot_general=self.qkv_dot_general,\n dot_general_cls=self.qkv_dot_general_cls,\n )\n # project inputs_q to multi-headed q/k/v\n # dimensions are then [batch..., length, n_heads, n_features_per_head]\n self.query = linear_general(rngs=rngs)\n self.key = linear_general(rngs=rngs)\n self.value = linear_general(rngs=rngs)\n\n self.query_ln: LayerNorm | None\n self.key_ln: LayerNorm | None\n if self.normalize_qk:\n # Normalizing query and key projections stabilizes training with higher\n # LR. See ViT-22B paper http://arxiv.org/abs/2302.05442 for analysis.\n self.query_ln = LayerNorm(\n self.head_dim,\n use_bias=False,\n dtype=self.dtype,\n param_dtype=self.param_dtype,\n rngs=rngs,\n )\n self.key_ln = LayerNorm(\n self.head_dim,\n use_bias=False,\n dtype=self.dtype,\n param_dtype=self.param_dtype,\n rngs=rngs,\n )\n else:\n self.query_ln = None\n self.key_ln = None\n\n self.out = LinearGeneral(\n in_features=(self.num_heads, self.head_dim),\n out_features=self.out_features,\n axis=(-2, -1),\n kernel_init=self.out_kernel_init or self.kernel_init,\n bias_init=self.out_bias_init or self.bias_init,\n use_bias=self.use_bias,\n dtype=self.dtype,\n param_dtype=self.param_dtype,\n precision=self.precision,\n dot_general=self.out_dot_general,\n dot_general_cls=self.out_dot_general_cls,\n rngs=rngs,\n )\n self.rngs = rngs if dropout_rate > 0.0 else None\n\n self.cached_key: nnx.Cache[Array] | None = None\n self.cached_value: nnx.Cache[Array] | None = None\n self.cache_index: nnx.Cache[Array] | None = None\n\n def __call__(\n self,\n inputs_q: Array,\n inputs_k: Array | None = None,\n inputs_v: Array | None = None,\n *,\n mask: Array | None = None,\n deterministic: bool | None = None,\n rngs: rnglib.Rngs | None = None,\n sow_weights: bool = False,\n decode: bool | None = None,\n ):\n """"""Applies multi-head dot product attention on the input data.\n\n Projects the inputs into multi-headed query, key, and value vectors,\n applies dot-product attention and project the results to an output vector.\n\n If both inputs_k and inputs_v are None, they will both copy the value of\n inputs_q (self attention).\n If only inputs_v is None, it will copy the value of inputs_k.\n\n Args:\n inputs_q: input queries of shape `[batch_sizes..., length, features]`.\n inputs_k: key of shape `[batch_sizes..., length, features]`. If None,\n inputs_k will copy the value of inputs_q.\n inputs_v: values of shape `[batch_sizes..., length, features]`. If None,\n inputs_v will copy the value of inputs_k.\n mask: attention mask of shape `[batch_sizes..., num_heads, query_length,\n key/value_length]`. Attention weights are masked out if their\n corresponding mask value is `False`.\n deterministic: if false, the attention weight is masked randomly using\n dropout, whereas if true, the attention weights are deterministic. The\n ``deterministic`` flag passed into the call method will take precedence\n over the ``deterministic`` flag passed into the constructor.\n rngs: rng key. The rng key passed into the call method will take\n precedence over the rng key passed into the constructor.\n sow_weights: if ``True``, the attention weights are sowed into the\n 'intermediates' collection.\n decode: whether to prepare and use an autoregressive cache. The ``decode``\n flag passed into the call method will take precedence over the ``decode``\n flag passed into the constructor.\n\n Returns:\n output of shape `[batch_sizes..., length, features]`.\n """"""\n if rngs is None:\n rngs = self.rngs\n\n if inputs_k is None:\n if inputs_v is not None:\n raise ValueError(\n '`inputs_k` cannot be None if `inputs_v` is not None. '\n 'To have both `inputs_k` and `inputs_v` be the same value, pass in the '\n 'value to `inputs_k` and leave `inputs_v` as None.'\n )\n inputs_k = inputs_q\n if inputs_v is None:\n inputs_v = inputs_k\n\n if inputs_q.shape[-1] != self.in_features:\n raise ValueError(\n f'Incompatible input dimension, got {inputs_q.shape[-1]} '\n f'but module expects {self.in_features}.'\n )\n\n query = self.query(inputs_q)\n key = self.key(inputs_k)\n value = self.value(inputs_v)\n\n if self.normalize_qk:\n assert self.query_ln is not None and self.key_ln is not None\n # Normalizing query and key projections stabilizes training with higher\n # LR. See ViT-22B paper http://arxiv.org/abs/2302.05442 for analysis.\n query = self.query_ln(query)\n key = self.key_ln(key)\n\n # During fast autoregressive decoding, we feed one position at a time,\n # and cache the keys and values step by step.\n decode = first_from(\n decode,\n self.decode,\n error_msg=""""""No `decode` argument was provided to MultiHeadAttention\n as either a __call__ argument, class attribute, or nnx.flag."""""",\n )\n\n if decode:\n if (\n self.cached_key is None\n or self.cached_value is None\n or self.cache_index is None\n ):\n raise ValueError(\n 'Autoregressive cache not initialized, call ``init_cache`` first.'\n )\n (\n *batch_dims,\n max_length,\n num_heads,\n depth_per_head,\n ) = self.cached_key.value.shape\n # shape check of cached keys against query input\n expected_shape = tuple(batch_dims) + (1, num_heads, depth_per_head)\n if expected_shape != query.shape:\n raise ValueError(\n 'Autoregressive cache shape error, '\n 'expected query shape %s instead got %s.'\n % (expected_shape, query.shape)\n )\n # update key, value caches with our new 1d spatial slices\n cur_index = self.cache_index.value\n zero = jnp.array(0, dtype=lax.dtype(cur_index.dtype))\n indices = (zero,) * len(batch_dims) + (cur_index, zero, zero)\n key = lax.dynamic_update_slice(self.cached_key.value, key, indices)\n value = lax.dynamic_update_slice(self.cached_value.value, value, indices)\n self.cached_key.value = key\n self.cached_value.value = value\n self.cache_index.value += 1\n # causal mask for cached decoder self-attention:\n # our single query position should only attend to those key\n # positions that have already been generated and cached,\n # not the remaining zero elements.\n mask = combine_masks(\n mask,\n jnp.broadcast_to(\n jnp.arange(max_length) <= cur_index,\n tuple(batch_dims) + (1, 1, max_length),\n ),\n )\n\n if (\n self.dropout_rate > 0.0\n ): # Require `deterministic` only if using dropout.\n deterministic = first_from(\n deterministic,\n self.deterministic,\n error_msg=""""""No `deterministic` argument was provided to MultiHeadAttention\n as either a __call__ argument, class attribute, or nnx.flag."""""",\n )\n if not deterministic:\n if rngs is None:\n raise ValueError(\n ""'rngs' must be provided if 'dropout_rng' is not given.""\n )\n dropout_rng = rngs.dropout()\n else:\n dropout_rng = None\n else:\n deterministic = True\n dropout_rng = None\n\n # apply attention\n x = self.attention_fn(\n query,\n key,\n value,\n mask=mask,\n dropout_rng=dropout_rng,\n dropout_rate=self.dropout_rate,\n broadcast_dropout=self.broadcast_dropout,\n deterministic=deterministic,\n dtype=self.dtype,\n precision=self.precision,\n module=self if sow_weights else None,\n )\n # back to the original inputs dimensions\n out = self.out(x)\n return out\n\n def init_cache(self, input_shape: Shape, dtype: Dtype = jnp.float32):\n """"""Initializes cache for fast autoregressive decoding. When\n ``decode=True``, this method must be called first before performing\n forward inference. When in decode mode, only one token must be passed\n at a time.\n\n Example usage::\n\n >>> from flax import nnx\n >>> import jax.numpy as jnp\n ...\n >>> batch_size = 5\n >>> embed_dim = 3\n >>> x = jnp.ones((batch_size, 1, embed_dim)) # single token\n ...\n >>> model_nnx = nnx.MultiHeadAttention(\n ... num_heads=2,\n ... in_features=3,\n ... qkv_features=6,\n ... out_features=6,\n ... decode=True,\n ... rngs=nnx.Rngs(42),\n ... )\n ...\n >>> # out_nnx = model_nnx(x) <-- throws an error because cache isn't initialized\n ...\n >>> model_nnx.init_cache(x.shape)\n >>> out_nnx = model_nnx(x)\n """"""\n cache_shape = (*input_shape[:-1], self.num_heads, self.head_dim)\n self.cached_key = nnx.Cache(jnp.zeros(cache_shape, dtype))\n self.cached_value = nnx.Cache(jnp.zeros(cache_shape, dtype))\n self.cache_index = nnx.Cache(jnp.array(0, dtype=jnp.int32))\n\n\n# mask-making utility functions\n\n\ndef make_attention_mask(\n query_input: Array,\n key_input: Array,\n pairwise_fn: Callable[..., Any] = jnp.multiply,\n extra_batch_dims: int = 0,\n dtype: Dtype = jnp.float32,\n):\n """"""Mask-making helper for attention weights.\n\n In case of 1d inputs (i.e., `[batch..., len_q]`, `[batch..., len_kv]`, the\n attention weights will be `[batch..., heads, len_q, len_kv]` and this\n function will produce `[batch..., 1, len_q, len_kv]`.\n\n Args:\n query_input: a batched, flat input of query_length size\n key_input: a batched, flat input of key_length size\n pairwise_fn: broadcasting elementwise comparison function\n extra_batch_dims: number of extra batch dims to add singleton axes for, none\n by default\n dtype: mask return dtype\n\n Returns:\n A `[batch..., 1, len_q, len_kv]` shaped mask for 1d attention.\n """"""\n mask = pairwise_fn(\n jnp.expand_dims(query_input, axis=-1), jnp.expand_dims(key_input, axis=-2)\n )\n mask = jnp.expand_dims(mask, axis=-3)\n mask = jnp.expand_dims(mask, axis=tuple(range(extra_batch_dims)))\n return mask.astype(dtype)\n\n\ndef make_causal_mask(\n x: Array, extra_batch_dims: int = 0, dtype: Dtype = jnp.float32\n) -> Array:\n """"""Make a causal mask for self-attention.\n\n In case of 1d inputs (i.e., `[batch..., len]`, the self-attention weights\n will be `[batch..., heads, len, len]` and this function will produce a\n causal mask of shape `[batch..., 1, len, len]`.\n\n Args:\n x: input array of shape `[batch..., len]`\n extra_batch_dims: number of batch dims to add singleton axes for, none by\n default\n dtype: mask return dtype\n\n Returns:\n A `[batch..., 1, len, len]` shaped causal mask for 1d attention.\n """"""\n idxs = jnp.broadcast_to(jnp.arange(x.shape[-1], dtype=jnp.int32), x.shape)\n return make_attention_mask(\n idxs,\n idxs,\n jnp.greater_equal,\n extra_batch_dims=extra_batch_dims,\n dtype=dtype,\n )\n\n\ndef combine_masks(\n *masks: Array | None, dtype: Dtype = jnp.float32\n) -> Array | None:\n """"""Combine attention masks.\n\n Args:\n *masks: set of attention mask arguments to combine, some can be None.\n dtype: dtype for the returned mask.\n\n Returns:\n Combined mask, reduced by logical and, returns None if no masks given.\n """"""\n masks_list = [m for m in masks if m is not None]\n if not masks_list:\n return None\n assert all(\n map(lambda x: x.ndim == masks_list[0].ndim, masks_list)\n ), f'masks must have same rank: {tuple(map(lambda x: x.ndim, masks_list))}'\n mask, *other_masks = masks_list\n for other_mask in other_masks:\n mask = jnp.logical_and(mask, other_mask)\n return mask.astype(dtype)\n",python,tab +1829,13325904,".venv/lib/python3.10/site-packages/flax/nnx/nn/attention.py",23317,5,"Cache",python,selection_command +1830,13326108,".venv/lib/python3.10/site-packages/flax/nnx/nn/attention.py",23321,0,"",python,selection_command +1831,13374556,".venv/lib/python3.10/site-packages/flax/nnx/variablelib.py",0,0,"",python,tab +1832,13374558,".venv/lib/python3.10/site-packages/flax/nnx/variablelib.py",22343,0,"",python,selection_command +1833,13374967,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",0,0,"",python,tab +1834,13374970,".venv/lib/python3.10/site-packages/flax/nnx/__init__.py",7490,0,"",python,selection_command +1835,13643341,".venv/lib/python3.10/site-packages/flax/nnx/variablelib.py",0,0,"",python,tab +1836,13644713,".venv/lib/python3.10/site-packages/flax/nnx/nn/attention.py",0,0,"",python,tab +1837,13646841,"train_dynamics.py",0,0,"",python,tab +1838,13648102,"train_dynamics.py",389,0,"",python,selection_command +1839,13648251,"train_dynamics.py",402,0,"",python,selection_command +1840,13648509,"train_dynamics.py",1650,0,"",python,selection_command +1841,13649145,"train_dynamics.py",3010,0,"",python,selection_command +1842,13649305,"train_dynamics.py",4881,0,"",python,selection_command +1843,13649512,"train_dynamics.py",6338,0,"",python,selection_command +1844,13650053,"train_dynamics.py",7608,0,"",python,selection_command +1845,13650310,"train_dynamics.py",9117,0,"",python,selection_command +1846,13650578,"train_dynamics.py",10607,0,"",python,selection_command +1847,13653705,"train_dynamics.py",12706,0,"",python,selection_command +1848,13660473,"train_dynamics.py",12500,0,"",python,selection_command +1849,13660747,"train_dynamics.py",4346,0,"",python,selection_command +1850,13670735,"train_dynamics.py",4362,0,"",python,selection_command +1851,13670983,"train_dynamics.py",4419,0,"",python,selection_command +1852,13671019,"train_dynamics.py",4459,0,"",python,selection_command +1853,13671058,"train_dynamics.py",4498,0,"",python,selection_command +1854,13671101,"train_dynamics.py",4503,0,"",python,selection_command +1855,13671569,"train_dynamics.py",4578,0,"",python,selection_command +1856,13671719,"train_dynamics.py",4582,0,"",python,selection_command +1857,13671879,"train_dynamics.py",4589,0,"",python,selection_command +1858,13672054,"train_dynamics.py",4605,0,"",python,selection_command +1859,13672350,"train_dynamics.py",4589,0,"",python,selection_command +1860,13672495,"train_dynamics.py",2472,0,"",python,selection_command +1861,13679116,"train_dynamics.py",4589,0,"",python,selection_command +1862,13681055,"train_dynamics.py",12500,0,"",python,selection_command +1863,13683219,"train_dynamics.py",12439,0,"",python,selection_command +1864,13685829,"train_dynamics.py",12500,0,"",python,selection_command +1865,13686719,"train_dynamics.py",4346,0,"",python,selection_command +1866,13687621,"train_dynamics.py",4362,0,"",python,selection_command +1867,13687753,"train_dynamics.py",4419,0,"",python,selection_command +1868,13687899,"train_dynamics.py",4459,0,"",python,selection_command +1869,13688150,"train_dynamics.py",4498,0,"",python,selection_command +1870,13688248,"train_dynamics.py",4503,0,"",python,selection_command +1871,13688397,"train_dynamics.py",4578,0,"",python,selection_command +1872,13688529,"train_dynamics.py",4582,0,"",python,selection_command +1873,13688720,"train_dynamics.py",4507,0,"",python,selection_command +1874,13689342,"train_dynamics.py",4582,0,"",python,selection_command +1875,13689451,"train_dynamics.py",4589,0,"",python,selection_command +1876,13689601,"train_dynamics.py",4605,0,"",python,selection_command +1877,13689965,"train_dynamics.py",4589,0,"",python,selection_command +1878,13690634,"train_dynamics.py",2472,0,"",python,selection_command +1879,13692474,"train_dynamics.py",2494,0,"",python,selection_command +1880,13692724,"train_dynamics.py",2525,0,"",python,selection_command +1881,13692759,"train_dynamics.py",2572,0,"",python,selection_command +1882,13692795,"train_dynamics.py",2611,0,"",python,selection_command +1883,13693146,"train_dynamics.py",2677,0,"",python,selection_command +1884,13694885,"train_dynamics.py",2756,0,"",python,selection_command +1885,13696247,"train_dynamics.py",2755,0,"",python,selection_command +1886,13696696,"train_dynamics.py",2750,0,"",python,selection_command +1887,13698147,"train_dynamics.py",2494,0,"",python,selection_command +1888,13699003,"train_dynamics.py",2499,0,"",python,selection_command +1889,13699166,"train_dynamics.py",2501,0,"",python,selection_command +1890,13699606,"genie.py",0,0,"from typing import Dict\n\nimport einops\nimport jax\nimport jax.numpy as jnp\nimport flax.nnx as nnx\nimport orbax.checkpoint as ocp\n\nfrom models.dynamics import DynamicsMaskGIT, DynamicsCausal\nfrom models.lam import LatentActionModel\nfrom models.tokenizer import TokenizerVQVAE\n\n\nclass Genie(nnx.Module):\n """"""Genie model""""""\n\n def __init__(\n self,\n in_dim: int,\n tokenizer_dim: int,\n tokenizer_ffn_dim: int,\n latent_patch_dim: int,\n num_patch_latents: int,\n patch_size: int,\n tokenizer_num_blocks: int,\n tokenizer_num_heads: int,\n lam_dim: int,\n lam_ffn_dim: int,\n latent_action_dim: int,\n num_latent_actions: int,\n lam_patch_size: int,\n lam_num_blocks: int,\n lam_num_heads: int,\n lam_co_train: bool,\n dyna_type: str,\n dyna_dim: int,\n dyna_ffn_dim: int,\n dyna_num_blocks: int,\n dyna_num_heads: int,\n param_dtype: jnp.dtype,\n dtype: jnp.dtype,\n use_flash_attention: bool,\n decode: bool,\n rngs: nnx.Rngs,\n dropout: float = 0.0,\n mask_limit: float = 0.0,\n ):\n # --- Tokenizer ---\n self.in_dim = in_dim\n self.tokenizer_dim = tokenizer_dim\n self.tokenizer_ffn_dim = tokenizer_ffn_dim\n self.latent_patch_dim = latent_patch_dim\n self.num_patch_latents = num_patch_latents\n self.patch_size = patch_size\n self.tokenizer_num_blocks = tokenizer_num_blocks\n self.tokenizer_num_heads = tokenizer_num_heads\n # --- LAM ---\n self.lam_dim = lam_dim\n self.lam_ffn_dim = lam_ffn_dim\n self.latent_action_dim = latent_action_dim\n self.num_latent_actions = num_latent_actions\n self.lam_patch_size = lam_patch_size\n self.lam_num_blocks = lam_num_blocks\n self.lam_num_heads = lam_num_heads\n self.lam_co_train = lam_co_train\n # --- Dynamics ---\n self.dyna_type = dyna_type\n self.dyna_dim = dyna_dim\n self.dyna_ffn_dim = dyna_ffn_dim\n self.dyna_num_blocks = dyna_num_blocks\n self.dyna_num_heads = dyna_num_heads\n self.param_dtype = param_dtype\n self.dtype = dtype\n self.use_flash_attention = use_flash_attention\n self.dropout = dropout\n self.mask_limit = mask_limit\n\n self.tokenizer = TokenizerVQVAE(\n in_dim=self.in_dim,\n model_dim=self.tokenizer_dim,\n ffn_dim=self.tokenizer_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_patch_latents,\n patch_size=self.patch_size,\n num_blocks=self.tokenizer_num_blocks,\n num_heads=self.tokenizer_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n self.lam = LatentActionModel(\n in_dim=self.in_dim,\n model_dim=self.lam_dim,\n ffn_dim=self.lam_ffn_dim,\n latent_dim=self.latent_patch_dim,\n num_latents=self.num_latent_actions,\n patch_size=self.lam_patch_size,\n num_blocks=self.lam_num_blocks,\n num_heads=self.lam_num_heads,\n dropout=0.0,\n codebook_dropout=0.0,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n if self.dyna_type == ""maskgit"":\n self.dynamics = DynamicsMaskGIT(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n mask_limit=self.mask_limit,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n elif self.dyna_type == ""causal"":\n self.dynamics = DynamicsCausal(\n model_dim=self.dyna_dim,\n ffn_dim=self.dyna_ffn_dim,\n num_latents=self.num_patch_latents,\n latent_action_dim=self.latent_action_dim,\n num_blocks=self.dyna_num_blocks,\n num_heads=self.dyna_num_heads,\n dropout=self.dropout,\n param_dtype=self.param_dtype,\n dtype=self.dtype,\n use_flash_attention=self.use_flash_attention,\n decode=decode,\n rngs=rngs,\n )\n else:\n raise ValueError(f""Invalid dynamics type: {self.dyna_type}"")\n\n def __call__(\n self, batch: Dict[str, jax.Array], training: bool = True\n ) -> Dict[str, jax.Array]:\n videos_BTHWC = batch[""videos""]\n tokenizer_outputs = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_indices_BTN = tokenizer_outputs[""indices""]\n lam_outputs = self.lam.vq_encode(videos_BTHWC, training=False)\n z_q_BTm11L = lam_outputs[""z_q""]\n action_indices_E = lam_outputs[""indices""]\n latent_actions_BTm11L = jax.lax.cond(\n self.lam_co_train,\n lambda: z_q_BTm11L,\n lambda: jax.lax.stop_gradient(z_q_BTm11L),\n )\n outputs = dict(\n video_tokens=jax.lax.stop_gradient(token_indices_BTN),\n latent_actions=latent_actions_BTm11L,\n )\n outputs[""mask_rng""] = batch[""mask_rng""]\n dyna_logits_BTNV, dyna_mask = self.dynamics(outputs, training)\n outputs[""token_logits""] = dyna_logits_BTNV\n if dyna_mask is not None:\n outputs[""mask""] = dyna_mask\n mle_indices_BTN = jnp.argmax(outputs[""token_logits""], axis=-1)\n H, W = batch[""videos""].shape[2:4]\n outputs[""recon""] = self.tokenizer.decode(mle_indices_BTN, (H, W))\n outputs[""lam_indices""] = action_indices_E\n return outputs\n\n def sample(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n steps: int = 25,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> jax.Array:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n P: S * N\n """"""\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n latent_actions_E = batch[""latent_actions""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n\n def maskgit_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array], step: jax.Array\n ) -> tuple[tuple[jax.Array, jax.Array, jax.Array, jax.Array], None]:\n rng, token_idxs_BSN, mask_BSN, action_tokens_EL = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # --- Construct + encode video ---\n vid_embed_BSNM = self.dynamics.patch_embed(token_idxs_BSN)\n mask_token_111M = self.dynamics.mask_token.value\n mask_expanded_BSN1 = mask_BSN[..., None]\n vid_embed_BSNM = jnp.where(mask_expanded_BSN1, mask_token_111M, vid_embed_BSNM)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = self.dynamics.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1]))\n vid_embed_BSNM += act_embed_BS1M\n unmasked_ratio = jnp.cos(jnp.pi * (step + 1) / (steps * 2))\n step_temp = temperature * (1.0 - unmasked_ratio)\n final_logits_BSNV = self.dynamics.transformer(vid_embed_BSNM) / step_temp\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_BSN = jnp.argmax(final_logits_BSNV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_BSN = jax.random.categorical(_rng, final_logits_BSNV)\n gather_fn = jax.vmap(jax.vmap(jax.vmap(lambda x, y: x[y])))\n final_token_probs_BSN = gather_fn(\n jax.nn.softmax(final_logits_BSNV), sampled_token_idxs_BSN\n )\n final_token_probs_BSN += ~mask_BSN\n # Update masked tokens only\n token_idxs_BSN = jnp.where(mask_BSN, sampled_token_idxs_BSN, token_idxs_BSN)\n\n # --- Update mask ---\n num_unmasked_tokens = jnp.round(N * (1.0 - unmasked_ratio)).astype(int)\n final_token_probs_flat_BP = einops.rearrange(final_token_probs_BSN, ""b s n -> b (s n)"")\n idx_mask_P = jnp.arange(final_token_probs_flat_BP.shape[-1]) <= N - num_unmasked_tokens\n sorted_idxs_BP = jnp.argsort(final_token_probs_flat_BP, axis=-1)\n mask_update_fn = jax.vmap(lambda msk, ids: msk.at[ids].set(idx_mask_P))\n mask_flat_BP = einops.rearrange(mask_BSN, ""b s n -> b (s n)"")\n new_mask_flat_BP = mask_update_fn(mask_flat_BP, sorted_idxs_BP)\n new_mask_BSN = einops.rearrange(new_mask_flat_BP, ""b (s n) -> b s n"", n=N)\n\n new_carry = (rng, token_idxs_BSN, new_mask_BSN, action_tokens_EL)\n return new_carry, None\n\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[tuple[jax.Array, jax.Array], None]:\n rng, current_token_idxs_BSN = carry\n rng, step_rng = jax.random.split(rng)\n\n # Mask current frame (i.e., t == step_t)\n mask_S = jnp.arange(seq_len) == step_t\n mask_BSN = jnp.broadcast_to(mask_S[None, :, None], (B, seq_len, N)).astype(\n bool\n )\n masked_token_idxs_BSN = current_token_idxs_BSN * ~mask_BSN\n\n # --- Initialize and run MaskGIT loop ---\n init_carry_maskgit = (\n step_rng,\n masked_token_idxs_BSN,\n mask_BSN,\n action_tokens_EL,\n )\n final_carry_maskgit, _ = jax.lax.scan(\n maskgit_step_fn, init_carry_maskgit, jnp.arange(steps)\n )\n updated_token_idxs_BSN = final_carry_maskgit[1]\n new_carry = (rng, updated_token_idxs_BSN)\n return new_carry, None\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry, _ = jax.lax.scan(\n generation_step_fn, initial_carry, timesteps_to_scan\n )\n final_token_idxs_BSN = final_carry[1]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC\n\n def sample_causal(\n self,\n batch: Dict[str, jax.Array],\n seq_len: int,\n temperature: float = 1,\n sample_argmax: bool = False,\n ) -> jax.Array:\n """"""\n Autoregressively samples up to `seq_len` future frames, following Figure 8 of the paper.\n\n - Input frames are tokenized once.\n - Future frames are generated autoregressively in token space.\n - All frames are detokenized in a single pass.\n\n Note:\n - For interactive or step-wise sampling, detokenization should occur after each action.\n - To maintain consistent tensor shapes across timesteps, all current and future frames are decoded at every step.\n - Temporal causal structure is preserved by\n a) reapplying the mask before each decoding step.\n b) a temporal causal mask is applied within each ST-transformer block.\n\n Dimension keys:\n B: batch size\n T: number of input (conditioning) frames\n N: number of patches per frame\n M: model dimension\n S: sequence length\n H: height\n W: width\n E: B * (S - 1)\n """"""\n assert isinstance(self.dynamics, DynamicsCausal)\n # --- Encode videos and actions ---\n videos_BTHWC = batch[""videos""]\n latent_actions_E = batch[""latent_actions""]\n tokenizer_out = self.tokenizer.vq_encode(videos_BTHWC, training=False)\n token_idxs_BTN = tokenizer_out[""indices""]\n B, T, N = token_idxs_BTN.shape\n pad_shape = (B, seq_len - T, N)\n pad = jnp.zeros(pad_shape, dtype=token_idxs_BTN.dtype)\n token_idxs_BSN = jnp.concatenate([token_idxs_BTN, pad], axis=1)\n action_tokens_EL = self.lam.vq.get_codes(latent_actions_E)\n dynamics_causal: DynamicsCausal = self.dynamics\n\n def causal_step_fn(\n carry: tuple[jax.Array, jax.Array, jax.Array, jax.Array], step_n: jax.Array\n ) -> tuple[tuple[jax.Array, jax.Array, jax.Array, jax.Array], None]:\n rng, token_idxs_BSN, action_tokens_EL, step_t = carry\n S, N = token_idxs_BSN.shape[1:]\n L = action_tokens_EL.shape[-1]\n\n # --- Construct + encode video ---\n vid_embed_BSNM = dynamics_causal.patch_embed(token_idxs_BSN)\n\n # --- Predict transition ---\n action_tokens_BSm1L = jnp.reshape(action_tokens_EL, (B, S - 1, L))\n act_embed_BSm1M = dynamics_causal.action_up(action_tokens_BSm1L)\n act_embed_BSM = jnp.pad(act_embed_BSm1M, ((0, 0), (1, 0), (0, 0)))\n act_embed_BS1M = jnp.reshape(act_embed_BSM, (B, S, 1, act_embed_BSM.shape[-1]))\n vid_embed_BSNp1M = jnp.concatenate([act_embed_BS1M, vid_embed_BSNM], axis=2)\n final_logits_BTNp1V = dynamics_causal.transformer(vid_embed_BSNp1M, (step_t, step_n)) / temperature\n final_logits_BV = final_logits_BTNp1V[:, step_t, step_n, :]\n\n # --- Sample new tokens for final frame ---\n if sample_argmax:\n sampled_token_idxs_B = jnp.argmax(final_logits_BV, axis=-1)\n else:\n rng, _rng = jax.random.split(rng)\n sampled_token_idxs_B = jax.random.categorical(_rng, final_logits_BV)\n # Update next tokens only\n token_idxs_BSN = token_idxs_BSN.at[:, step_t, step_n].set(sampled_token_idxs_B)\n\n new_carry = (rng, token_idxs_BSN, action_tokens_EL, step_t)\n return new_carry, None\n\n def generation_step_fn(\n carry: tuple[jax.Array, jax.Array], step_t: jax.Array\n ) -> tuple[tuple[jax.Array, jax.Array], None]:\n rng, current_token_idxs_BSN = carry\n rng, step_rng = jax.random.split(rng)\n\n # --- Initialize and run causal loop ---\n init_carry_causal = (\n step_rng,\n current_token_idxs_BSN,\n action_tokens_EL,\n step_t,\n )\n final_carry_causal, _ = jax.lax.scan(\n causal_step_fn, init_carry_causal, jnp.arange(N)\n )\n updated_token_idxs_BSN = final_carry_causal[1]\n new_carry = (rng, updated_token_idxs_BSN)\n return new_carry, None\n\n # --- Run the autoregressive generation using jax.lax.scan ---\n initial_carry = (batch[""rng""], token_idxs_BSN)\n timesteps_to_scan = jnp.arange(T, seq_len)\n final_carry, _ = jax.lax.scan(\n generation_step_fn, initial_carry, timesteps_to_scan\n )\n final_token_idxs_BSN = final_carry[1]\n\n # --- Decode all tokens at once at the end ---\n H, W = batch[""videos""].shape[2:4]\n final_frames_BSHWC = self.tokenizer.decode(\n final_token_idxs_BSN,\n video_hw=(H, W),\n )\n return final_frames_BSHWC\n\n def vq_encode(self, batch: Dict[str, jax.Array], training: bool) -> jax.Array:\n # --- Preprocess videos ---\n video_BTHWC = batch[""videos""]\n lam_output = self.lam.vq_encode(video_BTHWC, training=training)\n lam_indices_E = lam_output[""indices""]\n return lam_indices_E\n\n# FIXME (f.srambical): add conversion script for old checkpoints\ndef restore_genie_components(\n optimizer: nnx.Optimizer,\n sharding: jax.sharding.NamedSharding,\n rng: jax.Array,\n args,\n) -> nnx.Optimizer:\n """"""Restore pre-trained Genie components""""""\n rngs = nnx.Rngs(rng)\n\n tx = optimizer.tx\n model = optimizer.model\n handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n handler_registry.add(\n ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n )\n\n checkpoint_options = ocp.CheckpointManagerOptions(\n step_format_fixed_length=6,\n )\n tokenizer_checkpoint_manager = ocp.CheckpointManager(\n directory=args.tokenizer_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_tokenizer = TokenizerVQVAE(\n in_dim=args.image_channels,\n model_dim=args.tokenizer_dim,\n ffn_dim=args.tokenizer_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_patch_latents,\n patch_size=args.patch_size,\n num_blocks=args.tokenizer_num_blocks,\n num_heads=args.tokenizer_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n )\n dummy_tokenizer_optimizer = nnx.Optimizer(dummy_tokenizer, tx)\n dummy_tokenizer_optimizer_state = nnx.state(dummy_tokenizer_optimizer)\n abstract_sharded_tokenizer_optimizer_state = _create_abstract_sharded_pytree(\n dummy_tokenizer_optimizer_state, sharding\n )\n restored_tokenizer = tokenizer_checkpoint_manager.restore(\n step=tokenizer_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_tokenizer_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_tokenizer_optimizer.model, restored_tokenizer.model)\n model.tokenizer = dummy_tokenizer_optimizer.model\n tokenizer_checkpoint_manager.close()\n\n if args.lam_checkpoint:\n lam_checkpoint_manager = ocp.CheckpointManager(\n directory=args.lam_checkpoint,\n options=checkpoint_options,\n handler_registry=handler_registry,\n )\n dummy_lam = LatentActionModel(\n in_dim=args.image_channels,\n model_dim=args.lam_dim,\n ffn_dim=args.lam_ffn_dim,\n latent_dim=args.latent_patch_dim,\n num_latents=args.num_latent_actions,\n patch_size=args.lam_patch_size,\n num_blocks=args.lam_num_blocks,\n num_heads=args.lam_num_heads,\n dropout=args.dropout,\n codebook_dropout=args.dropout,\n param_dtype=args.param_dtype,\n dtype=args.dtype,\n use_flash_attention=args.use_flash_attention,\n rngs=rngs,\n )\n dummy_lam_optimizer = nnx.Optimizer(dummy_lam, tx)\n dummy_lam_optimizer_state = nnx.state(dummy_lam_optimizer)\n abstract_sharded_lam_optimizer_state = _create_abstract_sharded_pytree(\n dummy_lam_optimizer_state, sharding\n )\n restored_lam_optimizer = lam_checkpoint_manager.restore(\n step=lam_checkpoint_manager.latest_step(),\n args=ocp.args.Composite(\n model_state=ocp.args.PyTreeRestore( # type: ignore\n abstract_sharded_lam_optimizer_state # type: ignore\n ),\n ),\n )[""model_state""]\n nnx.update(dummy_lam_optimizer.model, restored_lam_optimizer.model)\n model.lam = dummy_lam_optimizer.model\n # Remove the LAM decoder to save memory and avoid unnecessary computation.\n del model.lam.decoder\n lam_checkpoint_manager.close()\n \n # Reinitialize the optimizer states\n optimizer = nnx.Optimizer(model, tx)\n return optimizer\n\n\ndef _create_abstract_sharded_pytree(\n pytree_template: nnx.GraphState, sharding_spec: jax.sharding.NamedSharding\n) -> jax.Array:\n """"""Replaces arrays in a pytree with ShapeDtypeStructs having the given sharding.""""""\n\n def map_fn(leaf_template):\n if hasattr(leaf_template, ""shape"") and hasattr(leaf_template, ""dtype""):\n return jax.ShapeDtypeStruct(\n leaf_template.shape, leaf_template.dtype, sharding=sharding_spec\n )\n return leaf_template\n\n return jax.tree_util.tree_map(map_fn, pytree_template)\n",python,tab +1891,13699614,"genie.py",282,0,"",python,selection_command +1892,13702678,"genie.py",4920,0,"",python,selection_command +1893,13704560,"genie.py",4938,0,"",python,selection_command +1894,13704687,"genie.py",5003,0,"",python,selection_command +1895,13704828,"genie.py",5034,0,"",python,selection_command +1896,13736800,"genie.py",5073,0,"",python,selection_command +1897,13738932,"genie.py",5089,0,"",python,selection_command +1898,13740378,"genie.py",5109,0,"",python,selection_command +1899,13740629,"models/tokenizer.py",0,0,"from typing import Dict, Tuple\n\nimport flax.nnx as nnx\nimport jax.numpy as jnp\nimport jax\n\nfrom utils.preprocess import patchify, unpatchify\nfrom utils.nn import STTransformer, VectorQuantizer\n\n\nclass TokenizerVQVAE(nnx.Module):\n """"""\n ST-ViVit VQ-VAE\n\n Dimension keys:\n B: batch size\n T: sequence length\n N: number of patches per frame\n L: latent dimension\n D: B * T * N\n H: height\n W: width\n C: number of channels\n P: patch token dimension (patch_size^2 * C)\n """"""\n\n def __init__(\n self,\n in_dim: int,\n model_dim: int,\n ffn_dim: int,\n latent_dim: int,\n num_latents: int,\n patch_size: int,\n num_blocks: int,\n num_heads: int,\n dropout: float,\n codebook_dropout: float,\n param_dtype: jnp.dtype,\n dtype: jnp.dtype,\n use_flash_attention: bool,\n rngs: nnx.Rngs,\n ):\n self.in_dim = in_dim\n self.model_dim = model_dim\n self.ffn_dim = ffn_dim\n self.latent_dim = latent_dim\n self.num_latents = num_latents\n self.patch_size = patch_size\n self.num_blocks = num_blocks\n self.num_heads = num_heads\n self.dropout = dropout\n self.codebook_dropout = codebook_dropout\n self.param_dtype = param_dtype\n self.dtype = dtype\n self.use_flash_attention = use_flash_attention\n\n self.encoder = STTransformer(\n self.in_dim * self.patch_size**2,\n self.model_dim,\n self.ffn_dim,\n self.latent_dim,\n self.num_blocks,\n self.num_heads,\n self.dropout,\n self.param_dtype,\n self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n self.vq = VectorQuantizer(\n self.latent_dim,\n self.num_latents,\n self.codebook_dropout,\n rngs=rngs,\n )\n self.out_dim = self.in_dim * self.patch_size**2\n self.decoder = STTransformer(\n self.latent_dim,\n self.model_dim,\n self.ffn_dim,\n self.out_dim,\n self.num_blocks,\n self.num_heads,\n self.dropout,\n self.param_dtype,\n self.dtype,\n use_flash_attention=self.use_flash_attention,\n rngs=rngs,\n )\n\n def __call__(\n self, batch: Dict[str, jax.Array], training: bool = True\n ) -> Dict[str, jax.Array]:\n H, W = batch[""videos""].shape[2:4]\n videos_BTHWC = batch[""videos""]\n outputs = self.vq_encode(videos_BTHWC, training)\n z_q_BTNL = outputs[""z_q""]\n recon_BTHWC = self.decoder(z_q_BTNL)\n recon_BTHWC = recon_BTHWC.astype(jnp.float32)\n recon_BTHWC = nnx.sigmoid(recon_BTHWC)\n recon_BTHWC = recon_BTHWC.astype(self.dtype)\n recon_BTHWC = unpatchify(recon_BTHWC, self.patch_size, H, W)\n outputs[""recon""] = recon_BTHWC\n return outputs\n\n def vq_encode(\n self, videos: jax.Array, training: bool = True\n ) -> Dict[str, jax.Array]:\n # --- Preprocess + encode ---\n B, T = videos.shape[:2]\n patch_BTNP = patchify(videos, self.patch_size)\n N = patch_BTNP.shape[2]\n x_BTNL = self.encoder(patch_BTNP)\n\n # --- Vector quantize ---\n x_DL = x_BTNL.reshape(B * T * N, self.latent_dim)\n z_q_DL, z_DL, emb_DL, indices_D = self.vq(x_DL, training)\n z_q_BTNL = z_q_DL.reshape(B, T, N, self.latent_dim)\n indices_BTN = indices_D.reshape(B, T, N)\n return dict(z_q=z_q_BTNL, z=z_DL, emb=emb_DL, indices=indices_BTN)\n\n def decode(self, indices_BTN: jax.Array, video_hw: Tuple[int, int]) -> jax.Array:\n z_BTNL = self.vq.codebook[indices_BTN]\n recon_BTNP = self.decoder(z_BTNL)\n recon_BTNP = recon_BTNP.astype(jnp.float32)\n recon_BTNP = nnx.sigmoid(recon_BTNP)\n recon_BTNP = recon_BTNP.astype(self.dtype)\n return unpatchify(recon_BTNP, self.patch_size, *video_hw)\n",python,tab +1900,13740636,"models/tokenizer.py",3042,0,"",python,selection_command +1901,13743188,"genie.py",0,0,"",python,tab +1902,13743190,"genie.py",5109,0,"",python,selection_command +1903,13745757,"genie.py",5192,0,"",python,selection_command +1904,13745998,"genie.py",5154,0,"",python,selection_command +1905,13757319,"genie.py",5211,0,"",python,selection_command +1906,13760195,"genie.py",5282,0,"",python,selection_command +1907,13794215,"genie.py",5211,0,"",python,selection_command +1908,13794343,"genie.py",5154,0,"",python,selection_command +1909,13795121,"genie.py",5609,0,"",python,selection_command +1910,13799589,"genie.py",5560,0,"",python,selection_command +1911,13799831,"genie.py",5536,0,"",python,selection_command +1912,13799866,"genie.py",5520,0,"",python,selection_command +1913,13799940,"genie.py",5471,0,"",python,selection_command +1914,13799946,"genie.py",5439,0,"",python,selection_command +1915,13800173,"genie.py",5408,0,"",python,selection_command +1916,13800189,"genie.py",5361,0,"",python,selection_command +1917,13800299,"genie.py",5312,0,"",python,selection_command +1918,13800404,"genie.py",5250,0,"",python,selection_command +1919,13800573,"genie.py",5193,0,"",python,selection_command +1920,13801331,"genie.py",5154,0,"",python,selection_command +1921,13808046,"genie.py",5071,0,"",python,selection_command +1922,13808462,"genie.py",5062,0,"\n ",python,content +1923,13808569,"genie.py",5071,0,"#",python,content +1924,13808572,"genie.py",5072,0,"",python,selection_keyboard +1925,13808607,"genie.py",5072,0," ",python,content +1926,13808611,"genie.py",5073,0,"",python,selection_keyboard +1927,13808981,"genie.py",5073,0,"F",python,content +1928,13808986,"genie.py",5074,0,"",python,selection_keyboard +1929,13809045,"genie.py",5074,0,"I",python,content +1930,13809048,"genie.py",5075,0,"",python,selection_keyboard +1931,13809234,"genie.py",5075,0,"X",python,content +1932,13809239,"genie.py",5076,0,"",python,selection_keyboard +1933,13809855,"genie.py",5076,0,"M",python,content +1934,13809862,"genie.py",5077,0,"",python,selection_keyboard +1935,13809949,"genie.py",5077,0,"E",python,content +1936,13809952,"genie.py",5078,0,"",python,selection_keyboard +1937,13810054,"genie.py",5078,0,":",python,content +1938,13810057,"genie.py",5079,0,"",python,selection_keyboard +1939,13810181,"genie.py",5079,0," ",python,content +1940,13810185,"genie.py",5080,0,"",python,selection_keyboard +1941,13811050,"genie.py",5079,1,"",python,content +1942,13811179,"genie.py",5078,1,"",python,content +1943,13811266,"genie.py",5078,0," ",python,content +1944,13811272,"genie.py",5079,0,"",python,selection_keyboard +1945,13811404,"genie.py",5079,0,"()",python,content +1946,13811411,"genie.py",5080,0,"",python,selection_keyboard +1947,13811590,"genie.py",5080,0,"f",python,content +1948,13811601,"genie.py",5081,0,"",python,selection_keyboard +1949,13811815,"genie.py",5081,0,"s",python,content +1950,13811819,"genie.py",5082,0,"",python,selection_keyboard +1951,13811888,"genie.py",5082,0,"r",python,content +1952,13811892,"genie.py",5083,0,"",python,selection_keyboard +1953,13811980,"genie.py",5083,0,"a",python,content +1954,13811984,"genie.py",5084,0,"",python,selection_keyboard +1955,13812024,"genie.py",5084,0,"m",python,content +1956,13812029,"genie.py",5085,0,"",python,selection_keyboard +1957,13812596,"genie.py",5084,1,"",python,content +1958,13812735,"genie.py",5083,1,"",python,content +1959,13812891,"genie.py",5082,1,"",python,content +1960,13813038,"genie.py",5081,1,"",python,content +1961,13813189,"genie.py",5081,0,".",python,content +1962,13813193,"genie.py",5082,0,"",python,selection_keyboard +1963,13813216,"genie.py",5082,0,"s",python,content +1964,13813218,"genie.py",5083,0,"",python,selection_keyboard +1965,13813314,"genie.py",5083,0,"r",python,content +1966,13813317,"genie.py",5084,0,"",python,selection_keyboard +1967,13813362,"genie.py",5084,0,"a",python,content +1968,13813365,"genie.py",5085,0,"",python,selection_keyboard +1969,13813458,"genie.py",5085,0,"m",python,content +1970,13813500,"genie.py",5086,0,"",python,selection_keyboard +1971,13813613,"genie.py",5086,0,"b",python,content +1972,13813616,"genie.py",5087,0,"",python,selection_keyboard +1973,13813732,"genie.py",5087,0,"i",python,content +1974,13813735,"genie.py",5088,0,"",python,selection_keyboard +1975,13813833,"genie.py",5088,0,"c",python,content +1976,13813840,"genie.py",5089,0,"",python,selection_keyboard +1977,13813881,"genie.py",5089,0,"a",python,content +1978,13813886,"genie.py",5090,0,"",python,selection_keyboard +1979,13813953,"genie.py",5090,0,"l",python,content +1980,13813974,"genie.py",5091,0,"",python,selection_keyboard +1981,13814165,"genie.py",5091,1,")",python,content +1982,13814173,"genie.py",5092,0,"",python,selection_keyboard +1983,13814327,"genie.py",5092,0,":",python,content +1984,13814333,"genie.py",5093,0,"",python,selection_keyboard +1985,13814449,"genie.py",5093,0," ",python,content +1986,13814455,"genie.py",5094,0,"",python,selection_keyboard +1987,13814663,"genie.py",5094,0,"t",python,content +1988,13814666,"genie.py",5095,0,"",python,selection_keyboard +1989,13814683,"genie.py",5095,0,"h",python,content +1990,13814685,"genie.py",5096,0,"",python,selection_keyboard +1991,13814782,"genie.py",5096,0,"i",python,content +1992,13814785,"genie.py",5097,0,"",python,selection_keyboard +1993,13815245,"genie.py",5097,0,"s",python,content +1994,13815249,"genie.py",5098,0,"",python,selection_keyboard +1995,13815353,"genie.py",5098,0," ",python,content +1996,13815356,"genie.py",5099,0,"",python,selection_keyboard +1997,13815718,"genie.py",5099,0,"s",python,content +1998,13815721,"genie.py",5100,0,"",python,selection_keyboard +1999,13816049,"genie.py",5099,1,"",python,content +2000,13816194,"genie.py",5099,0,"i",python,content +2001,13816196,"genie.py",5100,0,"",python,selection_keyboard +2002,13816251,"genie.py",5100,0,"s",python,content +2003,13816257,"genie.py",5101,0,"",python,selection_keyboard +2004,13816305,"genie.py",5101,0," ",python,content +2005,13816308,"genie.py",5102,0,"",python,selection_keyboard +2006,13816388,"genie.py",5102,0,"t",python,content +2007,13816391,"genie.py",5103,0,"",python,selection_keyboard +2008,13816452,"genie.py",5103,0,"h",python,content +2009,13816464,"genie.py",5104,0,"",python,selection_keyboard +2010,13816515,"genie.py",5104,0,"e",python,content +2011,13816520,"genie.py",5105,0,"",python,selection_keyboard +2012,13816596,"genie.py",5105,0," ",python,content +2013,13816599,"genie.py",5106,0,"",python,selection_keyboard +2014,13816692,"genie.py",5106,0,"p",python,content +2015,13816694,"genie.py",5107,0,"",python,selection_keyboard +2016,13816773,"genie.py",5107,0,"a",python,content +2017,13816782,"genie.py",5108,0,"",python,selection_keyboard +2018,13816850,"genie.py",5108,0,"r",python,content +2019,13816857,"genie.py",5109,0,"",python,selection_keyboard +2020,13816998,"genie.py",5109,0,"t",python,content +2021,13817000,"genie.py",5110,0,"",python,selection_keyboard +2022,13817154,"genie.py",5110,0," ",python,content +2023,13817157,"genie.py",5111,0,"",python,selection_keyboard +2024,13817249,"genie.py",5111,0,"t",python,content +2025,13817294,"genie.py",5112,0,"",python,selection_keyboard +2026,13817414,"genie.py",5112,0,"h",python,content +2027,13817417,"genie.py",5113,0,"",python,selection_keyboard +2028,13817467,"genie.py",5113,0,"a",python,content +2029,13817471,"genie.py",5114,0,"",python,selection_keyboard +2030,13817534,"genie.py",5114,0,"t",python,content +2031,13817539,"genie.py",5115,0,"",python,selection_keyboard +2032,13817580,"genie.py",5115,0," ",python,content +2033,13817583,"genie.py",5116,0,"",python,selection_keyboard +2034,13817831,"genie.py",5116,0,"I",python,content +2035,13817834,"genie.py",5117,0,"",python,selection_keyboard +2036,13817864,"genie.py",5117,0," ",python,content +2037,13817866,"genie.py",5118,0,"",python,selection_keyboard +2038,13818014,"genie.py",5118,0,"s",python,content +2039,13818019,"genie.py",5119,0,"",python,selection_keyboard +2040,13818091,"genie.py",5119,0,"h",python,content +2041,13818100,"genie.py",5120,0,"",python,selection_keyboard +2042,13818179,"genie.py",5120,0,"o",python,content +2043,13818182,"genie.py",5121,0,"",python,selection_keyboard +2044,13818233,"genie.py",5121,0,"u",python,content +2045,13818237,"genie.py",5122,0,"",python,selection_keyboard +2046,13818341,"genie.py",5122,0,"l",python,content +2047,13818343,"genie.py",5123,0,"",python,selection_keyboard +2048,13818407,"genie.py",5123,0,"d",python,content +2049,13818411,"genie.py",5124,0,"",python,selection_keyboard +2050,13818488,"genie.py",5124,0," ",python,content +2051,13818493,"genie.py",5125,0,"",python,selection_keyboard +2052,13818941,"genie.py",5125,0,"o",python,content +2053,13818945,"genie.py",5126,0,"",python,selection_keyboard +2054,13818996,"genie.py",5126,0,"m",python,content +2055,13818999,"genie.py",5127,0,"",python,selection_keyboard +2056,13819160,"genie.py",5127,0,"i",python,content +2057,13819163,"genie.py",5128,0,"",python,selection_keyboard +2058,13819216,"genie.py",5128,0,"t",python,content +2059,13819220,"genie.py",5129,0,"",python,selection_keyboard +2060,13819279,"genie.py",5129,0," ",python,content +2061,13819282,"genie.py",5130,0,"",python,selection_keyboard +2062,13819375,"genie.py",5130,0,"i",python,content +2063,13819379,"genie.py",5131,0,"",python,selection_keyboard +2064,13819457,"genie.py",5131,0,"n",python,content +2065,13819461,"genie.py",5132,0,"",python,selection_keyboard +2066,13819889,"genie.py",5130,2,"",python,content +2067,13820365,"genie.py",5130,0,"i",python,content +2068,13820370,"genie.py",5131,0,"",python,selection_keyboard +2069,13820488,"genie.py",5131,0,"n",python,content +2070,13820500,"genie.py",5132,0,"",python,selection_keyboard +2071,13820524,"genie.py",5132,0," ",python,content +2072,13820540,"genie.py",5133,0,"",python,selection_keyboard +2073,13820639,"genie.py",5133,0,"o",python,content +2074,13820641,"genie.py",5134,0,"",python,selection_keyboard +2075,13820681,"genie.py",5134,0,"r",python,content +2076,13820687,"genie.py",5135,0,"",python,selection_keyboard +2077,13820786,"genie.py",5135,0,"d",python,content +2078,13820790,"genie.py",5136,0,"",python,selection_keyboard +2079,13820898,"genie.py",5136,0,"e",python,content +2080,13820901,"genie.py",5137,0,"",python,selection_keyboard +2081,13820963,"genie.py",5137,0,"r",python,content +2082,13820967,"genie.py",5138,0,"",python,selection_keyboard +2083,13821012,"genie.py",5138,0," ",python,content +2084,13821020,"genie.py",5139,0,"",python,selection_keyboard +2085,13821143,"genie.py",5139,0,"t",python,content +2086,13821146,"genie.py",5140,0,"",python,selection_keyboard +2087,13821180,"genie.py",5140,0,"o",python,content +2088,13821181,"genie.py",5141,0,"",python,selection_keyboard +2089,13821248,"genie.py",5141,0," ",python,content +2090,13821252,"genie.py",5142,0,"",python,selection_keyboard +2091,13821346,"genie.py",5142,0,"t",python,content +2092,13821348,"genie.py",5143,0,"",python,selection_keyboard +2093,13821379,"genie.py",5143,0,"e",python,content +2094,13821380,"genie.py",5144,0,"",python,selection_keyboard +2095,13821455,"genie.py",5144,0,"s",python,content +2096,13821460,"genie.py",5145,0,"",python,selection_keyboard +2097,13821562,"genie.py",5145,0,"t",python,content +2098,13821569,"genie.py",5146,0,"",python,selection_keyboard +2099,13821662,"genie.py",5146,0," ",python,content +2100,13821673,"genie.py",5147,0,"",python,selection_keyboard +2101,13821896,"genie.py",5147,0,"c",python,content +2102,13821899,"genie.py",5148,0,"",python,selection_keyboard +2103,13821937,"genie.py",5148,0,"o",python,content +2104,13821941,"genie.py",5149,0,"",python,selection_keyboard +2105,13821981,"genie.py",5149,0,"m",python,content +2106,13821984,"genie.py",5150,0,"",python,selection_keyboard +2107,13822098,"genie.py",5150,0,"p",python,content +2108,13822101,"genie.py",5151,0,"",python,selection_keyboard +2109,13822157,"genie.py",5151,0,"u",python,content +2110,13822163,"genie.py",5152,0,"",python,selection_keyboard +2111,13822215,"genie.py",5152,0,"t",python,content +2112,13822218,"genie.py",5153,0,"",python,selection_keyboard +2113,13822339,"genie.py",5153,0,"e",python,content +2114,13822349,"genie.py",5154,0,"",python,selection_keyboard +2115,13822386,"genie.py",5154,0,"-",python,content +2116,13822390,"genie.py",5155,0,"",python,selection_keyboard +2117,13822511,"genie.py",5155,0,"b",python,content +2118,13822513,"genie.py",5156,0,"",python,selection_keyboard +2119,13822648,"genie.py",5156,0,"o",python,content +2120,13822652,"genie.py",5157,0,"",python,selection_keyboard +2121,13822745,"genie.py",5157,0,"u",python,content +2122,13822763,"genie.py",5158,0,"",python,selection_keyboard +2123,13822781,"genie.py",5158,0,"n",python,content +2124,13822782,"genie.py",5159,0,"",python,selection_keyboard +2125,13822812,"genie.py",5159,0,"d",python,content +2126,13822815,"genie.py",5160,0,"",python,selection_keyboard +2127,13822965,"genie.py",5160,0,"n",python,content +2128,13822968,"genie.py",5161,0,"",python,selection_keyboard +2129,13823023,"genie.py",5161,0,"e",python,content +2130,13823030,"genie.py",5162,0,"",python,selection_keyboard +2131,13823113,"genie.py",5162,0,"s",python,content +2132,13823117,"genie.py",5163,0,"",python,selection_keyboard +2133,13823168,"genie.py",5163,0,"s",python,content +2134,13823174,"genie.py",5164,0,"",python,selection_keyboard +2135,13823405,"genie.py",5163,0,"",python,selection_command +2136,13824280,"genie.py",5164,0,"",python,selection_command +2137,13824328,"genie.py",5164,0," ",python,content +2138,13824331,"genie.py",5165,0,"",python,selection_keyboard +2139,13824461,"genie.py",5165,0,"()",python,content +2140,13824467,"genie.py",5166,0,"",python,selection_keyboard +2141,13824500,"genie.py",5166,1,")",python,content +2142,13824507,"genie.py",5167,0,"",python,selection_keyboard +2143,13824677,"genie.py",5166,0,"",python,selection_command +2144,13824916,"genie.py",5166,0,"w",python,content +2145,13824920,"genie.py",5167,0,"",python,selection_keyboard +2146,13825168,"genie.py",5166,0,"",python,selection_command +2147,13825894,"genie.py",5165,0,"",python,selection_command +2148,13826042,"genie.py",5155,0,"",python,selection_command +2149,13826168,"genie.py",5154,0,"",python,selection_command +2150,13826431,"genie.py",5155,0,"",python,selection_command +2151,13826578,"genie.py",5154,0,"",python,selection_command +2152,13826727,"genie.py",5147,0,"",python,selection_command +2153,13827412,"genie.py",5147,0," ",python,content +2154,13827413,"genie.py",5148,0,"",python,selection_keyboard +2155,13827691,"genie.py",5147,1,"",python,content +2156,13827815,"genie.py",5147,0,"w",python,content +2157,13827817,"genie.py",5148,0,"",python,selection_keyboard +2158,13827897,"genie.py",5148,0,"h",python,content +2159,13827900,"genie.py",5149,0,"",python,selection_keyboard +2160,13827951,"genie.py",5149,0,"e",python,content +2161,13827952,"genie.py",5150,0,"",python,selection_keyboard +2162,13828091,"genie.py",5150,0,"t",python,content +2163,13828093,"genie.py",5151,0,"",python,selection_keyboard +2164,13828142,"genie.py",5151,0,"h",python,content +2165,13828146,"genie.py",5152,0,"",python,selection_keyboard +2166,13828265,"genie.py",5152,0,"e",python,content +2167,13828273,"genie.py",5153,0,"",python,selection_keyboard +2168,13828392,"genie.py",5153,0,"r",python,content +2169,13828403,"genie.py",5154,0,"",python,selection_keyboard +2170,13828462,"genie.py",5154,0," ",python,content +2171,13828465,"genie.py",5155,0,"",python,selection_keyboard +2172,13828603,"genie.py",5154,0,"",python,selection_command +2173,13828768,"genie.py",5155,0,"",python,selection_command +2174,13828852,"genie.py",5162,0,"",python,selection_command +2175,13829116,"genie.py",5162,14,"",python,content +2176,13829450,"genie.py",5161,0,"",python,selection_command +2177,13829566,"genie.py",5155,0,"",python,selection_command +2178,13829732,"genie.py",5155,7,"",python,content +2179,13830114,"genie.py",5155,0,"i",python,content +2180,13830118,"genie.py",5156,0,"",python,selection_keyboard +2181,13830190,"genie.py",5156,0,"n",python,content +2182,13830197,"genie.py",5157,0,"",python,selection_keyboard +2183,13830265,"genie.py",5157,0,"p",python,content +2184,13830268,"genie.py",5158,0,"",python,selection_keyboard +2185,13830360,"genie.py",5158,0,"u",python,content +2186,13830365,"genie.py",5159,0,"",python,selection_keyboard +2187,13830406,"genie.py",5159,0,"t",python,content +2188,13830413,"genie.py",5160,0,"",python,selection_keyboard +2189,13830561,"genie.py",5160,0,"-",python,content +2190,13830564,"genie.py",5161,0,"",python,selection_keyboard +2191,13830674,"genie.py",5161,0,"b",python,content +2192,13830678,"genie.py",5162,0,"",python,selection_keyboard +2193,13830795,"genie.py",5162,0,"o",python,content +2194,13830798,"genie.py",5163,0,"",python,selection_keyboard +2195,13830867,"genie.py",5163,0,"u",python,content +2196,13830883,"genie.py",5164,0,"",python,selection_keyboard +2197,13830919,"genie.py",5164,0,"n",python,content +2198,13830920,"genie.py",5165,0,"",python,selection_keyboard +2199,13830967,"genie.py",5165,0,"d",python,content +2200,13830973,"genie.py",5166,0,"",python,selection_keyboard +2201,13831105,"genie.py",5166,0,"n",python,content +2202,13831113,"genie.py",5167,0,"",python,selection_keyboard +2203,13831153,"genie.py",5167,0,"e",python,content +2204,13831156,"genie.py",5168,0,"",python,selection_keyboard +2205,13831267,"genie.py",5168,0,"s",python,content +2206,13831272,"genie.py",5169,0,"",python,selection_keyboard +2207,13831466,"genie.py",5169,0,"s",python,content +2208,13831469,"genie.py",5170,0,"",python,selection_keyboard +2209,13831542,"genie.py",5170,0," ",python,content +2210,13831546,"genie.py",5171,0,"",python,selection_keyboard +2211,13831668,"genie.py",5171,0,"i",python,content +2212,13831671,"genie.py",5172,0,"",python,selection_keyboard +2213,13831753,"genie.py",5172,0,"s",python,content +2214,13831762,"genie.py",5173,0,"",python,selection_keyboard +2215,13831832,"genie.py",5173,0," ",python,content +2216,13831836,"genie.py",5174,0,"",python,selection_keyboard +2217,13831948,"genie.py",5174,0,"t",python,content +2218,13831952,"genie.py",5175,0,"",python,selection_keyboard +2219,13832267,"genie.py",5175,0,"d",python,content +2220,13832270,"genie.py",5176,0,"",python,selection_keyboard +2221,13832395,"genie.py",5176,0,"u",python,content +2222,13832405,"genie.py",5177,0,"",python,selection_keyboard +2223,13832565,"genie.py",5176,1,"",python,content +2224,13832693,"genie.py",5175,1,"",python,content +2225,13832828,"genie.py",5174,1,"",python,content +2226,13833164,"genie.py",5174,0,"d",python,content +2227,13833166,"genie.py",5175,0,"",python,selection_keyboard +2228,13833265,"genie.py",5175,0,"u",python,content +2229,13833267,"genie.py",5176,0,"",python,selection_keyboard +2230,13833363,"genie.py",5176,0,"e",python,content +2231,13833365,"genie.py",5177,0,"",python,selection_keyboard +2232,13833427,"genie.py",5177,0," ",python,content +2233,13833430,"genie.py",5178,0,"",python,selection_keyboard +2234,13833550,"genie.py",5178,0,"t",python,content +2235,13833554,"genie.py",5179,0,"",python,selection_keyboard +2236,13833574,"genie.py",5179,0,"o",python,content +2237,13833580,"genie.py",5180,0,"",python,selection_keyboard +2238,13833698,"genie.py",5180,0," ",python,content +2239,13833702,"genie.py",5181,0,"",python,selection_keyboard +2240,13833848,"genie.py",5181,0,"u",python,content +2241,13833850,"genie.py",5182,0,"",python,selection_keyboard +2242,13833895,"genie.py",5182,0,"s",python,content +2243,13833897,"genie.py",5183,0,"",python,selection_keyboard +2244,13833981,"genie.py",5183,0," ",python,content +2245,13833983,"genie.py",5184,0,"",python,selection_keyboard +2246,13834164,"genie.py",5184,0,"b",python,content +2247,13834167,"genie.py",5185,0,"",python,selection_keyboard +2248,13834246,"genie.py",5185,0,"e",python,content +2249,13834251,"genie.py",5186,0,"",python,selection_keyboard +2250,13834379,"genie.py",5186,0,"i",python,content +2251,13834384,"genie.py",5187,0,"",python,selection_keyboard +2252,13834452,"genie.py",5187,0,"n",python,content +2253,13834454,"genie.py",5188,0,"",python,selection_keyboard +2254,13834522,"genie.py",5188,0,"g",python,content +2255,13834529,"genie.py",5189,0,"",python,selection_keyboard +2256,13834780,"genie.py",5189,0," ",python,content +2257,13834814,"genie.py",5190,0,"",python,selection_keyboard +2258,13834829,"genie.py",5190,0,"i",python,content +2259,13834831,"genie.py",5191,0,"",python,selection_keyboard +2260,13834849,"genie.py",5191,0,"n",python,content +2261,13834853,"genie.py",5192,0,"",python,selection_keyboard +2262,13834867,"genie.py",5192,0," ",python,content +2263,13834869,"genie.py",5193,0,"",python,selection_keyboard +2264,13834937,"genie.py",5193,0,"t",python,content +2265,13834942,"genie.py",5194,0,"",python,selection_keyboard +2266,13835007,"genie.py",5194,0,"h",python,content +2267,13835011,"genie.py",5195,0,"",python,selection_keyboard +2268,13835063,"genie.py",5195,0,"e",python,content +2269,13835066,"genie.py",5196,0,"",python,selection_keyboard +2270,13835095,"genie.py",5196,0," ",python,content +2271,13835097,"genie.py",5197,0,"",python,selection_keyboard +2272,13835229,"genie.py",5197,0,"v",python,content +2273,13835231,"genie.py",5198,0,"",python,selection_keyboard +2274,13835323,"genie.py",5198,0,"i",python,content +2275,13835347,"genie.py",5199,0,"",python,selection_keyboard +2276,13835429,"genie.py",5199,0,"d",python,content +2277,13835433,"genie.py",5200,0,"",python,selection_keyboard +2278,13835530,"genie.py",5200,0,"e",python,content +2279,13835534,"genie.py",5201,0,"",python,selection_keyboard +2280,13835630,"genie.py",5201,0,"o",python,content +2281,13835634,"genie.py",5202,0,"",python,selection_keyboard +2282,13835675,"genie.py",5202,0," ",python,content +2283,13835677,"genie.py",5203,0,"",python,selection_keyboard +2284,13836707,"genie.py",5203,0,"g",python,content +2285,13836709,"genie.py",5204,0,"",python,selection_keyboard +2286,13836801,"genie.py",5204,0,"e",python,content +2287,13836803,"genie.py",5205,0,"",python,selection_keyboard +2288,13836844,"genie.py",5205,0,"n",python,content +2289,13836845,"genie.py",5206,0,"",python,selection_keyboard +2290,13836912,"genie.py",5206,0,"e",python,content +2291,13836914,"genie.py",5207,0,"",python,selection_keyboard +2292,13836963,"genie.py",5207,0,"r",python,content +2293,13836972,"genie.py",5208,0,"",python,selection_keyboard +2294,13837073,"genie.py",5208,0,"a",python,content +2295,13837077,"genie.py",5209,0,"",python,selection_keyboard +2296,13837265,"genie.py",5209,0,"t",python,content +2297,13837268,"genie.py",5210,0,"",python,selection_keyboard +2298,13837379,"genie.py",5210,0,"i",python,content +2299,13837380,"genie.py",5211,0,"",python,selection_keyboard +2300,13837443,"genie.py",5211,0,"o",python,content +2301,13837446,"genie.py",5212,0,"",python,selection_keyboard +2302,13837512,"genie.py",5212,0,"n",python,content +2303,13837517,"genie.py",5213,0,"",python,selection_keyboard +2304,13837591,"genie.py",5213,0," ",python,content +2305,13837614,"genie.py",5214,0,"",python,selection_keyboard +2306,13837705,"genie.py",5214,0,"m",python,content +2307,13837707,"genie.py",5215,0,"",python,selection_keyboard +2308,13837753,"genie.py",5215,0,"o",python,content +2309,13837757,"genie.py",5216,0,"",python,selection_keyboard +2310,13837818,"genie.py",5216,0,"d",python,content +2311,13837824,"genie.py",5217,0,"",python,selection_keyboard +2312,13837880,"genie.py",5217,0,"e",python,content +2313,13837883,"genie.py",5218,0,"",python,selection_keyboard +2314,13837957,"genie.py",5218,0,"l",python,content +2315,13837961,"genie.py",5219,0,"",python,selection_keyboard +2316,13838023,"genie.py",5219,0," ",python,content +2317,13838027,"genie.py",5220,0,"",python,selection_keyboard +2318,13838148,"genie.py",5220,0,"r",python,content +2319,13838153,"genie.py",5221,0,"",python,selection_keyboard +2320,13838241,"genie.py",5221,0,"e",python,content +2321,13838247,"genie.py",5222,0,"",python,selection_keyboard +2322,13838381,"genie.py",5222,0,"g",python,content +2323,13838385,"genie.py",5223,0,"",python,selection_keyboard +2324,13838428,"genie.py",5223,0,"i",python,content +2325,13838431,"genie.py",5224,0,"",python,selection_keyboard +2326,13838473,"genie.py",5224,0,"m",python,content +2327,13838476,"genie.py",5225,0,"",python,selection_keyboard +2328,13838528,"genie.py",5225,0,"e",python,content +2329,13838531,"genie.py",5226,0,"",python,selection_keyboard +2330,13838679,"genie.py",5225,0,"",python,selection_command +2331,13838822,"genie.py",5220,0,"",python,selection_command +2332,13839034,"genie.py",5214,0,"",python,selection_command +2333,13839065,"genie.py",5203,0,"",python,selection_command +2334,13839120,"genie.py",5197,0,"",python,selection_command +2335,13839150,"genie.py",5193,0,"",python,selection_command +2336,13839171,"genie.py",5190,0,"",python,selection_command +2337,13839222,"genie.py",5184,0,"",python,selection_command +2338,13839256,"genie.py",5181,0,"",python,selection_command +2339,13839298,"genie.py",5178,0,"",python,selection_command +2340,13839317,"genie.py",5174,0,"",python,selection_command +2341,13839335,"genie.py",5171,0,"",python,selection_command +2342,13839360,"genie.py",5161,0,"",python,selection_command +2343,13839416,"genie.py",5160,0,"",python,selection_command +2344,13839437,"genie.py",5155,0,"",python,selection_command +2345,13839464,"genie.py",5147,0,"",python,selection_command +2346,13839519,"genie.py",5142,0,"",python,selection_command +2347,13839560,"genie.py",5139,0,"",python,selection_command +2348,13839586,"genie.py",5133,0,"",python,selection_command +2349,13839632,"genie.py",5130,0,"",python,selection_command +2350,13839688,"genie.py",5125,0,"",python,selection_command +2351,13839698,"genie.py",5118,0,"",python,selection_command +2352,13839730,"genie.py",5123,0,"",python,selection_command +2353,13839999,"genie.py",5128,0,"",python,selection_command +2354,13840019,"genie.py",5131,0,"",python,selection_command +2355,13840076,"genie.py",5137,0,"",python,selection_command +2356,13840335,"genie.py",5140,0,"",python,selection_command +2357,13840496,"genie.py",5145,0,"",python,selection_command +2358,13840648,"genie.py",5146,0,"",python,selection_command +2359,13840781,"genie.py",5146,0,"\n # ",python,content +2360,13840929,"genie.py",5156,0,"",python,selection_command +2361,13841264,"genie.py",5157,0,"",python,selection_command +2362,13841661,"genie.py",5156,0,"",python,selection_command +2363,13841740,"genie.py",5156,1,"",python,content +2364,13853631,"genie.py",5487,0,"",python,selection_mouse +2365,13853645,"genie.py",5486,0,"",python,selection_command +2366,13854468,"genie.py",5415,0,"",python,selection_command +2367,13854632,"genie.py",5358,0,"",python,selection_command +2368,13854761,"genie.py",5275,0,"",python,selection_command +2369,13855745,"genie.py",5487,0,"",python,selection_mouse +2370,13855756,"genie.py",5486,0,"",python,selection_command +2371,13856133,"genie.py",5415,0,"",python,selection_command +2372,13856270,"genie.py",5358,0,"",python,selection_command +2373,13856396,"genie.py",5275,0,"",python,selection_command +2374,13856533,"genie.py",5185,0,"",python,selection_command +2375,13856782,"genie.py",5147,0,"",python,selection_command +2376,13870998,"genie.py",5237,0,"",python,selection_command +2377,13876260,"genie.py",5245,0,"",python,selection_command +2378,13876322,"genie.py",5245,0,"#",python,content +2379,13876325,"genie.py",5246,0,"",python,selection_keyboard +2380,13876414,"genie.py",5246,0," ",python,content +2381,13876419,"genie.py",5247,0,"",python,selection_keyboard +2382,13876598,"genie.py",5246,0,"",python,selection_command +2383,13882029,"genie.py",5330,0,"",python,selection_command +2384,13882079,"genie.py",5330,0,"#",python,content +2385,13882083,"genie.py",5331,0,"",python,selection_keyboard +2386,13882113,"genie.py",5331,0," ",python,content +2387,13882116,"genie.py",5332,0,"",python,selection_keyboard +2388,13882305,"genie.py",5331,0,"",python,selection_command +2389,13882438,"genie.py",5380,0,"\n ",python,content +2390,13883541,"genie.py",5389,0,"t",python,content +2391,13883548,"genie.py",5390,0,"",python,selection_keyboard +2392,13883645,"genie.py",5390,0,"o",python,content +2393,13883652,"genie.py",5391,0,"",python,selection_keyboard +2394,13883712,"genie.py",5391,0,"k",python,content +2395,13883720,"genie.py",5392,0,"",python,selection_keyboard +2396,13883789,"genie.py",5392,0,"e",python,content +2397,13883799,"genie.py",5393,0,"",python,selection_keyboard +2398,13883947,"genie.py",5393,0,"n",python,content +2399,13883952,"genie.py",5394,0,"",python,selection_keyboard +2400,13884453,"genie.py",5394,0,"_indices_BTN = jnp.zeros((videos_BTHWC.shape[0], videos_BTHWC.shape[1], 1024), dtype=jnp.int32)",python,content +2401,13884687,"genie.py",5488,0,"",python,selection_command +2402,13885815,"genie.py",5381,0,"",python,selection_command +2403,13885913,"genie.py",5389,0,"",python,selection_command +2404,13898731,"genie.py",6365,0,"",python,selection_command +2405,13901149,"genie.py",6364,0,"",python,selection_command +2406,13901398,"genie.py",6360,0,"",python,selection_command +2407,13901432,"genie.py",6358,0,"",python,selection_command +2408,13901567,"genie.py",6355,0,"",python,selection_command +2409,13901696,"genie.py",6350,0,"",python,selection_command +2410,13901841,"genie.py",6348,0,"",python,selection_command +2411,13902267,"genie.py",6341,0,"",python,selection_command +2412,13906286,"train_dynamics.py",0,0,"",python,tab +2413,13907663,"train_dynamics.py",3283,0,"",python,selection_command +2414,13912102,"train_dynamics.py",3300,0,"",python,selection_command +2415,13912929,"train_dynamics.py",3341,0,"",python,selection_command +2416,13913227,"train_dynamics.py",3283,0,"",python,selection_command +2417,13913986,"train_dynamics.py",3364,0,"",python,selection_command +2418,13927697,"train_dynamics.py",3415,0,"",python,selection_command +2419,13929805,"train_dynamics.py",3447,0,"",python,selection_command +2420,13930543,"train_dynamics.py",4313,0,"",python,selection_command +2421,13936611,"train_dynamics.py",3447,0,"",python,selection_command +2422,13937044,"train_dynamics.py",3396,0,"",python,selection_command +2423,13937343,"train_dynamics.py",3315,0,"",python,selection_command +2424,13939347,"train_dynamics.py",3283,0,"",python,selection_command +2425,13939439,"train_dynamics.py",3283,0,"#",python,content +2426,13939446,"train_dynamics.py",3284,0,"",python,selection_keyboard +2427,13939473,"train_dynamics.py",3284,0," ",python,content +2428,13939477,"train_dynamics.py",3285,0,"",python,selection_keyboard +2429,13939772,"train_dynamics.py",3284,0,"",python,selection_command +2430,13941622,"train_dynamics.py",3418,0,"",python,selection_command +2431,13941897,"train_dynamics.py",3367,0,"",python,selection_command +2432,13942254,"train_dynamics.py",3366,0,"",python,selection_command +2433,13942313,"train_dynamics.py",3366,0,"#",python,content +2434,13942316,"train_dynamics.py",3367,0,"",python,selection_keyboard +2435,13942368,"train_dynamics.py",3367,0," ",python,content +2436,13942371,"train_dynamics.py",3368,0,"",python,selection_keyboard +2437,13942503,"train_dynamics.py",3367,0,"",python,selection_command +2438,13942628,"train_dynamics.py",3420,0,"",python,selection_command +2439,13942787,"train_dynamics.py",3419,0,"",python,selection_command +2440,13942914,"train_dynamics.py",3419,0,"#",python,content +2441,13942916,"train_dynamics.py",3420,0,"",python,selection_keyboard +2442,13942962,"train_dynamics.py",3420,0," ",python,content +2443,13942964,"train_dynamics.py",3421,0,"",python,selection_keyboard +2444,13943144,"train_dynamics.py",3420,0,"",python,selection_command +2445,13947565,"train_dynamics.py",4165,0," #",python,content +2446,13947565,"train_dynamics.py",4147,0,"# ",python,content +2447,13951145,"train_dynamics.py",4330,0,"e",python,content +2448,13951145,"train_dynamics.py",4328,2,"",python,content +2449,13951145,"train_dynamics.py",4326,0,"N",python,content +2450,13951145,"train_dynamics.py",4314,12,"",python,content +2451,13952907,"train_dynamics.py",3283,0,"",python,selection_command +2452,13953345,"train_dynamics.py",3233,0,"",python,selection_command +2453,13953594,"train_dynamics.py",3174,0,"",python,selection_command +2454,13953628,"train_dynamics.py",3132,0,"",python,selection_command +2455,13953662,"train_dynamics.py",3060,0,"",python,selection_command +2456,13953696,"train_dynamics.py",3010,0,"",python,selection_command +2457,13953729,"train_dynamics.py",3004,0,"",python,selection_command +2458,13953765,"train_dynamics.py",2947,0,"",python,selection_command +2459,13953796,"train_dynamics.py",2884,0,"",python,selection_command +2460,13953829,"train_dynamics.py",2810,0,"",python,selection_command +2461,13953862,"train_dynamics.py",2783,0,"",python,selection_command +2462,13953896,"train_dynamics.py",2740,0,"",python,selection_command +2463,13953932,"train_dynamics.py",2722,0,"",python,selection_command +2464,13953962,"train_dynamics.py",2677,0,"",python,selection_command +2465,13953996,"train_dynamics.py",2611,0,"",python,selection_command +2466,13954032,"train_dynamics.py",2572,0,"",python,selection_command +2467,13954070,"train_dynamics.py",2525,0,"",python,selection_command +2468,13954097,"train_dynamics.py",2494,0,"",python,selection_command +2469,13954237,"train_dynamics.py",2472,0,"",python,selection_command +2470,13954600,"train_dynamics.py",4587,0,"",python,selection_command +2471,13962396,"train_dynamics.py",4323,0,"",python,selection_command +2472,13969004,"train_dynamics.py",4587,0,"",python,selection_command +2473,13969678,"train_dynamics.py",4323,0,"",python,selection_command +2474,13973269,"train_dynamics.py",4587,0,"",python,selection_command +2475,13981416,"train_dynamics.py",4512,0,"",python,selection_command +2476,13981579,"train_dynamics.py",4505,0,"",python,selection_command +2477,13984132,"train_dynamics.py",4580,0,"",python,selection_command +2478,13984278,"train_dynamics.py",4619,0,"",python,selection_command +2479,13984393,"train_dynamics.py",4628,0,"",python,selection_command +2480,13984510,"train_dynamics.py",4629,0,"",python,selection_command +2481,13984815,"train_dynamics.py",4631,0,"",python,selection_command +2482,13985006,"train_dynamics.py",4632,0,"",python,selection_command +2483,13987195,"train_dynamics.py",4898,0,"",python,selection_command +2484,13989908,"train_dynamics.py",4879,0,"",python,selection_command +2485,13990156,"train_dynamics.py",4829,0,"",python,selection_command +2486,13990193,"train_dynamics.py",4779,0,"",python,selection_command +2487,13990226,"train_dynamics.py",4752,0,"",python,selection_command +2488,13990256,"train_dynamics.py",4724,0,"",python,selection_command +2489,13990296,"train_dynamics.py",4637,0,"",python,selection_command +2490,13990327,"train_dynamics.py",4619,0,"",python,selection_command +2491,13990362,"train_dynamics.py",4589,0,"",python,selection_command +2492,13990394,"train_dynamics.py",4514,0,"",python,selection_command +2493,13990426,"train_dynamics.py",4496,0,"",python,selection_command +2494,13990465,"train_dynamics.py",4470,0,"",python,selection_command +2495,13990613,"train_dynamics.py",4430,0,"",python,selection_command +2496,13990764,"train_dynamics.py",4373,0,"",python,selection_command +2497,13990916,"train_dynamics.py",4354,0,"",python,selection_command +2498,13991132,"train_dynamics.py",4344,0,"",python,selection_command +2499,13991354,"train_dynamics.py",12498,0,"",python,selection_command +2500,13992475,"train_dynamics.py",12496,0,"",python,selection_command +2501,13992614,"train_dynamics.py",12488,0,"",python,selection_command +2502,13992744,"train_dynamics.py",12486,0,"",python,selection_command +2503,13992893,"train_dynamics.py",12481,0,"",python,selection_command +2504,13995390,"train_dynamics.py",4632,0,"",python,selection_command +2505,13997398,"train_dynamics.py",12481,0,"",python,selection_command +2506,14000126,"train_dynamics.py",13189,0,"",python,selection_command +2507,14003277,"train_dynamics.py",13177,0,"",python,selection_command +2508,14005201,"train_dynamics.py",13100,0,"",python,selection_command +2509,14005339,"train_dynamics.py",13044,0,"",python,selection_command +2510,14007162,"train_dynamics.py",13100,1,"g",python,selection_command +2511,14007326,"train_dynamics.py",13100,1,"g",python,selection_command +2512,14007572,"train_dynamics.py",13100,1,"g",python,selection_command +2513,14007613,"train_dynamics.py",13100,1,"g",python,selection_command +2514,14007645,"train_dynamics.py",13100,1,"g",python,selection_command +2515,14007677,"train_dynamics.py",13100,1,"g",python,selection_command +2516,14007716,"train_dynamics.py",13100,1,"g",python,selection_command +2517,14007760,"train_dynamics.py",13100,1,"g",python,selection_command +2518,14007785,"train_dynamics.py",13100,1,"g",python,selection_command +2519,14007825,"train_dynamics.py",13100,1,"g",python,selection_command +2520,14007853,"train_dynamics.py",13100,1,"g",python,selection_command +2521,14007889,"train_dynamics.py",13100,1,"g",python,selection_command +2522,14007917,"train_dynamics.py",13100,1,"g",python,selection_command +2523,14007950,"train_dynamics.py",13100,1,"g",python,selection_command +2524,14008065,"train_dynamics.py",13100,1,"g",python,selection_command +2525,14008265,"train_dynamics.py",13100,1,"g",python,selection_command +2526,14008541,"train_dynamics.py",13100,1,"g",python,selection_command +2527,14008759,"train_dynamics.py",13100,0,"",python,selection_command +2528,14008867,"train_dynamics.py",13910,0,"#",python,content +2529,14008868,"train_dynamics.py",13884,0,"#",python,content +2530,14008868,"train_dynamics.py",13853,0,"#",python,content +2531,14008868,"train_dynamics.py",13777,0,"#",python,content +2532,14008868,"train_dynamics.py",13722,0,"#",python,content +2533,14008868,"train_dynamics.py",13634,0,"#",python,content +2534,14008868,"train_dynamics.py",13549,0,"#",python,content +2535,14008868,"train_dynamics.py",13506,0,"#",python,content +2536,14008868,"train_dynamics.py",13457,0,"#",python,content +2537,14008868,"train_dynamics.py",13435,0,"#",python,content +2538,14008868,"train_dynamics.py",13366,0,"#",python,content +2539,14008868,"train_dynamics.py",13311,0,"#",python,content +2540,14008868,"train_dynamics.py",13229,0,"#",python,content +2541,14008868,"train_dynamics.py",13177,0,"#",python,content +2542,14008868,"train_dynamics.py",13100,0,"#",python,content +2543,14008877,"train_dynamics.py",13101,0,"",python,selection_keyboard +2544,14008956,"train_dynamics.py",13925,0," ",python,content +2545,14008956,"train_dynamics.py",13898,0," ",python,content +2546,14008956,"train_dynamics.py",13866,0," ",python,content +2547,14008956,"train_dynamics.py",13789,0," ",python,content +2548,14008956,"train_dynamics.py",13733,0," ",python,content +2549,14008956,"train_dynamics.py",13644,0," ",python,content +2550,14008956,"train_dynamics.py",13558,0," ",python,content +2551,14008957,"train_dynamics.py",13514,0," ",python,content +2552,14008957,"train_dynamics.py",13464,0," ",python,content +2553,14008957,"train_dynamics.py",13441,0," ",python,content +2554,14008957,"train_dynamics.py",13371,0," ",python,content +2555,14008957,"train_dynamics.py",13315,0," ",python,content +2556,14008957,"train_dynamics.py",13232,0," ",python,content +2557,14008957,"train_dynamics.py",13179,0," ",python,content +2558,14008957,"train_dynamics.py",13101,0," ",python,content +2559,14008966,"train_dynamics.py",13102,0,"",python,selection_keyboard +2560,14009160,"train_dynamics.py",13101,0,"",python,selection_command +2561,14013364,"train_dynamics.py",13045,0,"",python,selection_command +2562,14013615,"train_dynamics.py",13022,0,"",python,selection_command +2563,14013647,"train_dynamics.py",12997,0,"",python,selection_command +2564,14013678,"train_dynamics.py",12958,0,"",python,selection_command +2565,14013711,"train_dynamics.py",12916,0,"",python,selection_command +2566,14013755,"train_dynamics.py",12874,0,"",python,selection_command +2567,14013798,"train_dynamics.py",12848,0,"",python,selection_command +2568,14013822,"train_dynamics.py",12817,0,"",python,selection_command +2569,14013865,"train_dynamics.py",12738,0,"",python,selection_command +2570,14013910,"train_dynamics.py",12713,0,"",python,selection_command +2571,14014019,"train_dynamics.py",12683,0,"",python,selection_command +2572,14014020,"train_dynamics.py",12661,0,"",python,selection_command +2573,14014038,"train_dynamics.py",12659,0,"",python,selection_command +2574,14014143,"train_dynamics.py",12612,0,"",python,selection_command +2575,14014287,"train_dynamics.py",12566,0,"",python,selection_command +2576,14014528,"train_dynamics.py",12484,0,"",python,selection_command +2577,14014795,"train_dynamics.py",12486,0,"",python,selection_command +2578,14015129,"train_dynamics.py",12488,0,"",python,selection_command +2579,14016816,"train_dynamics.py",12557,0,"",python,selection_command +2580,14017330,"train_dynamics.py",12967,0,"",python,selection_command +2581,14044458,"train_dynamics.py",13000,0,"",python,selection_command +2582,14044599,"train_dynamics.py",13022,0,"",python,selection_command +2583,14044752,"train_dynamics.py",13054,0,"",python,selection_command +2584,14044911,"train_dynamics.py",13110,0,"",python,selection_command +2585,14045224,"train_dynamics.py",13054,0,"",python,selection_command +2586,14046174,"train_dynamics.py",13079,0,"\n ",python,content +2587,14046324,"train_dynamics.py",13100,0,"p",python,content +2588,14046326,"train_dynamics.py",13101,0,"",python,selection_keyboard +2589,14046394,"train_dynamics.py",13101,0,"a",python,content +2590,14046398,"train_dynamics.py",13102,0,"",python,selection_keyboard +2591,14046481,"train_dynamics.py",13102,0,"s",python,content +2592,14046483,"train_dynamics.py",13103,0,"",python,selection_keyboard +2593,14046612,"train_dynamics.py",13103,0,"s",python,content +2594,14046614,"train_dynamics.py",13104,0,"",python,selection_keyboard +2595,14047013,"train_dynamics.py",13103,0,"",python,selection_command +2596,14050290,"train_dynamics.py",14294,0,"",python,selection_command +2597,14050911,"train_dynamics.py",14728,0,"",python,selection_command +2598,14051546,"train_dynamics.py",13902,0,"",python,selection_command +2599,14051811,"train_dynamics.py",14728,0,"",python,selection_command +2600,14052313,"train_dynamics.py",13902,0,"",python,selection_command +2601,14053692,"train_dynamics.py",13824,0,"",python,selection_command +2602,14053933,"train_dynamics.py",13767,0,"",python,selection_command +2603,14053970,"train_dynamics.py",13677,0,"",python,selection_command +2604,14054001,"train_dynamics.py",13590,0,"",python,selection_command +2605,14054396,"train_dynamics.py",13545,0,"",python,selection_command +2606,14054651,"train_dynamics.py",13494,0,"",python,selection_command +2607,14054677,"train_dynamics.py",13470,0,"",python,selection_command +2608,14054710,"train_dynamics.py",13399,0,"",python,selection_command +2609,14054743,"train_dynamics.py",13342,0,"",python,selection_command +2610,14054777,"train_dynamics.py",13258,0,"",python,selection_command +2611,14054810,"train_dynamics.py",13204,0,"",python,selection_command +2612,14054843,"train_dynamics.py",13125,0,"",python,selection_command +2613,14054876,"train_dynamics.py",13100,0,"",python,selection_command +2614,14054910,"train_dynamics.py",13044,0,"",python,selection_command +2615,14054949,"train_dynamics.py",13022,0,"",python,selection_command +2616,14054991,"train_dynamics.py",12996,0,"",python,selection_command +2617,14055040,"train_dynamics.py",12957,0,"",python,selection_command +2618,14055055,"train_dynamics.py",12915,0,"",python,selection_command +2619,14055091,"train_dynamics.py",12873,0,"",python,selection_command +2620,14055130,"train_dynamics.py",12847,0,"",python,selection_command +2621,14055167,"train_dynamics.py",12816,0,"",python,selection_command +2622,14055240,"train_dynamics.py",12737,0,"",python,selection_command +2623,14055257,"train_dynamics.py",12712,0,"",python,selection_command +2624,14055270,"train_dynamics.py",12682,0,"",python,selection_command +2625,14055293,"train_dynamics.py",12661,0,"",python,selection_command +2626,14055313,"train_dynamics.py",12659,0,"",python,selection_command +2627,14055355,"train_dynamics.py",12611,0,"",python,selection_command +2628,14055468,"train_dynamics.py",12565,0,"",python,selection_command +2629,14055653,"train_dynamics.py",12483,0,"",python,selection_command +2630,14072307,"train_dynamics.py",12463,0,"",python,selection_command +2631,14164063,"train_dynamics.py",12813,0,"",python,selection_mouse +2632,14165332,"train_dynamics.py",12826,0,"",python,selection_mouse +2633,14165339,"train_dynamics.py",12825,0,"",python,selection_command +2634,14167440,"train_dynamics.py",12507,0,"",python,selection_mouse +2635,14168239,"train_dynamics.py",12504,0,"",python,selection_mouse +2636,14172055,"train_dynamics.py",12540,0,"",python,selection_mouse +2637,14189707,"train_dynamics.py",12414,0,"",python,selection_command +2638,14190180,"train_dynamics.py",12421,0,"",python,selection_command +2639,14190360,"train_dynamics.py",12423,0,"",python,selection_command +2640,14190817,"train_dynamics.py",12421,0,"",python,selection_command +2641,14190958,"train_dynamics.py",12414,0,"",python,selection_command +2642,14191193,"train_dynamics.py",12421,0,"",python,selection_command +2643,14191412,"train_dynamics.py",12423,0,"",python,selection_command +2644,14191666,"train_dynamics.py",12427,0,"",python,selection_command +2645,14191808,"train_dynamics.py",12428,0,"",python,selection_command +2646,14193331,"train_dynamics.py",12434,0,"",python,selection_command +2647,14193487,"train_dynamics.py",12435,0,"",python,selection_command +2648,14195035,"train_dynamics.py",12434,0,"",python,selection_command +2649,14195159,"train_dynamics.py",12428,0,"",python,selection_command +2650,14195460,"train_dynamics.py",12428,1,"v",python,selection_command +2651,14195517,"train_dynamics.py",12428,6,"videos",python,selection_command +2652,14195677,"train_dynamics.py",12428,7,"videos=",python,selection_command +2653,14195825,"train_dynamics.py",12428,13,"videos=videos",python,selection_command +2654,14196587,"train_dynamics.py",12428,14,"videos=videos,",python,selection_command +2655,14196727,"train_dynamics.py",12428,15,"videos=videos, ",python,selection_command +2656,14196803,"train_dynamics.py",12428,15,"",python,content +2657,14197929,"train_dynamics.py",12374,0,"",python,selection_command +2658,14198144,"train_dynamics.py",12341,0,"",python,selection_command +2659,14198289,"train_dynamics.py",12307,0,"",python,selection_command +2660,14198704,"train_dynamics.py",12289,0,"",python,selection_command +2661,14198874,"train_dynamics.py",12293,0,"",python,selection_command +2662,14204292,"train_dynamics.py",12293,6,"",python,content +2663,14204555,"train_dynamics.py",12293,0,"_",python,content +2664,14204557,"train_dynamics.py",12294,0,"",python,selection_keyboard +2665,14204890,"train_dynamics.py",12293,0,"",python,selection_command +2666,14206374,"train_dynamics.py",12322,0,"",python,selection_command +2667,14206807,"train_dynamics.py",12293,1,"videos",python,content +2668,14206811,"train_dynamics.py",12293,0,"",python,selection_command +2669,14207402,"train_dynamics.py",12428,0,"videos=videos, ",python,content +2670,14207413,"train_dynamics.py",12428,0,"",python,selection_command +2671,14225293,"train_dynamics.py",12428,1,"v",python,selection_command +2672,14225330,"train_dynamics.py",12428,6,"videos",python,selection_command +2673,14225491,"train_dynamics.py",12428,7,"videos=",python,selection_command +2674,14225811,"train_dynamics.py",12428,8,"videos=v",python,selection_command +2675,14225960,"train_dynamics.py",12428,9,"videos=vi",python,selection_command +2676,14226211,"train_dynamics.py",12428,10,"videos=vid",python,selection_command +2677,14226240,"train_dynamics.py",12428,11,"videos=vide",python,selection_command +2678,14226394,"train_dynamics.py",12428,12,"videos=video",python,selection_command +2679,14226559,"train_dynamics.py",12428,13,"videos=videos",python,selection_command +2680,14226727,"train_dynamics.py",12428,14,"videos=videos,",python,selection_command +2681,14226862,"train_dynamics.py",12428,15,"videos=videos, ",python,selection_command +2682,14227022,"train_dynamics.py",12428,15,"",python,content +2683,14227740,"train_dynamics.py",12427,0,"",python,selection_command +2684,14227888,"train_dynamics.py",12423,0,"",python,selection_command +2685,14228031,"train_dynamics.py",12421,0,"",python,selection_command +2686,14228171,"train_dynamics.py",12414,0,"",python,selection_command +2687,14230220,"train_dynamics.py",12522,0,"",python,selection_command +2688,14230544,"train_dynamics.py",12520,0,"",python,selection_command +2689,14230796,"train_dynamics.py",12511,0,"",python,selection_command +2690,14230828,"train_dynamics.py",12509,0,"",python,selection_command +2691,14230860,"train_dynamics.py",12504,0,"",python,selection_command +2692,14230894,"train_dynamics.py",12503,0,"",python,selection_command +2693,14230927,"train_dynamics.py",12494,0,"",python,selection_command +2694,14230961,"train_dynamics.py",12493,0,"",python,selection_command +2695,14230999,"train_dynamics.py",12483,0,"",python,selection_command +2696,14231032,"train_dynamics.py",12481,0,"",python,selection_command +2697,14231432,"train_dynamics.py",12483,0,"",python,selection_command +2698,14232743,"train_dynamics.py",12437,0,"",python,selection_command +2699,14233473,"train_dynamics.py",12383,0,"",python,selection_command +2700,14233623,"train_dynamics.py",12346,0,"",python,selection_command +2701,14233772,"train_dynamics.py",12313,0,"",python,selection_command +2702,14234004,"train_dynamics.py",12303,0,"",python,selection_command +2703,14234272,"train_dynamics.py",12300,0,"",python,selection_command +2704,14234451,"train_dynamics.py",12293,0,"",python,selection_command +2705,14235009,"train_dynamics.py",12300,0,"",python,selection_command +2706,14235186,"train_dynamics.py",12303,0,"",python,selection_command +2707,14237758,"train_dynamics.py",12270,0,"",python,selection_command +2708,14238016,"train_dynamics.py",12218,0,"",python,selection_command +2709,14238044,"train_dynamics.py",12194,0,"",python,selection_command +2710,14238075,"train_dynamics.py",12177,0,"",python,selection_command +2711,14238387,"train_dynamics.py",12107,0,"",python,selection_command +2712,14239745,"train_dynamics.py",12131,0,"",python,selection_command +2713,14239888,"train_dynamics.py",12132,0,"",python,selection_command +2714,14240534,"train_dynamics.py",12085,0,"",python,selection_command +2715,14240606,"train_dynamics.py",12093,0,"",python,selection_command +2716,14240970,"train_dynamics.py",12096,0,"",python,selection_command +2717,14241154,"train_dynamics.py",12097,0,"",python,selection_command +2718,14241307,"train_dynamics.py",12131,0,"",python,selection_command +2719,14242027,"train_dynamics.py",12132,0,"",python,selection_command +2720,14242791,"train_dynamics.py",12147,0,"",python,selection_command +2721,14243004,"train_dynamics.py",12149,0,"",python,selection_command +2722,14246022,"train_dynamics.py",12083,0,"",python,selection_command +2723,14246191,"train_dynamics.py",12064,0,"",python,selection_command +2724,14246466,"train_dynamics.py",12083,0,"",python,selection_command +2725,14246803,"train_dynamics.py",12070,0,"",python,selection_command +2726,14247243,"train_dynamics.py",12070,1,"d",python,selection_command +2727,14247416,"train_dynamics.py",12070,1,"d",python,selection_command +2728,14247562,"train_dynamics.py",12070,1,"d",python,selection_command +2729,14247861,"train_dynamics.py",12070,1,"d",python,selection_command +2730,14248171,"train_dynamics.py",12070,0,"",python,selection_command +2731,14248270,"train_dynamics.py",12194,0,"#",python,content +2732,14248270,"train_dynamics.py",12159,0,"#",python,content +2733,14248270,"train_dynamics.py",12089,0,"#",python,content +2734,14248270,"train_dynamics.py",12070,0,"#",python,content +2735,14248273,"train_dynamics.py",12071,0,"",python,selection_keyboard +2736,14248309,"train_dynamics.py",12198,0," ",python,content +2737,14248309,"train_dynamics.py",12162,0," ",python,content +2738,14248309,"train_dynamics.py",12091,0," ",python,content +2739,14248309,"train_dynamics.py",12071,0," ",python,content +2740,14248311,"train_dynamics.py",12072,0,"",python,selection_keyboard +2741,14248501,"train_dynamics.py",12071,0,"",python,selection_command +2742,14252522,"train_dynamics.py",12092,0,"",python,selection_command +2743,14252708,"train_dynamics.py",12164,0,"",python,selection_command +2744,14252840,"train_dynamics.py",12201,0,"",python,selection_command +2745,14252984,"train_dynamics.py",12209,0,"",python,selection_command +2746,14253137,"train_dynamics.py",12261,0,"",python,selection_command +2747,14253273,"train_dynamics.py",12294,0,"",python,selection_command +2748,14253406,"train_dynamics.py",12297,0,"",python,selection_command +2749,14253790,"train_dynamics.py",12301,0,"",python,selection_command +2750,14256806,"train_dynamics.py",12301,6,"",python,content +2751,14257039,"train_dynamics.py",12301,0,"_",python,content +2752,14257041,"train_dynamics.py",12302,0,"",python,selection_keyboard +2753,14257389,"train_dynamics.py",12301,0,"",python,selection_command +2754,14257576,"train_dynamics.py",12302,0,"",python,selection_command +2755,14257787,"train_dynamics.py",12303,0,"",python,selection_command +2756,14257955,"train_dynamics.py",12304,0,"",python,selection_command +2757,14258124,"train_dynamics.py",12305,0,"",python,selection_command +2758,14258354,"train_dynamics.py",12306,0,"",python,selection_command +2759,14258691,"train_dynamics.py",12306,10,"",python,content +2760,14259008,"train_dynamics.py",12306,0,"r",python,content +2761,14259010,"train_dynamics.py",12307,0,"",python,selection_keyboard +2762,14259083,"train_dynamics.py",12307,0,"a",python,content +2763,14259089,"train_dynamics.py",12308,0,"",python,selection_keyboard +2764,14259170,"train_dynamics.py",12308,0,"n",python,content +2765,14259197,"train_dynamics.py",12309,0,"",python,selection_keyboard +2766,14259326,"train_dynamics.py",12309,0,"g",python,content +2767,14259331,"train_dynamics.py",12310,0,"",python,selection_keyboard +2768,14259421,"train_dynamics.py",12310,0,"e",python,content +2769,14259423,"train_dynamics.py",12311,0,"",python,selection_keyboard +2770,14259675,"train_dynamics.py",12311,0,"()",python,content +2771,14259682,"train_dynamics.py",12312,0,"",python,selection_keyboard +2772,14260223,"train_dynamics.py",12312,0,"1",python,content +2773,14260229,"train_dynamics.py",12313,0,"",python,selection_keyboard +2774,14260372,"train_dynamics.py",12313,0,"0",python,content +2775,14260388,"train_dynamics.py",12314,0,"",python,selection_keyboard +2776,14260625,"train_dynamics.py",12314,0,"0",python,content +2777,14260629,"train_dynamics.py",12315,0,"",python,selection_keyboard +2778,14260798,"train_dynamics.py",12314,0,"",python,selection_command +2779,14260815,"train_dynamics.py",12289,0,"",python,selection_command +2780,14262044,"train_dynamics.py",12316,0,"",python,selection_command +2781,14262374,"train_dynamics.py",12315,0,"",python,selection_command +2782,14262939,"train_dynamics.py",12315,0,"0",python,content +2783,14262941,"train_dynamics.py",12316,0,"",python,selection_keyboard +2784,14263106,"train_dynamics.py",12316,0,"0",python,content +2785,14263108,"train_dynamics.py",12317,0,"",python,selection_keyboard +2786,14263256,"train_dynamics.py",12317,0,"0",python,content +2787,14263258,"train_dynamics.py",12318,0,"",python,selection_keyboard +2788,14263401,"train_dynamics.py",12317,0,"",python,selection_command +2789,14266847,"train_dynamics.py",12349,0,"",python,selection_command +2790,14267175,"train_dynamics.py",12382,0,"",python,selection_command +2791,14267474,"train_dynamics.py",12436,0,"",python,selection_command +2792,14267726,"train_dynamics.py",12482,0,"",python,selection_command +2793,14268072,"train_dynamics.py",12487,0,"",python,selection_command +2794,14268256,"train_dynamics.py",12489,0,"",python,selection_command +2795,14268647,"train_dynamics.py",4344,0,"",python,selection_command +2796,14269751,"train_dynamics.py",4360,0,"",python,selection_command +2797,14269923,"train_dynamics.py",4365,0,"",python,selection_command +2798,14270458,"train_dynamics.py",4411,0,"",python,selection_command +2799,14270704,"train_dynamics.py",4408,0,"",python,selection_command +2800,14270803,"train_dynamics.py",4406,0,"",python,selection_command +2801,14270902,"train_dynamics.py",4400,0,"",python,selection_command +2802,14271640,"train_dynamics.py",4611,0,"",python,selection_command +2803,14273807,"train_dynamics.py",4580,0,"",python,selection_command +2804,14274022,"train_dynamics.py",4587,0,"",python,selection_command +2805,14274207,"train_dynamics.py",4603,0,"",python,selection_command +2806,14274468,"train_dynamics.py",4587,0,"",python,selection_command +2807,14274630,"train_dynamics.py",2472,0,"",python,selection_command +2808,14276233,"train_dynamics.py",2508,0,"",python,selection_command +2809,14277246,"train_dynamics.py",2628,0,"",python,selection_command +2810,14281494,"train_dynamics.py",2611,0,"",python,selection_command +2811,14281547,"train_dynamics.py",2611,0,"#",python,content +2812,14281551,"train_dynamics.py",2612,0,"",python,selection_keyboard +2813,14281608,"train_dynamics.py",2612,0," ",python,content +2814,14281611,"train_dynamics.py",2613,0,"",python,selection_keyboard +2815,14281870,"train_dynamics.py",2612,0,"",python,selection_command +2816,14282009,"train_dynamics.py",2680,0,"",python,selection_command +2817,14282330,"train_dynamics.py",2679,0,"",python,selection_command +2818,14282373,"train_dynamics.py",2679,0,"#",python,content +2819,14282376,"train_dynamics.py",2680,0,"",python,selection_keyboard +2820,14282437,"train_dynamics.py",2680,0," ",python,content +2821,14282443,"train_dynamics.py",2681,0,"",python,selection_keyboard +2822,14282690,"train_dynamics.py",2680,0,"",python,selection_command +2823,14286620,"train_dynamics.py",2727,0,"",python,selection_command +2824,14286887,"train_dynamics.py",2745,0,"",python,selection_command +2825,14287411,"train_dynamics.py",2752,0,"",python,selection_command +2826,14287588,"train_dynamics.py",2754,0,"",python,selection_command +2827,14287756,"train_dynamics.py",2759,0,"",python,selection_command +2828,14287927,"train_dynamics.py",2760,0,"",python,selection_command +2829,14296147,"train_dynamics.py",2759,0,"",python,selection_command +2830,14296314,"train_dynamics.py",2754,0,"",python,selection_command +2831,14297275,"train_dynamics.py",2736,0,"",python,selection_command +2832,14297527,"train_dynamics.py",2689,0,"",python,selection_command +2833,14297558,"train_dynamics.py",2621,0,"",python,selection_command +2834,14297592,"train_dynamics.py",2582,0,"",python,selection_command +2835,14297707,"train_dynamics.py",2535,0,"",python,selection_command +2836,14297892,"train_dynamics.py",2504,0,"",python,selection_command +2837,14298065,"train_dynamics.py",2482,0,"",python,selection_command +2838,14298408,"train_dynamics.py",4591,0,"",python,selection_command +2839,14298993,"train_dynamics.py",4607,0,"",python,selection_command +2840,14299171,"train_dynamics.py",4608,0,"",python,selection_command +2841,14299327,"train_dynamics.py",4613,0,"",python,selection_command +2842,14299508,"train_dynamics.py",4615,0,"",python,selection_command +2843,14299841,"train_dynamics.py",4404,0,"",python,selection_command +2844,14301218,"train_dynamics.py",4364,0,"",python,selection_command +2845,14301477,"train_dynamics.py",4348,0,"",python,selection_command +2846,14301884,"train_dynamics.py",12493,0,"",python,selection_command +2847,14302618,"train_dynamics.py",12503,0,"",python,selection_command +2848,14302867,"train_dynamics.py",12504,0,"",python,selection_command +2849,14302904,"train_dynamics.py",12513,0,"",python,selection_command +2850,14302930,"train_dynamics.py",12514,0,"",python,selection_command +2851,14302970,"train_dynamics.py",12519,0,"",python,selection_command +2852,14303194,"train_dynamics.py",12521,0,"",python,selection_command +2853,14303349,"train_dynamics.py",12530,0,"",python,selection_command +2854,14303505,"train_dynamics.py",12532,0,"",python,selection_command +2855,14305253,"train_dynamics.py",12470,0,"",python,selection_command +2856,14305433,"train_dynamics.py",12474,0,"",python,selection_command +2857,14305683,"train_dynamics.py",12476,0,"",python,selection_command +2858,14305723,"train_dynamics.py",12481,0,"",python,selection_command +2859,14305757,"train_dynamics.py",12483,0,"",python,selection_command +2860,14305915,"train_dynamics.py",12491,0,"",python,selection_command +2861,14306139,"train_dynamics.py",12493,0,"",python,selection_command +2862,14306438,"train_dynamics.py",4348,0,"",python,selection_command +2863,14307209,"train_dynamics.py",4364,0,"",python,selection_command +2864,14307454,"train_dynamics.py",4421,0,"",python,selection_command +2865,14307487,"train_dynamics.py",4461,0,"",python,selection_command +2866,14308257,"train_dynamics.py",4500,0,"",python,selection_command +2867,14308392,"train_dynamics.py",4505,0,"",python,selection_command +2868,14308554,"train_dynamics.py",4580,0,"",python,selection_command +2869,14308719,"train_dynamics.py",4584,0,"",python,selection_command +2870,14308904,"train_dynamics.py",4591,0,"",python,selection_command +2871,14309224,"train_dynamics.py",2472,0,"",python,selection_command +2872,14309733,"train_dynamics.py",2494,0,"",python,selection_command +2873,14309841,"train_dynamics.py",2525,0,"",python,selection_command +2874,14310094,"train_dynamics.py",2572,0,"",python,selection_command +2875,14310127,"train_dynamics.py",2611,0,"",python,selection_command +2876,14310159,"train_dynamics.py",2679,0,"",python,selection_command +2877,14310337,"train_dynamics.py",2726,0,"",python,selection_command +2878,14310526,"train_dynamics.py",2744,0,"",python,selection_command +2879,14311754,"train_dynamics.py",2752,0,"",python,selection_command +2880,14311956,"train_dynamics.py",2754,0,"",python,selection_command +2881,14313470,"train_dynamics.py",2736,0,"",python,selection_command +2882,14313639,"train_dynamics.py",2689,0,"",python,selection_command +2883,14313763,"train_dynamics.py",2621,0,"",python,selection_command +2884,14314672,"train_dynamics.py",2582,0,"",python,selection_command +2885,14314820,"train_dynamics.py",2535,0,"",python,selection_command +2886,14314924,"train_dynamics.py",2536,0,"",python,selection_command +2887,14315025,"train_dynamics.py",2505,0,"",python,selection_command +2888,14315255,"train_dynamics.py",2483,0,"",python,selection_command +2889,14315570,"train_dynamics.py",2488,0,"",python,selection_command +2890,14315915,"train_dynamics.py",2494,0,"",python,selection_command +2891,14316509,"train_dynamics.py",2499,0,"",python,selection_command +2892,14316675,"train_dynamics.py",2501,0,"",python,selection_command +2893,14316852,"train_dynamics.py",2506,0,"",python,selection_command +2894,14317057,"train_dynamics.py",2508,0,"",python,selection_command +2895,14317453,"train_dynamics.py",2630,0,"",python,selection_command +2896,14319573,"train_dynamics.py",2698,0,"",python,selection_command +2897,14319822,"train_dynamics.py",2738,0,"",python,selection_command +2898,14319854,"train_dynamics.py",2763,0,"",python,selection_command +2899,14319887,"train_dynamics.py",2806,0,"",python,selection_command +2900,14319932,"train_dynamics.py",2833,0,"",python,selection_command +2901,14319959,"train_dynamics.py",2907,0,"",python,selection_command +2902,14319997,"train_dynamics.py",2970,0,"",python,selection_command +2903,14320026,"train_dynamics.py",3008,0,"",python,selection_command +2904,14320063,"train_dynamics.py",3033,0,"",python,selection_command +2905,14320093,"train_dynamics.py",3083,0,"",python,selection_command +2906,14320127,"train_dynamics.py",3155,0,"",python,selection_command +2907,14320286,"train_dynamics.py",3197,0,"",python,selection_command +2908,14320449,"train_dynamics.py",3256,0,"",python,selection_command +2909,14320676,"train_dynamics.py",3237,0,"",python,selection_command +2910,14320770,"train_dynamics.py",3237,0,"#",python,content +2911,14320774,"train_dynamics.py",3238,0,"",python,selection_keyboard +2912,14320824,"train_dynamics.py",3238,0," ",python,content +2913,14320826,"train_dynamics.py",3239,0,"",python,selection_keyboard +2914,14321052,"train_dynamics.py",3238,0,"",python,selection_command +2915,14323878,"train_dynamics.py",3179,0,"",python,selection_command +2916,14324127,"train_dynamics.py",3137,0,"",python,selection_command +2917,14324155,"train_dynamics.py",3065,0,"",python,selection_command +2918,14324187,"train_dynamics.py",3015,0,"",python,selection_command +2919,14324219,"train_dynamics.py",3008,0,"",python,selection_command +2920,14324255,"train_dynamics.py",2952,0,"",python,selection_command +2921,14324305,"train_dynamics.py",2889,0,"",python,selection_command +2922,14324334,"train_dynamics.py",2815,0,"",python,selection_command +2923,14324366,"train_dynamics.py",2788,0,"",python,selection_command +2924,14324400,"train_dynamics.py",2745,0,"",python,selection_command +2925,14324440,"train_dynamics.py",2727,0,"",python,selection_command +2926,14324472,"train_dynamics.py",2680,0,"",python,selection_command +2927,14324506,"train_dynamics.py",2612,0,"",python,selection_command +2928,14324652,"train_dynamics.py",2573,0,"",python,selection_command +2929,14324803,"train_dynamics.py",2526,0,"",python,selection_command +2930,14324904,"train_dynamics.py",2573,0,"",python,selection_command +2931,14325104,"train_dynamics.py",2575,0,"",python,selection_command +2932,14325356,"train_dynamics.py",2583,0,"",python,selection_command +2933,14325687,"train_dynamics.py",2590,0,"",python,selection_command +2934,14325874,"train_dynamics.py",2543,0,"",python,selection_command +2935,14326106,"train_dynamics.py",2512,0,"",python,selection_command +2936,14326622,"train_dynamics.py",2630,0,"",python,selection_command +2937,14327958,"train_dynamics.py",2681,0,"",python,selection_command +2938,14328240,"train_dynamics.py",2760,0,"",python,selection_command +2939,14328610,"train_dynamics.py",4406,0,"",python,selection_command +2940,14329385,"train_dynamics.py",2760,0,"",python,selection_command +2941,14330589,"train_dynamics.py",2803,0,"",python,selection_command +2942,14330840,"train_dynamics.py",2830,0,"",python,selection_command +2943,14330872,"train_dynamics.py",2904,0,"",python,selection_command +2944,14330905,"train_dynamics.py",2967,0,"",python,selection_command +2945,14330939,"train_dynamics.py",3008,0,"",python,selection_command +2946,14330972,"train_dynamics.py",3030,0,"",python,selection_command +2947,14331007,"train_dynamics.py",3080,0,"",python,selection_command +2948,14331170,"train_dynamics.py",3152,0,"",python,selection_command +2949,14331322,"train_dynamics.py",3194,0,"",python,selection_command +2950,14334005,"train_dynamics.py",3239,0,"",python,selection_command +2951,14335026,"train_dynamics.py",3244,0,"",python,selection_command +2952,14338515,"train_dynamics.py",3185,0,"",python,selection_command +2953,14338768,"train_dynamics.py",3143,0,"",python,selection_command +2954,14338792,"train_dynamics.py",3071,0,"",python,selection_command +2955,14338832,"train_dynamics.py",3021,0,"",python,selection_command +2956,14338867,"train_dynamics.py",3008,0,"",python,selection_command +2957,14338899,"train_dynamics.py",2958,0,"",python,selection_command +2958,14338933,"train_dynamics.py",2895,0,"",python,selection_command +2959,14338966,"train_dynamics.py",2821,0,"",python,selection_command +2960,14339002,"train_dynamics.py",2794,0,"",python,selection_command +2961,14339032,"train_dynamics.py",2751,0,"",python,selection_command +2962,14339070,"train_dynamics.py",2733,0,"",python,selection_command +2963,14339111,"train_dynamics.py",2686,0,"",python,selection_command +2964,14339138,"train_dynamics.py",2618,0,"",python,selection_command +2965,14339203,"train_dynamics.py",2579,0,"",python,selection_command +2966,14339215,"train_dynamics.py",2532,0,"",python,selection_command +2967,14339386,"train_dynamics.py",2501,0,"",python,selection_command +2968,14339400,"train_dynamics.py",2479,0,"",python,selection_command +2969,14339405,"train_dynamics.py",2467,0,"",python,selection_command +2970,14339419,"train_dynamics.py",2466,0,"",python,selection_command +2971,14339424,"train_dynamics.py",2455,0,"",python,selection_command +2972,14339435,"train_dynamics.py",2443,0,"",python,selection_command +2973,14339458,"train_dynamics.py",2442,0,"",python,selection_command +2974,14339488,"train_dynamics.py",2430,0,"",python,selection_command +2975,14339536,"train_dynamics.py",2442,0,"",python,selection_command +2976,14339778,"train_dynamics.py",2443,0,"",python,selection_command +2977,14339813,"train_dynamics.py",2455,0,"",python,selection_command +2978,14339843,"train_dynamics.py",2466,0,"",python,selection_command +2979,14339880,"train_dynamics.py",2467,0,"",python,selection_command +2980,14339915,"train_dynamics.py",2479,0,"",python,selection_command +2981,14339948,"train_dynamics.py",2501,0,"",python,selection_command +2982,14339989,"train_dynamics.py",2532,0,"",python,selection_command +2983,14340033,"train_dynamics.py",2579,0,"",python,selection_command +2984,14340138,"train_dynamics.py",2618,0,"",python,selection_command +2985,14353875,"train_dynamics.py",2607,0,"",python,selection_command +2986,14353946,"train_dynamics.py",2675,0,"",python,selection_command +2987,14354087,"train_dynamics.py",2722,0,"",python,selection_command +2988,14354237,"train_dynamics.py",2740,0,"",python,selection_command +2989,14354387,"train_dynamics.py",2783,0,"",python,selection_command +2990,14354615,"train_dynamics.py",2740,0,"",python,selection_command +2991,14354748,"train_dynamics.py",2744,0,"",python,selection_command +2992,14354885,"train_dynamics.py",2752,0,"",python,selection_command +2993,14355024,"train_dynamics.py",2754,0,"",python,selection_command +2994,14355233,"train_dynamics.py",2759,0,"",python,selection_command +2995,14355429,"train_dynamics.py",2754,0,"",python,selection_command +2996,14355657,"train_dynamics.py",2494,0,"",python,selection_command +2997,14356524,"train_dynamics.py",2499,0,"",python,selection_command +2998,14356689,"train_dynamics.py",2501,0,"",python,selection_command +2999,14357014,"genie.py",0,0,"",python,tab +3000,14357020,"genie.py",282,0,"",python,selection_command +3001,14357873,"genie.py",4920,0,"",python,selection_command +3002,14361053,"genie.py",4938,0,"",python,selection_command +3003,14361934,"genie.py",4940,0,"",python,selection_command +3004,14362251,"genie.py",4942,0,"",python,selection_command +3005,14363141,"genie.py",5047,0,"",python,selection_command +3006,14365009,"genie.py",5032,0,"",python,selection_command +3007,14365082,"genie.py",5032,0,"#",python,content +3008,14365084,"genie.py",5033,0,"",python,selection_keyboard +3009,14365124,"genie.py",5033,0," ",python,content +3010,14365126,"genie.py",5034,0,"",python,selection_keyboard +3011,14365354,"genie.py",5033,0,"",python,selection_command +3012,14367390,"genie.py",5074,0,"",python,selection_command +3013,14367518,"genie.py",5158,0,"",python,selection_command +3014,14367671,"genie.py",5248,0,"",python,selection_command +3015,14367889,"genie.py",5333,0,"",python,selection_command +3016,14368047,"genie.py",5392,0,"",python,selection_command +3017,14371981,"genie.py",5383,0,"",python,selection_command +3018,14372043,"genie.py",5391,0,"",python,selection_command +3019,14372646,"genie.py",5409,0,"",python,selection_command +3020,14372896,"genie.py",5411,0,"",python,selection_command +3021,14372930,"genie.py",5414,0,"",python,selection_command +3022,14372954,"genie.py",5415,0,"",python,selection_command +3023,14372992,"genie.py",5420,0,"",python,selection_command +3024,14373023,"genie.py",5422,0,"",python,selection_command diff --git a/507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-b9559366-0d71-4ceb-9b37-1d3a0cf03cd61750867779082-2025_06_25-18.09.57.465/source.csv b/507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-b9559366-0d71-4ceb-9b37-1d3a0cf03cd61750867779082-2025_06_25-18.09.57.465/source.csv new file mode 100644 index 0000000000000000000000000000000000000000..ef3d3264ac00253a323ed880b770d5f54e78aab6 --- /dev/null +++ b/507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-b9559366-0d71-4ceb-9b37-1d3a0cf03cd61750867779082-2025_06_25-18.09.57.465/source.csv @@ -0,0 +1,64 @@ +Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type +1,6,"scripts/rename_mp4_files.py",0,0,"#!/usr/bin/env python3\n""""""\nScript to rename all MP4 files in a directory with a custom prefix.\n""""""\n\nimport os\nimport argparse\nimport re\nfrom pathlib import Path\n\n\ndef rename_mp4_files(directory, prefix, dry_run=False, pattern=None, start_number=1):\n """"""\n Rename all MP4 files in the specified directory with a custom prefix.\n \n Args:\n directory (str): Path to the directory containing MP4 files\n prefix (str): Custom prefix to add to filenames\n dry_run (bool): If True, only show what would be renamed without actually renaming\n pattern (str): Optional regex pattern to filter files\n start_number (int): Starting number for sequential naming\n """"""\n directory_path = Path(directory)\n \n if not directory_path.exists():\n print(f""Error: Directory '{directory}' does not exist."")\n return\n \n if not directory_path.is_dir():\n print(f""Error: '{directory}' is not a directory."")\n return\n \n # Find all MP4 files\n mp4_files = list(directory_path.glob(""*.mp4""))\n \n if not mp4_files:\n print(f""No MP4 files found in '{directory}'"")\n return\n \n # Filter by pattern if provided\n if pattern:\n regex = re.compile(pattern)\n mp4_files = [f for f in mp4_files if regex.search(f.name)]\n \n if not mp4_files:\n print(f""No MP4 files match the pattern '{pattern}' in '{directory}'"")\n return\n \n print(f""Found {len(mp4_files)} MP4 files to rename:"")\n \n # Sort files for consistent ordering\n mp4_files.sort()\n \n for i, file_path in enumerate(mp4_files, start=start_number):\n # Get the original filename without extension\n original_name = file_path.stem\n extension = file_path.suffix\n \n # Create new filename with prefix\n new_name = f""{prefix}_{i:03d}_{original_name}{extension}""\n new_path = file_path.parent / new_name\n \n # Check if new filename already exists\n if new_path.exists() and not dry_run:\n print(f""Warning: '{new_name}' already exists, skipping '{file_path.name}'"")\n continue\n \n if dry_run:\n print(f""Would rename: '{file_path.name}' -> '{new_name}'"")\n else:\n try:\n file_path.rename(new_path)\n print(f""Renamed: '{file_path.name}' -> '{new_name}'"")\n except OSError as e:\n print(f""Error renaming '{file_path.name}': {e}"")\n \n if dry_run:\n print(f""\nDry run completed. {len(mp4_files)} files would be renamed."")\n else:\n print(f""\nRenaming completed. {len(mp4_files)} files renamed."")\n\n\ndef main():\n parser = argparse.ArgumentParser(\n description=""Rename all MP4 files in a directory with a custom prefix"",\n formatter_class=argparse.RawDescriptionHelpFormatter,\n epilog=""""""\nExamples:\n %(prog)s /path/to/videos minecraft\n %(prog)s /path/to/videos player --dry-run\n %(prog)s /path/to/videos episode --pattern ""joost"" --start-number 10\n """"""\n )\n \n parser.add_argument(\n ""directory"",\n help=""Directory containing MP4 files to rename""\n )\n \n parser.add_argument(\n ""prefix"",\n help=""Custom prefix to add to filenames""\n )\n \n parser.add_argument(\n ""--dry-run"",\n action=""store_true"",\n help=""Show what would be renamed without actually renaming files""\n )\n \n parser.add_argument(\n ""--pattern"",\n help=""Regex pattern to filter files (only rename files matching this pattern)""\n )\n \n parser.add_argument(\n ""--start-number"",\n type=int,\n default=1,\n help=""Starting number for sequential naming (default: 1)""\n )\n \n args = parser.parse_args()\n \n # Validate start number\n if args.start_number < 1:\n print(""Error: start-number must be at least 1"")\n return\n \n rename_mp4_files(\n directory=args.directory,\n prefix=args.prefix,\n dry_run=args.dry_run,\n pattern=args.pattern,\n start_number=args.start_number\n )\n\n\nif __name__ == ""__main__"":\n main() ",python,tab +2,319,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"6:09:57 PM [info] Activating crowd-code\n6:09:57 PM [info] Welcome back tum_ind3695. Your user-id is '507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20'. Happy coding!\n6:09:57 PM [info] Recording started\n",Log,tab +3,2485,"scripts/rename_mp4_files.py",0,0,"",python,tab +4,2488,"scripts/rename_mp4_files.py",1333,0,"",python,selection_mouse +5,2493,"scripts/rename_mp4_files.py",1332,0,"",python,selection_command +6,4622,"TERMINAL",0,0,"",,terminal_focus +7,301525,"TERMINAL",0,0,"",,terminal_focus +8,419551,"TERMINAL",0,0,"cd /hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared//checkpoints/dyn/3289577",,terminal_command +9,419570,"TERMINAL",0,0,"]633;E;2025-06-25 18:16:56 cd /hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared//checkpoints/dyn/3289577;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577]633;D;0",,terminal_output +10,420075,"TERMINAL",0,0,"ls",,terminal_command +11,420099,"TERMINAL",0,0,"]633;E;2025-06-25 18:16:57 ls;ef772368-fdbb-4ff2-ad99-2db92190992c]633;Cgenie_1750786631_1000 genie_1750786631_11000 genie_1750786631_12500 genie_1750786631_2500 genie_1750786631_4000 genie_1750786631_5000 genie_1750786631_6500 genie_1750786631_8000 genie_1750786631_9500\r\ngenie_1750786631_10000 genie_1750786631_11500 genie_1750786631_1500 genie_1750786631_3000 genie_1750786631_4500 genie_1750786631_5500 genie_1750786631_7000 genie_1750786631_8500\r\ngenie_1750786631_10500 genie_1750786631_12000 genie_1750786631_2000 genie_1750786631_3500 genie_1750786631_500 genie_1750786631_6000 genie_1750786631_7500 genie_1750786631_9000\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577]633;D;0",,terminal_output +12,439480,"TERMINAL",0,0,"cd genie_1750786631_12500",,terminal_command +13,440497,"TERMINAL",0,0,"pwd",,terminal_command +14,440512,"TERMINAL",0,0,"]633;E;2025-06-25 18:17:17 pwd;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577/genie_1750786631_12500\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577/genie_1750786631_12500]633;D;0",,terminal_output +15,615482,"TERMINAL",0,0,"cd ../../..",,terminal_command +16,615494,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:12 cd ../../..;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints]633;D;0",,terminal_output +17,616076,"TERMINAL",0,0,"ls",,terminal_command +18,616111,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:13 ls;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C0000 3290283 3290284 3290295 3290296 3290366 3290367 3290391 3290392 3290439 3290440 3291405 dyn lam tokenizer\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints]633;D;0",,terminal_output +19,619493,"TERMINAL",0,0,"cd tokenizer/",,terminal_command +20,619506,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:16 cd tokenizer/;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer]633;D;0",,terminal_output +21,619896,"TERMINAL",0,0,"ls",,terminal_command +22,619948,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:17 ls;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C",,terminal_output +23,619997,"TERMINAL",0,0,"3255052 3255483 3256924 3257467 3258278 3259422 3261720 3273042 3273290 3273480 3273494 3273627 3273742 3273795 3273824 3273830 3273846 3276030 3276048 3276053 3285798\r\n3255466 3256873 3256926 3257633 3258283 3260527 3261722 3273174 3273348 3273488 3273496 3273681 3273743 3273816 3273828 3273831 3275950 3276039 3276051 3276058 3285811\r\n3255482 3256921 3256929 3257812 3259405 3260932 3273026 3273229 3273476 3273489 3273503 3273687 3273746 3273820 3273829 3273841 3275991 3276043 3276052 3285784 3286114\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer]633;D;0",,terminal_output +24,649663,"TERMINAL",0,0,"cd 3255052/",,terminal_command +25,649679,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:47 cd 3255052/;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer/3255052]633;D;0",,terminal_output +26,650176,"TERMINAL",0,0,"ls",,terminal_command +27,650194,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:47 ls;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer/3255052]633;D;0",,terminal_output +28,652601,"TERMINAL",0,0,"cd ..",,terminal_command +29,652609,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:50 cd ..;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer]633;D;0",,terminal_output +30,654392,"TERMINAL",0,0,"cd ..",,terminal_command +31,654398,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:51 cd ..;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints]633;D;0",,terminal_output +32,655918,"TERMINAL",0,0,"cd ..",,terminal_command +33,655926,"TERMINAL",0,0,"]633;E;2025-06-25 18:20:53 cd ..;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared]633;D;0",,terminal_output +34,1148074,"TERMINAL",0,0,"queue",,terminal_command +35,1148124,"TERMINAL",0,0,"]633;E;2025-06-25 18:29:05 queue;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C",,terminal_output +36,1148182,"TERMINAL",0,0,"[?1049h(B[?7hEvery 1.0s: squeue --mehkn1993.localdomain: Wed Jun 25 18:29:05 2025JOBID PARTITION NAME USER ST\tTIME NODES NODELIST(REASON)3292119 accelerat train_to tum_ind3 PD\t0:00\t 1 (Priority)3289577 accelerat train_dy tum_ind3 R 22:52:29\t 1 hkn0712",,terminal_output +37,1149242,"TERMINAL",0,0,"630",,terminal_output +38,1149475,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared]633;D;0",,terminal_output +39,1150350,"TERMINAL",0,0,"queue",,terminal_command +40,1150402,"TERMINAL",0,0,"]633;E;2025-06-25 18:29:07 queue;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C",,terminal_output +41,1150481,"TERMINAL",0,0,"[?1049h(B[?7hEvery 1.0s: squeue --mehkn1993.localdomain: Wed Jun 25 18:29:07 2025JOBID PARTITION NAME USER ST\tTIME NODES NODELIST(REASON)3292119 accelerat train_to tum_ind3 PD\t0:00\t 1 (Priority)3289577 accelerat train_dy tum_ind3 R 22:52:31\t 1 hkn0712",,terminal_output +42,1151212,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared]633;D;0",,terminal_output +43,1152587,"TERMINAL",0,0,"squeue",,terminal_command +44,1152616,"TERMINAL",0,0,"]633;E;2025-06-25 18:29:10 squeue;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)\r\n 3292119 accelerat train_to tum_ind3 PD 0:00 1 (Priority)\r\n 3289577 accelerat train_dy tum_ind3 R 22:52:34 1 hkn0712\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared]633;D;0]633;P;Cwd=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared",,terminal_output +45,2263248,"TERMINAL",0,0,"cd /hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared//checkpoints/dyn/3289577",,terminal_command +46,2263255,"TERMINAL",0,0,"]633;E;2025-06-25 18:47:40 cd /hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared//checkpoints/dyn/3289577;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577]633;D;0",,terminal_output +47,2264288,"TERMINAL",0,0,"ls",,terminal_command +48,2264341,"TERMINAL",0,0,"]633;E;2025-06-25 18:47:41 ls;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C",,terminal_output +49,2264401,"TERMINAL",0,0,"genie_1750786631_1000 genie_1750786631_11000 genie_1750786631_12500 genie_1750786631_2000 genie_1750786631_3500 genie_1750786631_500 genie_1750786631_6000 genie_1750786631_7500 genie_1750786631_9000\r\ngenie_1750786631_10000 genie_1750786631_11500 genie_1750786631_13000 genie_1750786631_2500 genie_1750786631_4000 genie_1750786631_5000 genie_1750786631_6500 genie_1750786631_8000 genie_1750786631_9500\r\ngenie_1750786631_10500 genie_1750786631_12000 genie_1750786631_1500 genie_1750786631_3000 genie_1750786631_4500 genie_1750786631_5500 genie_1750786631_7000 genie_1750786631_8500\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577]633;D;0",,terminal_output +50,2269289,"TERMINAL",0,0,"cd genie_1750786631_13000",,terminal_command +51,2269296,"TERMINAL",0,0,"]633;E;2025-06-25 18:47:46 cd genie_1750786631_13000;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577/genie_1750786631_13000]633;D;0",,terminal_output +52,2270468,"TERMINAL",0,0,"pwd",,terminal_command +53,2270477,"TERMINAL",0,0,"]633;E;2025-06-25 18:47:47 pwd;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577/genie_1750786631_13000\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577/genie_1750786631_13000]633;D;0",,terminal_output +54,2272125,"TERMINAL",0,0,"os",,terminal_command +55,2272199,"TERMINAL",0,0,"]633;E;2025-06-25 18:47:49 os;ef772368-fdbb-4ff2-ad99-2db92190992c]633;Cbash: os: command not found...\r\n",,terminal_output +56,2274189,"TERMINAL",0,0,"ls",,terminal_command +57,2274240,"TERMINAL",0,0,"]633;E;2025-06-25 18:47:51 ls;ef772368-fdbb-4ff2-ad99-2db92190992c]633;C",,terminal_output +58,2274298,"TERMINAL",0,0,"array_metadatas _CHECKPOINT_METADATA d manifest.ocdbt _METADATA ocdbt.process_0 ocdbt.process_1 ocdbt.process_2 ocdbt.process_3 _sharding\r\n]0;tum_ind3695@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/dyn/3289577/genie_1750786631_13000]633;D;0",,terminal_output +59,2708521,"scripts/file_duplicate_checker.py",0,0,"import os\nfrom collections import defaultdict\nfrom tqdm import tqdm\n\ndef find_duplicate_filenames(root_dir):\n filenames = defaultdict(list)\n file_count = 0\n\n # Use tqdm with manual update and no percentage/ETA bar\n pbar = tqdm(desc=""Files scanned"", unit=""file"", dynamic_ncols=True, bar_format=""{desc}: {n_fmt}"")\n\n # Walk the directory recursively\n for dirpath, _, files in os.walk(root_dir):\n for file in files:\n full_path = os.path.join(dirpath, file)\n if os.path.isfile(full_path):\n filenames[file].append(full_path)\n file_count += 1\n pbar.update(1)\n\n pbar.close()\n\n # Print duplicates\n duplicates = {name: paths for name, paths in filenames.items() if len(paths) > 1}\n if duplicates:\n print(""\nDuplicate filenames found:\n"")\n for name, paths in duplicates.items():\n print(f""Filename: {name}"")\n for path in paths:\n print(f"" - {path}"")\n print()\n else:\n print(""\nNo duplicate filenames found."")\n\nif __name__ == ""__main__"":\n import sys\n if len(sys.argv) < 2:\n print(""Usage: python find_duplicates.py "")\n else:\n find_duplicate_filenames(sys.argv[1])\n\n",python,tab +60,2729693,"scripts/file_duplicate_checker.py",991,0,"",python,selection_mouse +61,2729697,"scripts/file_duplicate_checker.py",990,0,"",python,selection_command +62,2729708,"scripts/file_duplicate_checker.py",990,1,")",python,selection_mouse +63,2729710,"scripts/file_duplicate_checker.py",991,0,"",python,selection_command