Add files using upload-large-folder tool

1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-5342e4d6-3c20-40cb-9bcb-64bf1931df6e1753973941916-2025_07_31-16.59.20.943/source.csv

@@ -0,0 +1,5 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,3,"utils/nn.py",0,0,"import math\nfrom typing import Tuple, Callable\n\nfrom flax import nnx\nimport jax\nimport jax.numpy as jnp\nimport einops\n\n\nclass PositionalEncoding(nnx.Module):\n    """"""https://uvadlc-notebooks.readthedocs.io/en/latest/tutorial_notebooks/JAX/tutorial6/Transformers_and_MHAttention.html""""""\n\n    def __init__(self, d_model: int, max_len: int = 5000):\n        self.d_model = d_model\n        self.max_len = max_len\n\n        pe = jnp.zeros((self.max_len, self.d_model))\n        position = jnp.arange(0, self.max_len, dtype=jnp.float32)[:, None]\n        div_term = jnp.exp(\n            jnp.arange(0, self.d_model, 2) * (-math.log(10000.0) / self.d_model)\n        )\n        pe = pe.at[:, 0::2].set(jnp.sin(position * div_term))\n        pe = pe.at[:, 1::2].set(jnp.cos(position * div_term))\n        self.pe = nnx.Variable(pe)\n\n    def __call__(self, x: jax.Array) -> jax.Array:\n        x = x + self.pe[: x.shape[2]]\n        return x\n\n\nclass STBlock(nnx.Module):\n    def __init__(\n        self,\n        dim: int,\n        ffn_dim: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        rngs: nnx.Rngs,\n    ):\n        self.dim = dim\n        self.ffn_dim = ffn_dim\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n\n        self.spatial_pos_enc = PositionalEncoding(self.dim)\n        self.spatial_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.spatial_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.dim,\n            qkv_features=self.dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=False\n            ),\n            rngs=rngs,\n            decode=False,\n        )\n\n        self.temporal_pos_enc = PositionalEncoding(self.dim)\n        self.temporal_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.temporal_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.dim,\n            qkv_features=self.dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=True\n            ),\n            rngs=rngs,\n            decode=False,\n        )\n\n        self.ffn_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_dense1 = nnx.Linear(\n            in_features=self.dim,\n            out_features=self.ffn_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_dense2 = nnx.Linear(\n            in_features=self.ffn_dim,\n            out_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    @nnx.remat\n    def __call__(self, x_BTNM: jax.Array) -> jax.Array:\n        # --- Spatial attention ---\n        z_BTNM = self.spatial_pos_enc(x_BTNM)\n        z_BTNM = self.spatial_norm(z_BTNM)\n        z_BTNM = self.spatial_attention(z_BTNM)\n        x_BTNM = x_BTNM + z_BTNM\n\n        # --- Temporal attention ---\n        x_BNTM = x_BTNM.swapaxes(1, 2)\n        z_BNTM = self.temporal_pos_enc(x_BNTM)\n        z_BNTM = self.temporal_norm(z_BNTM)\n        z_BNTM = self.temporal_attention(z_BNTM)\n        x_BNTM = x_BNTM + z_BNTM\n        x_BTNM = x_BNTM.swapaxes(1, 2)\n\n        # --- Feedforward ---\n        z_BTNM = self.ffn_norm(x_BTNM)\n        z_BTND = self.ffn_dense1(z_BTNM)\n        z_BTND = jax.nn.gelu(z_BTND)\n        z_BTNM = self.ffn_dense2(z_BTND)\n        x_BTNM = x_BTNM + z_BTNM\n\n        return x_BTNM\n\n\nclass STTransformer(nnx.Module):\n    """"""\n    Dimension keys:\n        B: batch size\n        T: number of frames\n        N: number of patches per frame\n        I: number of input features\n        M: model dimension\n        D: FFN dimension\n        O: number of output features\n    """"""\n    def __init__(\n        self,\n        input_dim: int,\n        model_dim: int,\n        ffn_dim: int,\n        out_dim: int,\n        num_blocks: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        rngs: nnx.Rngs,\n    ):\n        self.input_dim = input_dim\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.out_dim = out_dim\n        self.num_blocks = num_blocks\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n\n        self.input_norm1 = nnx.LayerNorm(\n            num_features=self.input_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_dense = nnx.Linear(\n            in_features=self.input_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_norm2 = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n        self.blocks = []\n        for _ in range(self.num_blocks):\n            self.blocks.append(\n                STBlock(\n                    dim=self.model_dim,\n                    ffn_dim=self.ffn_dim,\n                    num_heads=self.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                    rngs=rngs,\n                )\n            )\n\n        self.output_dense = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.out_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    def __call__(self, x_BTNI: jax.Array) -> jax.Array:\n        x_BTNI = self.input_norm1(x_BTNI)\n        x_BTNM = self.input_dense(x_BTNI)\n        x_BTNM = self.input_norm2(x_BTNM)\n\n        for block in self.blocks:\n            x_BTNM = block(x_BTNM)\n\n        x_BTNO = self.output_dense(x_BTNM)\n        return x_BTNO\n\nclass TransformerBlock(nnx.Module):\n    def __init__(\n        self,\n        model_dim: int,\n        ffn_dim: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        decode: bool,\n        rngs: nnx.Rngs,\n    ):\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n\n        self.temporal_pos_enc = PositionalEncoding(self.model_dim)\n        self.spatial_pos_enc = PositionalEncoding(self.model_dim)\n        self.temporal_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.spatial_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.temporal_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.model_dim,\n            qkv_features=self.model_dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=True\n            ),\n            rngs=rngs,\n            decode=decode,\n        )\n        self.spatial_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.model_dim,\n            qkv_features=self.model_dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=True\n            ),\n            rngs=rngs,\n            decode=decode,\n        )\n        self.ffn_dense1 = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.ffn_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_dense2 = nnx.Linear(\n            in_features=self.ffn_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    @nnx.remat\n    def __call__(self, x_BTNM: jax.Array) -> jax.Array:\n        # FIXME (f.srambical): this is exactly the same as STBlock (except for the positional encoding)\n        # --- Spatial attention ---\n        _, T, N, _ = x_BTNM.shape\n        z_FNM = einops.rearrange(x_BTNM, ""b t n m -> (b t) n m"")\n        z_FNM = self.spatial_norm(z_FNM)\n        # FIXME (f.srambical): only input last token\n        z_FNM = self.spatial_attention(z_FNM)\n        z_BTNM = einops.rearrange(z_FNM, ""(b t) n m -> b t n m"", t=T)\n        x_BTNM = x_BTNM + z_BTNM\n        # --- Temporal attention ---\n        z_PTM = einops.rearrange(x_BTNM, ""b t n m -> (b n) t m"")\n        z_PTM = self.temporal_norm(z_PTM)\n        # FIXME (f.srambical): only input last token\n        z_PTM = self.temporal_attention(z_PTM)\n        z_BTNM = einops.rearrange(z_PTM, ""(b n) t m -> b t n m"", n=N)\n        x_BTNM = x_BTNM + z_BTNM\n        # --- Feedforward ---\n        z_BTNM = self.ffn_norm(x_BTNM)\n        z_BTND = self.ffn_dense1(z_BTNM)\n        z_BTND = jax.nn.gelu(z_BTND)\n        z_BTNM = self.ffn_dense2(z_BTND)\n        x_BTNM = x_BTNM + z_BTNM\n\n        return x_BTNM\n\nclass Transformer(nnx.Module):\n    """"""\n    Dimension keys:\n        B: batch size\n        T: number of frames\n        N: number of patches per frame\n        I: number of input features\n        M: model dimension\n        D: FFN dimension\n        O: number of output features\n        F: number of frames in batch\n        P: number of patch positions in batch\n    """"""\n    def __init__(\n        self,\n        input_dim: int,\n        model_dim: int,\n        ffn_dim: int,\n        out_dim: int,\n        num_blocks: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        decode: bool,\n        rngs: nnx.Rngs,\n    ):\n        self.input_dim = input_dim\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.out_dim = out_dim\n        self.num_blocks = num_blocks\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n\n        self.pos_enc = PositionalEncoding(self.model_dim)\n        self.input_norm1 = nnx.LayerNorm(\n            num_features=self.input_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_dense = nnx.Linear(\n            in_features=self.input_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_norm2 = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n        self.blocks = []\n        for _ in range(self.num_blocks):\n            self.blocks.append(\n                TransformerBlock(\n                    model_dim=self.model_dim,\n                    ffn_dim=self.ffn_dim,\n                    num_heads=self.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            )\n        self.output_dense = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.out_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    def __call__(self, x_BTNI: jax.Array) -> jax.Array:\n        x_BTNI = self.input_norm1(x_BTNI)\n        x_BTNM = self.input_dense(x_BTNI)\n        x_BTNM = self.input_norm2(x_BTNM)\n        x_BTNM = self.pos_enc(x_BTNM)\n\n        for block in self.blocks:\n            x_BTNM = block(x_BTNM)\n\n        x_BTNV = self.output_dense(x_BTNM)\n        return x_BTNV\n\ndef normalize(x: jax.Array) -> jax.Array:\n    return x / (jnp.linalg.norm(x, ord=2, axis=-1, keepdims=True) + 1e-8)\n\n\nclass VectorQuantizer(nnx.Module):\n    """"""\n    Dimension keys:\n        D: B * T * N\n        K: number of latents\n        L: latent dimension\n    """"""\n    def __init__(\n        self, latent_dim: int, num_latents: int, dropout: float, rngs: nnx.Rngs\n    ):\n        self.latent_dim = latent_dim\n        self.num_latents = num_latents\n        self.dropout = dropout\n\n        self.codebook = nnx.Param(\n            normalize(\n                nnx.initializers.lecun_uniform()(\n                    rngs.params(), (self.num_latents, self.latent_dim)\n                )\n            )\n        )\n        self.drop = nnx.Dropout(self.dropout, rngs=rngs)\n\n    def __call__(\n        self, x_DL: jax.Array, training: bool\n    ) -> Tuple[jax.Array, jax.Array, jax.Array, jax.Array]:\n        # --- Compute distances ---\n        x_DL = normalize(x_DL)\n        normalized_codebook_KL = normalize(self.codebook.value)\n        distance_DK = -jnp.matmul(x_DL, normalized_codebook_KL.T)\n        if training:\n            distance_DK = self.drop(distance_DK)\n\n        # --- Get indices and embeddings ---\n        indices_D = jnp.argmin(distance_DK, axis=-1)\n        z_DL = self.codebook[indices_D]\n\n        # --- Straight through estimator ---\n        z_q_DL = x_DL + jax.lax.stop_gradient(z_DL - x_DL)\n        return z_q_DL, z_DL, x_DL, indices_D\n\n    def get_codes(self, indices_E: jax.Array) -> jax.Array:\n        return self.codebook[indices_E]\n\n\ndef _create_flash_attention_fn(use_flash_attention: bool, is_causal: bool) -> Callable:\n    """"""\n    Create an attention function that uses flash attention if enabled.\n\n    Flax MultiHeadAttention provides tensors with shape (batch..., length, num_heads, head_dim)\n    jax.nn.dot_product_attention expects (batch, length, num_heads, head_dim).\n\n    We need to reshape to ensure compatibility. cuDNN's flash attention additionally\n    requires a sequence length that is a multiple of 4. We pad the sequence length to the nearest\n    multiple of 4 and mask accordingly.\n    """"""\n\n    def attention_fn(query, key, value, bias=None, mask=None, **kwargs):\n        implementation = ""cudnn"" if use_flash_attention else None\n\n        def _rearrange(x):\n            return einops.rearrange(x, ""... l h d -> (...) l h d"")\n\n        def _pad(x):\n            return jnp.pad(x, ((0, 0), (0, pad_size), (0, 0), (0, 0)))\n\n        def _fuse_masks(mask: jax.Array, attention_mask: jax.Array) -> jax.Array:\n            mask_bool = mask.astype(jnp.bool_)\n            expanded_mask = jnp.pad(\n                mask_bool, ((0, pad_size), (0, pad_size)), constant_values=False\n            )\n            return jnp.logical_and(attention_mask, expanded_mask)\n\n        original_shape = query.shape\n        original_seq_len = query.shape[-3]\n\n        # Pad to nearest multiple of 4\n        target_seq_len = ((original_seq_len + 3) // 4) * 4\n        pad_size = target_seq_len - original_seq_len\n\n        query_4d = _pad(_rearrange(query))\n        key_4d = _pad(_rearrange(key))\n        value_4d = _pad(_rearrange(value))\n\n        attention_mask = jnp.ones((target_seq_len, target_seq_len), dtype=jnp.bool_)\n        attention_mask = attention_mask.at[original_seq_len:, :].set(False)\n        attention_mask = attention_mask.at[:, original_seq_len:].set(False)\n\n        mask_4d = (\n            _fuse_masks(mask, attention_mask) if mask is not None else attention_mask\n        )\n        mask_4d = mask_4d[jnp.newaxis, jnp.newaxis, :, :]  # (1, 1, seq_len, seq_len)\n\n        bias_4d = _pad(_rearrange(bias)) if bias is not None else None\n\n        # NOTE: jax.nn.dot_product_attention does not support dropout\n        output_4d = jax.nn.dot_product_attention(\n            query=query_4d,\n            key=key_4d,\n            value=value_4d,\n            bias=bias_4d,\n            mask=mask_4d,\n            implementation=implementation,\n            is_causal=is_causal,\n        )\n        return output_4d[..., :original_seq_len, :, :].reshape(original_shape)\n\n    return attention_fn\n",python,tab
+2,408,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"4:59:20 PM [info] Activating crowd-code\n4:59:20 PM [info] Recording started\n4:59:20 PM [info] Initializing git provider using file system watchers...\n",Log,tab
+3,558,"extension-output-pdoom-org.crowd-code-#1-crowd-code",150,0,"4:59:21 PM [info] Git repository found\n4:59:21 PM [info] Git provider initialized successfully\n4:59:21 PM [info] Initial git state: [object Object]\n",Log,content
+4,6295,"utils/nn.py",0,0,"",python,tab

1f15334ab7e6820c9fda17c961659882ef9853cc80f7356b9a9b22f286fd7389/crowd-code-f8f922dd-4d7b-4eee-9d87-0cdea457c7d91763046453495-2025_11_13-16.07.47.281/source.csv

@@ -0,0 +1,5 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,2,"slurm/dev/franz/berlin/crowd-pilot/generate_array_record_dataset_bash_version.sh",0,0,"#!/bin/bash\n\nset -uex\n\nOUTPUT_DIR=""/fast/project/HFMI_SynergyUnit/jafar_ws/data/crowd-pilot/crowd-code-0.1/bash_format_array_record/""\nCSV_ROOT=""/fast/project/HFMI_SynergyUnit/jafar_ws/data/crowd-pilot/crowd-code-0.1/csv/""\n\nuv run crowd-pilot/serialize_dataset_array_record.py --csv_root=$CSV_ROOT --output_dir=$OUTPUT_DIR",shellscript,tab
+2,304,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"4:07:47 PM [info] Activating crowd-code\n4:07:47 PM [info] Recording started\n4:07:47 PM [info] Initializing git provider using file system watchers...\n",Log,tab
+3,437,"extension-output-pdoom-org.crowd-code-#1-crowd-code",150,0,"4:07:47 PM [info] Git repository found\n4:07:47 PM [info] Git provider initialized successfully\n4:07:47 PM [info] Initial git state: [object Object]\n",Log,content
+4,9551,"slurm/dev/franz/berlin/crowd-pilot/generate_array_record_dataset_bash_version.sh",0,0,"",shellscript,tab

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-067ab08c-cbdc-4e8b-9fb5-785f5bdfc09f1750961858926-2025_06_26-20.18.07.393/source.csv

@@ -0,0 +1,297 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,14,".gitignore",0,0,"*.pyc\n*.npy\n*.png\n*.gif\n\nwandb_key\ncheckpoints/\nwandb/\n__pycache__/\n\n\n\nlogs/\ndata/\nsandbox\nsbatch_scripts/\nutils/clip_checker.py\nnotes\nrequirements_fran.txt\ntrain_tokenizer_normal\nshell_scripts/\n\n\n",ignore,tab
+2,6465,".gitignore",197,0,"",ignore,selection_command
+3,8296,"train_tokenizer.py",0,0,"from dataclasses import dataclass\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 dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\n\nfrom models.tokenizer import TokenizerVQVAE\nfrom utils.dataloader import get_dataloader\n\nts = int(time.time())\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 = ""data_tfrecords/coinrun""\n    checkpoint: str = """"\n    # Optimization\n    vq_beta: float = 0.25\n    batch_size: int = 48\n    min_lr: float = 3e-4\n    max_lr: float = 3e-4\n    warmup_steps: int = 10000\n    # Tokenizer\n    model_dim: int = 512\n    latent_dim: int = 32\n    num_latents: int = 1024\n    patch_size: int = 4\n    num_blocks: int = 8\n    num_heads: int = 8\n    dropout: float = 0.0\n    codebook_dropout: float = 0.01\n    # Logging\n    log: bool = False\n    entity: str = """"\n    project: str = """"\n    log_interval: int = 5\n    log_image_interval: int = 250\n    ckpt_dir: str = """"\n    log_checkpoint_interval: int = 10000\n    log_gradients: bool = False\n\n\nargs = tyro.cli(Args)\n\n\ndef tokenizer_loss_fn(params, state, inputs):\n    # --- Compute loss ---\n    outputs = state.apply_fn(\n        params,\n        inputs,\n        training=True,\n        rngs={""params"": inputs[""rng""], ""dropout"": inputs[""dropout_rng""]},\n    )\n    mse = jnp.square(inputs[""videos""] - 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 = inputs[""videos""].clip(0, 1).reshape(-1, *inputs[""videos""].shape[2:])\n    recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n    psnr = pix.psnr(gt, recon).mean()\n    ssim = 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@jax.jit\ndef train_step(state, inputs):\n    grad_fn = jax.value_and_grad(tokenizer_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[""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 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(entity=args.entity, project=args.project, group=""overfit"", config=args)\n\n    # --- Initialize model ---\n    tokenizer = TokenizerVQVAE(\n        in_dim=args.image_channels,\n        model_dim=args.model_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    )\n    rng, _rng = jax.random.split(rng)\n    image_shape = (args.image_height, args.image_width, args.image_channels)\n    inputs = dict(\n        videos=jnp.zeros(\n            (per_device_batch_size_for_init, args.seq_len, *image_shape),\n            dtype=jnp.float32,\n        ),\n    )\n    init_params = tokenizer.init(_rng, 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=tokenizer.apply, params=init_params, tx=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    train_state = jax.device_put(train_state, replicated_sharding)\n\n    # --- Load checkpoint ---\n    step = 0\n    if args.checkpoint:\n        restore_target = {""model"": train_state}\n        restore_args = orbax_utils.restore_args_from_target(restore_target)\n        train_state.params[""params""].update(\n            PyTreeCheckpointer()\n            .restore(args.checkpoint, item=restore_target, restore_args=restore_args)[\n                ""model""\n            ]\n            .params[""params""]\n        )\n        # Assume checkpoint is of the form tokenizer_<timestamp>_<step>\n        step += int(args.checkpoint.split(""_"")[-1])\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    print(f""Starting training from step {step}..."")\n    while step < args.num_steps:\n        for videos in dataloader:\n            # --- Train step ---\n            start_time = time.time()\n\n            rng, _rng, _rng_dropout = jax.random.split(rng, 3)\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(videos=videos, rng=_rng, dropout_rng=_rng_dropout)\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                    # 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            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""tokenizer_{ts}_{step}""),\n                    ckpt,\n                    save_args=save_args,\n                )\n            if step >= args.num_steps:\n                break\n",python,tab
+4,14643,"train_tokenizer.py",0,0,"",python,selection_command
+5,27564,"train_tokenizer.py",701,0,"",python,selection_mouse
+6,27576,"train_tokenizer.py",700,0,"",python,selection_command
+7,52541,"train_tokenizer.py",540,0,"",python,selection_mouse
+8,53876,"TERMINAL",0,0,"",,terminal_focus
+9,65320,"TERMINAL",0,0,"ls",,terminal_command
+10,65341,"TERMINAL",0,0,"]633;E;2025-06-26 20:19:12 ls;b8c83623-699a-4617-a1fd-25c482b22092]633;Cgenerate_dataset.py                        logs                    requirements.txt  single_batch.npy                train_tokenizer_logging.py\r\ngeneration_1750681785.3743937.gif          models                  sample.py         train_dynamics.py               train_tokenizer.py\r\ngeneration_video_0_1750688558.8735778.gif  notes.md                sample_results    train_dynamics_single_batch.py  train_tokenizer_single_batch.py\r\ngenie.py                                   __pycache__             sandbox           train_lam.py                    utils\r\njafar                                      README.md               sbatch_scripts    train_lam_single_batch.py       wandb\r\nLICENSE                                    requirements_franz.txt  shell_scripts     train_lam_tf_seeding.py\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;0",,terminal_output
+11,66867,"TERMINAL",0,0,"git clone git@github.com:p-doom/slurm.git",,terminal_command
+12,66932,"TERMINAL",0,0,"]633;E;2025-06-26 20:19:14 git clone git@github.com:p-doom/slurm.git;b8c83623-699a-4617-a1fd-25c482b22092]633;CCloning into 'slurm'...\r\n",,terminal_output
+13,68608,"TERMINAL",0,0,"remote: Enumerating objects: 167, done.\r\nremote: Counting objects:   0% (1/167)\rremote: Counting objects:   1% (2/167)\rremote: Counting objects:   2% (4/167)\rremote: Counting objects:   3% (6/167)\rremote: Counting objects:   4% (7/167)\rremote: Counting objects:   5% (9/167)\rremote: Counting objects:   6% (11/167)\rremote: Counting objects:   7% (12/167)\rremote: Counting objects:   8% (14/167)\rremote: Counting objects:   9% (16/167)\rremote: Counting objects:  10% (17/167)\rremote: Counting objects:  11% (19/167)\rremote: Counting objects:  12% (21/167)\rremote: Counting objects:  13% (22/167)\rremote: Counting objects:  14% (24/167)\rremote: Counting objects:  15% (26/167)\rremote: Counting objects:  16% (27/167)\rremote: Counting objects:  17% (29/167)\rremote: Counting objects:  18% (31/167)\rremote: Counting objects:  19% (32/167)\rremote: Counting objects:  20% (34/167)\rremote: Counting objects:  21% (36/167)\rremote: Counting objects:  22% (37/167)\rremote: Counting objects:  23% (39/167)\rremote: Counting objects:  24% (41/167)\rremote: Counting objects:  25% (42/167)\rremote: Counting objects:  26% (44/167)\rremote: Counting objects:  27% (46/167)\rremote: Counting objects:  28% (47/167)\rremote: Counting objects:  29% (49/167)\rremote: Counting objects:  30% (51/167)\rremote: Counting objects:  31% (52/167)\rremote: Counting objects:  32% (54/167)\rremote: Counting objects:  33% (56/167)\rremote: Counting objects:  34% (57/167)\rremote: Counting objects:  35% (59/167)\rremote: Counting objects:  36% (61/167)\rremote: Counting objects:  37% (62/167)\rremote: Counting objects:  38% (64/167)\rremote: Counting objects:  39% (66/167)\rremote: Counting objects:  40% (67/167)\rremote: Counting objects:  41% (69/167)\rremote: Counting objects:  42% (71/167)\rremote: Counting objects:  43% (72/167)\rremote: Counting objects:  44% (74/167)\rremote: Counting objects:  45% (76/167)\rremote: Counting objects:  46% (77/167)\rremote: Counting objects:  47% (79/167)\rremote: Counting objects:  48% (81/167)\rremote: Counting objects:  49% (82/167)\rremote: Counting objects:  50% (84/167)\rremote: Counting objects:  51% (86/167)\rremote: Counting objects:  52% (87/167)\rremote: Counting objects:  53% (89/167)\rremote: Counting objects:  54% (91/167)\rremote: Counting objects:  55% (92/167)\rremote: Counting objects:  56% (94/167)\rremote: Counting objects:  57% (96/167)\rremote: Counting objects:  58% (97/167)\rremote: Counting objects:  59% (99/167)\rremote: Counting objects:  60% (101/167)\rremote: Counting objects:  61% (102/167)\rremote: Counting objects:  62% (104/167)\rremote: Counting objects:  63% (106/167)\rremote: Counting objects:  64% (107/167)\rremote: Counting objects:  65% (109/167)\rremote: Counting objects:  66% (111/167)\rremote: Counting objects:  67% (112/167)\rremote: Counting objects:  68% (114/167)\rremote: Counting objects:  69% (116/167)\rremote: Counting objects:  70% (117/167)\rremote: Counting objects:  71% (119/167)\rremote: Counting objects:  72% (121/167)\rremote: Counting objects:  73% (122/167)\rremote: Counting objects:  74% (124/167)\rremote: Counting objects:  75% (126/167)\rremote: Counting objects:  76% (127/167)\rremote: Counting objects:  77% (129/167)\rremote: Counting objects:  78% (131/167)\rremote: Counting objects:  79% (132/167)\rremote: Counting objects:  80% (134/167)\rremote: Counting objects:  81% (136/167)\rremote: Counting objects:  82% (137/167)\rremote: Counting objects:  83% (139/167)\rremote: Counting objects:  84% (141/167)\rremote: Counting objects:  85% (142/167)\rremote: Counting objects:  86% (144/167)\rremote: Counting objects:  87% (146/167)\rremote: Counting objects:  88% (147/167)\rremote: Counting objects:  89% (149/167)\rremote: Counting objects:  90% (151/167)\rremote: Counting objects:  91% (152/167)\rremote: Counting objects:  92% (154/167)\rremote: Counting objects:  93% (156/167)\rremote: Counting objects:  94% (157/167)\rremote: Counting objects:  95% (159/167)\rremote: Counting objects:  96% (161/167)\rremote: Counting objects:  97% (162/167)\rremote: Counting objects:  98% (164/167)\rremote: Counting objects:  99% (166/167)\rremote: Counting objects: 100% (167/167)\rremote: Counting objects: 100% (167/167), done.\r\nremote: Compressing objects:   1% (1/76)\rremote: Compressing objects:   2% (2/76)\rremote: Compressing objects:   3% (3/76)\rremote: Compressing objects:   5% (4/76)\rremote: Compressing objects:   6% (5/76)\rremote: Compressing objects:   7% (6/76)\rremote: Compressing objects:   9% (7/76)\rremote: Compressing objects:  10% (8/76)\r",,terminal_output
+14,68791,"TERMINAL",0,0,"remote: Compressing objects:  11% (9/76)\rremote: Compressing objects:  13% (10/76)\rremote: Compressing objects:  14% (11/76)\rremote: Compressing objects:  15% (12/76)\rremote: Compressing objects:  17% (13/76)\rremote: Compressing objects:  18% (14/76)\rremote: Compressing objects:  19% (15/76)\rremote: Compressing objects:  21% (16/76)\rremote: Compressing objects:  22% (17/76)\rremote: Compressing objects:  23% (18/76)\rremote: Compressing objects:  25% (19/76)\rremote: Compressing objects:  26% (20/76)\rremote: Compressing objects:  27% (21/76)\rremote: Compressing objects:  28% (22/76)\rremote: Compressing objects:  30% (23/76)\rremote: Compressing objects:  31% (24/76)\rremote: Compressing objects:  32% (25/76)\rremote: Compressing objects:  34% (26/76)\rremote: Compressing objects:  35% (27/76)\rremote: Compressing objects:  36% (28/76)\rremote: Compressing objects:  38% (29/76)\rremote: Compressing objects:  39% (30/76)\rremote: Compressing objects:  40% (31/76)\rremote: Compressing objects:  42% (32/76)\rremote: Compressing objects:  43% (33/76)\rremote: Compressing objects:  44% (34/76)\rremote: Compressing objects:  46% (35/76)\rremote: Compressing objects:  47% (36/76)\rremote: Compressing objects:  48% (37/76)\rremote: Compressing objects:  50% (38/76)\rremote: Compressing objects:  51% (39/76)\rremote: Compressing objects:  52% (40/76)\rremote: Compressing objects:  53% (41/76)\rremote: Compressing objects:  55% (42/76)\rremote: Compressing objects:  56% (43/76)\rremote: Compressing objects:  57% (44/76)\rremote: Compressing objects:  59% (45/76)\rremote: Compressing objects:  60% (46/76)\rremote: Compressing objects:  61% (47/76)\rremote: Compressing objects:  63% (48/76)\rremote: Compressing objects:  64% (49/76)\rremote: Compressing objects:  65% (50/76)\rremote: Compressing objects:  67% (51/76)\rremote: Compressing objects:  68% (52/76)\rremote: Compressing objects:  69% (53/76)\rremote: Compressing objects:  71% (54/76)\rremote: Compressing objects:  72% (55/76)\rremote: Compressing objects:  73% (56/76)\rremote: Compressing objects:  75% (57/76)\rremote: Compressing objects:  76% (58/76)\rremote: Compressing objects:  77% (59/76)\rremote: Compressing objects:  78% (60/76)\rremote: Compressing objects:  80% (61/76)\rremote: Compressing objects:  81% (62/76)\rremote: Compressing objects:  82% (63/76)\rremote: Compressing objects:  84% (64/76)\rremote: Compressing objects:  85% (65/76)\rremote: Compressing objects:  86% (66/76)\rremote: Compressing objects:  88% (67/76)\rremote: Compressing objects:  89% (68/76)\rremote: Compressing objects:  90% (69/76)\rremote: Compressing objects:  92% (70/76)\rremote: Compressing objects:  93% (71/76)\rremote: Compressing objects:  94% (72/76)\rremote: Compressing objects:  96% (73/76)\rremote: Compressing objects:  97% (74/76)\rremote: Compressing objects:  98% (75/76)\rremote: Compressing objects: 100% (76/76)\rremote: Compressing objects: 100% (76/76), done.\r\nReceiving objects:   0% (1/167)\rReceiving objects:   1% (2/167)\rReceiving objects:   2% (4/167)\rReceiving objects:   3% (6/167)\rReceiving objects:   4% (7/167)\rReceiving objects:   5% (9/167)\rReceiving objects:   6% (11/167)\rReceiving objects:   7% (12/167)\rReceiving objects:   8% (14/167)\rReceiving objects:   9% (16/167)\rReceiving objects:  10% (17/167)\rReceiving objects:  11% (19/167)\rReceiving objects:  12% (21/167)\rReceiving objects:  13% (22/167)\rReceiving objects:  14% (24/167)\rReceiving objects:  15% (26/167)\rReceiving objects:  16% (27/167)\rReceiving objects:  17% (29/167)\rReceiving objects:  18% (31/167)\rReceiving objects:  19% (32/167)\rReceiving objects:  20% (34/167)\rReceiving objects:  21% (36/167)\rReceiving objects:  22% (37/167)\rReceiving objects:  23% (39/167)\rReceiving objects:  24% (41/167)\rReceiving objects:  25% (42/167)\rReceiving objects:  26% (44/167)\rReceiving objects:  27% (46/167)\rReceiving objects:  28% (47/167)\rReceiving objects:  29% (49/167)\rReceiving objects:  30% (51/167)\rReceiving objects:  31% (52/167)\rReceiving objects:  32% (54/167)\rReceiving objects:  33% (56/167)\rReceiving objects:  34% (57/167)\rReceiving objects:  35% (59/167)\rReceiving objects:  36% (61/167)\rReceiving objects:  37% (62/167)\rReceiving objects:  38% (64/167)\rReceiving objects:  39% (66/167)\rReceiving objects:  40% (67/167)\rReceiving objects:  41% (69/167)\rReceiving objects:  42% (71/167)\rReceiving objects:  43% (72/167)\rReceiving objects:  44% (74/167)\rReceiving objects:  45% (76/167)\rReceiving objects:  46% (77/167)\rReceiving objects:  47% (79/167)\rReceiving objects:  48% (81/167)\rReceiving objects:  49% (82/167)\rReceiving objects:  50% (84/167)\rReceiving objects:  51% (86/167)\rReceiving objects:  52% (87/167)\rReceiving objects:  53% (89/167)\rReceiving objects:  54% (91/167)\rReceiving objects:  55% (92/167)\rReceiving objects:  56% (94/167)\rReceiving objects:  57% (96/167)\rReceiving objects:  58% (97/167)\rReceiving objects:  59% (99/167)\rReceiving objects:  60% (101/167)\rReceiving objects:  61% (102/167)\rReceiving objects:  62% (104/167)\rReceiving objects:  63% (106/167)\rReceiving objects:  64% (107/167)\rReceiving objects:  65% (109/167)\rReceiving objects:  66% (111/167)\rReceiving objects:  67% (112/167)\rReceiving objects:  68% (114/167)\rReceiving objects:  69% (116/167)\rReceiving objects:  70% (117/167)\rReceiving objects:  71% (119/167)\rReceiving objects:  72% (121/167)\rReceiving objects:  73% (122/167)\rReceiving objects:  74% (124/167)\rReceiving objects:  75% (126/167)\rReceiving objects:  76% (127/167)\rReceiving objects:  77% (129/167)\rReceiving objects:  78% (131/167)\rReceiving objects:  79% (132/167)\rremote: Total 167 (delta 93), reused 158 (delta 85), pack-reused 0 (from 0)\r\nReceiving objects:  80% (134/167)\rReceiving objects:  81% (136/167)\rReceiving objects:  82% (137/167)\rReceiving objects:  83% (139/167)\rReceiving objects:  84% (141/167)\rReceiving objects:  85% (142/167)\rReceiving objects:  86% (144/167)\rReceiving objects:  87% (146/167)\rReceiving objects:  88% (147/167)\rReceiving objects:  89% (149/167)\rReceiving objects:  90% (151/167)\rReceiving objects:  91% (152/167)\rReceiving objects:  92% (154/167)\rReceiving objects:  93% (156/167)\rReceiving objects:  94% (157/167)\rReceiving objects:  95% (159/167)\rReceiving objects:  96% (161/167)\rReceiving objects:  97% (162/167)\rReceiving objects:  98% (164/167)\rReceiving objects:  99% (166/167)\rReceiving objects: 100% (167/167)\rReceiving objects: 100% (167/167), 29.24 KiB | 332.00 KiB/s, done.\r\nResolving deltas:   0% (0/93)\rResolving deltas:   1% (1/93)\rResolving deltas:   2% (2/93)\rResolving deltas:   3% (3/93)\rResolving deltas:   4% (4/93)\rResolving deltas:   5% (5/93)\rResolving deltas:   6% (6/93)\rResolving deltas:   7% (7/93)\rResolving deltas:   8% (8/93)\rResolving deltas:   9% (9/93)\rResolving deltas:  10% (10/93)\rResolving deltas:  11% (11/93)\rResolving deltas:  12% (12/93)\rResolving deltas:  13% (13/93)\rResolving deltas:  16% (15/93)\rResolving deltas:  17% (16/93)\rResolving deltas:  18% (17/93)\rResolving deltas:  19% (18/93)\rResolving deltas:  20% (19/93)\rResolving deltas:  21% (20/93)\rResolving deltas:  24% (23/93)\rResolving deltas:  25% (24/93)\rResolving deltas:  26% (25/93)\rResolving deltas:  27% (26/93)\rResolving deltas:  29% (27/93)\rResolving deltas:  30% (28/93)\rResolving deltas:  31% (29/93)\rResolving deltas:  32% (30/93)\rResolving deltas:  35% (33/93)\rResolving deltas:  38% (36/93)\rResolving deltas:  39% (37/93)\rResolving deltas:  40% (38/93)\rResolving deltas:  41% (39/93)\rResolving deltas:  43% (40/93)\rResolving deltas:  44% (41/93)\rResolving deltas:  46% (43/93)\rResolving deltas:  48% (45/93)\rResolving deltas:  49% (46/93)\rResolving deltas:  50% (47/93)\rResolving deltas:  51% (48/93)\rResolving deltas:  56% (53/93)\rResolving deltas:  58% (54/93)\rResolving deltas:  59% (55/93)\rResolving deltas:  63% (59/93)\rResolving deltas:  64% (60/93)\rResolving deltas:  65% (61/93)\rResolving deltas:  66% (62/93)\rResolving deltas:  67% (63/93)\rResolving deltas:  68% (64/93)\rResolving deltas:  72% (67/93)\rResolving deltas:  73% (68/93)\rResolving deltas:  74% (69/93)\rResolving deltas:  75% (70/93)\rResolving deltas:  76% (71/93)\rResolving deltas:  77% (72/93)\rResolving deltas:  78% (73/93)\rResolving deltas:  79% (74/93)\rResolving deltas:  80% (75/93)\rResolving deltas:  81% (76/93)\rResolving deltas:  82% (77/93)\rResolving deltas:  83% (78/93)\rResolving deltas:  84% (79/93)\rResolving deltas:  86% (80/93)\rResolving deltas:  87% (81/93)\rResolving deltas:  88% (82/93)\rResolving deltas:  89% (83/93)\rResolving deltas:  90% (84/93)\rResolving deltas:  91% (85/93)\rResolving deltas:  92% (86/93)\rResolving deltas:  93% (87/93)\rResolving deltas:  94% (88/93)\rResolving deltas:  95% (89/93)\rResolving deltas:  96% (90/93)\rResolving deltas:  97% (91/93)\rResolving deltas:  98% (92/93)\rResolving deltas: 100% (93/93)\rResolving deltas: 100% (93/93), done.\r\n",,terminal_output
+15,68984,"TERMINAL",0,0,"]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;0",,terminal_output
+16,111140,"TERMINAL",0,0,"ls",,terminal_command
+17,111153,"TERMINAL",0,0,"]633;E;2025-06-26 20:19:58 ls;b8c83623-699a-4617-a1fd-25c482b22092]633;Cgenerate_dataset.py                        logs                    requirements.txt  single_batch.npy                train_lam_tf_seeding.py\r\ngeneration_1750681785.3743937.gif          models                  sample.py         slurm                           train_tokenizer_logging.py\r\ngeneration_video_0_1750688558.8735778.gif  notes.md                sample_results    train_dynamics.py               train_tokenizer.py\r\ngenie.py                                   __pycache__             sandbox           train_dynamics_single_batch.py  train_tokenizer_single_batch.py\r\njafar                                      README.md               sbatch_scripts    train_lam.py                    utils\r\nLICENSE                                    requirements_franz.txt  shell_scripts     train_lam_single_batch.py       wandb\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;0",,terminal_output
+18,112454,"TERMINAL",0,0,"gs",,terminal_command
+19,112492,"TERMINAL",0,0,"]633;E;2025-06-26 20:19:59 gs;b8c83623-699a-4617-a1fd-25c482b22092]633;COn branch log-time-train-step\r\nYour branch is up to date with 'origin/log-time-train-step'.\r\n\r\nChanges not staged for commit:\r\n  (use ""git add <file>..."" to update what will be committed)\r\n  (use ""git restore <file>..."" to discard changes in working directory)\r\n\tmodified:   .gitignore\r\n\tmodified:   train_lam.py\r\n\tmodified:   train_tokenizer.py\r\n\r\nUntracked files:\r\n  (use ""git add <file>..."" to include in what will be committed)\r\n\tjafar/\r\n\tnotes.md\r\n\trequirements_franz.txt\r\n\tslurm/\r\n\ttrain_dynamics_single_batch.py\r\n\ttrain_lam_single_batch.py\r\n\ttrain_lam_tf_seeding.py\r\n\ttrain_tokenizer_logging.py\r\n\ttrain_tokenizer_single_batch.py\r\n\r\nno changes added to commit (use ""git add"" and/or ""git commit -a"")\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;0",,terminal_output
+20,127523,"TERMINAL",0,0,"rm -r jafar/",,terminal_command
+21,127575,"TERMINAL",0,0,"]633;E;2025-06-26 20:20:14 rm -r jafar/;b8c83623-699a-4617-a1fd-25c482b22092]633;C",,terminal_output
+22,127808,"TERMINAL",0,0,"rm: remove write-protected regular file 'jafar/.git/objects/pack/pack-c0a30aea3a054e61ce30b8f1ee49202e4a180479.pack'? ",,terminal_output
+23,128903,"TERMINAL",0,0,"^C\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;130",,terminal_output
+24,131538,"TERMINAL",0,0,"rm -rf jafar/",,terminal_command
+25,131584,"TERMINAL",0,0,"]633;E;2025-06-26 20:20:18 rm -rf jafar/;b8c83623-699a-4617-a1fd-25c482b22092]633;C]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;0",,terminal_output
+26,137483,"TERMINAL",0,0,"ls",,terminal_command
+27,137491,"TERMINAL",0,0,"]633;E;2025-06-26 20:20:24 ls;b8c83623-699a-4617-a1fd-25c482b22092]633;Cgenerate_dataset.py                        models                  sample.py         slurm                           train_tokenizer_logging.py\r\ngeneration_1750681785.3743937.gif          notes.md                sample_results    train_dynamics.py               train_tokenizer.py\r\ngeneration_video_0_1750688558.8735778.gif  __pycache__             sandbox           train_dynamics_single_batch.py  train_tokenizer_single_batch.py\r\ngenie.py                                   README.md               sbatch_scripts    train_lam.py                    utils\r\nLICENSE                                    requirements_franz.txt  shell_scripts     train_lam_single_batch.py       wandb\r\nlogs                                       requirements.txt        single_batch.npy  train_lam_tf_seeding.py\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;0",,terminal_output
+28,140264,"TERMINAL",0,0,"cd j^C",,terminal_command
+29,140272,"TERMINAL",0,0,"^C[?2004l\r[?2004h[?2004l\r\r\n]633;E;;b8c83623-699a-4617-a1fd-25c482b22092]633;C]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D",,terminal_output
+30,140471,"TERMINAL",0,0,"^C",,terminal_command
+31,140485,"TERMINAL",0,0,"^C[?2004l\r[?2004h[?2004l\r\r\n]633;E;;b8c83623-699a-4617-a1fd-25c482b22092]633;C]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D",,terminal_output
+32,140616,"TERMINAL",0,0,"^C",,terminal_command
+33,140621,"TERMINAL",0,0,"^C[?2004l\r[?2004h[?2004l\r\r\n]633;E;;b8c83623-699a-4617-a1fd-25c482b22092]633;C]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D",,terminal_output
+34,146736,"TERMINAL",0,0,"git clone git@github.com:p-doom/slurm.git",,terminal_command
+35,146752,"TERMINAL",0,0,"]633;E;2025-06-26 20:20:34 git clone git@github.com:p-doom/slurm.git;b8c83623-699a-4617-a1fd-25c482b22092]633;Cfatal: destination path 'slurm' already exists and is not an empty directory.\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_run_overfit]633;D;128",,terminal_output
+36,176180,"sbatch_scripts/train_dyn/train_dyn_overfit_tfrecord_10.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=4\n#SBATCH --time=24:00:00\n#SBATCH --partition=accelerated\n#SBATCH --account=hk-project-p0023960\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:4\n#SBATCH --output=logs/logs_training_dyn/%x_%j.log\n#SBATCH --error=logs/logs_training_dyn/%x_%j.log\n#SBATCH --mail-user=avocadoaling@gmail.com\n#SBATCH --job-name=train_dyn_overfit_tfrecord_10\n#SBATCH --mail-type=ALL\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv_jafar/bin/activate\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\n\nCHECKPOINT_DIR=$ws_dir/checkpoints/dyn/$job_name_$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\n\ndata_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_tfrecords_500_shards_overfit_10'\n\nlam_checkpoint='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/lam/3286113/lam_1750693337_13500'\ntokenizer_checkpoint='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer/3286114/tokenizer_1750693337_13500'\n\nsrun python train_dynamics.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --lam_checkpoint $lam_checkpoint \\n    --tokenizer_checkpoint $tokenizer_checkpoint \\n    --num_steps=300000 \\n    --warmup_steps=10000 \\n    --batch_size=48 \\n    --log_checkpoint_interval=500 \\n    --log_image_interval=10 \\n    --seed=0 \\n    --min_lr=0.0000866 \\n    --max_lr=0.0000866 \\n    --log \\n    --entity instant-uv \\n    --data_dir $data_dir \\n    --project jafar\n\n\n",shellscript,tab
+37,199579,"slurm/dev/alfred/train_dyn/train_dyn_knoms_full.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=4\n#SBATCH --time=24:00:00\n#SBATCH --partition=accelerated\n#SBATCH --account=hk-project-p0023960\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:4\n#SBATCH --output=logs/logs_training_dyn/%x_%j.log\n#SBATCH --error=logs/logs_training_dyn/%x_%j.log\n#SBATCH --mail-user=avocadoaling@gmail.com\n#SBATCH --job-name=train_dyn_knoms_full\n#SBATCH --mail-type=ALL\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv_jafar/bin/activate\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\n\nCHECKPOINT_DIR=$ws_dir/checkpoints/dyn/$job_name_$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\n\ndata_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_tfrecords_500_shards'\n\n\nlam_checkpoint='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/lam/3277318/lam_1750444032_10000'\ntokenizer_checkpoint='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer/3273828/tokenizer_1750269881_15500'\n\nsrun python train_dynamics.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --lam_checkpoint $lam_checkpoint \\n    --tokenizer_checkpoint $tokenizer_checkpoint \\n    --batch_size=48 \\n    --log_checkpoint_interval=500 \\n    --log_image_interval=50 \\n    --seed=0 \\n    --min_lr=0.0000866 \\n    --max_lr=0.0000866 \\n    --log \\n    --entity instant-uv \\n    --data_dir $data_dir \\n    --project jafar\n\n",shellscript,tab
+38,753939,"train_tokenizer.py",0,0,"",python,tab
+39,754710,"train_tokenizer.py",425,0,"",python,selection_mouse
+40,754715,"train_tokenizer.py",424,0,"",python,selection_command
+41,755029,"TERMINAL",0,0,"bash",,terminal_focus
+42,762041,"train_tokenizer.py",6505,0,"",python,selection_mouse
+43,762640,"train_tokenizer.py",6538,0,"",python,selection_mouse
+44,763556,"train_tokenizer.py",6537,0,"",python,selection_command
+45,769994,"train_tokenizer.py",6537,0,"#",python,content
+46,770001,"train_tokenizer.py",6538,0,"",python,selection_keyboard
+47,770142,"train_tokenizer.py",6537,0,"",python,selection_command
+48,770505,"train_tokenizer.py",6572,0,"",python,selection_command
+49,770770,"train_tokenizer.py",6563,0,"\n        ",python,content
+50,770774,"train_tokenizer.py",6572,0,"    ",python,content
+51,771351,"train_tokenizer.py",6572,4,"",python,content
+52,772273,"train_tokenizer.py",6572,0,"s",python,content
+53,772274,"train_tokenizer.py",6573,0,"",python,selection_keyboard
+54,772309,"train_tokenizer.py",6573,0,"i",python,content
+55,772311,"train_tokenizer.py",6574,0,"",python,selection_keyboard
+56,772377,"train_tokenizer.py",6574,0,"n",python,content
+57,772379,"train_tokenizer.py",6575,0,"",python,selection_keyboard
+58,772459,"train_tokenizer.py",6575,0,"g",python,content
+59,772460,"train_tokenizer.py",6576,0,"",python,selection_keyboard
+60,772575,"train_tokenizer.py",6576,0,"l",python,content
+61,772577,"train_tokenizer.py",6577,0,"",python,selection_keyboard
+62,772691,"train_tokenizer.py",6577,0,"e",python,content
+63,772693,"train_tokenizer.py",6578,0,"",python,selection_keyboard
+64,774046,"train_tokenizer.py",6578,0,"_video = next(dataloader)",python,content
+65,774826,"train_tokenizer.py",6602,0,"",python,selection_command
+66,774916,"train_tokenizer.py",6592,0,"",python,selection_command
+67,775307,"train_tokenizer.py",6591,0,"",python,selection_command
+68,775754,"train_tokenizer.py",6587,0,"",python,selection_command
+69,775990,"train_tokenizer.py",6587,16,"",python,content
+70,775994,"train_tokenizer.py",6586,0,"",python,selection_command
+71,776097,"train_tokenizer.py",6587,0,"",python,selection_command
+72,776303,"train_tokenizer.py",6587,0,"n",python,content
+73,776304,"train_tokenizer.py",6588,0,"",python,selection_keyboard
+74,776428,"train_tokenizer.py",6588,0,"o",python,content
+75,776429,"train_tokenizer.py",6589,0,"",python,selection_keyboard
+76,777056,"train_tokenizer.py",6588,1,"",python,content
+77,777200,"train_tokenizer.py",6587,1,"",python,content
+78,778198,"train_tokenizer.py",6585,2,"",python,content
+79,778339,"train_tokenizer.py",6572,13,"",python,content
+80,778813,"train_tokenizer.py",6572,0,"v",python,content
+81,778815,"train_tokenizer.py",6573,0,"",python,selection_keyboard
+82,778880,"train_tokenizer.py",6573,0,"i",python,content
+83,778881,"train_tokenizer.py",6574,0,"",python,selection_keyboard
+84,779007,"train_tokenizer.py",6574,0,"d",python,content
+85,779008,"train_tokenizer.py",6575,0,"",python,selection_keyboard
+86,779146,"train_tokenizer.py",6575,0,"e",python,content
+87,779147,"train_tokenizer.py",6576,0,"",python,selection_keyboard
+88,779249,"train_tokenizer.py",6576,0,"o",python,content
+89,779251,"train_tokenizer.py",6577,0,"",python,selection_keyboard
+90,779789,"train_tokenizer.py",6577,0,"s",python,content
+91,779790,"train_tokenizer.py",6578,0,"",python,selection_keyboard
+92,780167,"train_tokenizer.py",6578,0," ",python,content
+93,780168,"train_tokenizer.py",6579,0,"",python,selection_keyboard
+94,781454,"train_tokenizer.py",6579,0,"-",python,content
+95,781456,"train_tokenizer.py",6580,0,"",python,selection_keyboard
+96,781879,"train_tokenizer.py",6579,1,"",python,content
+97,782402,"train_tokenizer.py",6579,0,"=",python,content
+98,782403,"train_tokenizer.py",6580,0,"",python,selection_keyboard
+99,782492,"train_tokenizer.py",6580,0," ",python,content
+100,782494,"train_tokenizer.py",6581,0,"",python,selection_keyboard
+101,782920,"train_tokenizer.py",6581,0,"n",python,content
+102,782921,"train_tokenizer.py",6582,0,"",python,selection_keyboard
+103,783049,"train_tokenizer.py",6582,0,"p",python,content
+104,783050,"train_tokenizer.py",6583,0,"",python,selection_keyboard
+105,783366,"train_tokenizer.py",6583,0,".",python,content
+106,783368,"train_tokenizer.py",6584,0,"",python,selection_keyboard
+107,783601,"train_tokenizer.py",6584,0,"l",python,content
+108,783601,"train_tokenizer.py",6585,0,"",python,selection_keyboard
+109,783739,"train_tokenizer.py",6585,0,"o",python,content
+110,783741,"train_tokenizer.py",6586,0,"",python,selection_keyboard
+111,783764,"train_tokenizer.py",6586,0,"a",python,content
+112,783765,"train_tokenizer.py",6587,0,"",python,selection_keyboard
+113,783908,"train_tokenizer.py",6587,0,"d",python,content
+114,783910,"train_tokenizer.py",6588,0,"",python,selection_keyboard
+115,784474,"train_tokenizer.py",6588,0,"()",python,content
+116,784476,"train_tokenizer.py",6589,0,"",python,selection_keyboard
+117,785970,"train_tokenizer.py",6589,0,"'/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_tfrecords_500_shards_overfit_10/shard_00000.npy'",python,content
+118,786550,"train_tokenizer.py",6704,0,"",python,selection_command
+119,787393,"train_tokenizer.py",6738,0,"",python,selection_command
+120,787617,"train_tokenizer.py",6707,0,"",python,selection_command
+121,787893,"train_tokenizer.py",6719,0,"",python,selection_command
+122,788376,"train_tokenizer.py",6719,1,"#",python,selection_command
+123,788741,"train_tokenizer.py",6707,13,"            #",python,selection_command
+124,789290,"train_tokenizer.py",6707,0,"",python,selection_command
+125,789930,"train_tokenizer.py",9402,1,"",python,content
+126,789931,"train_tokenizer.py",9380,1,"",python,content
+127,789931,"train_tokenizer.py",9341,1,"",python,content
+128,789931,"train_tokenizer.py",9323,1,"",python,content
+129,789931,"train_tokenizer.py",9282,1,"",python,content
+130,789931,"train_tokenizer.py",9256,1,"",python,content
+131,789931,"train_tokenizer.py",9168,1,"",python,content
+132,789931,"train_tokenizer.py",9127,1,"",python,content
+133,789931,"train_tokenizer.py",9059,1,"",python,content
+134,789931,"train_tokenizer.py",8984,1,"",python,content
+135,789931,"train_tokenizer.py",8938,1,"",python,content
+136,789931,"train_tokenizer.py",8881,1,"",python,content
+137,789932,"train_tokenizer.py",8835,1,"",python,content
+138,789932,"train_tokenizer.py",8809,1,"",python,content
+139,789932,"train_tokenizer.py",8778,1,"",python,content
+140,789932,"train_tokenizer.py",8702,1,"",python,content
+141,789932,"train_tokenizer.py",8647,1,"",python,content
+142,789932,"train_tokenizer.py",8574,1,"",python,content
+143,789932,"train_tokenizer.py",8504,1,"",python,content
+144,789932,"train_tokenizer.py",8461,1,"",python,content
+145,789932,"train_tokenizer.py",8412,1,"",python,content
+146,789932,"train_tokenizer.py",8335,1,"",python,content
+147,789932,"train_tokenizer.py",8257,1,"",python,content
+148,789932,"train_tokenizer.py",8177,1,"",python,content
+149,789932,"train_tokenizer.py",8155,1,"",python,content
+150,789932,"train_tokenizer.py",8086,1,"",python,content
+151,789933,"train_tokenizer.py",8031,1,"",python,content
+152,789933,"train_tokenizer.py",7949,1,"",python,content
+153,789933,"train_tokenizer.py",7897,1,"",python,content
+154,789933,"train_tokenizer.py",7848,1,"",python,content
+155,789933,"train_tokenizer.py",7792,1,"",python,content
+156,789933,"train_tokenizer.py",7770,1,"",python,content
+157,789933,"train_tokenizer.py",7744,1,"",python,content
+158,789933,"train_tokenizer.py",7705,1,"",python,content
+159,789933,"train_tokenizer.py",7647,1,"",python,content
+160,789933,"train_tokenizer.py",7605,1,"",python,content
+161,789933,"train_tokenizer.py",7563,1,"",python,content
+162,789933,"train_tokenizer.py",7537,1,"",python,content
+163,789933,"train_tokenizer.py",7506,1,"",python,content
+164,789933,"train_tokenizer.py",7427,1,"",python,content
+165,789934,"train_tokenizer.py",7402,1,"",python,content
+166,789934,"train_tokenizer.py",7372,1,"",python,content
+167,789934,"train_tokenizer.py",7371,1,"",python,content
+168,789934,"train_tokenizer.py",7349,1,"",python,content
+169,789934,"train_tokenizer.py",7272,1,"",python,content
+170,789934,"train_tokenizer.py",7211,1,"",python,content
+171,789934,"train_tokenizer.py",7131,1,"",python,content
+172,789934,"train_tokenizer.py",7054,1,"",python,content
+173,789934,"train_tokenizer.py",7053,1,"",python,content
+174,789934,"train_tokenizer.py",6968,1,"",python,content
+175,789934,"train_tokenizer.py",6954,1,"",python,content
+176,789934,"train_tokenizer.py",6886,1,"",python,content
+177,789935,"train_tokenizer.py",6841,1,"",python,content
+178,789935,"train_tokenizer.py",6840,1,"",python,content
+179,789935,"train_tokenizer.py",6777,1,"",python,content
+180,789935,"train_tokenizer.py",6776,1,"",python,content
+181,789935,"train_tokenizer.py",6739,1,"",python,content
+182,789935,"train_tokenizer.py",6706,1,"",python,content
+183,790488,"train_tokenizer.py",6705,0,"",python,selection_command
+184,790590,"train_tokenizer.py",9345,0,"\n",python,content
+185,790591,"train_tokenizer.py",9324,0,"\n",python,content
+186,790593,"train_tokenizer.py",9286,0,"\n",python,content
+187,790594,"train_tokenizer.py",9269,0,"\n",python,content
+188,790597,"train_tokenizer.py",9229,0,"\n",python,content
+189,790601,"train_tokenizer.py",9204,0,"\n",python,content
+190,790602,"train_tokenizer.py",9117,0,"\n",python,content
+191,790603,"train_tokenizer.py",9077,0,"\n",python,content
+192,790604,"train_tokenizer.py",9010,0,"\n",python,content
+193,790605,"train_tokenizer.py",8936,0,"\n",python,content
+194,790606,"train_tokenizer.py",8891,0,"\n",python,content
+195,790607,"train_tokenizer.py",8835,0,"\n",python,content
+196,790608,"train_tokenizer.py",8790,0,"\n",python,content
+197,790609,"train_tokenizer.py",8765,0,"\n",python,content
+198,790613,"train_tokenizer.py",8735,0,"\n",python,content
+199,790614,"train_tokenizer.py",8660,0,"\n",python,content
+200,790615,"train_tokenizer.py",8606,0,"\n",python,content
+201,790616,"train_tokenizer.py",8534,0,"\n",python,content
+202,790617,"train_tokenizer.py",8465,0,"\n",python,content
+203,790618,"train_tokenizer.py",8423,0,"\n",python,content
+204,790620,"train_tokenizer.py",8375,0,"\n",python,content
+205,790621,"train_tokenizer.py",8299,0,"\n",python,content
+206,790622,"train_tokenizer.py",8222,0,"\n",python,content
+207,790623,"train_tokenizer.py",8143,0,"\n",python,content
+208,790624,"train_tokenizer.py",8122,0,"\n",python,content
+209,790632,"train_tokenizer.py",8054,0,"\n",python,content
+210,790634,"train_tokenizer.py",8000,0,"\n",python,content
+211,790635,"train_tokenizer.py",7919,0,"\n",python,content
+212,790637,"train_tokenizer.py",7868,0,"\n",python,content
+213,790640,"train_tokenizer.py",7820,0,"\n",python,content
+214,790652,"train_tokenizer.py",7765,0,"\n",python,content
+215,790653,"train_tokenizer.py",7744,0,"\n",python,content
+216,790654,"train_tokenizer.py",7719,0,"\n",python,content
+217,790657,"train_tokenizer.py",7681,0,"\n",python,content
+218,790661,"train_tokenizer.py",7624,0,"\n",python,content
+219,790662,"train_tokenizer.py",7583,0,"\n",python,content
+220,790663,"train_tokenizer.py",7542,0,"\n",python,content
+221,790664,"train_tokenizer.py",7517,0,"\n",python,content
+222,790669,"train_tokenizer.py",7487,0,"\n",python,content
+223,790671,"train_tokenizer.py",7409,0,"\n",python,content
+224,790673,"train_tokenizer.py",7385,0,"\n",python,content
+225,790674,"train_tokenizer.py",7356,0,"\n\n",python,content
+226,790679,"train_tokenizer.py",7335,0,"\n",python,content
+227,790680,"train_tokenizer.py",7259,0,"\n",python,content
+228,790682,"train_tokenizer.py",7199,0,"\n",python,content
+229,790682,"train_tokenizer.py",7120,0,"\n",python,content
+230,790683,"train_tokenizer.py",7044,0,"\n\n",python,content
+231,790684,"train_tokenizer.py",6960,0,"\n",python,content
+232,790685,"train_tokenizer.py",6947,0,"\n",python,content
+233,790685,"train_tokenizer.py",6880,0,"\n",python,content
+234,790686,"train_tokenizer.py",6836,0,"\n\n",python,content
+235,790687,"train_tokenizer.py",6774,0,"\n\n",python,content
+236,790688,"train_tokenizer.py",6738,0,"\n",python,content
+237,790688,"train_tokenizer.py",6706,0,"\n",python,content
+238,790689,"train_tokenizer.py",6707,0,"",python,selection_command
+239,791209,"train_tokenizer.py",6719,0,"",python,selection_command
+240,791637,"train_tokenizer.py",6719,1,"#",python,selection_command
+241,792183,"train_tokenizer.py",6719,0,"",python,selection_command
+242,792887,"train_tokenizer.py",6576,0,"",python,selection_command
+243,793410,"train_tokenizer.py",6572,0,"",python,selection_command
+244,793664,"train_tokenizer.py",6572,0,"    ",python,content
+245,794097,"train_tokenizer.py",6575,0,"",python,selection_command
+246,794388,"train_tokenizer.py",6540,0,"",python,selection_command
+247,795966,"train_tokenizer.py",6529,35,"",python,content
+248,795978,"train_tokenizer.py",6541,0,"",python,selection_command
+249,797805,"train_tokenizer.py",6528,0,"\n        ",python,content
+250,797815,"train_tokenizer.py",6537,0,"    ",python,content
+251,798655,"train_tokenizer.py",6529,12,"",python,content
+252,798768,"train_tokenizer.py",6529,1,"",python,content
+253,798771,"train_tokenizer.py",6541,0,"",python,selection_command
+254,799153,"train_tokenizer.py",6537,0,"#for videos in dataloader:\n        ",python,content
+255,799160,"train_tokenizer.py",6540,0,"",python,selection_command
+256,799721,"train_tokenizer.py",6575,0,"",python,selection_command
+257,800089,"train_tokenizer.py",6563,0,"\n        ",python,content
+258,800104,"train_tokenizer.py",6572,0,"    ",python,content
+259,800648,"train_tokenizer.py",6572,4,"",python,content
+260,801051,"train_tokenizer.py",6572,0,"w",python,content
+261,801052,"train_tokenizer.py",6573,0,"",python,selection_keyboard
+262,801113,"train_tokenizer.py",6573,0,"h",python,content
+263,801115,"train_tokenizer.py",6574,0,"",python,selection_keyboard
+264,801186,"train_tokenizer.py",6574,0,"i",python,content
+265,801188,"train_tokenizer.py",6575,0,"",python,selection_keyboard
+266,801281,"train_tokenizer.py",6575,0,"l",python,content
+267,801283,"train_tokenizer.py",6576,0,"",python,selection_keyboard
+268,801421,"train_tokenizer.py",6576,0,"e",python,content
+269,801424,"train_tokenizer.py",6577,0,"",python,selection_keyboard
+270,802046,"train_tokenizer.py",6577,0," True:",python,content
+271,802214,"train_tokenizer.py",6582,0,"",python,selection_command
+272,802397,"train_tokenizer.py",6602,0,"",python,selection_command
+273,802763,"train_tokenizer.py",6749,0,"",python,selection_command
+274,803187,"train_tokenizer.py",6602,0,"",python,selection_command
+275,804448,"train_tokenizer.py",6584,147,"",python,content
+276,804453,"train_tokenizer.py",6596,0,"",python,selection_command
+277,804513,"train_tokenizer.py",6576,0,"",python,selection_command
+278,804750,"train_tokenizer.py",6564,0,"            videos = np.load('/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_tfrecords_500_shards_overfit_10/shard_00000.npy')\n",python,content
+279,804758,"train_tokenizer.py",6576,0,"",python,selection_command
+280,805486,"train_tokenizer.py",6572,4,"",python,content
+281,805699,"train_tokenizer.py",6571,0,"",python,selection_command
+282,805848,"train_tokenizer.py",6714,0,"",python,selection_command
+283,806682,"train_tokenizer.py",6706,0,"\n        ",python,content
+284,806887,"train_tokenizer.py",6707,8,"",python,content
+285,807036,"train_tokenizer.py",6708,0,"",python,selection_command
+286,808254,"train_tokenizer.py",6707,0,"",python,selection_command
+287,808623,"train_tokenizer.py",6707,1,"",python,content
+288,808626,"train_tokenizer.py",6715,0,"",python,selection_command
+289,811768,"train_tokenizer.py",6726,0,"",python,selection_mouse
+290,811770,"train_tokenizer.py",6725,0,"",python,selection_command
+291,812074,"train_tokenizer.py",6604,0,"",python,selection_mouse
+292,821039,"train_tokenizer.py",6721,0,"",python,selection_mouse
+293,823556,"train_tokenizer.py",6573,0,"",python,selection_mouse
+294,823738,"train_tokenizer.py",6572,6,"videos",python,selection_mouse
+295,839480,"train_tokenizer.py",6796,0,"",python,selection_mouse
+296,839485,"train_tokenizer.py",6795,0,"",python,selection_command

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-07efaead-a430-42b3-b659-119d2814db601751307487336-2025_06_30-20.19.49.655/source.csv

@@ -0,0 +1,65 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,17,"train_lam.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 dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\n\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    checkpoint: str = """"\n    # Optimization\n    batch_size: int = 36\n    vq_beta: float = 0.25\n    min_lr: float = 3e-6\n    max_lr: float = 3e-5\n    warmup_steps: int = 5000\n    vq_reset_thresh: int = 50\n    # LAM\n    model_dim: int = 512\n    latent_dim: int = 32\n    num_latents: int = 6\n    patch_size: int = 16\n    num_blocks: int = 8\n    num_heads: int = 8\n    dropout: float = 0.0\n    codebook_dropout: float = 0.0\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\n\nargs = tyro.cli(Args)\n\n\ndef lam_loss_fn(params, state, inputs):\n    # --- Compute loss ---\n    outputs = state.apply_fn(\n        params, inputs, training=True, rngs={""dropout"": inputs[""rng""]}\n    )\n    gt_future_frames = inputs[""videos""][:, 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 = pix.psnr(gt, recon).mean()\n    ssim = 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@jax.jit\ndef train_step(state, inputs, action_last_active):\n    # --- Update model ---\n    rng, inputs[""rng""] = jax.random.split(inputs[""rng""])\n    grad_fn = jax.value_and_grad(lam_loss_fn, has_aux=True, allow_int=True)\n    (loss, (recon, idx_counts, metrics)), grads = grad_fn(state.params, state, inputs)\n    state = state.apply_gradients(grads=grads)\n\n    # --- Reset inactive latent actions ---\n    codebook = state.params[""params""][""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\n    )\n    state.params[""params""][""vq""][""codebook""] = new_codebook\n    action_last_active = jnp.where(do_reset, 0, action_last_active)\n    return state, 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.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    lam = LatentActionModel(\n        in_dim=args.image_channels,\n        model_dim=args.model_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    )\n    # Track when each action was last sampled\n    action_last_active = jnp.zeros(args.num_latents)\n    image_shape = (args.image_height, args.image_width, args.image_channels)\n    rng, _rng = jax.random.split(rng)\n    inputs = dict(\n        videos=jnp.zeros(\n            (per_device_batch_size_for_init, args.seq_len, *image_shape),\n            dtype=jnp.float32,\n        ),\n        rng=_rng,\n    )\n    rng, _rng = jax.random.split(rng)\n    init_params = lam.init(_rng, 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=lam.apply, params=init_params, tx=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    train_state = jax.device_put(train_state, replicated_sharding)\n    action_last_active = jax.device_put(action_last_active, replicated_sharding)\n\n    # --- Load checkpoint ---\n    step = 0\n    if args.checkpoint:\n        restore_target = {""model"": train_state}\n        restore_args = orbax_utils.restore_args_from_target(restore_target)\n        train_state.params[""params""].update(\n            PyTreeCheckpointer()\n            .restore(args.checkpoint, item=restore_target, restore_args=restore_args)[\n                ""model""\n            ]\n            .params[""params""]\n        )\n        # Assume checkpoint is of the form tokenizer_<timestamp>_<step>\n        step += int(args.checkpoint.split(""_"")[-1])\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    print(f""Starting training from step {step}..."")\n    while step < args.num_steps:\n        for videos in dataloader:\n            # --- Train step ---\n            rng, _rng = jax.random.split(rng)\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(videos=videos, rng=_rng)\n            train_state, loss, recon, action_last_active, metrics = train_step(\n                train_state, inputs, action_last_active\n            )\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:]\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[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            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""lam_{ts}_{step}""),\n                    ckpt,\n                    save_args=save_args,\n                )\n            if step >= args.num_steps:\n                break\n",python,tab
+2,3802,"train_lam.py",1,0,"/home/hk-project-pai00039/tum_ind3695/tmp_crowd_source",python,content
+3,3813,"train_lam.py",54,0,"",python,selection_command
+4,5555,".gitignore",0,0,"*.pyc\n*.npy\n*.png\n*.gif\n\nwandb_key\ncheckpoints/\nwandb/\n__pycache__/\n",ignore,tab
+5,16626,".gitignore",68,0,"",ignore,selection_command
+6,16760,".gitignore",68,0,"\n",ignore,content
+7,17144,".gitignore",69,0,"\nlogs/\nsandbox/\nsbatch_scripts/\nvs-code-recorder/\ndata/\ndata_tfrecords/\nsh_scripts/\nutils/clip_checker.py\nutils/dataloader_seeding.py\nutils/preprocess_video_splitter_tmp.py\nrequirements_franz.txt\nsample_resolution_batches.py\ntrain_dynamics_*\ntrain_lam_*\ntrain_tokenizer_*\nnotes.md\nshell_scripts/\nslurm/\n\n",ignore,content
+8,17150,".gitignore",69,0,"",ignore,selection_command
+9,17970,".gitignore",69,1,"",ignore,content
+10,18341,".gitignore",372,0,"",ignore,selection_command
+11,18756,".gitignore",371,1,"",ignore,content
+12,19484,".gitignore",370,1,"",ignore,content
+13,19494,".gitignore",364,0,"",ignore,selection_command
+14,20220,"train_lam.py",0,0,"",python,tab
+15,34536,"train_lam.py",1,54,"",python,content
+16,34548,"train_lam.py",0,0,"",python,selection_command
+17,37422,"train_lam.py",0,0,"",python,tab
+18,37997,".gitignore",0,0,"",ignore,tab
+19,38057,"train_lam.py",0,0,"",python,tab
+20,38756,".gitignore",0,0,"",ignore,tab
+21,39088,"train_lam.py",0,0,"",python,tab
+22,41367,"train_lam.py",9640,0,"",python,selection_command
+23,42033,"train_lam.py",7904,0,"",python,selection_command
+24,42185,"train_lam.py",7632,0,"",python,selection_command
+25,42349,"train_lam.py",7419,0,"",python,selection_command
+26,42625,"train_lam.py",6793,0,"",python,selection_command
+27,42825,"train_lam.py",6234,0,"",python,selection_command
+28,43412,"train_lam.py",6793,0,"",python,selection_command
+29,43565,"train_lam.py",7419,0,"",python,selection_command
+30,43711,"train_lam.py",7632,0,"",python,selection_command
+31,46933,".gitignore",0,0,"",ignore,tab
+32,48165,"TERMINAL",0,0,"gs",,terminal_command
+33,48185,"TERMINAL",0,0,"]633;E;2025-06-30 20:20:37 gs;6979a12f-f382-4ba6-ae05-07968467e789]633;C",,terminal_output
+34,54816,"TERMINAL",0,0,"",,terminal_focus
+35,55117,"TERMINAL",0,0,"bash",,terminal_focus
+36,55380,"TERMINAL",0,0,"bash",,terminal_focus
+37,57408,"TERMINAL",0,0,"gs",,terminal_command
+38,57456,"TERMINAL",0,0,"]633;E;2025-06-30 20:20:47 gs;6fdaa671-1998-4205-8f9d-888450c2f397]633;COn branch main\r\nYour branch is behind 'origin/main' by 2 commits, and can be fast-forwarded.\r\n  (use ""git pull"" to update your local branch)\r\n\r\nChanges not staged for commit:\r\n  (use ""git add <file>..."" to update what will be committed)\r\n  (use ""git restore <file>..."" to discard changes in working directory)\r\n\tmodified:   .gitignore\r\n\r\nUntracked files:\r\n  (use ""git add <file>..."" to include in what will be committed)\r\n\tslurm_tmp/\r\n\tutils/dataloader_bak.py\r\n\r\nno changes added to commit (use ""git add"" and/or ""git commit -a"")\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_batch_size_scaling]633;D;0",,terminal_output
+39,64001,"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, seed):\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        seed: The seed for the random number generator.\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, seed=seed\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        seed=seed,\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
+40,66824,"utils/dataloader.py",2784,0,"",python,selection_command
+41,67354,"utils/dataloader.py",2784,1,"",python,content
+42,67492,"utils/dataloader.py",2784,1,"",python,content
+43,67653,"utils/dataloader.py",2784,1,"",python,content
+44,68510,"utils/dataloader.py",2784,0,"5",python,content
+45,68514,"utils/dataloader.py",2785,0,"",python,selection_keyboard
+46,68676,"utils/dataloader.py",2784,0,"",python,selection_command
+47,70131,"TERMINAL",0,0,"bash",,terminal_focus
+48,70135,".gitignore",0,0,"",ignore,tab
+49,83949,".gitignore",0,0,"",ignore,selection_command
+50,84161,".gitignore",6,0,"",ignore,selection_command
+51,84411,".gitignore",12,0,"",ignore,selection_command
+52,88051,"TERMINAL",0,0,"",,terminal_focus
+53,90004,"TERMINAL",0,0,"cd slurm/",,terminal_command
+54,90007,"TERMINAL",0,0,"]633;E;2025-06-30 20:21:19 cd slurm/;5160b2c5-8f6e-406f-95ef-45827d2d74ec]633;C]0;tum_ind3695@hkn1993:~/projects/jafar_batch_size_scaling/slurm]633;D;0",,terminal_output
+55,90543,"TERMINAL",0,0,"ls",,terminal_command
+56,90568,"TERMINAL",0,0,"]633;E;2025-06-30 20:21:20 ls;5160b2c5-8f6e-406f-95ef-45827d2d74ec]633;Ccommon  dev  jobs  README.md  templates  utils\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_batch_size_scaling/slurm]633;D;0",,terminal_output
+57,93183,"TERMINAL",0,0,"git pull",,terminal_command
+58,93234,"TERMINAL",0,0,"]633;E;2025-06-30 20:21:22 git pull;5160b2c5-8f6e-406f-95ef-45827d2d74ec]633;C",,terminal_output
+59,94727,"TERMINAL",0,0,"remote: Enumerating objects: 108, done.\r\nremote: Counting objects:   1% (1/85)\rremote: Counting objects:   2% (2/85)\rremote: Counting objects:   3% (3/85)\rremote: Counting objects:   4% (4/85)\rremote: Counting objects:   5% (5/85)\rremote: Counting objects:   7% (6/85)\rremote: Counting objects:   8% (7/85)\rremote: Counting objects:   9% (8/85)\rremote: Counting objects:  10% (9/85)\rremote: Counting objects:  11% (10/85)\rremote: Counting objects:  12% (11/85)\rremote: Counting objects:  14% (12/85)\rremote: Counting objects:  15% (13/85)\rremote: Counting objects:  16% (14/85)\rremote: Counting objects:  17% (15/85)\rremote: Counting objects:  18% (16/85)\rremote: Counting objects:  20% (17/85)\rremote: Counting objects:  21% (18/85)\rremote: Counting objects:  22% (19/85)\rremote: Counting objects:  23% (20/85)\rremote: Counting objects:  24% (21/85)\rremote: Counting objects:  25% (22/85)\rremote: Counting objects:  27% (23/85)\rremote: Counting objects:  28% (24/85)\rremote: Counting objects:  29% (25/85)\rremote: Counting objects:  30% (26/85)\rremote: Counting objects:  31% (27/85)\rremote: Counting objects:  32% (28/85)\rremote: Counting objects:  34% (29/85)\rremote: Counting objects:  35% (30/85)\rremote: Counting objects:  36% (31/85)\rremote: Counting objects:  37% (32/85)\rremote: Counting objects:  38% (33/85)\rremote: Counting objects:  40% (34/85)\rremote: Counting objects:  41% (35/85)\rremote: Counting objects:  42% (36/85)\rremote: Counting objects:  43% (37/85)\rremote: Counting objects:  44% (38/85)\rremote: Counting objects:  45% (39/85)\rremote: Counting objects:  47% (40/85)\rremote: Counting objects:  48% (41/85)\rremote: Counting objects:  49% (42/85)\rremote: Counting objects:  50% (43/85)\rremote: Counting objects:  51% (44/85)\rremote: Counting objects:  52% (45/85)\rremote: Counting objects:  54% (46/85)\rremote: Counting objects:  55% (47/85)\rremote: Counting objects:  56% (48/85)\rremote: Counting objects:  57% (49/85)\rremote: Counting objects:  58% (50/85)\rremote: Counting objects:  60% (51/85)\rremote: Counting objects:  61% (52/85)\rremote: Counting objects:  62% (53/85)\rremote: Counting objects:  63% (54/85)\rremote: Counting objects:  64% (55/85)\rremote: Counting objects:  65% (56/85)\rremote: Counting objects:  67% (57/85)\rremote: Counting objects:  68% (58/85)\rremote: Counting objects:  69% (59/85)\rremote: Counting objects:  70% (60/85)\rremote: Counting objects:  71% (61/85)\rremote: Counting objects:  72% (62/85)\rremote: Counting objects:  74% (63/85)\rremote: Counting objects:  75% (64/85)\rremote: Counting objects:  76% (65/85)\rremote: Counting objects:  77% (66/85)\rremote: Counting objects:  78% (67/85)\rremote: Counting objects:  80% (68/85)\rremote: Counting objects:  81% (69/85)\rremote: Counting objects:  82% (70/85)\rremote: Counting objects:  83% (71/85)\rremote: Counting objects:  84% (72/85)\rremote: Counting objects:  85% (73/85)\rremote: Counting objects:  87% (74/85)\rremote: Counting objects:  88% (75/85)\rremote: Counting objects:  89% (76/85)\rremote: Counting objects:  90% (77/85)\rremote: Counting objects:  91% (78/85)\rremote: Counting objects:  92% (79/85)\rremote: Counting objects:  94% (80/85)\rremote: Counting objects:  95% (81/85)\rremote: Counting objects:  96% (82/85)\rremote: Counting objects:  97% (83/85)\rremote: Counting objects:  98% (84/85)\rremote: Counting objects: 100% (85/85)\rremote: Counting objects: 100% (85/85), done.\r\nremote: Compressing objects:   3% (1/32)\rremote: Compressing objects:   6% (2/32)\rremote: Compressing objects:   9% (3/32)\rremote: Compressing objects:  12% (4/32)\r",,terminal_output
+60,95049,"TERMINAL",0,0,"remote: Compressing objects:  15% (5/32)\rremote: Compressing objects:  18% (6/32)\rremote: Compressing objects:  21% (7/32)\rremote: Compressing objects:  25% (8/32)\rremote: Compressing objects:  28% (9/32)\rremote: Compressing objects:  31% (10/32)\rremote: Compressing objects:  34% (11/32)\rremote: Compressing objects:  37% (12/32)\rremote: Compressing objects:  40% (13/32)\rremote: Compressing objects:  43% (14/32)\rremote: Compressing objects:  46% (15/32)\rremote: Compressing objects:  50% (16/32)\rremote: Compressing objects:  53% (17/32)\rremote: Compressing objects:  56% (18/32)\rremote: Compressing objects:  59% (19/32)\rremote: Compressing objects:  62% (20/32)\rremote: Compressing objects:  65% (21/32)\rremote: Compressing objects:  68% (22/32)\rremote: Compressing objects:  71% (23/32)\rremote: Compressing objects:  75% (24/32)\rremote: Compressing objects:  78% (25/32)\rremote: Compressing objects:  81% (26/32)\rremote: Compressing objects:  84% (27/32)\rremote: Compressing objects:  87% (28/32)\rremote: Compressing objects:  90% (29/32)\rremote: Compressing objects:  93% (30/32)\rremote: Compressing objects:  96% (31/32)\rremote: Compressing objects: 100% (32/32)\rremote: Compressing objects: 100% (32/32), done.\r\nremote: Total 64 (delta 38), reused 56 (delta 31), pack-reused 0 (from 0)\r\nUnpacking objects:   1% (1/64)\rUnpacking objects:   3% (2/64)\rUnpacking objects:   4% (3/64)\rUnpacking objects:   6% (4/64)\rUnpacking objects:   7% (5/64)\rUnpacking objects:   9% (6/64)\rUnpacking objects:  10% (7/64)\rUnpacking objects:  12% (8/64)\rUnpacking objects:  14% (9/64)\rUnpacking objects:  15% (10/64)\rUnpacking objects:  17% (11/64)\rUnpacking objects:  18% (12/64)\rUnpacking objects:  20% (13/64)\rUnpacking objects:  21% (14/64)\rUnpacking objects:  23% (15/64)\rUnpacking objects:  25% (16/64)\rUnpacking objects:  26% (17/64)\rUnpacking objects:  28% (18/64)\rUnpacking objects:  29% (19/64)\rUnpacking objects:  31% (20/64)\rUnpacking objects:  32% (21/64)\rUnpacking objects:  34% (22/64)\rUnpacking objects:  35% (23/64)\rUnpacking objects:  37% (24/64)\rUnpacking objects:  39% (25/64)\rUnpacking objects:  40% (26/64)\rUnpacking objects:  42% (27/64)\rUnpacking objects:  43% (28/64)\rUnpacking objects:  45% (29/64)\rUnpacking objects:  46% (30/64)\rUnpacking objects:  48% (31/64)\rUnpacking objects:  50% (32/64)\rUnpacking objects:  51% (33/64)\rUnpacking objects:  53% (34/64)\rUnpacking objects:  54% (35/64)\rUnpacking objects:  56% (36/64)\rUnpacking objects:  57% (37/64)\rUnpacking objects:  59% (38/64)\rUnpacking objects:  60% (39/64)\rUnpacking objects:  62% (40/64)\rUnpacking objects:  64% (41/64)\rUnpacking objects:  65% (42/64)\rUnpacking objects:  67% (43/64)\rUnpacking objects:  68% (44/64)\rUnpacking objects:  70% (45/64)\rUnpacking objects:  71% (46/64)\rUnpacking objects:  73% (47/64)\rUnpacking objects:  75% (48/64)\rUnpacking objects:  76% (49/64)\rUnpacking objects:  78% (50/64)\rUnpacking objects:  79% (51/64)\rUnpacking objects:  81% (52/64)\rUnpacking objects:  82% (53/64)\rUnpacking objects:  84% (54/64)\rUnpacking objects:  85% (55/64)\rUnpacking objects:  87% (56/64)\rUnpacking objects:  89% (57/64)\rUnpacking objects:  90% (58/64)\rUnpacking objects:  92% (59/64)\rUnpacking objects:  93% (60/64)\rUnpacking objects:  95% (61/64)\rUnpacking objects:  96% (62/64)\rUnpacking objects:  98% (63/64)\rUnpacking objects: 100% (64/64)\rUnpacking objects: 100% (64/64), 8.30 KiB | 35.00 KiB/s, done.\r\n",,terminal_output
+61,95170,"TERMINAL",0,0,"From github.com:p-doom/slurm\r\n   06dec29..705b23d  main                 -> origin/main\r\n * [new branch]      create-utils-subdirs -> origin/create-utils-subdirs\r\nUpdating 06dec29..705b23d\r\n",,terminal_output
+62,95438,"TERMINAL",0,0,"Fast-forward\r\n",,terminal_output
+63,95470,"TERMINAL",0,0," dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sbatch                                          |  67 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sh                                              |  53 +++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_sample/train_dynamics_overfit_sample.sbatch                                                           |   2 +-\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0.6_mio.sbatch                                             |  45 +++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0_5.sbatch                                                 |  39 ---------------------------------------\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0_5.sh                                                     |  43 +++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_21_mio.sbatch                                              |  45 +++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_2_mio.sbatch                                               |  45 +++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_9_mio.sbatch                                               |  45 +++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_small_mio.sbatch                                           |  45 +++++++++++++++++++++++++++++++++++++++++++++\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/logs/logs_training/train_tokenizer_batch_size_scaling_2_node_3292213.log | 136 ----------------------------------------------------------------------------------------------------------------------------------------\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_16_nodes.sbatch                                          |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_1_nodes.sbatch                                           |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_2_nodes.sbatch                                           |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_32_nodes.sbatch                                          |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_4_nodes.sbatch                                           |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_8_nodes.sbatch                                           |   2 +-\r\n jobs/mihir/horeka/overfit_sample_tiny/tester.sh                                                                          |  38 ++++++++++++++++++++++++++++++++++++++\r\n jobs/mihir/horeka/overfit_sample_tiny/train_dynamics_overfit_sample.sbatch                                               |  58 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n jobs/mihir/horeka/overfit_sample_tiny/train_lam_overfit_sample.sbatch                                                    |  54 ++++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n jobs/mihir/horeka/overfit_sample_tiny/train_tokenizer_overfit_sample.sbatch                                              |  53 +++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n utils/alfred/alfred_placeholder.txt                                                                                      |   1 +\r\n utils/franz/franz_placeholder.txt                                                                                        |   1 +\r\n 23 files changed, 600 insertions(+), 182 deletions(-)\r\n create mode 100644 dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sbatch\r\n create mode 100755 dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sh\r\n create mode 100644 dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0.6_mio.sbatch\r\n delete mode 100644 dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0_5.sbatch\r\n create mode 100755 dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0_5.sh\r\n create mode 100644 dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_21_mio.sbatch\r\n create mode 100644 dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_2_mio.sbatch\r\n create mode 100644 dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_9_mio.sbatch\r\n create mode 100644 dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_small_mio.sbatch\r\n delete mode 100644 jobs/mihir/horeka/batchsize_scaling/adjusted_lr/logs/logs_training/train_tokenizer_batch_size_scaling_2_node_3292213.log\r\n create mode 100644 jobs/mihir/horeka/overfit_sample_tiny/tester.sh\r\n create mode 100644 jobs/mihir/horeka/overfit_sample_tiny/train_dynamics_overfit_sample.sbatch\r\n create mode 100644 jobs/mihir/horeka/overfit_sample_tiny/train_lam_overfit_sample.sbatch\r\n create mode 100644 jobs/mihir/horeka/overfit_sample_tiny/train_tokenizer_overfit_sample.sbatch\r\n create mode 100644 utils/alfred/alfred_placeholder.txt\r\n create mode 100644 utils/franz/franz_placeholder.txt\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_batch_size_scaling/slurm]633;D;0",,terminal_output
+64,294322,"TERMINAL",0,0,"cd ..",,terminal_command

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-147cbc16-b49b-4a51-93fe-1d098f282eba1750965675085-2025_06_26-21.21.29.79/source.csv

@@ -0,0 +1,5 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,4,"/home/hk-project-pai00039/tum_ind3695/projects/jafar_dyn/sbatch_scripts/coinrun/latent_action_ablation/train_tokenizer_coinrun.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=4\n#SBATCH --time=24:00:00\n#SBATCH --partition=accelerated\n#SBATCH --account=hk-project-p0023960\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:4\n#SBATCH --output=logs/logs_training_tokenizer/%x_%j.log\n#SBATCH --error=logs/logs_training_tokenizer/%x_%j.log\n#SBATCH --mail-user=avocadoaling@gmail.com\n#SBATCH --job-name=train_tokenizer_coinrun\n#SBATCH --mail-type=ALL\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv_jafar/bin/activate\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nws_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared'\n\nCHECKPOINT_DIR=$ws_dir/data/checkpoints/$job_name_$slurm_job_id\nLOG_DIR=$ws_dir/logs/$job_name_$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\nmkdir -p $LOG_DIR\n\ndata_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/coinrun/coinrun_tfrecords'\n\nsrun python train_tokenizer.py \\n    --batch_size=192 \\n    --experiment_name $job_name \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --log_checkpoint_interval=1000 \\n    --log_image_interval=10 \\n    --image_height 64 \\n    --image_width 64 \\n    --log \\n    --entity instant-uv \\n    --data_dir $data_dir \\n    --project jafar\n",shellscript,tab
+2,1355,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"9:21:29 PM [info] Activating crowd-code\n9:21:29 PM [info] Welcome back tum_ind3695. Your user-id is '507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20'. Happy coding!\n9:21:29 PM [info] Recording started\n",Log,tab
+3,3666,"TERMINAL",0,0,"/bin/python3 /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/printEnvVariablesToFile.py /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash/envVars.txt",,terminal_command
+4,3749,"TERMINAL",0,0,"]633;E;2025-06-26 21:21:32 /bin/python3 /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/printEnvVariablesToFile.py /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash/envVars.txt;75b7428c-8463-49f3-815b-b639f4309d7b]633;C]0;tum_ind3695@hkn1993:/hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash]633;D;0",,terminal_output

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-26b17221-07df-460b-8719-6668ecea44721750772995721-2025_06_24-16.00.55.611/source.csv

@@ -0,0 +1,8 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,12,"genie.py",0,0,"from typing import Dict, Any\n\nimport optax\nimport jax\nimport jax.numpy as jnp\nimport flax.linen as nn\nfrom jax import NamedSharding\nfrom flax.training.train_state import TrainState\nfrom flax.training import orbax_utils\nfrom orbax.checkpoint import PyTreeCheckpointer\n\nfrom models.dynamics import DynamicsMaskGIT\nfrom models.lam import LatentActionModel\nfrom models.tokenizer import TokenizerVQVAE\n\n\nclass Genie(nn.Module):\n    """"""Genie model""""""\n\n    # --- Tokenizer ---\n    in_dim: int\n    tokenizer_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 ---\n    lam_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    # --- Dynamics ---\n    dyna_dim: int\n    dyna_num_blocks: int\n    dyna_num_heads: int\n    dropout: float = 0.0\n    mask_limit: float = 0.0\n\n    def setup(self):\n        self.tokenizer = TokenizerVQVAE(\n            in_dim=self.in_dim,\n            model_dim=self.tokenizer_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        )\n        self.lam = LatentActionModel(\n            in_dim=self.in_dim,\n            model_dim=self.lam_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        )\n        self.dynamics = DynamicsMaskGIT(\n            model_dim=self.dyna_dim,\n            num_latents=self.num_patch_latents,\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        )\n\n    def __call__(self, batch: Dict[str, Any], training: bool = True) -> Dict[str, Any]:\n        tokenizer_outputs = self.tokenizer.vq_encode(batch[""videos""], training=False)\n        lam_outputs = self.lam.vq_encode(batch[""videos""], training=False)\n        outputs = dict(\n            video_tokens=jax.lax.stop_gradient(tokenizer_outputs[""indices""]),\n            latent_actions=jax.lax.stop_gradient(lam_outputs[""z_q""]),\n        )\n        outputs[""mask_rng""] = batch[""mask_rng""]\n        dyna_outputs = self.dynamics(outputs, training)\n        outputs.update(dyna_outputs)\n        mle_indices = jnp.argmax(outputs[""token_logits""], axis=-1)\n        outputs[""recon""] = self.tokenizer.decode(\n            mle_indices, batch[""videos""].shape[2:4]\n        )\n        return outputs\n\n    @nn.compact\n    def sample(\n        self,\n        batch: Dict[str, Any],\n        steps: int = 25,\n        temperature: int = 1,\n        sample_argmax: bool = False,\n    ) -> Any:\n        # --- Encode videos and actions ---\n        tokenizer_out = self.tokenizer.vq_encode(batch[""videos""], training=False)\n        token_idxs = tokenizer_out[""indices""]\n        new_frame_idxs = jnp.zeros_like(token_idxs)[:, 0]\n        action_tokens = self.lam.vq.get_codes(batch[""latent_actions""])\n\n        # --- Initialize MaskGIT ---\n        init_mask = jnp.ones_like(token_idxs, dtype=bool)[:, 0]\n        init_carry = (\n            batch[""rng""],\n            new_frame_idxs,\n            init_mask,\n            token_idxs,\n            action_tokens,\n        )\n        MaskGITLoop = nn.scan(\n            MaskGITStep,\n            variable_broadcast=""params"",\n            split_rngs={""params"": False},\n            in_axes=0,\n            out_axes=0,\n            length=steps,\n        )\n\n        # --- Run MaskGIT loop ---\n        loop_fn = MaskGITLoop(\n            dynamics=self.dynamics,\n            tokenizer=self.tokenizer,\n            temperature=temperature,\n            sample_argmax=sample_argmax,\n            steps=steps,\n        )\n        final_carry, _ = loop_fn(init_carry, jnp.arange(steps))\n        new_frame_idxs = final_carry[1]\n        new_frame_pixels = self.tokenizer.decode(\n            jnp.expand_dims(new_frame_idxs, 1),\n            video_hw=batch[""videos""].shape[2:4],\n        )\n        return new_frame_pixels\n\n    def vq_encode(self, batch, training) -> Dict[str, Any]:\n        # --- Preprocess videos ---\n        lam_output = self.lam.vq_encode(batch[""videos""], training=training)\n        return lam_output[""indices""]\n\n\nclass MaskGITStep(nn.Module):\n    dynamics: nn.Module\n    tokenizer: nn.Module\n    temperature: float\n    sample_argmax: bool\n    steps: int\n\n    @nn.compact\n    def __call__(self, carry, x):\n        rng, final_token_idxs, mask, token_idxs, action_tokens = carry\n        step = x\n        B, T, N = token_idxs.shape[:3]\n\n        # --- Construct + encode video ---\n        vid_token_idxs = jnp.concatenate(\n            (token_idxs, jnp.expand_dims(final_token_idxs, 1)), axis=1\n        )\n        vid_embed = self.dynamics.patch_embed(vid_token_idxs)\n        curr_masked_frame = jnp.where(\n            jnp.expand_dims(mask, -1),\n            self.dynamics.mask_token[0],\n            vid_embed[:, -1],\n        )\n        vid_embed = vid_embed.at[:, -1].set(curr_masked_frame)\n\n        # --- Predict transition ---\n        act_embed = self.dynamics.action_up(action_tokens)\n        vid_embed += jnp.pad(act_embed, ((0, 0), (1, 0), (0, 0), (0, 0)))\n        unmasked_ratio = jnp.cos(jnp.pi * (step + 1) / (self.steps * 2))\n        step_temp = self.temperature * (1.0 - unmasked_ratio)\n        final_logits = self.dynamics.dynamics(vid_embed)[:, -1] / step_temp\n\n        # --- Sample new tokens for final frame ---\n        if self.sample_argmax:\n            sampled_token_idxs = jnp.argmax(final_logits, axis=-1)\n        else:\n            rng, _rng = jax.random.split(rng)\n            sampled_token_idxs = jnp.where(\n                step == self.steps - 1,\n                jnp.argmax(final_logits, axis=-1),\n                jax.random.categorical(_rng, final_logits),\n            )\n        gather_fn = jax.vmap(jax.vmap(lambda x, y: x[y]))\n        final_token_probs = gather_fn(jax.nn.softmax(final_logits), sampled_token_idxs)\n        final_token_probs += ~mask\n        # Update masked tokens only\n        new_token_idxs = jnp.where(mask, sampled_token_idxs, final_token_idxs)\n\n        # --- Update mask ---\n        num_unmasked_tokens = jnp.round(N * (1.0 - unmasked_ratio)).astype(int)\n        idx_mask = jnp.arange(final_token_probs.shape[-1]) > num_unmasked_tokens\n        sorted_idxs = jnp.argsort(final_token_probs, axis=-1, descending=True)\n        mask_update_fn = jax.vmap(lambda msk, ids: msk.at[ids].set(idx_mask))\n        new_mask = mask_update_fn(mask, sorted_idxs)\n\n        new_carry = (rng, new_token_idxs, new_mask, token_idxs, action_tokens)\n        return new_carry, None\n\n\ndef restore_genie_components(\n    train_state: TrainState,\n    sharding: NamedSharding,\n    inputs: Dict[str, jax.Array],\n    rng: jax.Array,\n    args,\n):\n    """"""Restore pre-trained Genie components""""""\n    rng, _rng = jax.random.split(rng)\n\n    dummy_tokenizer = TokenizerVQVAE(\n        in_dim=args.image_channels,\n        model_dim=args.tokenizer_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    )\n    dummy_lam = LatentActionModel(\n        in_dim=args.image_channels,\n        model_dim=args.lam_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    )\n    tokenizer_init_params = dummy_tokenizer.init(_rng, inputs)\n    lam_init_params = dummy_lam.init(_rng, inputs)\n\n    # dummy values since we only use tx to initialize the dummy train states\n    dummy_tx = optax.adamw(\n        learning_rate=optax.constant_schedule(args.max_lr),\n        b1=0.9,\n        b2=0.9,\n        weight_decay=1e-4,\n    )\n\n    dummy_tokenizer_train_state = TrainState.create(\n        apply_fn=dummy_tokenizer.apply, params=tokenizer_init_params, tx=dummy_tx\n    )\n    dummy_lam_train_state = TrainState.create(\n        apply_fn=dummy_lam.apply, params=lam_init_params, tx=dummy_tx\n    )\n\n    def create_abstract_sharded_pytree(pytree_template, sharding_spec):\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\n    abstract_sharded_tokenizer_state = create_abstract_sharded_pytree(\n        dummy_tokenizer_train_state, sharding\n    )\n    abstract_sharded_lam_state = create_abstract_sharded_pytree(\n        dummy_lam_train_state, sharding\n    )\n\n    tokenizer_restore_target = {""model"": abstract_sharded_tokenizer_state}\n    lam_restore_target = {""model"": abstract_sharded_lam_state}\n\n    tokenizer_restore_args = orbax_utils.restore_args_from_target(\n        tokenizer_restore_target\n    )\n    lam_restore_args = orbax_utils.restore_args_from_target(lam_restore_target)\n\n    restored_tokenizer_params = (\n        PyTreeCheckpointer()\n        .restore(\n            args.tokenizer_checkpoint,\n            item=tokenizer_restore_target,\n            restore_args=tokenizer_restore_args,\n        )[""model""]\n        .params[""params""]\n    )\n    restored_lam_params = (\n        PyTreeCheckpointer()\n        .restore(\n            args.lam_checkpoint, item=lam_restore_target, restore_args=lam_restore_args\n        )[""model""]\n        .params[""params""]\n    )\n    # Genie does not initialize all LAM modules, thus we omit those extra modules during restoration\n    # (f.srambical) FIXME: Currently, this is a small HBM memory crunch since the LAM's decoder is loaded into HBM and immediately dicarded.\n    # A workaround would be to restore to host memory first, and only move the weights to HBM after pruning the decoder\n    restored_lam_params = {\n        k: v\n        for k, v in restored_lam_params.items()\n        if k in train_state.params[""params""][""lam""]\n    }\n\n    train_state.params[""params""][""tokenizer""].update(restored_tokenizer_params)\n    train_state.params[""params""][""lam""].update(restored_lam_params)\n\n    return train_state\n",python,tab
+2,2295,"genie.py",218,0,"",python,selection_mouse
+3,2306,"genie.py",217,0,"",python,selection_command
+4,15131,"TERMINAL",0,0,"cursor jafar_run_overfit/",,terminal_command
+5,15182,"TERMINAL",0,0,"]633;E;2025-06-24 16:01:10 cursor jafar_run_overfit/;3785e992-afee-48a3-a5a8-2f86d34a9f49]633;C",,terminal_output
+6,15322,"TERMINAL",0,0,"]0;tum_ind3695@hkn1993:~/projects]633;D;0]633;P;Cwd=/home/hk-project-pai00039/tum_ind3695/projects",,terminal_output
+7,17837,"genie.py",0,0,"",python,tab

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-37cbf512-b8df-48c0-837d-65f771724c0f1750986496490-2025_06_27-03.08.56.360/source.csv

@@ -0,0 +1,199 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,6,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=4\n#SBATCH --time=24:00:00\n#SBATCH --partition=accelerated\n#SBATCH --account=hk-project-p0023960\n#SBATCH --cpus-per-task=8\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_alfred/logs_training_dyn/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_alfred/logs_training_dyn/%x_%j.log\n#SBATCH --mail-user=avocadoaling@gmail.com\n#SBATCH --job-name=train_dyn_coinrun\n#SBATCH --mail-type=ALL\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv_jafar/bin/activate\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nws_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared'\n\nCHECKPOINT_DIR=$ws_dir/data/checkpoints/$job_name_$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ndata_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/coinrun/coinrun_tfrecords'\n\nlam_checkpoint='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/lam_ckpt_dir/lam_1748080132_50000'\ntokenizer_checkpoint='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/tokenizer_ckpt_dir/tokenizer_1748080132_10000'\n\n\nsrun python train_dynamics.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --lam_checkpoint $lam_checkpoint \\n    --image_height 64 \\n    --image_width 64 \\n    --tokenizer_checkpoint $tokenizer_checkpoint \\n    --batch_size=192 \\n    --log_checkpoint_interval=500 \\n    --log_image_interval=50 \\n    --seed=0 \\n    --name=$job_name \\n    --min_lr=0.0000866 \\n    --max_lr=0.0000866 \\n    --log \\n    --entity instant-uv \\n    --data_dir $data_dir \\n    --project jafar\n\n",shellscript,tab
+2,2430,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"3:08:56 AM [info] Activating crowd-code\n3:08:56 AM [info] Welcome back tum_ind3695. Your user-id is '507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20'. Happy coding!\n3:08:56 AM [info] Recording started\n",Log,tab
+3,98818,"extension-output-pdoom-org.crowd-code-#1-crowd-code",218,0,"",Log,selection_mouse
+4,99370,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,selection_command
+5,100079,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",0,0,"",shellscript,tab
+6,107485,"TERMINAL",0,0,"./slurm/dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0_5.sh ^C",,terminal_command
+7,107507,"TERMINAL",0,0,"^C[?2004l\r[?2004h[?2004l\r\r\n]633;E;;8ce264ca-66f0-4798-9e7f-85f974fbab1a]633;C",,terminal_output
+8,108125,"TERMINAL",0,0,"^C",,terminal_command
+9,108137,"TERMINAL",0,0,"^C[?2004l\r[?2004h[?2004l\r\r\n]633;E;;8ce264ca-66f0-4798-9e7f-85f974fbab1a]633;C]0;tum_ind3695@hkn1993:~/projects/jafar_dyn]633;D",,terminal_output
+10,108378,"TERMINAL",0,0,"^C",,terminal_command
+11,108386,"TERMINAL",0,0,"^C[?2004l\r[?2004h[?2004l\r\r\n]633;E;;8ce264ca-66f0-4798-9e7f-85f974fbab1a]633;C]0;tum_ind3695@hkn1993:~/projects/jafar_dyn]633;D",,terminal_output
+12,134258,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",500,0,"",shellscript,selection_command
+13,134904,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",677,0,"",shellscript,selection_command
+14,135987,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",702,0,"",shellscript,selection_command
+15,136228,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",725,0,"",shellscript,selection_command
+16,136262,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",730,0,"",shellscript,selection_command
+17,136285,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",791,0,"",shellscript,selection_command
+18,136318,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",796,0,"",shellscript,selection_command
+19,136359,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",860,0,"",shellscript,selection_command
+20,136387,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",881,0,"",shellscript,selection_command
+21,136419,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",886,0,"",shellscript,selection_command
+22,136453,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",980,0,"",shellscript,selection_command
+23,136487,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",985,0,"",shellscript,selection_command
+24,136522,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1104,0,"",shellscript,selection_command
+25,136555,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1237,0,"",shellscript,selection_command
+26,136588,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1238,0,"",shellscript,selection_command
+27,136625,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1243,0,"",shellscript,selection_command
+28,136655,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1275,0,"",shellscript,selection_command
+29,136698,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1308,0,"",shellscript,selection_command
+30,136725,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1347,0,"",shellscript,selection_command
+31,136756,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1371,0,"",shellscript,selection_command
+32,136790,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1394,0,"",shellscript,selection_command
+33,136824,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1445,0,"",shellscript,selection_command
+34,136858,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1468,0,"",shellscript,selection_command
+35,136915,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1445,0,"",shellscript,selection_command
+36,137176,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1394,0,"",shellscript,selection_command
+37,137199,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1371,0,"",shellscript,selection_command
+38,137230,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1347,0,"",shellscript,selection_command
+39,137265,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1308,0,"",shellscript,selection_command
+40,137298,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1275,0,"",shellscript,selection_command
+41,137335,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1243,0,"",shellscript,selection_command
+42,137366,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1238,0,"",shellscript,selection_command
+43,137398,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1237,0,"",shellscript,selection_command
+44,137434,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1104,0,"",shellscript,selection_command
+45,137465,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",985,0,"",shellscript,selection_command
+46,137501,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",980,0,"",shellscript,selection_command
+47,137534,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",886,0,"",shellscript,selection_command
+48,138389,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",881,0,"",shellscript,selection_command
+49,138635,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",860,0,"",shellscript,selection_command
+50,138657,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",796,0,"",shellscript,selection_command
+51,138686,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",791,0,"",shellscript,selection_command
+52,138718,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",730,0,"",shellscript,selection_command
+53,138926,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",725,0,"",shellscript,selection_command
+54,139286,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",724,0,"\n",shellscript,content
+55,141358,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",727,0,"",shellscript,selection_command
+56,141580,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",792,0,"",shellscript,selection_command
+57,142263,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",727,0,"",shellscript,selection_command
+58,143135,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",792,0,"",shellscript,selection_command
+59,143303,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",793,0,"",shellscript,selection_command
+60,146200,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",792,0,"",shellscript,selection_command
+61,146862,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",792,1,"",shellscript,content
+62,147494,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",727,0,"",shellscript,selection_command
+63,147672,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",726,0,"",shellscript,selection_command
+64,147816,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",725,0,"",shellscript,selection_command
+65,148099,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",725,0,"\n",shellscript,content
+66,148855,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",725,0,"",shellscript,selection_command
+67,149414,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",725,0,"t",shellscript,content
+68,149517,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",726,0,"a",shellscript,content
+69,149518,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",727,0,"",shellscript,selection_keyboard
+70,149635,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",727,0,"g",shellscript,content
+71,149636,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",728,0,"",shellscript,selection_keyboard
+72,149806,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",728,0,"s",shellscript,content
+73,149807,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",729,0,"",shellscript,selection_keyboard
+74,151522,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",729,0,"='latent_action_ablation'",shellscript,content
+75,151522,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",728,1,"",shellscript,content
+76,152489,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",752,0,"",shellscript,selection_command
+77,153552,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",754,0,"",shellscript,selection_command
+78,154193,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",725,0,"",shellscript,selection_command
+79,154604,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",726,0,"",shellscript,selection_command
+80,154775,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",727,0,"",shellscript,selection_command
+81,155094,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",728,0,"",shellscript,selection_command
+82,155370,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",728,0,"s",shellscript,content
+83,155371,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",729,0,"",shellscript,selection_keyboard
+84,155964,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",728,0,"",shellscript,selection_command
+85,156119,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",729,0,"",shellscript,selection_command
+86,156284,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",730,0,"",shellscript,selection_command
+87,156822,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",730,0,"()",shellscript,content
+88,156823,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",731,0,"",shellscript,selection_keyboard
+89,157970,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",756,0,")",shellscript,content
+90,157970,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",731,1,"",shellscript,content
+91,158278,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",730,0,"",shellscript,selection_command
+92,160762,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",732,0,"",shellscript,selection_command
+93,162759,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",732,22,"",shellscript,content
+94,163517,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",732,0,"t",shellscript,content
+95,163518,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",733,0,"",shellscript,selection_keyboard
+96,163577,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",733,0,"r",shellscript,content
+97,163578,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",734,0,"",shellscript,selection_keyboard
+98,163640,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",734,0,"a",shellscript,content
+99,163641,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",735,0,"",shellscript,selection_keyboard
+100,163725,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",735,0,"i",shellscript,content
+101,163727,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",736,0,"",shellscript,selection_keyboard
+102,164285,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",735,1,"",shellscript,content
+103,164436,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",734,1,"",shellscript,content
+104,164578,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",733,1,"",shellscript,content
+105,164726,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",732,1,"",shellscript,content
+106,164878,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",732,0,"c",shellscript,content
+107,164882,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",733,0,"",shellscript,selection_keyboard
+108,164975,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",733,0,"o",shellscript,content
+109,164976,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",734,0,"",shellscript,selection_keyboard
+110,165098,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",734,0,"i",shellscript,content
+111,165099,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",735,0,"",shellscript,selection_keyboard
+112,165175,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",735,0,"n",shellscript,content
+113,165176,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",736,0,"",shellscript,selection_keyboard
+114,166003,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",736,0,"r",shellscript,content
+115,166004,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",737,0,"",shellscript,selection_keyboard
+116,166141,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",737,0,"u",shellscript,content
+117,166142,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",738,0,"",shellscript,selection_keyboard
+118,166232,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",738,0,"n",shellscript,content
+119,166233,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",739,0,"",shellscript,selection_keyboard
+120,166831,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",738,0,"",shellscript,selection_command
+121,166930,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",739,0,"",shellscript,selection_command
+122,167148,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",740,0,"",shellscript,selection_command
+123,168183,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",740,0," ",shellscript,content
+124,168184,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",741,0,"",shellscript,selection_keyboard
+125,169259,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",741,0,"'dynamics' 'latent_action_ablation'",shellscript,content
+126,169701,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",775,0,"",shellscript,selection_command
+127,171187,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",753,0,"",shellscript,selection_command
+128,171358,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",752,0,"",shellscript,selection_command
+129,172047,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",753,0,"",shellscript,selection_command
+130,172383,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",753,22,"",shellscript,content
+131,173612,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",753,0,"p",shellscript,content
+132,173613,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",754,0,"",shellscript,selection_keyboard
+133,174632,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",754,0,"i",shellscript,content
+134,174634,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",755,0,"",shellscript,selection_keyboard
+135,175104,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",755,0,"e",shellscript,content
+136,175105,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",756,0,"",shellscript,selection_keyboard
+137,175682,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",755,1,"",shellscript,content
+138,175887,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",755,0,"p",shellscript,content
+139,175888,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",756,0,"",shellscript,selection_keyboard
+140,176356,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",756,0,"e",shellscript,content
+141,176357,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",757,0,"",shellscript,selection_keyboard
+142,176612,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",757,0,"l",shellscript,content
+143,176613,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",758,0,"",shellscript,selection_keyboard
+144,176733,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",758,0,"i",shellscript,content
+145,176734,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",759,0,"",shellscript,selection_keyboard
+146,176771,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",759,0,"n",shellscript,content
+147,176773,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",760,0,"",shellscript,selection_keyboard
+148,176873,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",760,0,"e",shellscript,content
+149,176875,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",761,0,"",shellscript,selection_keyboard
+150,177010,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",761,0," ",shellscript,content
+151,177011,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",762,0,"",shellscript,selection_keyboard
+152,177508,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",761,1,"",shellscript,content
+153,177889,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",761,0,"_",shellscript,content
+154,177890,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",762,0,"",shellscript,selection_keyboard
+155,178118,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",762,0,"t",shellscript,content
+156,178119,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",763,0,"",shellscript,selection_keyboard
+157,178182,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",763,0,"e",shellscript,content
+158,178184,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",764,0,"",shellscript,selection_keyboard
+159,178313,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",764,0,"s",shellscript,content
+160,178314,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",765,0,"",shellscript,selection_keyboard
+161,178406,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",765,0,"t",shellscript,content
+162,178407,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",766,0,"",shellscript,selection_keyboard
+163,179684,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",767,0," 'latent_action_ablation'",shellscript,content
+164,179685,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",792,0,"",shellscript,selection_command
+165,179897,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",791,0,"",shellscript,selection_command
+166,180187,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",794,0,"",shellscript,selection_command
+167,180441,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",795,0,"",shellscript,selection_command
+168,180456,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",859,0,"",shellscript,selection_command
+169,180492,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",923,0,"",shellscript,selection_command
+170,180523,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",948,0,"",shellscript,selection_command
+171,180556,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",950,0,"",shellscript,selection_command
+172,180593,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1017,0,"",shellscript,selection_command
+173,180624,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1049,0,"",shellscript,selection_command
+174,180659,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1116,0,"",shellscript,selection_command
+175,180757,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1235,0,"",shellscript,selection_command
+176,180762,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1306,0,"",shellscript,selection_command
+177,180764,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1307,0,"",shellscript,selection_command
+178,180838,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1338,0,"",shellscript,selection_command
+179,180848,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1371,0,"",shellscript,selection_command
+180,180861,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1410,0,"",shellscript,selection_command
+181,180894,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1434,0,"",shellscript,selection_command
+182,180928,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1457,0,"",shellscript,selection_command
+183,180962,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1508,0,"",shellscript,selection_command
+184,180995,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1531,0,"",shellscript,selection_command
+185,181030,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1567,0,"",shellscript,selection_command
+186,181064,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1597,0,"",shellscript,selection_command
+187,181096,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1612,0,"",shellscript,selection_command
+188,181131,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1635,0,"",shellscript,selection_command
+189,181165,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1660,0,"",shellscript,selection_command
+190,181198,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1685,0,"",shellscript,selection_command
+191,181232,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1697,0,"",shellscript,selection_command
+192,181267,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",1723,0,"",shellscript,selection_command
+193,184178,"TERMINAL",0,0,"sbatch slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",,terminal_command
+194,184222,"TERMINAL",0,0,"]633;E;2025-06-27 03:12:00 sbatch slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch;8ce264ca-66f0-4798-9e7f-85f974fbab1a]633;CSubmitted batch job 3297726\r\n]0;tum_ind3695@hkn1993:~/projects/jafar_dyn]633;D;0",,terminal_output
+195,352753,"train_dynamics.py",0,0,"from dataclasses import dataclass\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    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    name: str = """"\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(entity=args.entity, project=args.project, group=""debug"", name=args.name, config=args)\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
+196,357793,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",0,0,"",shellscript,tab
+197,408583,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",924,0,"",shellscript,selection_mouse
+198,408586,"slurm/dev/alfred/coinrun/latent_action_ablation/train_dynamics_coinrun.sbatch",923,0,"",shellscript,selection_command

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-9ff54a43-2a59-41a8-96bc-f7e46d5244651750887279734-2025_06_25-23.36.32.560/source.csv

@@ -0,0 +1,3 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,18,"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 <directory_path>"")\n    else:\n        find_duplicate_filenames(sys.argv[1])\n\n",python,tab
+2,521,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"11:36:32 PM [info] Activating crowd-code\n11:36:32 PM [info] Welcome back tum_ind3695. Your user-id is '507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20'. Happy coding!\n11:36:32 PM [info] Recording started\n",Log,tab

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-aa8be5f9-c447-4faf-b9c6-7142909b3c591750719092446-2025_06_24-00.51.37.15/source.csv

@@ -0,0 +1,6 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,12,"todos.md",0,0,"# Todo's\n\n## Misc:\n- [ ] Steuererklärung \n- [ ] jan wlan geld internet und paket\n- [ ] thanh geld 22 club\n- [ ] Arzt termine\n  - [ ] normal\n  - [ ] zahn\n- [ ] Proteinshake\n- [ ] Protein shaker\n- [ ] Zahnbürste\n- [ ] laptop?? ask stefan what to do with it\n\n\n## 23.06 Monday\n\n- [x] PR step duration\n- [x] generate samples from dyn\n- [x] single batch training overfit\n  - [x] lam\n  - [x] tokenizer\n  - [ ] dynamics model\n\n- [ ] lr finding \n- [ ] pr: update sampling method for resolution \n\n- [ ] retrain lam until convergence?\n- [ ] see if data parallel is better than single gpu?\n\n- [ ] issue: make sampling faster\n- [ ] blog post crowd source\n\n- [ ] Preprocess entire minecraft dataset\n- [ ] Dont let tf see gpus\n- [ ] Look into dataloader\n\nQuestions:\n- optimal lr\n- optimal batch size\n- how to scale lr with batch size\n- how to test/evaluate lam performance?\n  - how good are the actions for the lam\n  - how many actions do we have?\n  - how many actions do we need\n\n\n## 22.06 Sunday\nNotes:\n- How should tokenizer training behave?\n- How should lam training behave?\n- How should dynamics model training behave?\n- \n\nTODOS:\n- [x] look at prs jafar\n- [x] extension version update\n- [x] make big run directory\n- [x] look at the run of lam\n- [ ] train the dynamics model on the new models\n\n- [ ] start a coin run baseline\n  - [ ] tokenizer\n  - [ ] lam \n  - [ ] dynamics model\n\n- [ ] move tfrecord to huggingface\n\n- [x] tierlist for thanh\n- [ ] move tfrecord to helmholtz\n- [ ] helmholtz setup\n- [ ] fix dataloader seeding\n\n- [ ] FFM homework\n- [ ] Tennis\n- [ ] gym\n\n\n\n## 20.06 Friday\n- [x] wäsche\n- [x] flaschen wegbringen\n- [x] zhs \n- [x] cursor setup\n- [x] run overfiting runs on lam\n\n## 19.06 Thursday\n- [x] extension ace palenight\n\n\n- [x] run overfiting runs on dynamics model (got aborted bc of branch switching)\n  - [x] test multi gpu overfit on full tfrecord\n\n\n## 18.06 Wednesday\nJobs:\n- [x] run overfiting runs for tokenizer \n  - [x] test multi gpu overfit on single batch\n  - [x] test multi gpu overfit on single sample\n  - [x] test multi gpu overfit on 10 tfrecord\n  - [x] test multi gpu overfit on full tfrecord\n\n- [x] create tf record batch (1,4,8,16) sample from dataloader saven \n\n\n## 17.06 Tuesday\n- [x] cleanup Home\n- [x] cleanup ws shared\n\n## 13.06 Friday\n- [ ] Start job for single batch training (overfit dataset)\n- [ ] Make slides for presentation \n\n- [ ] helmholtz setup\n- [ ] move one tf record to helmholts\n\n## 12.06 Thursday\n- [x] Fix oom issue\n  - Dataloader was caching\n- [x] find out biggest batch size for now\n- [x] start a run on one node with coinrun for 12h sbatch\n- [x] start a run on one node with minecraft \n- [x] cleanup ws\n- [x] cleanup wandb\n\n\n\nQuestions Dataloader:\n- [ ] What is one 'element' in the dataloader (Is it an episode or mutlitple episodes? Is it clips from an episode?)\n- [ ] What is the number of elements in the dataloader?\n- [ ] Why is shuffling so slow? Why does it have to shuffle the data and not the indices?\n- [ ] Does the dataloader currently shuffle per episode or per clip?\n- [ ] How do we get the best shuffle? (Optimally uniformly over all clips?)\n- [ ] Do we have to use chaching? Does it improve performance? If yes how much? Is it worth it?\n\nQuestions Training:\n- [ ] What is the best batch size? (What is the best way to get the optimal batch size? Just running it?? Can we calculate/estimate it?)\n- [ ] Can we just use the Genie hyp params or does our data paralelism change the optimal params? (Is the setup good enough?)\n\n\n## 11.06 Wednesday\n- [ ] Start coinrun training\n- [ ] start minecraft training\n\n## 10.06 Tuesday\n- [ ] First run with new pipeline\n- [ ] Zahnarzt\n- [ ] Flaschen wegbringen\n- [ ] \n- [ ] Blutbild und Schulter termin\n\n## 09.06 Monday\n- [x] Zahnarzt termin machen \n- [ ] Presenation Vorbereitung\n\nOffice:\n  - [x] jan ikea\n  - [x] ben geld sae trip\n  - [x] namitha geld\n\n\nHome:\n  - [ ] pa \n  - [ ] jan wlan  \n  - [ ] Barmer geld impfung\n  - [ ] physio geld\n\n\n## 08.06 Sunday\n- [ ] zimmer aufräumen\n  - [ ] klamotten sofa\n  - [ ] müll\n- [ ] küche \n  - [ ] boden\n  - [ ] müll\n- [ ] \n\n- [ ] video pipeline\n  - [ ] clip the npy videos in 16 sec chunks with the designated ranges\n  - [ ] 16sec video to npy?\n\n- [ ] readup data pipeline\n  - [ ] read vpt \n  - [ ] mineworld \n  - [ ] genieN\n  - [ ] 5k openai\n\n## 07.06 Saturday\n- [x] einkaufen \n  - [x] soap \n  - [x] öl \n\n## 06.06 Friday\n- [x] log idling gpus every 30mins\n- [x] make website for it on lxhalle\n\n- [ ] video pipeline  \n  - [x] split into 16 sec videos for datalabeling (training labeler)\n    - [ ] verify results\n    - [x] some videos are not 16sec\n  \n- [x] move videos to new shared workspace\n\nNotes:\n- ffmpeg is bad at keeping length of videos if not re-encoding\n- takes the next best key frame if not re-encoding \n- encoding is super slow (20min video in 40 min instead of 11sec)\n\n## 05.06 Thursday\n- [x] write email to stefan for cluster group \n  - [x] mihir \n  - [x] me \n- [x] video pipeline\n  - [x] convert mp4 to npy\n  - [x] verify results\n\n\n## 03.06 Tuesday\n- [x] kit report \n- [x] karte sap\n\n\n## 30.05 Friday\n- [x] Empty bottles\n- [x] Groceries\n\n- [ ] random dataset for trainign\n- [ ] Macbook?\n- [ ] SAP card\n- [ ] Access to cluster\n- [ ] report for KIT\n\n\n## 28.05 Wednesday\n- [x] Data paralelism \n- [x] Sampling with mihir\n\n## 27.05 Tuesday\n- [ ] agents on envs\n- [ ] video dataset\n- [ ] Crafter/Cratax\n\n- [ ] other envs\n  - [ ] procgen\n  - [ ] gymretro\n  - [ ] minecraft\n\n\n\n## 26.05 Monday\n- [x] data sharing\n- [x] 16 procgen envs\n  - [x] 10mio frames pro env\n\nTraining times:\n1 node 1 env 10 tage \n1.5b param  5b frames 50 atari env 72k gpuh \n64gpus need 12 days\n\n\n\n## 24.05 Saturday\n- [x] email stefan compute access\n- [ ] data gen for the old envs\n\n- [ ] images from gymnax\n- [ ] craftax images\n\n- [ ] 1 page report for KIT\n- [ ] make BA presentation plan\n\n## 23.05 Friday\n- [x] Kit setup\n\n\n\n## 22.05 Thursday\n- [x] BBH\n  - [x] Lemon grass\n  - [x] koriander\n  - [x] topf -> tam\n  - [x] gym?\n- [x] Pr for data generation\n- [x] setup and run gymnax\n\n\n## 21.05 Wednesday\n\n- [ ] datageneration\n  - [x] what do other ppl use? (jafar, openai, craftax)\n  - [x] what is the format (compatibility)\n  - [x] find good environments (easy ones for a good baseline)\n  - [x] implement data generation script (so it jafar compatible)\n  - [x] paramteer for env agents in procgen\n- [x] job for 2 simple environments\n\n*Notes*:\n- jafar uses procgen: https://github.com/openai/procgen\n  - can easily generate 16 different environments\n- gym, gym3, ALE (atari)\n- backwards compatibility was broken by gym in version 0.26.*\n   - have to downgrade to 0.25.* for gym3 to work with gym\n- todo: might have to save the seed into the meta data??\n\n*Sources*:\nGym: https://gymnasium.farama.org/\nGymnax: https://github.com/RobertTLange/gymnax#implemented-accelerated-environments-%EF%B8%8F\nCraftax: https://github.com/MichaelTMatthews/Craftax\n\n\n## 20.05 Tuesday\n- [ ] write email for the kit cluster access\n- [ ] local hpc access\n- [ ] florian bauer macbook \n\n- [ ] look through code\n  - [ ] how does the training work?\n  - [ ] how is it different\n\n  - [ ] Geld for SEA trip\n- [ ] Barmer\n\n\n## 19.05 Monday \n- [x] read through papers for DLM \n- [x] BA upload\n  - [x] notes from below\n  - [x] signation\n\nBA notes:\ntraining-time\ninference-time\nfine-tune\nfinetune\nllm \npython\nin context\nfigure/FIG references\ntransformation appendix??\n\n\n## 18.05 Sunday \n- [x] Email Meinel Lärmstuff\n- [x] Verivox email schreiben\n- [x] BA grammarly \n\n\n\n## 16.05 Friday\n- [x] request Helmholtz HPC access\n- [x] Horeka access\n- [x] Franz fragen \n  - [x] setup reward distribution\n  - [ ] setup genie\n- [x] bashscripts\n- [x] Befragung TUM RDI\n\n## BA\n- [x] Acknowledgements\n- [ ] Read requirements\n  - [x] Tum logos on the front page?\n- [ ] Figures\n  - [x] architecture\n  - [x] inference strat\n  - [x] computation to performance\n- [ ] Appendix?\n  - [x] Comprehensive list of tranformations \n  - [ ] exmaple prompt\n- [ ] Fig 4.1:\n  - Say that we remove the test task and use one of the other tasks as test task\n\n- [x] Figure layout\n- [x] Conclusion put the results in there \n\n\n## Week in Hotze\nMisc:\n- [ ] Steuererklärung \n- [ ] jan wlan geld internet und paket\n- [ ] thanh geld 22 club\n- [ ] Britts rezension \n- [ ] Rezension ha giang\n- [ ] Befragung TUM RDI\n- [ ] Arzt termine\n  - [ ] normal\n  - [ ] zahn\n- [ ] Email Meinel Lärmstuff\n\n## 10.05.25 Saturday\n- [x] BA Figures\n\n\n## 05.05.25 Monday\n- [x] theo paypalen 86 Euro Amazon\n- [x] Email Isabelle Helmholz\n\n\n## DONE\n- [x] Dokument Helmholz\n- [x] stuff bestellen\n  - [x] topper\n  - [x] picknickdecke\n  - [x] reis\n\n- [x] Tasse für PA \n\n\n## 26.03.25 Tuesday\n- [ ] Kreissparkasse\n- [x] Singapore arrival card\n- [ ] Barmer\n  - [ ] impfung\n- [ ] Verivox\n- [ ] \n\n- [x] Packen\n- [x] Meds\n- [x] Müll\n- [x] Toilette\n- [x] Strom ausstecken\n- [x] Fenster schließen\n\n\n\n\n\n\n## 14.03.25 Friday\n\n- [x] Masterbwerbung random\n- [x] emails \n\n- [ ] methods\n\n\n## 12.03.25 Wednesday\nSidequests:\n- [ ] Perso\n- [ ] Führerschein\n- [ ] Barmer karte\n- [ ] hostel singapur\n- [ ] arbeiten\n- [ ] mac abgeben\n\nBA:\n- [ ] background section\n- [ ] methods section\n- [ ] evaluation pipeline\n  - [ ] get the submission files\n  - [ ] evaluate the results (easy, med, hard)\n  - [ ] plot\n- [ ] ds/ep scaling for the tasks that where not solved\n- [ ]\n\n## 11.03.25 Tuesday\nHiwi:\n- [x] Application HIWI\n- [x] master bewerbung\n\nBA:\n- [x] anmeldung BA\n\n- [x] Sparkasse Rückzahlung\n\nSide Quests:\n- [x] paket zurückgeben\n- [x] telekom/verivox thingy\n- [x] master bewerbung\n- [x] jacke flicken\n- [x] arbeiten\n\n## 10.03.25 Monday\n\n## 08.03.25 Saturday\n\n\n## 05.03.25 Wednesday\n\n\n\n\n## 04.03.25 Tuesday\nCleanup:\n- [ ] Wäsche\n- [x] Zimmer random\n- [x] Küche random\n- [x] Staubsaugen\n- [ ] Fenster putzen\n- [ ] Badezimmer putzen\n\nGeld:\n- [ ] Impfung\n- [ ] Hose\n- [ ] Verivox \n\nWork:\n- [ ] Basti\n- [ ] Urlaubausbezahlen?\n- [ ] \n\nBA:\n- [ ] fix inference????\n- [ ] fix transformation???\n\n\n## 03.03.23\n- [x] run predictions for epoch scaling\n- [x] run predictions for epoch scaling\n\n## 02.03.23 Sunday\n- [ ] look for  errors in adapter creation\n  - [ ] some transformation error\n\n- [x] run the predictions\n  - [x] vllm setup\n  - [x] run predictions for epoch scaling\n  - [x] run predictions for epoch scaling\n  - [x] run predictions for epoch scaling\n\n- [x] evaluate epoch scaling batch 1\n\n## 01.03.25 Saturday\n- [x] Run ttt for the experiments\n  - [x] scaling epoch batch 2\n  - [x] scaling data \n  - [x] scaling data + base llama\n  - [x] scaling epoch batch 1\n- [x] run prediction for the experiments\n  - [x] all the scripts setup the scripts\n- [x] notes background section\n\n\n## 27.02.25 Thursday:\nJuwels Cluster setup:\n- [x]  \n\n\nHoreka Cluster setup:\n- [x] get the repo up an running \n  - [x] vpn\n  - [x] clone (repo clean up)\n  - [x] environment\n  - [x] scp the adapters?\n  - [x] run the creation of the adapters from scratch\n  - [x] run predict\n\n\nArzt: \n- [x] Impfung\n\n\n## 24.02.25 Monday\n- [x] start adapter generation \n- [x] debugging the max seq length problem\n- [x] move adapters to scratch?\n\n- [ ] ai fasttrack\n- [ ] mcp\n\n\n\n## 18.02.25 Tuesday\n- [] Laundry (Towels)\n- [x] Groceries\n\n- [ ] Slides holen \n  - [ ] eist \n  - [ ] gad\n  - [ ] \n- [ ] Cleanup room (30min)\n\n\n## 17.02.25 Monday\nBA:\n- [x] start prediction job\n\nAP:\n- [ ] ap test\n\n## 16.02.25 Sunday\n- [x] analyze the training jsons\n  - [x] min/max\n  - [x] debug\n\n- [x] run ttt with 2k samples\n    - [x] which training jsons\n    - [x] where are the ttt adapters save\n\n- [x] test the training/predict piepline\n    - [x] run predict with 4 processes (one per gpu)\n  - [x] evaluation pipeline?\n  - [x] only time for creating the all adapters\n  - [x] how to measure inference time?\n  - [x] how to measure training time?\n\n- [x] Fix transformations \n  - [x] debug the one that are too big?\n  - [x] make it work until 1000\n  - [x] make the transformations more random\n  - [x] look for other transformations\n  - [x] plot all the ones under 500?\n\n- [x] buy pants\n\n## 12.02.25 Wednesday\n- [ ]  \n\n\n## 11.02.25 Dienstag\n- [x] Stefan Bauer schreiben für meeting\n\n- [x] Plan für die nächsten 6 Woche\n  - [x] Aptitude Test\n  - [ ] Bachelor Arbeit\n  - [ ] Arbeit\n\nI’ll be unavailable for an hour because of a doctor's appointment from 10:30 to 11:30 later\n\nVietnam\n- [x] mama anrufen wegen vietnam\n- [x] Impfugen vietnam egerndorfer\n- [x] singapur flüge \n- [x] yu ling fragen wegen referal\n\n## 10.02.25 Monday\n- [x] Table of contents für Bachelorarbeit\n- [x] Repo + Template für Bachelorarbeit\n- [x] Plan für die nächsten 6 Woche\n  - [x] Aptitude Test\n  - [ ] Bachelor Arbeit\n  - [ ] Arbeit\n\n## 09.02.25 Sonntag\n- [x] PA\n- [x] Wäsche \n- [x] Putzen\n  - [x] Fenster Küche\n  - [x] Badezimmer\n  - [x] Zimmer\n- [x] Plan für vietnam\n  - [x] Mia\n  - [x] Ben\n- [x] Machine Learning HA\n\n## 07.02.25 Friday\n- [x] Paper submisssion\n\n## 06.02.25 Thursday\n- [x] sparkasse rückzahlung email  schreibn\n- [x] zusage test\n- [x] barmer geld \n  - [x] sepa lastschrift\n\n### Estimating difficulty\nFor var, ent and std\n- [x] linear regression for the plot (with L1 and L2)\n- [x] ground truth line/regression\n\n- [x] loop through sample length and create plots\n- [x] create one combined plot of all the ablations\n- [x] later do the prm as well \n- [ ] buckets at the end with accuracy?\n\n\n- N: number of samples\n- exp for samples (2^n)\n- linear for seq_len (stepsize 2)\n- for var entr std\n- put all lin regressions in one plot\n\n### Metrics\n- L1 (ground truth)\n- L2 (ground truth)\n\nWhen done \n- Buckets + accuracy\n- per class accuracy\n\n## 04.02.25 Tuesday\n- [x] Deep eval for search\n  - [x] generate golden\n    - [x] convert .md to .txt\n    - [x] run some tests\n\n- [x] geschenk für Joe\n- [x] nach muc\n- [x] arbeiten\n- [x] wäsche\n\n## 03.02.25 Monday\n- [x] Evaluate the results\n  - [x] reinstall vllm\n- [x] aptitude test questioning\n\n\n## 02.02.25 Sunday\n- [x] Zweitsteuer \n- [x] ICE-ticket\n- [x] tum geld semesterbeitrag\n- [x] zulip eignungstest\n\n## 31.01.25 Friday\n- [x] might have to adapt the *.bin_{adapter_num} to *_{adapter_num}.bin\n\n## 30.01.25 Thursday\n- [x] fix error in predict.py\n- [x] create adapter per iteration\n  - [x] every 10 iterations till 100\n  - [x] every 40 iterations till 500\n  - [x] every 100 iterations till 1000\n  - [x] every 200 iterations till 2000\n\n- [x] setup per iteration checkpointing\n- [x] write down ideas\n\n## 29.01.25 Wednesday\n\n- [x] started jobs for one epoch\n\n\n## 28.01.25 Tuesday\n- [x] put the models stats in the output.json\n- [x] run predict on 2000k 1 epoch and 2000k 2 epoch\n\n\nHypothesis:  \n- [ ] harder problem need more samples\n- [ ] peaks for the problems are at different points\n- [ ] add the peaks together\n- [ ] how to predict these peaks\n\n- how to estimate the peak beforehand?\n-> stefan bauer \n\n\n- [ ] epoch 1 \n- [ ] check pointed while traingin\n- [ ] how does test time traing affect the performance?\n\n- [ ] how does it affect \n  - [ ] othter test time scaling\n  - [ ] HER\n  - [ ] sampling size\n  - [ ] test time search\n  - [ ] get the logic\n  - [ ] how could longer ttt affect the performance\n  - [ ] test time search methods\n\n- [ ] easy is solving other tasks\n- [ ] how good is grouping?\n- [ ] costs\n\n\n- [ ] estimate the difficulty differently?\n- [ ] how does this difficulty scale \n\n\nbefore doing anything:\n- questions\n- what do I have to do\n- what do I have to investigate\n- \n\n- [ ] von den einzelen tasks wie verhält sich das?\n- [ ]\n\n\n- [ ] make a road map of all the things I need to do\n- [ ] change the epochs and test for one \n  - [ ] then run the same thing but with 2000 samples and checkpoint\n- [ ] run with deepseek distill\n- [ ] run with deepseek distill on transduction tasks\n\n- [ ] deploy r1\n\n\n- [ ] for the 30 \n- [ ] model chart with total solved tasks\n- [ ] change the lora adapters numbers\n- [ ] change the amount of transformations\n- [ ] change the type of transformations\n- [ ] what is the inference startegy rn?\n- [ ] \n\n\n- [ ] other methods of test time scaling?\n  - [ ] cot on reasoning traces\n  - [ ] more sampling\n  - [ ] look for more\n\n- [ ] fix the gpu errors on some tasks\n- [ ] see if more transformations are needed\n- [ ] distilled models on hard tasks?\n- [ ] might need some reasoning cues....\n- [ ] mehr samplen for inference\n\n\n## 27.01.25 Monday\n- [x] saving results with wandb (redundant)\n  - llm typically trained on 1 epoch\n  - validation would be next token prediction\n  - arc uses accuracy as only metric (free verification and thats the task)\n- [x] move adapters to one folder\n- [x] check how the training is going when changeing the learning set\n- [x] create a dataset of only the hard tasks\n- [x] run training on the hard tasks\n\n- [x] viusalize hard/easy tasks\n  - [x] find out hard task (solved by <40% of the modes)\n- [ ] see if the model transformation is working\n  - get the train data size into the task.csv as well\n## 26.01.25 Sunday\n- [x] find out the duplicates???\n\n## 25.01.25 Saturday\n- [x] multi job on one node \n- [x] find out which tasks are not solved\n  - [x] list of solved/unsolved tasks\n\n## 23.01.25 Thursday\n- [x] chrissie schreiben für getränke\n- [x] und list für ne gruppe machen \n- [x] Franzi schreiben\n- [x] get multigpu training to run (hell nah they rewrote torchtune)\n- [x] get dataset for FT\n- [x] Cluster setup mihir\n- [x] get BARC setup \n  - [x] barc ft\n  - [x] barc adapters\n- [x] Mihir 5e\n- [x] Sushi geld\n\n## 21.01.25 Tuesday\n- [x] work\n- [x] food with ching chongs\n- [x] get the stats\n- [x] start all finetuning jobs\n\n\n## 20.01.25 Monday\n- [ ] Start some lora finetuning\n- [ ] write stefan bauer\n- [ ] \n\n## 19.01.25 Sunday\n- [ ] find out which one got solved\n  - [ ] from baseline 250\n  - [ ] from ours\n  - [ ] get list of adapters\n  - [ ] get list of solved tasks\n\n- [ ] create lora adapters for all sizes\n  - [x] verify number of training samples\n  - [x] verify number of successfully created adapters\n  - [ ] start jobs for create adpaters for 10, 50, 100, 200, 500, 1000 tasks\n\n- [ ] run prediction on all adapters\n  - [ ] 10 \n  - [ ] verify stats: solved/not solved\n\n\n- [ ] fix the spiky behavior\n- [ ] clean up repo\n\n- [ ] spikey behavior\n\n- [ ] Wlan rechnung -> jan\n- [ ] train 1b model on 250 tasks each, if possible\n  - [ ] get the adapters\n  - [ ] run the predict\n  - [ ] see which one are not solve\n  - [ ] run a loop on [10, 50, 100, 200, 500, 1000]\n- [ ] putzen\n- [ ] ML hausuaufgaben\n\n\n## 18.01.25 Saturday  \n- [x] einkauf rechnung\n- [x] rechnungen circula\n- [x] email lmu\n- [x] stefan bauer email für recommendation letter for datathon\n- [x] email osapiens\n- [ ] bewerbungen\n  - [ ] aws\n  - [ ] \n\n\n## 17.01.25 Friday\n- [x] Gym Beine\n- [x] Email draften for mehul\n- [x] linkedin stuff\n- [x] ramakant stuff\n- [x] email for tech support \n\n## 16.01.25 Thursday\n- [x] salloc 4h \n- [x] are we training on json?\n [ ] \n\n## 15.01.25 Wednesday\n- [x] arbeiten\n- [x] raum meetup\n- [ ] \n\n## 14.01.25 Tuesday\n- [x] Erasmus bewerbung\n- [x] Arbeiten\n- [x] Email stefan bauer (cluster access, helping with writing, test runs, i have 60000h no 6000h)\n- [x] Email eikyun\n\n\nHi Stefan, \n\nwäre es möglich Mihir noch Cluster access zu geben? Das würde uns rein vom setup und zusammen arbeiten mega viel Zeit ersparen. \nAußerdem ist mir aufgefallen, dass es sich um 60.000 und nicht um 6.000 cpu stunden handelt. Ich habe noch ca 58.000h übrig. Das sollte erstmal ausreichen denke ich.\nIch würde mich melden falls ich mehr brauche:)\n\nLG Alfred\n\nKurzes Update:\nIch habe jetzt über das Wochende ein paar inference Tasks mit den gegebenen Lora Adaptern laufen lassen. Gerade schreibe ich die Repo um für multi-gpu finetuning damit wir unsere eigenen Adaptere trainieren können. \nAm Freitag haben wir noch ein kurzes Meeting mit Mehul Daminan, für den adaptive compute part (https://arxiv.org/abs/2410.04707, er hat auch am TTT Paper mitgearbeitet). \n\nShort update:\nI was ran some inference tasks with the given Lora Adapters. I am currently rewriting the repo to support multi-gpu finetuning so we can train our own adapters.\nOn Friday we had a short meeting with Mehul Daminan, for the adaptive compute part (https://arxiv.org/abs/2410.04707, he also contributed to the TTT paper).\n\n\n## 13.01.25 Monday\n- [ ] how to generate 5-1000 new tasks?\n- [ ] how transformations work?\n- [ ] Ranking for erasmus\n\n\n## 11.01.25 Saturday\n- [x] Machine learning HA hochladen\n- [x] ttt repo run\n\n- [x] ask stefan for cluster access (mihir and franz)\n- [x] put avocadoaling on iphone\n\n\n- [ ] read paper\n  - [ ] other ttt\n  - [ ] self improvement\n\n- [x] Text stefan bauer\n  - relative confident in idee, but might need supervision and tips : biweekly, help with writing, experiments, etc.\n\n- [x] write core hypothesis and experiments\n- [x] filtern for notes\n\n\n## 10.01.25 Friday\n\n- [x] get TTT to work\n\n- [x] Machine learning hausaufgabe\n\n- [x] read paper\n  - [x] stefan bauer\n  - [x] ttt arc\n\n- [x] ergebnisse vom alten projekt\n- [x] Stefan Bauer meeting\n  - [x] eventuell hiwi\n  - [x] idee schicken\n  - [x] paper schicken\n  - [x] is it ok to be two ppl for the project?\n  - [x] hat er plan\n  - [x] ob er jmd kennt der plan hat?\n  - [x] iclr workshop\n  - [x] ob er leute kennt die ahnung hat \n  - [x] already texted the mit dudes\n  - [x] scaling is ass with arc approaches\n- [x] get access to slack again\n- [x] reward hacking?\n\n## 09.01.25 Thursday\n- [x] write fabian \n- [x] handwerker\n- [x] ramakant\n- [x] presentation of semantic search\n- [x] telekom magenta eins\n- [x] barmer gek stuff\n- [x] clothes for washing tmrw\n- [x] read paper: how hard to think\n- [x] jonas essen \n\n## 08.01.25 Wednesday\n- [x] project idea machen zum doc einladen\n- [x] nachricht an stefan schreiben \n- [x] TTT repo installation setup \n- [x] arbeiten\n- [x] project idea first draft\n\n## 07.01.25 Tuesday\n- [x] merging done\n- [x] Gym\n- [x] Fenstersanierung\n\n## 06.01.25 Monday\n- [x] Gym\n\n- [x] caching and get some results\n\n- [ ] clean up room \n  - [x] couch \n  - [x] table\n  - [x] floor\n  - [x] kitchenstuff\n  - [x] vacuum floor\n- [x] clean up kitchen\n  - [x] dishes \n  - [x] table\n  - [x] vacuum floor\n- [x] clean up bathroom\n  - [x] vacuum floor\n\n- [x] Complete test run with deep  - not possible took too long\n- [x] medium - """"""\n- [x] for the teststats\n- [ ] add test time training in there for the tasks that were not solved\n\n- [ ] why 10 min between runs?\n- [ ] some tasks are way easier than others\n- [ ] setup for running background processes on the compute node\n- [ ] write a visualization for the results\n  - [ ] intermediategrid representation?\n  - [ ] store solutions/reasoning?\n  - [ ] jupyter notebook for visualization\n- [ ] how are the solutions stored?\n- [ ] make one full test run and compare it to the og setup with claude\n  - [ ] get the times for one run\n  - [ ] get the time for the whole dataset\n  - [ ] log it somewhere?\n  - [ ] get the numbers for claude setup\n  - [ ] get the numbers for qwen70b\n- [ ] play around with setup \n  - [ ] differne models\n  - [ ] depth\n    - [ ] deep\n    - [ ] medium\n    - [ ] shallow\n  - [ ] representation\n  - [ ] look at the solved/unsolved ratio and tasks\n  - [ ] any insights?\n- [ ] test out with smaller models\n  - [ ] lamma7b\n  - [ ] qwencoder 32b\n  - [ ] ...\n  - [ ] might be able to run them in parralle on one node\n  - [ ] test out with finetuned models\n- [ ] finetune some model on it?\n- [ ] think about other approaches \n  - [ ] natural language approach?\n  - [ ] use the finetuned model of the winners\n  - [ ] other generation schemes\n- [ ] get claude bedrock running\n- [ ] clean up\n\n## 05.01.25 Sunday\n- [x] gym\n- [x] send jan his datev letter\n- [x] return adidas stuff\n\n- [x] list for mom\n  - [x] keyboard \n  - [x] tshirt\n\n## 04.01.2025\n- [x] write update for stefan \n\n## 03.01.25 Friday\n\nMaiborn:\n- [x] get the circular app\n\nBachelor:\n- [ ] implement bedrock in the pipeline\n- [ ] one test run with claude\n- [ ] debug error??\n- [ ] figure out: intermediate results / tmp files\n  - [ ] saving the python scripts somewhere not /tmp/...\n  - [ ] manual testing of runnning python script\n  - [ ] see what the prompts are getting from the representation\n\nBewerbung:\n- [ ] erasmus\n  - [ ] bewerbung rausschicken\n- [ ] arbeit @aws @google @nvidia @other_big_tech @lab?\n\n\n## 02.01.25 Thursday\nMaiborn:\n- [x] Times\n- [x] Mobility Budget entry\n\nBachelor:\n- [x] setup aws account\n- [x] get bedrock running\n\nMisc: \n- [x] Sehtest Brille  und brille kaufen\n- [x] email wandstreichen\n\n\n## 31.12.24 Monday\nBestellung:\n- [x] Schuhe \n- [x] Socken\n- [ ] \n\nSteuererklärung:\n- [ ] Antrag\n- [ ] App holen\n\nMaiborn:\n- [ ] Times\n- [ ] Mobility Budget\n- [ ] \n\nBachelor:\n- [x] setup aws account\n- [ ] one test run with claude\n- [ ] debug error??\n- [ ] figure out: intermediate results / tmp files\n  - [ ] saving the python scripts somewhere not /tmp/...\n  - [ ] manual testing of runnning python script\n  - [ ] see what the prompts are getting from the representation\n\n\nMisc:\n- [ ] Wlan rechnung -> jan\n- [ ] termin für wandstreichen\n\nBewerbung:\n- [x] master bei lmu\n- [ ] erasmus\n  - [x] info\n  - [ ] bewerbung rausschicken\n- [ ] arbeit @aws @google @nvidia @other_big_tech @lab?\n\n\n\n## Monday \n- [ ] play around with:\n  - [ ] cot\n  - [ ] ascii representation\n  - [ ] prompts\n  - [ ] \n\n## Tuesday\n- [ ]\n\n\n## Wednesday\n- [ ] \n\n\n## Thursday\n- [x] change logging \n  - [x] normal logger class?\n  - [x] save to other format and then logfire?\n\n- [x] trying to run qwen72B -> too big bc of quantization?\n\n- [ ] download other models\n  - [x] vision models are ass\n  - [x] chat models\n\nGiven this iq test. What common transformation do these rows follow?\n\nI have this iq test for you. \nInstructions: \n\nGiven these pairs of grid patterns where each pair shows a 'before' (left) and 'after' (right) transformation, please:\n 1. Identify the rule that defines which how to transform the 'before' pattern into the 'after' pattern\n 2. provide the common transformation occurring to the geometric shapes\n 3. explain your reasoning on how you came to this conclusion\n\n\n- [x] issues fixing linebreak\n- [x] trying to run qwen coder 32B\n  - [x] figuring out the output format\n- [x] figure out how to change tree structure\n\n\n## Friday \n- [ ] maybe use a smaller model?\n- [ ] finetune it on vision tasks\n- [ ] use the other finetune for generation?\n- [ ] self learning approach?\n- [ ] use the dsl and mix to create new stuff\n- [ ] see what kind of tasks can only be solve transductively\n- [ ] what did the top \n\n## Sunday \n- [x] telekom handyvertrag\n- [x] telekom wlan vertrag\n\n\nhttps://www.youtube.com/watch?v=WK5XYG-dH-k&list=PL0oJ2_Q2jPrfkO2Bo8ljN10cShkjkaWzr\n\nhttps://www.youtube.com/watch?v=3yQqNCOYfJo&list=PLbFBnggbJ1rnaXljgzhGm0p_Co9kwfs3t\n\nhttps://www.youtube.com/watch?v=bUudx1cPiAA&list=PLgENJ0iY3XBiJ0jZ53HT8v9Qa3cch7YEV\n\nhttps://www.youtube.com/watch?v=BsJJUAGoFBc&list=PLu0hRahvlQEahuFlF_Dc0_1AMmI_UCLz8\n\nhttps://www.youtube.com/watch?v=0PfLQkUBgcI&list=PL45ZeKlPnPvB6UGmZAJoH57ukARvRNZv3\n\nhttps://www.youtube.com/watch?v=ASfVaQH_1kI&list=PL0oJ2_Q2jPrdt6JFbZtTi8H_XwgYXVRfI\n\nhttps://www.youtube.com/watch?v=DM52HxaLK-Y&list=PLI-n-55RUT--saxVQngjQA3er4QXM67Mt\n\n## Ikea\n\nTotal: 375,78\n\nJan:\n- Bratpfanne Seong: 14,99€\nSum: 14,99€ \n\nAlfred: \n- Schwein 7,99€\n- Dröna Fach 3x1,99€=5,97€\nSum: 13,96€\n\nSplitting:\n375,78 - 14,99 - 13,96 = 346,83\n346,83 / 2 = 173,41\n\nTotal Alfred: 173,41 + 13,96 = 187,37€\nTotal Jan: 173,41 + 14,99 = 188,40€\n\n## SEA ben trip geld\n\n\nHostels:\nHa Long: 1,81 (11,60 total)\nHo Chi Minh: 2.76 (17,30 total)\nBangkok: 7.31 (45,94 total)\nKoh Samui: 4,02 (25,30 total)\n\nHanoi: \nLake View Hostel: 1.04.000d (37,39e)\nLake View Hostel: 130.000d (4,68e)\n\n\nHue:\nSecret Garden Hostel: 269.00d (9,53e)\nGrab: 149.00d (5,27e)\n1995s Hostel: 63.000d (2,18e)\nGrab: 33.000d (1,14e)\nGrab: 111.000d (3,87e)\nGrab: 113.000d (3,94e)\nGrab: 26.000d (0,91e)\n\nHoi An:\nHostel Scalo Villa: 535.000d (18.61e)\n\n\n\n\n\n\n\n## Go Asia\n\nBun: 1.99\n\n\n\n## Meeting Kungwoo\nBehaviour cloning:\n- How much data is needed to properly train the model?\n- \n  \nOther gameplay datasets:\n- pubg dataset\n- biggest dataset\n- 1k hours\n\nData collection\n- bot for data collection\n- nvid\n\n- complexity \n- different worlds\n- not focusing on video games\n  -> only use one game (depending on game)\n  -> not many games \n\n\n\nTier 0:\nhttps://www.youtube.com/results?search_query=alle+meine+entchen+klavier\n\n\n\nI was asked to play piano for someones graduation ceremoy. How much money should I ask for?\nIn the past i got 200 euros for playing Fantasie Impromptu by Chopin and the Waldstein Sonata by Beethoven.\n\nMy current options would be:\nWinter Wind by Choping\nBallad No 4 by Chopin (too long)\nThe Wii Mii Theme\nTwinkle Twinkle Little Star\nAlle meine Entchen\nDepartures by Animenz\nGlimpse of Us arranged by Birru\n\nWhat are some other options that I could play and should I ask for less or more money depending on the piece?\nShould I make a tier list?\n\nBelow is a summary table with the YouTube search URLs for each piece, and an explanation of how you would approach getting the direct video URL if you had access to YouTube’s Data API or a scraping library like youtube-search-python.\n\n| **Tier** | **Piece**                         | **Composer / Arranger** | **Difficulty**                         | **Suggested Fee** | **URL**                                     |\n| -------- | --------------------------------- | ----------------------- | -------------------------------------- | ----------------- | ------------------------------------------- |\n| SS       | Ballade No. 4                     | Chopin                  | Extremely demanding, long, expressive  | 800 €             | https://www.youtube.com/watch?v=pe-GrRQz8pk |\n| S        | Winter Wind Étude (Op. 25 No. 11) | Chopin                  | Very fast, technically virtuosic       | 600 €             | https://www.youtube.com/watch?v=pHlqEvAwdVc |\n| A        | Glimpse of Us                     | Birru (arr.)            | Modern ballad, emotionally nuanced     | 400 €             | https://www.youtube.com/watch?v=Xe2Pr4omHMs |\n| A        | Departures                        | Animenz (arr.)          | Fast, cinematic anime arrangement      | 400 €             | https://www.youtube.com/watch?v=5hft807EJ6o |\n| A        | Aeolian Harp Étude (Op. 25 No. 1) | Chopin                  | Flowing arpeggios, control and clarity | 400 €             | https://www.youtube.com/watch?v=Ob0AQLp3a5s |\n| B        | Clair de Lune                     | Debussy                 | Voicing finesse, sensitive rubato      | 370 €             | https://www.youtube.com/watch?v=U3u4pQ4WKOk |\n| C        | Nocturne Op. 9 No. 2              | Chopin                  | Elegant phrasing, ornamentation        | 300 €             | https://www.youtube.com/watch?v=JVBzE0mUlSs |\n| D        | Twinkle Twinkle Little Star       | Traditional             | Beginner level, simple melody          | 15€               | https://www.youtube.com/watch?v=yCjJyiqpAuU |\n| E        | Wii Mii Theme                     | Nintendo (arr.)         | Nostalgic, light and rhythmic          | 10€               | https://www.youtube.com/watch?v=xo8H6e2Afi4 |\n| F        | Alle meine Entchen                | Traditional             | Basic children's song                  | 5€                | https://www.youtube.com/watch?v=5CDLwq8UgIE |\n\n\nPerfect! Here's how you could add Fantaisie-Impromptu in the A-tier with a fitting description:\n\nTier\tPiece\tComposer / Arranger\tDifficulty\tSuggested Fee\tURL\nA\tFantaisie-Impromptu (Op. 66)\tChopin\tFast, rhythmically complex, technically demanding but less intense than top Chopin Études\t450 €\thttps://www.youtube.com/watch?v=IvUU8joBb1Q\n\nIf you want, I can help you refine the difficulty wording or suggest a fee based on your current pricing scale!\n\n\n\n\n\n\n\n\n\n",markdown,tab
+2,98,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab
+3,851,"TERMINAL",0,0,"python /Users/alfred/.cursor/extensions/ms-python.python-2025.6.1-darwin-arm64/python_files/printEnvVariablesToFile.py /Users/alfred/.cursor/extensions/ms-python.python-2025.6.1-darwin-arm64/python_files/deactivate/bash/envVars.txt",,terminal_command
+4,867,"TERMINAL",0,0,"]633;E;python /Users/alfred/.cursor/extensions/ms-python.python-2025.6.1-darwin-arm64/python_files/printEnvVariablesToFile.py /Users/alfred/.cursor/extensions/ms-python.python-2025.6.1-darwin-arm64/python_files/deactivate/bash/envVars.txt;94f1541c-df11-407c-9963-13cd6466203f]633;C",,terminal_output
+5,876,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"12:51:37 AM [info] Activating crowd-code\n12:51:37 AM [info] Welcome back alfred. Your user-id is '05d9d5da933137c5402a176a469b618685c7e9142aa8972616ca5cdf0f6e53d1'. Happy coding!\n12:51:37 AM [info] Recording started\n",Log,content

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-cd3adcdf-8370-4269-848e-4350f71afc211751306079061-2025_06_30-19.54.55.948/source.csv

@@ -0,0 +1,65 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,4,"train_lam.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 dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\n\nfrom models.lam import LatentActionModel\nfrom utils.dataloader import get_dataloader\nfrom utils.parameter_utils import count_parameters_by_component\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    checkpoint: str = """"\n    # Optimization\n    batch_size: int = 36\n    vq_beta: float = 0.25\n    min_lr: float = 3e-6\n    max_lr: float = 3e-5\n    warmup_steps: int = 5000\n    vq_reset_thresh: int = 50\n    # LAM\n    model_dim: int = 512\n    latent_dim: int = 32\n    num_latents: int = 6\n    patch_size: int = 16\n    num_blocks: int = 8\n    num_heads: int = 8\n    dropout: float = 0.0\n    codebook_dropout: float = 0.0\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\n\nargs = tyro.cli(Args)\n\n\ndef lam_loss_fn(params, state, inputs):\n    # --- Compute loss ---\n    outputs = state.apply_fn(\n        params, inputs, training=True, rngs={""dropout"": inputs[""rng""]}\n    )\n    gt_future_frames = inputs[""videos""][:, 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 = pix.psnr(gt, recon).mean()\n    ssim = 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@jax.jit\ndef train_step(state, inputs, action_last_active):\n    # --- Update model ---\n    rng, inputs[""rng""] = jax.random.split(inputs[""rng""])\n    grad_fn = jax.value_and_grad(lam_loss_fn, has_aux=True, allow_int=True)\n    (loss, (recon, idx_counts, metrics)), grads = grad_fn(state.params, state, inputs)\n    state = state.apply_gradients(grads=grads)\n\n    # --- Reset inactive latent actions ---\n    codebook = state.params[""params""][""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\n    )\n    state.params[""params""][""vq""][""codebook""] = new_codebook\n    action_last_active = jnp.where(do_reset, 0, action_last_active)\n    return state, 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.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    lam = LatentActionModel(\n        in_dim=args.image_channels,\n        model_dim=args.model_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    )\n    # Track when each action was last sampled\n    action_last_active = jnp.zeros(args.num_latents)\n    image_shape = (args.image_height, args.image_width, args.image_channels)\n    rng, _rng = jax.random.split(rng)\n    inputs = dict(\n        videos=jnp.zeros(\n            (per_device_batch_size_for_init, args.seq_len, *image_shape),\n            dtype=jnp.float32,\n        ),\n        rng=_rng,\n    )\n    rng, _rng = jax.random.split(rng)\n    init_params = lam.init(_rng, inputs)\n\n    param_counts = count_parameters_by_component(init_params)\n    print(""Parameter counts:"")\n    print(param_counts)\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=lam.apply, params=init_params, tx=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    train_state = jax.device_put(train_state, replicated_sharding)\n    action_last_active = jax.device_put(action_last_active, replicated_sharding)\n\n    # --- Load checkpoint ---\n    step = 0\n    if args.checkpoint:\n        restore_target = {""model"": train_state}\n        restore_args = orbax_utils.restore_args_from_target(restore_target)\n        train_state.params[""params""].update(\n            PyTreeCheckpointer()\n            .restore(args.checkpoint, item=restore_target, restore_args=restore_args)[\n                ""model""\n            ]\n            .params[""params""]\n        )\n        # Assume checkpoint is of the form tokenizer_<timestamp>_<step>\n        step += int(args.checkpoint.split(""_"")[-1])\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    print(f""Starting training from step {step}..."")\n    while step < args.num_steps:\n        for videos in dataloader:\n            # --- Train step ---\n            rng, _rng = jax.random.split(rng)\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(videos=videos, rng=_rng)\n            start_time = time.time()\n            train_state, loss, recon, action_last_active, metrics = train_step(\n                train_state, inputs, action_last_active\n            )\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][1:]\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[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            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""lam_{ts}_{step}""),\n                    ckpt,\n                    save_args=save_args,\n                )\n            if step >= args.num_steps:\n                break\n",python,tab
+2,1326,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"7:54:55 PM [info] Activating crowd-code\n7:54:55 PM [info] Welcome back tum_ind3695. Your user-id is '507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20'. Happy coding!\n7:54:55 PM [info] Recording started\n",Log,tab
+3,1935,"train_lam.py",0,0,"",python,tab
+4,3854,"TERMINAL",0,0,"/bin/python /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/printEnvVariablesToFile.py /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash/envVars.txt",,terminal_command
+5,3900,"TERMINAL",0,0,"]633;E;2025-06-30 19:54:59 /bin/python /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/printEnvVariablesToFile.py /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash/envVars.txt;e30aeaa0-56ca-42c9-9301-5e91a5c354f0]633;C",,terminal_output
+6,3979,"TERMINAL",0,0,"]0;tum_ind3695@hkn1993:/hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash]633;D;0",,terminal_output
+7,59185,"slurm/dev/alfred/train_lam_dev/train_lam_single_batch.sh",0,0,"module unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv_jafar/bin/activate\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\n\nCHECKPOINT_DIR=$ws_dir/checkpoints/lam/$job_name_$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\n\n# data_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_tfrecords_1_shard_overfit'\ndata_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_tfrecords_500_shards'\n# data_dir = '/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_tfrecords_500_shards_overfit_1'\n\n# srun python train_lam.py \\npython train_lam.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=48 \\n    --log_checkpoint_interval=1000 \\n    --log_image_interval=10 \\n    --seed=0 \\n    --min_lr=0.0000433 \\n    --max_lr=0.0000433 \\n    --log \\n    --entity instant-uv \\n    --data_dir $data_dir \\n    --project jafar\n",shellscript,tab
+8,59197,"slurm/dev/alfred/preprocess/preprocess_video_to_npy.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=01:30:00\n#SBATCH --partition=cpuonly\n#SBATCH --account=hk-project-pai00039\n#SBATCH --cpus-per-task=72\n#SBATCH --output=logs/logs_preprocessing/%x_%j.log\n#SBATCH --error=logs/logs_preprocessing/%x_%j.log\n#SBATCH --mail-user=avocadoaling@gmail.com\n#SBATCH --job-name=preprocess_video_to_npy\n#SBATCH --mail-type=ALL\n\n# Log the sbatch script\ncat $0\n\nsource .venv/bin/activate\n\ninput_path=""/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms/""\noutput_path=""/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/knoms_npy_tmp""\n\nstart_time=$(date +%s)  \npython utils/preprocess_video_to_npy.py --input_path $input_path --output_path $output_path\nend_time=$(date +%s)\necho ""Time taken: $((end_time - start_time)) seconds""\n\n\n\n\npython utils/preprocess_video_to_npy.py --input_path $input_path --output_path $output_path",shellscript,tab
+9,59199,"TERMINAL",0,0,"bash",,terminal_focus
+10,59200,"TERMINAL",0,0,"fsacct_week",,terminal_command
+11,59201,"TERMINAL",0,0,"bash",,terminal_focus
+12,59203,"slurm/dev/alfred/preprocess/preprocess_video_to_npy.sbatch",0,0,"",shellscript,tab
+13,59207,"slurm/dev/alfred/train_lam_dev/train_lam_single_batch.sh",0,0,"",shellscript,tab
+14,125352,"TERMINAL",0,0,"",,terminal_focus
+15,128390,"TERMINAL",0,0,"idle",,terminal_command
+16,128434,"TERMINAL",0,0,"]633;E;2025-06-30 19:57:04 idle;63556177-5250-4c5a-a767-4cf70a02e8b0]633;CPartition dev_cpuonly             :      6 nodes idle\r\nPartition cpuonly                 :      8 nodes idle\r\nPartition dev_accelerated         :      2 nodes idle\r\nPartition accelerated             :     11 nodes idle\r\nPartition dev_accelerated-h100    :      0 nodes idle\r\nPartition accelerated-h100        :      1 nodes idle\r\nPartition large                   :      8 nodes idle\r\n]0;tum_ind3695@hkn1993:~/projects/jafar]633;D;0",,terminal_output
+17,170201,"TERMINAL",0,0,"cd slurm/",,terminal_command
+18,170209,"TERMINAL",0,0,"]633;E;2025-06-30 19:57:46 cd slurm/;63556177-5250-4c5a-a767-4cf70a02e8b0]633;C]0;tum_ind3695@hkn1993:~/projects/jafar/slurm]633;D;0]633;P;Cwd=/home/hk-project-pai00039/tum_ind3695/projects/jafar/slurm",,terminal_output
+19,171637,"TERMINAL",0,0,"git pull",,terminal_command
+20,171690,"TERMINAL",0,0,"]633;E;2025-06-30 19:57:47 git pull;63556177-5250-4c5a-a767-4cf70a02e8b0]633;C",,terminal_output
+21,173600,"TERMINAL",0,0,"remote: Enumerating objects: 59, done.\r\nremote: Counting objects:   2% (1/44)\rremote: Counting objects:   4% (2/44)\rremote: Counting objects:   6% (3/44)\rremote: Counting objects:   9% (4/44)\rremote: Counting objects:  11% (5/44)\rremote: Counting objects:  13% (6/44)\rremote: Counting objects:  15% (7/44)\rremote: Counting objects:  18% (8/44)\rremote: Counting objects:  20% (9/44)\rremote: Counting objects:  22% (10/44)\rremote: Counting objects:  25% (11/44)\rremote: Counting objects:  27% (12/44)\rremote: Counting objects:  29% (13/44)\rremote: Counting objects:  31% (14/44)\rremote: Counting objects:  34% (15/44)\rremote: Counting objects:  36% (16/44)\rremote: Counting objects:  38% (17/44)\rremote: Counting objects:  40% (18/44)\rremote: Counting objects:  43% (19/44)\rremote: Counting objects:  45% (20/44)\rremote: Counting objects:  47% (21/44)\rremote: Counting objects:  50% (22/44)\rremote: Counting objects:  52% (23/44)\rremote: Counting objects:  54% (24/44)\rremote: Counting objects:  56% (25/44)\rremote: Counting objects:  59% (26/44)\rremote: Counting objects:  61% (27/44)\rremote: Counting objects:  63% (28/44)\rremote: Counting objects:  65% (29/44)\rremote: Counting objects:  68% (30/44)\rremote: Counting objects:  70% (31/44)\rremote: Counting objects:  72% (32/44)\rremote: Counting objects:  75% (33/44)\rremote: Counting objects:  77% (34/44)\rremote: Counting objects:  79% (35/44)\rremote: Counting objects:  81% (36/44)\rremote: Counting objects:  84% (37/44)\rremote: Counting objects:  86% (38/44)\rremote: Counting objects:  88% (39/44)\rremote: Counting objects:  90% (40/44)\rremote: Counting objects:  93% (41/44)\rremote: Counting objects:  95% (42/44)\rremote: Counting objects:  97% (43/44)\rremote: Counting objects: 100% (44/44)\rremote: Counting objects: 100% (44/44), done.\r\nremote: Compressing objects:  10% (1/10)\rremote: Compressing objects:  20% (2/10)\rremote: Compressing objects:  30% (3/10)\rremote: Compressing objects:  40% (4/10)\rremote: Compressing objects:  50% (5/10)\rremote: Compressing objects:  60% (6/10)\rremote: Compressing objects:  70% (7/10)\rremote: Compressing objects:  80% (8/10)\rremote: Compressing objects:  90% (9/10)\rremote: Compressing objects: 100% (10/10)\rremote: Compressing objects: 100% (10/10), done.\r\nUnpacking objects:   3% (1/27)\rUnpacking objects:   7% (2/27)\rUnpacking objects:  11% (3/27)\rUnpacking objects:  14% (4/27)\rUnpacking objects:  18% (5/27)\rUnpacking objects:  22% (6/27)\rUnpacking objects:  25% (7/27)\rUnpacking objects:  29% (8/27)\rUnpacking objects:  33% (9/27)\rUnpacking objects:  37% (10/27)\rUnpacking objects:  40% (11/27)\rUnpacking objects:  44% (12/27)\rUnpacking objects:  48% (13/27)\rUnpacking objects:  51% (14/27)\rUnpacking objects:  55% (15/27)\rUnpacking objects:  59% (16/27)\rremote: Total 27 (delta 18), reused 26 (delta 17), pack-reused 0 (from 0)\r\nUnpacking objects:  62% (17/27)\rUnpacking objects:  66% (18/27)\rUnpacking objects:  70% (19/27)\rUnpacking objects:  74% (20/27)\rUnpacking objects:  77% (21/27)\rUnpacking objects:  81% (22/27)\rUnpacking objects:  85% (23/27)\rUnpacking objects:  88% (24/27)\rUnpacking objects:  92% (25/27)\rUnpacking objects:  96% (26/27)\rUnpacking objects: 100% (27/27)\rUnpacking objects: 100% (27/27), 3.07 KiB | 23.00 KiB/s, done.\r\nFrom github.com:p-doom/slurm\r\n   00c01e0..705b23d  main       -> origin/main\r\nUpdating 00c01e0..705b23d\r\nFast-forward\r\n dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sbatch                                          |  67 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sh                                              |  53 +++++++++++++++++++++++++++++++++++++++++++++++++++++\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0.6_mio.sbatch                                             |   2 +-\r\n dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0_5.sh                                                     |  22 ++++++++++++++++------\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/logs/logs_training/train_tokenizer_batch_size_scaling_2_node_3292213.log | 136 ----------------------------------------------------------------------------------------------------------------------------------------\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_16_nodes.sbatch                                          |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_1_nodes.sbatch                                           |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_2_nodes.sbatch                                           |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_32_nodes.sbatch                                          |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_4_nodes.sbatch                                           |   2 +-\r\n jobs/mihir/horeka/batchsize_scaling/adjusted_lr/train_tokenizer_8_nodes.sbatch                                           |   2 +-\r\n 11 files changed, 143 insertions(+), 149 deletions(-)\r\n create mode 100644 dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sbatch\r\n create mode 100755 dev/alfred/overfit_minecraft_single_sample/train_dynamics_overfit_sample.sh\r\n delete mode 100644 jobs/mihir/horeka/batchsize_scaling/adjusted_lr/logs/logs_training/train_tokenizer_batch_size_scaling_2_node_3292213.log\r\n]0;tum_ind3695@hkn1993:~/projects/jafar/slurm]633;D;0",,terminal_output
+22,555751,"TERMINAL",0,0,"bash",,terminal_focus
+23,555946,"TERMINAL",0,0,"bash",,terminal_focus
+24,559191,"TERMINAL",0,0,"cd ..",,terminal_command
+25,559198,"TERMINAL",0,0,"]633;E;2025-06-30 20:04:15 cd ..;63556177-5250-4c5a-a767-4cf70a02e8b0]633;C]0;tum_ind3695@hkn1993:~/projects/jafar]633;D;0",,terminal_output
+26,560343,"TERMINAL",0,0,"git pull",,terminal_command
+27,560388,"TERMINAL",0,0,"]633;E;2025-06-30 20:04:16 git pull;63556177-5250-4c5a-a767-4cf70a02e8b0]633;C",,terminal_output
+28,562127,"TERMINAL",0,0,"remote: Enumerating objects: 8, done.\r\nremote: Counting objects:  12% (1/8)\rremote: Counting objects:  25% (2/8)\rremote: Counting objects:  37% (3/8)\rremote: Counting objects:  50% (4/8)\rremote: Counting objects:  62% (5/8)\rremote: Counting objects:  75% (6/8)\rremote: Counting objects:  87% (7/8)\rremote: Counting objects: 100% (8/8)\rremote: Counting objects: 100% (8/8), done.\r\nremote: Compressing objects:  12% (1/8)\rremote: Compressing objects:  25% (2/8)\rremote: Compressing objects:  37% (3/8)\rremote: Compressing objects:  50% (4/8)\rremote: Compressing objects:  62% (5/8)\rremote: Compressing objects:  75% (6/8)\rremote: Compressing objects:  87% (7/8)\rremote: Compressing objects: 100% (8/8)\rremote: Compressing objects: 100% (8/8), done.\r\n",,terminal_output
+29,562246,"TERMINAL",0,0,"remote: Total 8 (delta 2), reused 0 (delta 0), pack-reused 0 (from 0)\r\nUnpacking objects:  12% (1/8)\rUnpacking objects:  25% (2/8)\rUnpacking objects:  37% (3/8)\rUnpacking objects:  50% (4/8)\rUnpacking objects:  62% (5/8)\rUnpacking objects:  75% (6/8)\rUnpacking objects:  87% (7/8)\rUnpacking objects: 100% (8/8)\rUnpacking objects: 100% (8/8), 6.44 KiB | 219.00 KiB/s, done.\r\n",,terminal_output
+30,562305,"TERMINAL",0,0,"From github.com:p-doom/jafar\r\n   278d4b1..1a7ac42  main       -> origin/main\r\nAlready up to date.\r\n]0;tum_ind3695@hkn1993:~/projects/jafar]633;D;0",,terminal_output
+31,739290,".gitignore",0,0,"*.pyc\n*.npy\n*.png\n*.gif\n\nwandb_key\ncheckpoints/\nwandb/\n__pycache__/\n\nlogs/\nsandbox/\nsbatch_scripts/\nvs-code-recorder/\ndata/\ndata_tfrecords/\nsh_scripts/\nutils/clip_checker.py\nutils/dataloader_seeding.py\nutils/preprocess_video_splitter_tmp.py\nrequirements_franz.txt\nsample_resolution_batches.py\ntrain_dynamics_*\ntrain_lam_*\ntrain_tokenizer_*\nnotes.md\nshell_scripts/\nslurm/\n",ignore,tab
+32,740900,".gitignore",0,371,"*.pyc\n*.npy\n*.png\n*.gif\n\nwandb_key\ncheckpoints/\nwandb/\n__pycache__/\n\nlogs/\nsandbox/\nsbatch_scripts/\nvs-code-recorder/\ndata/\ndata_tfrecords/\nsh_scripts/\nutils/clip_checker.py\nutils/dataloader_seeding.py\nutils/preprocess_video_splitter_tmp.py\nrequirements_franz.txt\nsample_resolution_batches.py\ntrain_dynamics_*\ntrain_lam_*\ntrain_tokenizer_*\nnotes.md\nshell_scripts/\nslurm/\n",ignore,selection_command
+33,741057,".gitignore",0,0,"",ignore,selection_command
+34,947658,".gitignore",371,0,"",ignore,selection_mouse
+35,948143,"TERMINAL",0,0,"bash",,terminal_focus
+36,950955,"TERMINAL",0,0,"bash",,terminal_focus
+37,951736,"TERMINAL",0,0,"gs",,terminal_command
+38,951791,"TERMINAL",0,0,"]633;E;2025-06-30 20:10:47 gs;63556177-5250-4c5a-a767-4cf70a02e8b0]633;COn branch feature/model-parameter-count-utils\r\nYour branch is up to date with 'origin/feature/model-parameter-count-utils'.\r\n\r\nnothing to commit, working tree clean\r\n]0;tum_ind3695@hkn1993:~/projects/jafar]633;D;0",,terminal_output
+39,956657,"TERMINAL",0,0,"git branch",,terminal_command
+40,956693,"TERMINAL",0,0,"]633;E;2025-06-30 20:10:52 git branch;63556177-5250-4c5a-a767-4cf70a02e8b0]633;C[?1h=\r  dataset-generation-procgen-gym\r\n* feature/model-parameter-count-utils\r\n  fix-dataloader-caching\r\n  fix-image-resolution-args\r\n  fix-image-resolution-lam-dynamics\r\n  fix-seeding-stateless-sampling\r\n  fix_preprocess_video_paths\r\n  infra-slurm-dev-scripts\r\n  main\r\n  preprocess_video\r\n  quickfix-all-gather-induced-idling\r\n  refactor/precommit-cleanup\r\n  seeded-episode-sampling\r\n  single-batch-training\r\n\r[?1l>]0;tum_ind3695@hkn1993:~/projects/jafar]633;D;0",,terminal_output
+41,963169,"TERMINAL",0,0,"git ps",,terminal_command
+42,963198,"TERMINAL",0,0,"]633;E;2025-06-30 20:10:59 git ps;63556177-5250-4c5a-a767-4cf70a02e8b0]633;Cgit: 'ps' is not a git command. See 'git --help'.\r\n\r\nThe most similar command is\r\n\tpush\r\n]0;tum_ind3695@hkn1993:~/projects/jafar]633;D;1",,terminal_output
+43,964175,"TERMINAL",0,0,"gs",,terminal_command
+44,964207,"TERMINAL",0,0,"]633;E;2025-06-30 20:11:00 gs;63556177-5250-4c5a-a767-4cf70a02e8b0]633;COn branch feature/model-parameter-count-utils\r\nYour branch is up to date with 'origin/feature/model-parameter-count-utils'.\r\n\r\nnothing to commit, working tree clean\r\n]0;tum_ind3695@hkn1993:~/projects/jafar]633;D;0",,terminal_output
+45,1008477,".gitignore",0,0,"",ignore,tab
+46,1009557,".gitignore",68,303,"\nlogs/\nsandbox/\nsbatch_scripts/\nvs-code-recorder/\ndata/\ndata_tfrecords/\nsh_scripts/\nutils/clip_checker.py\nutils/dataloader_seeding.py\nutils/preprocess_video_splitter_tmp.py\nrequirements_franz.txt\nsample_resolution_batches.py\ntrain_dynamics_*\ntrain_lam_*\ntrain_tokenizer_*\nnotes.md\nshell_scripts/\nslurm/\n",ignore,selection_command
+47,1009880,".gitignore",68,0,"",ignore,selection_command
+48,1356023,"slurm/dev/alfred/overfit_sample/train_tokenizer_overfit_sample_size_0_5.sh",0,0,"# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv_jafar/bin/activate\n\nws_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/'\ntf_records_dir=$ws_dir/data/knoms_tfrecords_200_shards\n\nslurm_job_id=$SLURM_JOB_ID\n\ntags=""overfit_sample tokenizer debug alfred""\n\nCHECKPOINT_DIR=$ws_dir/checkpoints/$job_name_$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\nenv | grep SLURM\n\n# model_dim=32\n# num_blocks=4\nlr=1e-4\n\nmodel_dim=256\nnum_blocks=8\n\njob_name=cointok_mod_dim_${model_dim}_num_blocks_${num_blocks}_lr_${lr}\n\npython train_tokenizer_single_sample.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=1 \\n    --min_lr=$lr \\n    --max_lr=$lr \\n    --log_image_interval=500 \\n    --log \\n    --entity instant-uv \\n    --project jafar \\n    --name $job_name \\n    --tags $tags \\n    --model_dim $model_dim \\n    --num_blocks $num_blocks \\n    --data_dir $tf_records_dir\n\n",shellscript,tab
+49,1356534,".gitignore",0,0,"",ignore,tab
+50,1361542,"TERMINAL",0,0,"bash",,terminal_focus
+51,1361546,"slurm/dev/alfred/train_lam_dev/train_lam_single_batch.sh",0,0,"",shellscript,tab
+52,2876080,"/home/hk-project-pai00039/tum_ind3695/.bashrc",0,0,"# .bashrc\n\n# Source global definitions\nif [ -f /etc/bashrc ]; then\n    . /etc/bashrc\nfi\n\n# User specific environment\nif ! [[ ""$PATH"" =~ ""$HOME/.local/bin:$HOME/bin:"" ]]\nthen\n    PATH=""$HOME/.local/bin:$HOME/bin:$PATH""\nfi\nexport PATH\n\n\nalias queue='watch -n 1 squeue --me'\nalias fqueue='squeue -o ""%.10i %.16P %.60j %.8u %.8T %.10M %.9l %.6D %R""'\nalias idle='sinfo_t_idle'\nalias smi='watch -n 1 nvidia-smi'\nalias fsacct_week='sacct --format=""JobID%15,JobName%30,Partition%16,AllocCPUS%3,State%12,Elapsed%10,Timelimit%10"" --starttime $(date -d ""last week"" +%Y-%m-%d) | grep -vE ""*.batch|*.extern|*.inter|bash|python|CANCELLED|echo""'\nalias fsacct_today='sacct --format=""JobID%15,JobName%55,Partition%16,AllocCPUS%3,State%12,Elapsed%10,Timelimit%10"" --starttime $(date +%Y-%m-%d) | grep -vE ""*.batch|*.extern|*.inter|bash""'\nalias execute_order_66_gpu_1='salloc --account=hk-project-p0023960 --time=01:00:00 --partition=accelerated --nodes=1 --gres=gpu:1  --cpus-per-task=8 --mem=450G --ntasks-per-node=1'\nalias execute_order_66_gpu_4='salloc --account=hk-project-p0023960 --time=01:00:00 --partition=accelerated --nodes=1 --gres=gpu:4  --cpus-per-task=8 --ntasks-per-node=4'\nalias execute_order_66_cpuonly='salloc --account=hk-project-p0023960 --time=1:00:00 --partition=dev_cpuonly --nodes=1 --cpus-per-task=128 --ntasks-per-node=1'\n\nexport ws_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/'\nexport HF_HOME=""$ws_dir/huggingface/""\n\nalias binger='cd $ws_dir/'\nalias dinger='cd /home/hk-project-pai00039/tum_ind3695/projects/jafar'\n\nalias gs='git status'\n\n\nexport PATH=/home/hk-project-pai00039/tum_ind3695/projects/ffmpeg-7.0.2-amd64-static:$PATH            \n\noverlapper() {\n    if [ -z ""$1"" ]; then\n        echo ""Usage: overlap <jobid>""\n    else\n        srun --jobid=""$1"" --pty /usr/bin/bash\n    fi\n}\n\n",shellscript,tab
+53,2877004,"/home/hk-project-pai00039/tum_ind3695/.bashrc",0,0,"",shellscript,selection_command
+54,2877701,"/home/hk-project-pai00039/tum_ind3695/.bashrc",10,0,"",shellscript,selection_command
+55,2877857,"/home/hk-project-pai00039/tum_ind3695/.bashrc",88,0,"",shellscript,selection_command
+56,2877990,"/home/hk-project-pai00039/tum_ind3695/.bashrc",233,0,"",shellscript,selection_command
+57,2878132,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1330,0,"",shellscript,selection_command
+58,2878273,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1442,0,"",shellscript,selection_command
+59,2878422,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1541,0,"",shellscript,selection_command
+60,2879188,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1564,0,"",shellscript,selection_command
+61,2879333,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1669,0,"",shellscript,selection_command
+62,2879473,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1812,0,"",shellscript,selection_command
+63,2879924,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1669,0,"",shellscript,selection_command
+64,2880085,"/home/hk-project-pai00039/tum_ind3695/.bashrc",1565,0,"",shellscript,selection_command

507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20/crowd-code-fe1de2ac-919a-4753-a5bb-5c68f3cb240b1750773379086-2025_06_24-15.56.40.593/source.csv

@@ -0,0 +1,32 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,9,"data_download/download_scripts/download_6xx.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=6:00:00\n#SBATCH --partition=cpuonly\n#SBATCH --account=hk-project-p0023960\n#SBATCH --output=logs/%x_%j.log\n#SBATCH --error=logs/%x_%j.log\n#SBATCH --mail-user=avocadoaling@gmail.com\n#SBATCH --job-name=download_6xx\n#SBATCH --mail-type=ALL\n\nsource ../.venv/bin/activate\n\n\nindex_file='index_json/all_6xx_Jun_29.json'\noutput_dir='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/open_ai_minecraft'\n\npython download_scripts/download_videos.py --index_file_path $index_file --output_dir $output_dir \n",shellscript,tab
+2,855,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"3:56:40 PM [info] Activating crowd-code\n3:56:40 PM [info] Welcome back tum_ind3695. Your user-id is '507ab0ec0dfe0c18ad7778dd15e072f92367194c94623114de802c8ed9c52e20'. Happy coding!\n3:56:40 PM [info] Recording started\n",Log,tab
+3,1707,"data_download/download_scripts/download_6xx.sbatch",0,0,"",shellscript,tab
+4,2927,"TERMINAL",0,0,"/bin/python3 /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/printEnvVariablesToFile.py /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash/envVars.txt",,terminal_command
+5,2981,"TERMINAL",0,0,"]633;E;2025-06-24 15:56:43 /bin/python3 /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/printEnvVariablesToFile.py /hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash/envVars.txt;31cb0e7a-7016-4ee5-abfe-86d22d42bb62]633;C",,terminal_output
+6,2993,"TERMINAL",0,0,"]0;tum_ind3695@hkn1993:/hkfs/home/project/hk-project-pai00039/tum_ind3695/.cursor-server/extensions/ms-python.python-2025.6.1-linux-x64/python_files/deactivate/bash]633;D;0",,terminal_output
+7,21309,"data_download/download_scripts/download_6xx.sbatch",586,0,"",shellscript,selection_mouse
+8,45294,"data_download/download_scripts/download_videos.py",0,0,"import json\nimport requests\nimport os\nimport tyro\nimport logging\nfrom urllib.parse import urljoin\nfrom dataclasses import dataclass\nfrom pathlib import Path\nfrom tqdm import tqdm\nfrom multiprocessing import Pool, cpu_count\nimport time\n\n@dataclass\nclass DownloadVideos:\n    index_file_path: str ='index_json/all_6xx_Jun_29.json'\n    output_dir: str ='/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/open_ai_minecraft'\n\ndef download_single_file(args):\n    """"""Download a single file - designed to be used with multiprocessing""""""\n    url, output_path, relpath = args\n    \n    if os.path.exists(output_path):\n        return f""Skipped {relpath} (already exists)""\n    \n    # Create parent directories if they don't exist\n    os.makedirs(os.path.dirname(output_path), exist_ok=True)\n    \n    try:\n        response = requests.get(url, stream=True, timeout=30)\n        if response.status_code == 200:\n            file_size = 0\n            with open(output_path, 'wb') as f:\n                for chunk in response.iter_content(chunk_size=8192):\n                    if chunk:\n                        f.write(chunk)\n                        file_size += len(chunk)\n            \n            # Convert to MB for logging\n            file_size_mb = file_size / (1024 * 1024)\n            return f""Downloaded {relpath} ({file_size_mb:.2f} MB)""\n        else:\n            return f""Failed to download {relpath}: HTTP {response.status_code}""\n    except requests.exceptions.RequestException as e:\n        return f""Request failed for {relpath}: {e}""\n    except Exception as e:\n        return f""Unexpected error downloading {relpath}: {e}""\n\ndef download_dataset(index_file_path, output_dir, num_workers=64):\n    # Load the index file\n    with open(index_file_path, 'r') as f:\n        index_data = json.load(f)\n    \n    basedir = index_data['basedir']\n    relpaths = index_data['relpaths']\n    \n    # Filter for mp4 files only\n    mp4_files = [(relpath, urljoin(basedir, relpath), os.path.join(output_dir, relpath)) \n                 for relpath in relpaths if relpath.endswith('.mp4')]\n    \n    print(f""Found {len(mp4_files)} MP4 files to download"")\n    print(f""Using {num_workers} workers for parallel downloads"")\n    \n    # Use multiprocessing pool to download files in parallel\n    start_time = time.time()\n    \n    if num_workers > len(mp4_files):\n        num_workers = len(mp4_files)\n    \n    # Create a progress bar for overall progress\n    with tqdm(total=len(mp4_files), desc=""Overall Download Progress"", unit=""files"") as pbar:\n        with Pool(processes=num_workers) as pool:\n            # Map the download function to all files (without passing pbar)\n            results = []\n            for result in pool.imap_unordered(download_single_file, \n                                             [(url, output_path, relpath) for relpath, url, output_path in mp4_files]):\n                results.append(result)\n                pbar.update(1)\n                # Print every 100th result to avoid overwhelming output\n                if len(results) % 100 == 0:\n                    print(f""Completed {len(results)} downloads..."")\n    \n    # Print final results summary\n    successful_downloads = sum(1 for r in results if ""Downloaded"" in r)\n    skipped_files = sum(1 for r in results if ""Skipped"" in r)\n    failed_downloads = len(results) - successful_downloads - skipped_files\n    \n    print(f""\nDownload Summary:"")\n    print(f""  Successful downloads: {successful_downloads}"")\n    print(f""  Skipped files: {skipped_files}"")\n    print(f""  Failed downloads: {failed_downloads}"")\n    \n    end_time = time.time()\n    total_time = end_time - start_time\n    print(f""Download completed in {total_time:.2f} seconds"")\n\nif __name__ == ""__main__"":\n    args = tyro.cli(DownloadVideos)\n\n    output_dir = os.path.join(args.output_dir, os.path.basename(args.index_file_path).split('.')[0])\n    os.makedirs(output_dir, exist_ok=True)\n\n    num_workers = os.cpu_count()\n\n    # print all args\n    print(f""Index file path: {args.index_file_path}"")\n    print(f""Output directory: {output_dir}"")\n    print(f""Number of workers: {num_workers}"")\n\n    # get cpu count\n    download_dataset(args.index_file_path, output_dir, num_workers)\n",python,tab
+9,46501,"data_download/download_scripts/download_videos.py",4131,0,"",python,selection_command
+10,49029,"data_download/download_scripts/download_videos.py",4133,0,"",python,selection_mouse
+11,50432,"data_download/download_scripts/download_videos.py",1625,0,"",python,selection_mouse
+12,50797,"data_download/download_scripts/download_videos.py",1671,0,"",python,selection_mouse
+13,50953,"data_download/download_scripts/download_videos.py",1664,10,"output_dir",python,selection_mouse
+14,57528,"data_download/download_scripts/download_videos.py",2692,0,"",python,selection_mouse
+15,58027,"data_download/download_scripts/download_videos.py",2703,0,"",python,selection_mouse
+16,59001,"data_download/download_scripts/download_videos.py",2929,0,"",python,selection_mouse
+17,62123,"data_download/download_scripts/download_videos.py",3388,0,"",python,selection_mouse
+18,64346,"data_download/download_scripts/download_videos.py",3119,0,"",python,selection_mouse
+19,64355,"data_download/download_scripts/download_videos.py",3118,0,"",python,selection_command
+20,68771,"data_download/download_scripts/download_videos.py",3117,0,"",python,selection_command
+21,75051,"data_download/download_scripts/download_videos.py",458,0,"",python,selection_mouse
+22,75174,"data_download/download_scripts/download_videos.py",456,4,"args",python,selection_mouse
+23,78234,"data_download/download_scripts/download_videos.py",2098,0,"",python,selection_mouse
+24,80870,"data_download/download_scripts/download_videos.py",2710,0,"",python,selection_mouse
+25,81326,"data_download/download_scripts/download_videos.py",2720,0,"",python,selection_mouse
+26,81703,"data_download/download_scripts/download_videos.py",2718,20,"download_single_file",python,selection_mouse
+27,83165,"data_download/download_scripts/download_videos.py",2721,0,"",python,selection_mouse
+28,84631,"data_download/download_scripts/download_videos.py",1000,0,"",python,selection_mouse
+29,85027,"data_download/download_scripts/download_videos.py",1099,0,"",python,selection_mouse
+30,85437,"data_download/download_scripts/download_videos.py",1239,0,"",python,selection_mouse
+31,86711,"data_download/download_scripts/download_videos.py",1302,0,"",python,selection_mouse