Add b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-0162b81a-7b28-4ef5-b0e9-c56355c44e161765287461281-2025_12_09-14.41.45.741/source.csv

@@ -0,0 +1,28 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,4,"scripts/run-qwen3-0.6B-base-sft.sh",0,0,"#!/bin/bash\n\n# for rerun the task\npkill -9 sglang\nsleep 3\nray stop --force\npkill -9 ray\npkill -9 python\nsleep 3\npkill -9 ray\npkill -9 python\n\nset -ex\n\n# will prevent ray from buffering stdout/stderr\nexport PYTHONBUFFERED=16\n\nNVLINK_COUNT=$(nvidia-smi topo -m 2>/dev/null | grep -o 'NV[0-9][0-9]*' | wc -l)\nif [ ""$NVLINK_COUNT"" -gt 0 ]; then\n    HAS_NVLINK=1\nelse\n    HAS_NVLINK=0\nfi\necho ""HAS_NVLINK: $HAS_NVLINK (detected $NVLINK_COUNT NVLink references)""\n\nSCRIPT_DIR=""$(cd -- ""$(dirname -- ""${BASH_SOURCE[0]}"")"" &>/dev/null && pwd)""\necho ""SCRIPT_DIR: $SCRIPT_DIR""\nsource ""${SCRIPT_DIR}/models/qwen3-0.6B.sh""\n\nCKPT_ARGS=(\n   --hf-checkpoint /fast/project/HFMI_SynergyUnit/tab_model/huggingface/Qwen3-0.6B/\n   --ref-load /fast/project/HFMI_SynergyUnit/tab_model/huggingface/Qwen3-0.6B_torch_dist\n   --load /fast/project/HFMI_SynergyUnit/tab_model/huggingface/Qwen3-0.6B_miles/\n   --save /fast/project/HFMI_SynergyUnit/tab_model/huggingface/Qwen3-0.6B_miles/\n   --save-interval 1000\n)\n\nSFT_ARGS=(\n   --rollout-function-path miles.rollout.sft_rollout.generate_rollout\n   --prompt-data /fast/project/HFMI_SynergyUnit/tab_model/huggingface/openhermes2_5.parquet\n   --apply-chat-template\n   --input-key messages\n   --rollout-shuffle\n   --num-epoch 3\n   --rollout-batch-size 1\n   --global-batch-size 1\n\n   --loss-type sft_loss\n   --calculate-per-token-loss\n   --disable-compute-advantages-and-returns\n   --debug-train-only\n)\n\nPERF_ARGS=(\n   --tensor-model-parallel-size 1\n   --sequence-parallel\n   --pipeline-model-parallel-size 1\n   --context-parallel-size 1\n   --expert-model-parallel-size 1\n   --expert-tensor-parallel-size 1\n\n   --recompute-granularity full\n   --recompute-method uniform\n   --recompute-num-layers 1\n\n   # --micro-batch-size 1\n   --use-dynamic-batch-size\n   --max-tokens-per-gpu 9216\n)\n\nOPTIMIZER_ARGS=(\n   --optimizer adam\n   --lr 1e-5\n   --lr-decay-style cosine\n   --min-lr 1e-6\n   --lr-warmup-fraction 0.1\n   --weight-decay 0.1\n   --adam-beta1 0.9\n   --adam-beta2 0.95\n)\n\nWANDB_ARGS=(\n   # --use-wandb\n   # --wandb-project miles-dev\n   # --wandb-group qwen3-4B-base-sft\n   # --wandb-key ${WANDB_KEY}\n)\n\nMISC_ARGS=(\n   # default dropout in megatron is 0.1\n   --attention-dropout 0.0\n   --hidden-dropout 0.0\n   # should be good for model performance\n   --accumulate-allreduce-grads-in-fp32\n   --attention-softmax-in-fp32\n   # need to comment this when using model with MLA\n   --attention-backend flash\n)\n\n# launch the master node of ray in container\nexport MASTER_ADDR=${MASTER_ADDR:-""10.86.2.50""}\nexport no_proxy=""10.86.2.50,${MASTER_ADDR}""\nray start --head --node-ip-address ${MASTER_ADDR} --num-gpus 2 --disable-usage-stats --dashboard-host=0.0.0.0 --dashboard-port=8265\n\n\n# Build the runtime environment JSON with proper variable substitution\nRUNTIME_ENV_JSON=""{\n  \""env_vars\"": {\n    \""PYTHONPATH\"": \""/fast/project/HFMI_SynergyUnit/tab_model/Megatron-LM/\"",\n    \""CUDA_DEVICE_MAX_CONNECTIONS\"": \""1\"",\n    \""NCCL_NVLS_ENABLE\"": \""${HAS_NVLINK}\"",\n    \""PYTORCH_CUDA_ALLOC_CONF\"": \""expandable_segments:True\""\n  }\n}""\n\nray job submit --address=""http://10.86.2.50:8265"" \\n   --runtime-env-json=""${RUNTIME_ENV_JSON}"" \\n   -- python3 train_async.py \\n   --actor-num-nodes 1 \\n   --actor-num-gpus-per-node 2 \\n   ${MODEL_ARGS[@]} \\n   ${CKPT_ARGS[@]} \\n   ${SFT_ARGS[@]} \\n   ${OPTIMIZER_ARGS[@]} \\n   ${WANDB_ARGS[@]} \\n   ${PERF_ARGS[@]} \\n   ${EVAL_ARGS[@]} \\n   ${MISC_ARGS[@]}\n",shellscript,tab
+2,28880,"scripts/run-qwen3-0.6B-base-sft.sh",1553,0,"",shellscript,selection_mouse
+3,32556,"scripts/run-qwen3-0.6B-base-sft.sh",1552,0,"",shellscript,selection_command
+4,54151,"TERMINAL",0,0,"",,terminal_focus
+5,109805,"scripts/run-qwen3-0.6B-base-sft.sh",1623,0,"",shellscript,selection_mouse
+6,144315,"TERMINAL",0,0,"queue",,terminal_command
+7,145344,"TERMINAL",0,0,"]633;C[?1049h(B[?7hEvery 1.0s: squeue --mehai-login1.haicore.berlin: Tue Dec  9 14:43:59 2025JOBIDUSER PARTITION NODES  CPUS STSUBMIT_TIMESTART_TIME     TIME TIME_LIMIT NODELIST(REASON)",,terminal_output
+8,146052,"TERMINAL",0,0,"4:00\t ",,terminal_output
+9,147444,"TERMINAL",0,0,"1\t ",,terminal_output
+10,148382,"TERMINAL",0,0,"2\t ",,terminal_output
+11,148682,"TERMINAL",0,0,"3\t ",,terminal_output
+12,149104,"TERMINAL",0,0,"4\t 5\t 6\t 7\t Every 1.0s: squeue --mehai-login1.haicore.berlin: Tue Dec  9 14:44:07 2025JOBIDUSER PARTITION NODES  CPUS STSUBMIT_TIMESTART_TIME     TIME TIME_LIMIT NODELIST(REASON)8\t ",,terminal_output
+13,149394,"TERMINAL",0,0,"9\t 10\t 1\t 2\t ",,terminal_output
+14,149474,"TERMINAL",0,0,"3\t ",,terminal_output
+15,149922,"TERMINAL",0,0,"4\t ",,terminal_output
+16,165760,"TERMINAL",0,0,"5\t ",,terminal_output
+17,165891,"TERMINAL",0,0,"6\t 7\t ",,terminal_output
+18,166638,"TERMINAL",0,0,"8\t ",,terminal_output
+19,166802,"TERMINAL",0,0,"9\t 20\t ",,terminal_output
+20,166978,"TERMINAL",0,0,"1\t ",,terminal_output
+21,176592,"TERMINAL",0,0,"2\t 3\t 4\t 5\t 7\t 8\t 9\t 30\t [?1049l\r[?1l>]0;mihir.mahajan@hai-login1:~/Projects/miles",,terminal_output
+22,372641,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"2:41:45 PM [info] Activating crowd-code\n2:41:45 PM [info] Recording started\n2:41:45 PM [info] Initializing git provider using file system watchers...\n2:41:47 PM [info] Retrying git provider initialization...\n2:42:21 PM [info] Git repository found\n2:42:21 PM [info] Git provider initialized successfully\n2:42:21 PM [info] Initial git state: [object Object]\n2:42:29 PM [info] Git repository found\n2:42:29 PM [info] Git provider initialized successfully\n2:42:29 PM [info] Initial git state: [object Object]\n",Log,tab
+23,445141,"scripts/run-qwen3-0.6B-base-sft.sh",0,0,"",shellscript,tab
+24,453380,"TERMINAL",0,0,"source /home/mihir.mahajan/Projects/miles/.venv/bin/activate",,terminal_command
+25,454711,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login1:~/Projects/miles",,terminal_output
+26,558063,"scripts/run-qwen3-0.6B-base-sft.sh",1524,0,"",shellscript,selection_mouse
+27,562108,"scripts/run-qwen3-0.6B-base-sft.sh",1523,0,"",shellscript,selection_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-09e95c4f-d571-45a4-a506-a80e6e398a3f1761597703208-2025_10_27-21.42.06.986/source.csv

@@ -0,0 +1,3 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,3,"/fast/home/mihir.mahajan/Projects/jasmine/jasmine/baselines/diffusion/train_tokenizer_mae.py",0,0,"import os\n\nos.environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ""0.98"")\n\nfrom dataclasses import dataclass, field\nfrom typing import cast, Optional\n\nimport einops\nimport itertools\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax.checkpoint as ocp\nimport numpy as np\nimport dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\nimport grain\nimport flax.nnx as nnx\n\nfrom jasmine.models.tokenizer import TokenizerMAE\nfrom jasmine.utils.dataloader import get_dataloader\nfrom jasmine.utils.train_utils import (\n    get_lr_schedule,\n    count_parameters_by_component,\n    print_mem_stats,\n    print_compiled_memory_stats,\n    print_compiled_cost_analysis,\n)\n\n\n@dataclass\nclass Args:\n    # Experiment\n    num_steps: int = 300_000\n    seed: int = 0\n    seq_len: int = 16\n    image_channels: int = 3\n    image_height: int = 64\n    image_width: int = 64\n    data_dir: str = """"\n    save_ckpt: bool = False\n    restore_ckpt: bool = False\n    # Optimization\n    batch_size: int = 48\n    init_lr: float = 0.0\n    max_lr: float = 3e-4\n    decay_end: float = 0.0\n    wsd_decay_steps: int = (\n        30_000  # NOTE: wsd_decay_steps will only be used when using a wsd-schedule\n    )\n    lr_schedule: str = ""wsd""  # supported options: wsd, cos\n    warmup_steps: int = 10000\n    # Tokenizer\n    model_dim: int = 512\n    ffn_dim: int = 2048\n    latent_dim: int = 32\n    num_latents: int = 1024\n    patch_size: int = 16\n    num_blocks: int = 4\n    num_heads: int = 8\n    dropout: float = 0.0\n    max_mask_ratio: float = 0.9\n    param_dtype = jnp.float32\n    dtype = jnp.bfloat16\n    use_flash_attention: bool = True\n    # Logging\n    log: bool = True\n    entity: str = """"\n    project: str = """"\n    name: str = ""train_tokenizer_mae""\n    tags: list[str] = field(default_factory=lambda: [""tokenizer"", ""mae""])\n    log_interval: int = 50\n    log_image_interval: int = 1000\n    ckpt_dir: str = """"\n    log_checkpoint_interval: int = 1000\n    log_checkpoint_keep_period: int = 20_000\n    log_gradients: bool = False\n    val_data_dir: str = """"\n    val_interval: int = 20_000\n    val_steps: int = 50\n    wandb_id: str = """"\n\n\ndef build_model(args: Args, rng: jax.Array) -> tuple[TokenizerMAE, jax.Array]:\n    rng, _rng = jax.random.split(rng)\n    rngs = nnx.Rngs(_rng)\n    tokenizer = TokenizerMAE(\n        in_dim=args.image_channels,\n        model_dim=args.model_dim,\n        ffn_dim=args.ffn_dim,\n        latent_dim=args.latent_dim,\n        num_latents=args.num_latents,\n        patch_size=args.patch_size,\n        num_blocks=args.num_blocks,\n        num_heads=args.num_heads,\n        dropout=args.dropout,\n        max_mask_ratio=args.max_mask_ratio,\n        param_dtype=args.param_dtype,\n        dtype=args.dtype,\n        use_flash_attention=args.use_flash_attention,\n        rngs=rngs,\n    )\n    return tokenizer, rng\n\n\ndef build_optimizer(model: TokenizerMAE, args: Args) -> nnx.ModelAndOptimizer:\n    lr_schedule = get_lr_schedule(\n        args.lr_schedule,\n        args.init_lr,\n        args.max_lr,\n        args.decay_end,\n        args.num_steps,\n        args.warmup_steps,\n        args.wsd_decay_steps,\n    )\n    tx = optax.adamw(\n        learning_rate=lr_schedule,\n        b1=0.9,\n        b2=0.9,\n        weight_decay=1e-4,\n        mu_dtype=args.param_dtype,  # moments in full precision\n    )\n    optimizer = nnx.ModelAndOptimizer(model, tx)\n    return optimizer\n\n\ndef build_mesh_and_sharding(\n    num_devices: int,\n) -> tuple[Mesh, NamedSharding, NamedSharding]:\n    device_mesh_arr = create_device_mesh((num_devices,))\n    mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n    replicated_sharding = NamedSharding(mesh, PartitionSpec())\n    videos_sharding = NamedSharding(mesh, PartitionSpec(""data"", None, None, None, None))\n    return mesh, replicated_sharding, videos_sharding\n\n\ndef shard_optimizer_states(\n    optimizer: nnx.ModelAndOptimizer, replicated_sharding: NamedSharding\n) -> None:\n    model_state = nnx.state(optimizer.model)\n    model_sharded_state = jax.lax.with_sharding_constraint(\n        model_state, replicated_sharding\n    )\n    nnx.update(optimizer.model, model_sharded_state)\n    optimizer_state = nnx.state(optimizer, nnx.optimizer.OptState)\n    optimizer_sharded_state = jax.lax.with_sharding_constraint(\n        optimizer_state, replicated_sharding\n    )\n    nnx.update(optimizer, optimizer_sharded_state)\n\n\ndef build_dataloader(args: Args, data_dir: str) -> grain.DataLoaderIterator:\n    image_shape = (args.image_height, args.image_width, args.image_channels)\n    array_record_files = [\n        os.path.join(data_dir, x)\n        for x in os.listdir(data_dir)\n        if x.endswith("".array_record"")\n    ]\n    grain_dataloader = get_dataloader(\n        array_record_files,\n        args.seq_len,\n        # NOTE: We deliberately pass the global batch size\n        # The dataloader shards the dataset across all processes\n        args.batch_size,\n        *image_shape,\n        num_workers=8,\n        prefetch_buffer_size=1,\n        seed=args.seed,\n    )\n    initial_state = grain_dataloader._create_initial_state()\n    grain_iterator = grain.DataLoaderIterator(grain_dataloader, initial_state)\n    return grain_iterator\n\n\ndef build_checkpoint_manager(args: Args) -> Optional[ocp.CheckpointManager]:\n    if args.restore_ckpt or args.save_ckpt:\n        handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n        handler_registry.add(\n            ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n        )\n        handler_registry.add(\n            ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n        )\n        handler_registry.add(\n            ""train_dataloader_state"",\n            grain.checkpoint.CheckpointSave,\n            cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n        )\n        handler_registry.add(\n            ""train_dataloader_state"",\n            grain.checkpoint.CheckpointRestore,\n            cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n        )\n        if args.val_data_dir:\n            handler_registry.add(\n                ""val_dataloader_state"",\n                grain.checkpoint.CheckpointSave,\n                cast(\n                    ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler\n                ),\n            )\n            handler_registry.add(\n                ""val_dataloader_state"",\n                grain.checkpoint.CheckpointRestore,\n                cast(\n                    ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler\n                ),\n            )\n        checkpoint_options = ocp.CheckpointManagerOptions(\n            save_interval_steps=args.log_checkpoint_interval,\n            max_to_keep=3,\n            keep_period=args.log_checkpoint_keep_period,\n            step_format_fixed_length=6,\n            cleanup_tmp_directories=True,\n        )\n        checkpoint_manager = ocp.CheckpointManager(\n            args.ckpt_dir,\n            options=checkpoint_options,\n            handler_registry=handler_registry,\n        )\n        return checkpoint_manager\n    else:\n        return None\n\n\ndef restore_checkpoint_if_needed(\n    args: Args,\n    checkpoint_manager: Optional[ocp.CheckpointManager],\n    optimizer: nnx.ModelAndOptimizer,\n    train_iterator: grain.DataLoaderIterator,\n    val_iterator: Optional[grain.DataLoaderIterator],\n    restore_step: Optional[int] = None,\n) -> tuple[\n    int, nnx.ModelAndOptimizer, grain.DataLoaderIterator, grain.DataLoaderIterator\n]:\n    step = 0\n    if checkpoint_manager and restore_step is None:\n        restore_step = checkpoint_manager.latest_step()\n    if args.restore_ckpt:\n        assert checkpoint_manager is not None\n        abstract_optimizer = nnx.eval_shape(lambda: optimizer)\n        abstract_optimizer_state = nnx.state(abstract_optimizer)\n        if val_iterator:\n            restore_args = ocp.args.Composite(\n                model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n                train_dataloader_state=grain.checkpoint.CheckpointRestore(train_iterator),  # type: ignore\n                val_dataloader_state=grain.checkpoint.CheckpointRestore(val_iterator),  # type: ignore\n            )\n        else:\n            restore_args = ocp.args.Composite(\n                model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n                train_dataloader_state=grain.checkpoint.CheckpointRestore(train_iterator),  # type: ignore\n            )\n        restored = checkpoint_manager.restore(restore_step, args=restore_args)\n        restored_optimizer_state = restored[""model_state""]\n        nnx.update(optimizer, restored_optimizer_state)\n        train_iterator = restored[""train_dataloader_state""]\n        if val_iterator:\n            val_iterator = restored[""val_dataloader_state""]\n        step = restore_step or 0\n        print(f""Restored dataloader and model state from step {step}"")\n    return step, optimizer, train_iterator, val_iterator\n\n\ndef main(args: Args) -> None:\n    jax.distributed.initialize()\n    num_devices = jax.device_count()\n    if num_devices == 0:\n        raise ValueError(""No JAX devices found."")\n    print(f""Running on {num_devices} devices."")\n\n    if args.batch_size % num_devices != 0:\n        raise ValueError(\n            f""Global batch size {args.batch_size} must be divisible by ""\n            f""number of devices {num_devices}.""\n        )\n\n    rng = jax.random.key(args.seed)\n\n    # --- Initialize model ---\n    tokenizer, rng = build_model(args, rng)\n\n    _, params, _ = nnx.split(tokenizer, nnx.Param, ...)\n    param_counts = count_parameters_by_component(params)\n\n    if args.log and jax.process_index() == 0:\n        wandb_init_kwargs = {\n            ""entity"": args.entity,\n            ""project"": args.project,\n            ""name"": args.name,\n            ""tags"": args.tags,\n            ""group"": ""debug"",\n            ""config"": args,\n        }\n\n        if args.wandb_id:\n            wandb_init_kwargs.update(\n                {\n                    ""id"": args.wandb_id,\n                    ""resume"": ""allow"",\n                }\n            )\n        wandb.init(**wandb_init_kwargs)\n\n        wandb.config.update({""model_param_count"": param_counts})\n\n    print(""Parameter counts:"")\n    print(param_counts)\n\n    # --- Initialize optimizer ---\n    optimizer = build_optimizer(tokenizer, args)\n    del tokenizer\n\n    # FIXME: switch to create_hybrid_device_mesh for runs spanning multiple nodes\n    _, replicated_sharding, videos_sharding = build_mesh_and_sharding(num_devices)\n\n    shard_optimizer_states(optimizer, replicated_sharding)\n\n    # --- Initialize checkpoint manager ---\n    checkpoint_manager = build_checkpoint_manager(args)\n\n    # --- Create DataLoaderIterator from dataloader ---\n    train_iterator = build_dataloader(args, args.data_dir)\n    val_iterator = None\n    if args.val_data_dir:\n        val_iterator = build_dataloader(args, args.val_data_dir)\n\n    # --- Restore checkpoint ---\n    step, optimizer, train_iterator, val_iterator = restore_checkpoint_if_needed(\n        args, checkpoint_manager, optimizer, train_iterator, val_iterator\n    )\n\n    # --- Define loss and train step (close over args) ---\n    def tokenizer_loss_fn(\n        model: TokenizerMAE, inputs: dict, training: bool = False\n    ) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n        gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n        inputs[""videos""] = gt.astype(args.dtype)\n        outputs = model(inputs, training=training)\n        outputs[""recon""] = outputs[""recon""].astype(jnp.float32)\n        mse = jnp.square(gt - outputs[""recon""]).mean()\n\n        gt_clipped = gt.clip(0, 1).reshape(-1, *gt.shape[2:])\n        recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n        psnr = jnp.asarray(pix.psnr(gt_clipped, recon)).mean()\n        ssim = jnp.asarray(pix.ssim(gt_clipped, recon)).mean()\n\n        metrics = dict(\n            mse=mse,\n            psnr=psnr,\n            ssim=ssim,\n            loss=mse,\n        )\n\n        return mse, (outputs[""recon""], metrics)\n\n    @nnx.jit(donate_argnums=0)\n    def train_step(\n        optimizer: nnx.ModelAndOptimizer, inputs: dict\n    ) -> tuple[jax.Array, jax.Array, dict]:\n        def loss_fn(\n            model: TokenizerMAE,\n        ) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n            model.train()\n            return tokenizer_loss_fn(model, inputs, training=True)\n\n        (loss, (recon, metrics)), grads = nnx.value_and_grad(loss_fn, has_aux=True)(\n            optimizer.model\n        )\n        optimizer.update(grads)\n        if args.log_gradients:\n            metrics[""encoder_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""encoder""]\n            )\n            metrics[""decoder_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""decoder""]\n            )\n        return loss, recon, metrics\n\n    @nnx.jit\n    def val_step(\n        tokenizer: TokenizerMAE, inputs: dict\n    ) -> tuple[jax.Array, jax.Array, dict]:\n        tokenizer.eval()\n        (loss, (recon, metrics)) = tokenizer_loss_fn(tokenizer, inputs, training=False)\n        return loss, recon, metrics\n\n    def calculate_validation_metrics(val_dataloader, tokenizer, rng):\n        step = 0\n        loss_per_step = []\n        metrics_per_step = []\n        batch = None\n        recon = None\n        for batch in val_dataloader:\n            rng, _rng_mask = jax.random.split(rng, 2)\n            batch[""rng""] = _rng_mask\n            loss, recon, metrics = val_step(tokenizer, batch)\n            loss_per_step.append(loss)\n            metrics_per_step.append(metrics)\n            step += 1\n            if step > args.val_steps:\n                break\n\n        if step < args.val_steps:\n            print(\n                f""Warning: Your validation dataset is too small to make val_steps many steps. Made {step} steps, expected {args.val_steps}""\n            )\n\n        val_loss = np.mean(loss_per_step)\n        val_metrics = {\n            f""val_{key}"": np.mean([float(m[key]) for m in metrics_per_step])\n            for key in metrics_per_step[0].keys()\n        }\n        val_metrics[""val_loss""] = val_loss\n        return val_metrics, batch, recon\n\n    # --- TRAIN LOOP ---\n    dataloader_train = (\n        {\n            ""videos"": jax.make_array_from_process_local_data(\n                videos_sharding, elem[""videos""]\n            ),\n        }\n        for elem in train_iterator\n    )\n    dataloader_val = None\n    if val_iterator:\n        dataloader_val = (\n            {\n                ""videos"": jax.make_array_from_process_local_data(\n                    videos_sharding, elem[""videos""]\n                ),\n            }\n            for elem in val_iterator\n        )\n    if jax.process_index() == 0:\n        rng, _rng = jax.random.split(rng)\n        first_batch = next(dataloader_train)\n        first_batch[""rng""] = _rng\n        compiled = train_step.lower(optimizer, first_batch).compile()\n        print_compiled_memory_stats(compiled.memory_analysis())\n        print_compiled_cost_analysis(compiled.cost_analysis())\n        # Do not skip the first batch during training\n        dataloader_train = itertools.chain([first_batch], dataloader_train)\n    print(f""Starting training from step {step}..."")\n    first_step = step\n    while step < args.num_steps:\n        for batch in dataloader_train:\n            # --- Train step ---\n            rng, _rng = jax.random.split(rng)\n            batch[""rng""] = _rng\n            loss, recon, metrics = train_step(optimizer, batch)\n            if step == first_step:\n                print_mem_stats(""After params initialized"")\n            step += 1\n\n            # --- Validation loss ---\n            val_results = {}\n            if dataloader_val and step % args.val_interval == 0:\n                print(""Calculating validation metrics..."")\n                rng, _rng_mask_val = jax.random.split(rng, 2)\n                val_metrics, val_gt_batch, val_recon = calculate_validation_metrics(\n                    dataloader_val, optimizer.model, _rng_mask_val\n                )\n                print(f""Step {step}, validation loss: {val_metrics['val_loss']}"")\n                val_results = {\n                    ""metrics"": val_metrics,\n                    ""gt_batch"": val_gt_batch,\n                    ""recon"": val_recon,\n                }\n\n            # --- Logging ---\n            if args.log:\n                if step % args.log_interval == 0 and jax.process_index() == 0:\n                    log_dict = {""loss"": loss, ""step"": step, **metrics}\n                    if val_results:\n                        log_dict.update(val_results[""metrics""])\n                    wandb.log(log_dict)\n                if step % args.log_image_interval == 0:\n                    gt_seq = batch[""videos""][0].astype(jnp.float32) / 255.0\n                    recon_seq = recon[0].clip(0, 1)\n                    comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n                    comparison_seq = einops.rearrange(\n                        comparison_seq * 255, ""t h w c -> h (t w) c""\n                    )\n                    if val_results and step % args.val_interval == 0:\n                        val_results[""gt_seq_val""] = (\n                            val_results[""gt_batch""][""videos""][0].astype(jnp.float32)\n                            / 255.0\n                        )\n                        val_results[""recon_seq_val""] = val_results[""recon""][0].clip(\n                            0, 1\n                        )\n                        val_results[""val_comparison_seq""] = jnp.concatenate(\n                            (val_results[""gt_seq_val""], val_results[""recon_seq_val""]),\n                            axis=1,\n                        )\n                        val_results[""val_comparison_seq""] = einops.rearrange(\n                            val_results[""val_comparison_seq""] * 255,\n                            ""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                        if val_results and step % args.val_interval == 0:\n                            log_images.update(\n                                dict(\n                                    val_image=wandb.Image(\n                                        np.asarray(val_results[""gt_seq_val""][0])\n                                    ),\n                                    val_recon=wandb.Image(\n                                        np.asarray(val_results[""recon_seq_val""][0])\n                                    ),\n                                    val_true_vs_recon=wandb.Image(\n                                        np.asarray(\n                                            val_results[""val_comparison_seq""].astype(\n                                                np.uint8\n                                            )\n                                        )\n                                    ),\n                                )\n                            )\n                        wandb.log(log_images)\n            # --- Checkpointing ---\n            if args.save_ckpt and step % args.log_checkpoint_interval == 0:\n                assert checkpoint_manager is not None\n                optimizer_state = nnx.state(optimizer)\n                if val_iterator:\n                    ckpt_manager_args = ocp.args.Composite(\n                        model_state=ocp.args.PyTreeSave(optimizer_state),  # type: ignore\n                        train_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            train_iterator  # type: ignore\n                        ),\n                        val_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            val_iterator  # type: ignore\n                        ),\n                    )\n                else:\n                    ckpt_manager_args = ocp.args.Composite(\n                        model_state=ocp.args.PyTreeSave(optimizer_state),  # type: ignore\n                        train_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            train_iterator  # type: ignore\n                        ),\n                    )\n                checkpoint_manager.save(step, args=ckpt_manager_args)\n                print(f""Saved checkpoint at step {step}"")\n            if step >= args.num_steps:\n                break\n\n    if checkpoint_manager:\n        checkpoint_manager.close()\n\n\nif __name__ == ""__main__"":\n    args = tyro.cli(Args)\n    main(args)\n",python,tab
+2,924197,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"9:42:06 PM [info] Activating crowd-code\n9:42:06 PM [info] Recording started\n9:42:06 PM [info] Initializing git provider using file system watchers...\n9:42:07 PM [info] Git repository found\n9:42:07 PM [info] Git provider initialized successfully\n9:42:07 PM [info] Initial git state: [object Object]\n",Log,tab

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-0f1ad218-d2d6-4ace-9150-2ff8d2109f6d1762606110433-2025_11_08-13.49.11.795/source.csv

@@ -0,0 +1,27 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,9721,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"1:49:11 PM [info] Activating crowd-code\n1:49:11 PM [info] Recording started\n1:49:11 PM [info] Initializing git provider using file system watchers...\n1:49:13 PM [info] Retrying git provider initialization...\n1:49:15 PM [info] Git repository found\n1:49:15 PM [info] Git provider initialized successfully\n1:49:15 PM [info] Initial git state: [object Object]\n1:49:15 PM [info] Git repository found\n1:49:15 PM [info] Git provider initialized successfully\n1:49:15 PM [info] Initial git state: [object Object]\n",Log,tab
+3,56495,"TERMINAL",0,0,"",,terminal_focus
+4,96627,"TERMINAL",0,0,"",,terminal_focus
+5,167415,"TERMINAL",0,0,"git clone https://github.com/p-doom/slurm",,terminal_command
+6,167477,"TERMINAL",0,0,"]633;CCloning into 'slurm'...\r\n",,terminal_output
+7,172593,"TERMINAL",0,0,"^[^[",,terminal_output
+8,175222,"TERMINAL",0,0,"remote: Repository not found.\r\nfatal: Authentication failed for 'https://github.com/p-doom/slurm/'\r\n]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot",,terminal_output
+9,192444,"TERMINAL",0,0,"git clone git@github.com:p-doom/slurm.git",,terminal_command
+10,192494,"TERMINAL",0,0,"]633;CCloning into 'slurm'...\r\n",,terminal_output
+11,194019,"TERMINAL",0,0,"remote: Enumerating objects: 5247, done.\r\nremote: Counting objects:   0% (1/443)\rremote: Counting objects:   1% (5/443)\rremote: Counting objects:   2% (9/443)\rremote: Counting objects:   3% (14/443)\rremote: Counting objects:   4% (18/443)\rremote: Counting objects:   5% (23/443)\rremote: Counting objects:   6% (27/443)\rremote: Counting objects:   7% (32/443)\rremote: Counting objects:   8% (36/443)\rremote: Counting objects:   9% (40/443)\rremote: Counting objects:  10% (45/443)\rremote: Counting objects:  11% (49/443)\rremote: Counting objects:  12% (54/443)\rremote: Counting objects:  13% (58/443)\rremote: Counting objects:  14% (63/443)\rremote: Counting objects:  15% (67/443)\rremote: Counting objects:  16% (71/443)\rremote: Counting objects:  17% (76/443)\rremote: Counting objects:  18% (80/443)\rremote: Counting objects:  19% (85/443)\rremote: Counting objects:  20% (89/443)\rremote: Counting objects:  21% (94/443)\rremote: Counting objects:  22% (98/443)\rremote: Counting objects:  23% (102/443)\rremote: Counting objects:  24% (107/443)\rremote: Counting objects:  25% (111/443)\rremote: Counting objects:  26% (116/443)\rremote: Counting objects:  27% (120/443)\rremote: Counting objects:  28% (125/443)\rremote: Counting objects:  29% (129/443)\rremote: Counting objects:  30% (133/443)\rremote: Counting objects:  31% (138/443)\rremote: Counting objects:  32% (142/443)\rremote: Counting objects:  33% (147/443)\rremote: Counting objects:  34% (151/443)\rremote: Counting objects:  35% (156/443)\rremote: Counting objects:  36% (160/443)\rremote: Counting objects:  37% (164/443)\rremote: Counting objects:  38% (169/443)\rremote: Counting objects:  39% (173/443)\r",,terminal_output
+12,194213,"TERMINAL",0,0,"remote: Counting objects:  40% (178/443)\rremote: Counting objects:  41% (182/443)\rremote: Counting objects:  42% (187/443)\rremote: Counting objects:  43% (191/443)\rremote: Counting objects:  44% (195/443)\rremote: Counting objects:  45% (200/443)\rremote: Counting objects:  46% (204/443)\rremote: Counting objects:  47% (209/443)\rremote: Counting objects:  48% (213/443)\rremote: Counting objects:  49% (218/443)\rremote: Counting objects:  50% (222/443)\rremote: Counting objects:  51% (226/443)\rremote: Counting objects:  52% (231/443)\rremote: Counting objects:  53% (235/443)\rremote: Counting objects:  54% (240/443)\rremote: Counting objects:  55% (244/443)\rremote: Counting objects:  56% (249/443)\rremote: Counting objects:  57% (253/443)\rremote: Counting objects:  58% (257/443)\rremote: Counting objects:  59% (262/443)\rremote: Counting objects:  60% (266/443)\rremote: Counting objects:  61% (271/443)\rremote: Counting objects:  62% (275/443)\rremote: Counting objects:  63% (280/443)\rremote: Counting objects:  64% (284/443)\rremote: Counting objects:  65% (288/443)\rremote: Counting objects:  66% (293/443)\rremote: Counting objects:  67% (297/443)\rremote: Counting objects:  68% (302/443)\rremote: Counting objects:  69% (306/443)\rremote: Counting objects:  70% (311/443)\rremote: Counting objects:  71% (315/443)\rremote: Counting objects:  72% (319/443)\rremote: Counting objects:  73% (324/443)\rremote: Counting objects:  74% (328/443)\rremote: Counting objects:  75% (333/443)\rremote: Counting objects:  76% (337/443)\rremote: Counting objects:  77% (342/443)\rremote: Counting objects:  78% (346/443)\rremote: Counting objects:  79% (350/443)\rremote: Counting objects:  80% (355/443)\rremote: Counting objects:  81% (359/443)\rremote: Counting objects:  82% (364/443)\rremote: Counting objects:  83% (368/443)\rremote: Counting objects:  84% (373/443)\rremote: Counting objects:  85% (377/443)\rremote: Counting objects:  86% (381/443)\rremote: Counting objects:  87% (386/443)\rremote: Counting objects:  88% (390/443)\rremote: Counting objects:  89% (395/443)\rremote: Counting objects:  90% (399/443)\rremote: Counting objects:  91% (404/443)\rremote: Counting objects:  92% (408/443)\rremote: Counting objects:  93% (412/443)\rremote: Counting objects:  94% (417/443)\rremote: Counting objects:  95% (421/443)\rremote: Counting objects:  96% (426/443)\rremote: Counting objects:  97% (430/443)\rremote: Counting objects:  98% (435/443)\rremote: Counting objects:  99% (439/443)\rremote: Counting objects: 100% (443/443)\rremote: Counting objects: 100% (443/443), done.\r\nremote: Compressing objects:   0% (1/310)\rremote: Compressing objects:   1% (4/310)\rremote: Compressing objects:   2% (7/310)\rremote: Compressing objects:   3% (10/310)\rremote: Compressing objects:   4% (13/310)\rremote: Compressing objects:   5% (16/310)\rremote: Compressing objects:   6% (19/310)\rremote: Compressing objects:   7% (22/310)\rremote: Compressing objects:   8% (25/310)\rremote: Compressing objects:   9% (28/310)\rremote: Compressing objects:  10% (31/310)\rremote: Compressing objects:  11% (35/310)\rremote: Compressing objects:  12% (38/310)\rremote: Compressing objects:  13% (41/310)\rremote: Compressing objects:  14% (44/310)\rremote: Compressing objects:  15% (47/310)\rremote: Compressing objects:  16% (50/310)\rremote: Compressing objects:  17% (53/310)\rremote: Compressing objects:  18% (56/310)\rremote: Compressing objects:  19% (59/310)\rremote: Compressing objects:  20% (62/310)\rremote: Compressing objects:  21% (66/310)\rremote: Compressing objects:  22% (69/310)\rremote: Compressing objects:  23% (72/310)\rremote: Compressing objects:  24% (75/310)\rremote: Compressing objects:  25% (78/310)\rremote: Compressing objects:  26% (81/310)\rremote: Compressing objects:  27% (84/310)\rremote: Compressing objects:  28% (87/310)\rremote: Compressing objects:  29% (90/310)\rremote: Compressing objects:  30% (93/310)\rremote: Compressing objects:  31% (97/310)\rremote: Compressing objects:  32% (100/310)\rremote: Compressing objects:  33% (103/310)\rremote: Compressing objects:  34% (106/310)\rremote: Compressing objects:  35% (109/310)\rremote: Compressing objects:  36% (112/310)\rremote: Compressing objects:  37% (115/310)\rremote: Compressing objects:  38% (118/310)\rremote: Compressing objects:  39% (121/310)\rremote: Compressing objects:  40% (124/310)\rremote: Compressing objects:  41% (128/310)\rremote: Compressing objects:  42% (131/310)\rremote: Compressing objects:  43% (134/310)\rremote: Compressing objects:  44% (137/310)\rremote: Compressing objects:  45% (140/310)\rremote: Compressing objects:  46% (143/310)\rremote: Compressing objects:  47% (146/310)\rremote: Compressing objects:  48% (149/310)\rremote: Compressing objects:  49% (152/310)\rremote: Compressing objects:  50% (155/310)\rremote: Compressing objects:  51% (159/310)\rremote: Compressing objects:  52% (162/310)\rremote: Compressing objects:  53% (165/310)\rremote: Compressing objects:  54% (168/310)\rremote: Compressing objects:  55% (171/310)\rremote: Compressing objects:  56% (174/310)\rremote: Compressing objects:  57% (177/310)\rremote: Compressing objects:  58% (180/310)\rremote: Compressing objects:  59% (183/310)\rremote: Compressing objects:  60% (186/310)\rremote: Compressing objects:  61% (190/310)\rremote: Compressing objects:  62% (193/310)\rremote: Compressing objects:  63% (196/310)\rremote: Compressing objects:  64% (199/310)\rremote: Compressing objects:  65% (202/310)\rremote: Compressing objects:  66% (205/310)\rremote: Compressing objects:  67% (208/310)\rremote: Compressing objects:  68% (211/310)\rremote: Compressing objects:  69% (214/310)\rremote: Compressing objects:  70% (217/310)\rremote: Compressing objects:  71% (221/310)\rremote: Compressing objects:  72% (224/310)\rremote: Compressing objects:  73% (227/310)\rremote: Compressing objects:  74% (230/310)\rremote: Compressing objects:  75% (233/310)\rremote: Compressing objects:  76% (236/310)\rremote: Compressing objects:  77% (239/310)\rremote: Compressing objects:  78% (242/310)\rremote: Compressing objects:  79% (245/310)\rremote: Compressing objects:  80% (248/310)\rremote: Compressing objects:  81% (252/310)\rremote: Compressing objects:  82% (255/310)\rremote: Compressing objects:  83% (258/310)\rremote: Compressing objects:  84% (261/310)\rremote: Compressing objects:  85% (264/310)\rremote: Compressing objects:  86% (267/310)\rremote: Compressing objects:  87% (270/310)\rremote: Compressing objects:  88% (273/310)\rremote: Compressing objects:  89% (276/310)\rremote: Compressing objects:  90% (279/310)\rremote: Compressing objects:  91% (283/310)\rremote: Compressing objects:  92% (286/310)\rremote: Compressing objects:  93% (289/310)\rremote: Compressing objects:  94% (292/310)\rremote: Compressing objects:  95% (295/310)\rremote: Compressing objects:  96% (298/310)\rremote: Compressing objects:  97% (301/310)\rremote: Compressing objects:  98% (304/310)\rremote: Compressing objects:  99% (307/310)\rremote: Compressing objects: 100% (310/310)\rremote: Compressing objects: 100% (310/310), done.\r\nReceiving objects:   0% (1/5247)\rReceiving objects:   1% (53/5247)\rReceiving objects:   2% (105/5247)\rReceiving objects:   3% (158/5247)\rReceiving objects:   4% (210/5247)\rReceiving objects:   5% (263/5247)\rReceiving objects:   6% (315/5247)\rReceiving objects:   7% (368/5247)\rReceiving objects:   8% (420/5247)\r",,terminal_output
+13,194432,"TERMINAL",0,0,"Receiving objects:   9% (473/5247)\rReceiving objects:  10% (525/5247)\rReceiving objects:  11% (578/5247)\rReceiving objects:  12% (630/5247)\rReceiving objects:  13% (683/5247)\rReceiving objects:  14% (735/5247)\rReceiving objects:  15% (788/5247)\rReceiving objects:  16% (840/5247)\rReceiving objects:  17% (892/5247)\rReceiving objects:  18% (945/5247)\rReceiving objects:  19% (997/5247)\rReceiving objects:  20% (1050/5247)\rReceiving objects:  21% (1102/5247)\rReceiving objects:  22% (1155/5247)\rReceiving objects:  23% (1207/5247)\rReceiving objects:  24% (1260/5247)\rReceiving objects:  25% (1312/5247)\rReceiving objects:  26% (1365/5247)\rReceiving objects:  27% (1417/5247)\rReceiving objects:  28% (1470/5247)\rReceiving objects:  29% (1522/5247)\rReceiving objects:  30% (1575/5247)\rReceiving objects:  31% (1627/5247)\rReceiving objects:  32% (1680/5247)\rReceiving objects:  33% (1732/5247)\rReceiving objects:  34% (1784/5247)\rReceiving objects:  35% (1837/5247)\rReceiving objects:  36% (1889/5247)\rReceiving objects:  37% (1942/5247)\rReceiving objects:  38% (1994/5247)\rReceiving objects:  39% (2047/5247)\rReceiving objects:  40% (2099/5247)\rReceiving objects:  41% (2152/5247)\rReceiving objects:  42% (2204/5247)\rReceiving objects:  43% (2257/5247)\rReceiving objects:  44% (2309/5247)\r",,terminal_output
+14,194711,"TERMINAL",0,0,"Receiving objects:  45% (2362/5247)\rReceiving objects:  46% (2414/5247)\rReceiving objects:  47% (2467/5247)\rReceiving objects:  48% (2519/5247)\rReceiving objects:  49% (2572/5247)\rReceiving objects:  50% (2624/5247)\rReceiving objects:  51% (2676/5247)\rReceiving objects:  52% (2729/5247)\rReceiving objects:  53% (2781/5247)\rReceiving objects:  54% (2834/5247)\rReceiving objects:  55% (2886/5247)\rReceiving objects:  56% (2939/5247)\rReceiving objects:  57% (2991/5247)\rReceiving objects:  58% (3044/5247)\rReceiving objects:  59% (3096/5247)\rReceiving objects:  60% (3149/5247)\rReceiving objects:  61% (3201/5247)\rReceiving objects:  62% (3254/5247)\rReceiving objects:  63% (3306/5247)\rReceiving objects:  64% (3359/5247)\rReceiving objects:  65% (3411/5247)\rReceiving objects:  66% (3464/5247)\rReceiving objects:  67% (3516/5247)\rReceiving objects:  68% (3568/5247)\rReceiving objects:  69% (3621/5247)\rReceiving objects:  70% (3673/5247)\rReceiving objects:  71% (3726/5247)\rReceiving objects:  72% (3778/5247)\rReceiving objects:  73% (3831/5247)\rReceiving objects:  74% (3883/5247)\rReceiving objects:  75% (3936/5247)\rReceiving objects:  76% (3988/5247)\rReceiving objects:  77% (4041/5247)\rReceiving objects:  78% (4093/5247)\rReceiving objects:  79% (4146/5247)\rReceiving objects:  80% (4198/5247)\rReceiving objects:  81% (4251/5247)\rReceiving objects:  82% (4303/5247)\rReceiving objects:  83% (4356/5247)\rReceiving objects:  84% (4408/5247)\rReceiving objects:  85% (4460/5247)\rReceiving objects:  86% (4513/5247)\rReceiving objects:  87% (4565/5247)\rReceiving objects:  88% (4618/5247)\rReceiving objects:  89% (4670/5247)\rReceiving objects:  90% (4723/5247)\rremote: Total 5247 (delta 217), reused 334 (delta 125), pack-reused 4804 (from 1)\r\nReceiving objects:  91% (4775/5247)\rReceiving objects:  92% (4828/5247)\rReceiving objects:  93% (4880/5247)\rReceiving objects:  94% (4933/5247)\rReceiving objects:  95% (4985/5247)\rReceiving objects:  96% (5038/5247)\rReceiving objects:  97% (5090/5247)\rReceiving objects:  98% (5143/5247)\rReceiving objects:  99% (5195/5247)\rReceiving objects: 100% (5247/5247)\rReceiving objects: 100% (5247/5247), 659.26 KiB | 1.30 MiB/s, done.\r\nResolving deltas:   0% (0/3291)\rResolving deltas:   1% (33/3291)\rResolving deltas:   2% (68/3291)\rResolving deltas:   3% (100/3291)\rResolving deltas:   4% (132/3291)\rResolving deltas:   5% (165/3291)\rResolving deltas:   6% (199/3291)\rResolving deltas:   7% (232/3291)\rResolving deltas:   8% (264/3291)\rResolving deltas:   9% (299/3291)\rResolving deltas:  10% (330/3291)\rResolving deltas:  11% (363/3291)\rResolving deltas:  12% (395/3291)\rResolving deltas:  13% (429/3291)\rResolving deltas:  14% (461/3291)\rResolving deltas:  15% (494/3291)\rResolving deltas:  16% (528/3291)\rResolving deltas:  17% (560/3291)\rResolving deltas:  18% (594/3291)\rResolving deltas:  19% (626/3291)\rResolving deltas:  20% (661/3291)\rResolving deltas:  21% (693/3291)\rResolving deltas:  22% (726/3291)\rResolving deltas:  23% (757/3291)\rResolving deltas:  24% (790/3291)\rResolving deltas:  25% (823/3291)\rResolving deltas:  26% (856/3291)\rResolving deltas:  27% (889/3291)\rResolving deltas:  28% (922/3291)\rResolving deltas:  29% (955/3291)\rResolving deltas:  30% (988/3291)\rResolving deltas:  31% (1021/3291)\rResolving deltas:  32% (1054/3291)\rResolving deltas:  33% (1087/3291)\rResolving deltas:  34% (1120/3291)\rResolving deltas:  35% (1152/3291)\rResolving deltas:  36% (1185/3291)\rResolving deltas:  37% (1218/3291)\rResolving deltas:  38% (1251/3291)\rResolving deltas:  39% (1284/3291)\rResolving deltas:  40% (1319/3291)\rResolving deltas:  41% (1350/3291)\rResolving deltas:  42% (1383/3291)\rResolving deltas:  43% (1417/3291)\rResolving deltas:  44% (1450/3291)\rResolving deltas:  45% (1481/3291)\rResolving deltas:  46% (1515/3291)\rResolving deltas:  47% (1547/3291)\rResolving deltas:  48% (1580/3291)\rResolving deltas:  49% (1613/3291)\rResolving deltas:  50% (1646/3291)\rResolving deltas:  51% (1679/3291)\rResolving deltas:  52% (1713/3291)\rResolving deltas:  53% (1745/3291)\rResolving deltas:  54% (1778/3291)\rResolving deltas:  55% (1811/3291)\rResolving deltas:  56% (1843/3291)\rResolving deltas:  57% (1876/3291)\rResolving deltas:  58% (1909/3291)\rResolving deltas:  59% (1942/3291)\rResolving deltas:  60% (1975/3291)\rResolving deltas:  61% (2008/3291)\rResolving deltas:  62% (2041/3291)\rResolving deltas:  63% (2075/3291)\rResolving deltas:  64% (2107/3291)\rResolving deltas:  65% (2140/3291)\rResolving deltas:  66% (2173/3291)\rResolving deltas:  67% (2205/3291)\rResolving deltas:  68% (2238/3291)\rResolving deltas:  69% (2272/3291)\rResolving deltas:  70% (2304/3291)\rResolving deltas:  71% (2337/3291)\rResolving deltas:  72% (2370/3291)\rResolving deltas:  73% (2403/3291)\rResolving deltas:  74% (2436/3291)\rResolving deltas:  75% (2469/3291)\rResolving deltas:  76% (2502/3291)\rResolving deltas:  77% (2535/3291)\rResolving deltas:  78% (2567/3291)\rResolving deltas:  79% (2600/3291)\rResolving deltas:  80% (2636/3291)\rResolving deltas:  81% (2666/3291)\rResolving deltas:  82% (2699/3291)\rResolving deltas:  83% (2732/3291)\rResolving deltas:  84% (2765/3291)\rResolving deltas:  85% (2798/3291)\rResolving deltas:  86% (2831/3291)\rResolving deltas:  87% (2864/3291)\rResolving deltas:  88% (2897/3291)\rResolving deltas:  89% (2929/3291)\rResolving deltas:  90% (2963/3291)\rResolving deltas:  91% (2995/3291)\rResolving deltas:  92% (3031/3291)\rResolving deltas:  93% (3061/3291)\rResolving deltas:  94% (3095/3291)\rResolving deltas:  95% (3127/3291)\rResolving deltas:  96% (3160/3291)\rResolving deltas:  97% (3194/3291)\rResolving deltas:  98% (3227/3291)\rResolving deltas:  99% (3259/3291)\rResolving deltas: 100% (3291/3291)\rResolving deltas: 100% (3291/3291), done.\r\n",,terminal_output
+15,195903,"TERMINAL",0,0,"Updating files:  30% (408/1342)\rUpdating files:  31% (417/1342)\rUpdating files:  32% (430/1342)\r",,terminal_output
+16,196274,"TERMINAL",0,0,"Updating files:  33% (443/1342)\rUpdating files:  34% (457/1342)\rUpdating files:  35% (470/1342)\rUpdating files:  36% (484/1342)\rUpdating files:  37% (497/1342)\rUpdating files:  38% (510/1342)\rUpdating files:  39% (524/1342)\rUpdating files:  40% (537/1342)\r",,terminal_output
+17,196440,"TERMINAL",0,0,"Updating files:  41% (551/1342)\rUpdating files:  42% (564/1342)\rUpdating files:  43% (578/1342)\rUpdating files:  44% (591/1342)\r",,terminal_output
+18,196716,"TERMINAL",0,0,"Updating files:  45% (604/1342)\rUpdating files:  46% (618/1342)\rUpdating files:  47% (631/1342)\rUpdating files:  48% (645/1342)\rUpdating files:  49% (658/1342)\rUpdating files:  50% (671/1342)\rUpdating files:  51% (685/1342)\rUpdating files:  52% (698/1342)\rUpdating files:  53% (712/1342)\rUpdating files:  54% (725/1342)\rUpdating files:  55% (739/1342)\rUpdating files:  56% (752/1342)\rUpdating files:  57% (765/1342)\rUpdating files:  58% (779/1342)\rUpdating files:  59% (792/1342)\rUpdating files:  60% (806/1342)\rUpdating files:  61% (819/1342)\rUpdating files:  62% (833/1342)\rUpdating files:  63% (846/1342)\r",,terminal_output
+19,197013,"TERMINAL",0,0,"Updating files:  64% (859/1342)\rUpdating files:  65% (873/1342)\rUpdating files:  65% (874/1342)\r",,terminal_output
+20,197234,"TERMINAL",0,0,"Updating files:  66% (886/1342)\rUpdating files:  67% (900/1342)\rUpdating files:  68% (913/1342)\rUpdating files:  69% (926/1342)\rUpdating files:  70% (940/1342)\rUpdating files:  71% (953/1342)\rUpdating files:  72% (967/1342)\rUpdating files:  73% (980/1342)\rUpdating files:  74% (994/1342)\rUpdating files:  75% (1007/1342)\rUpdating files:  76% (1020/1342)\rUpdating files:  77% (1034/1342)\rUpdating files:  78% (1047/1342)\rUpdating files:  79% (1061/1342)\rUpdating files:  80% (1074/1342)\rUpdating files:  81% (1088/1342)\rUpdating files:  82% (1101/1342)\r",,terminal_output
+21,197682,"TERMINAL",0,0,"Updating files:  83% (1114/1342)\rUpdating files:  84% (1128/1342)\rUpdating files:  85% (1141/1342)\rUpdating files:  86% (1155/1342)\rUpdating files:  87% (1168/1342)\rUpdating files:  88% (1181/1342)\rUpdating files:  89% (1195/1342)\rUpdating files:  90% (1208/1342)\rUpdating files:  91% (1222/1342)\rUpdating files:  92% (1235/1342)\rUpdating files:  93% (1249/1342)\rUpdating files:  94% (1262/1342)\rUpdating files:  95% (1275/1342)\rUpdating files:  96% (1289/1342)\rUpdating files:  97% (1302/1342)\rUpdating files:  98% (1316/1342)\rUpdating files:  99% (1329/1342)\rUpdating files: 100% (1342/1342)\rUpdating files: 100% (1342/1342), done.\r\n]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot",,terminal_output
+22,219692,"TERMINAL",0,0,"cat slurm/dev/alfred/berlin/crowd_pilot/install_sglang.sh",,terminal_command
+23,219694,"TERMINAL",0,0,"]633;C#! /bin/bash\r\n\r\n# NOTE: Make sure that the project is set correctly using jutil\r\nmkdir $PROJECT/alfred/\r\nexport CARGO_HOME=$PROJECT/alfred/.cargo\r\nexport RUSTUP_HOME=$PROJECT/alfred/.rustup\r\n\r\nsource .venv/bin/activate\r\n\r\n# NOTE: A local rust installation already exists in the project root\r\ncurl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y\r\nuv pip install ""sglang"" --prerelease=allow\r\n]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot",,terminal_output
+24,234818,"TERMINAL",0,0,"cat slurm/dev/alfred/berlin/crowd_pilot/launch_sglang.sh",,terminal_command
+25,234836,"TERMINAL",0,0,"]633;CFLASHINFER_CACHE_DIR=$PROJECT/.cache/flashinfer\r\n\r\nsource .venv/bin/activate\r\n\r\npython3 -m sglang.launch_server --model-path qwen/qwen2.5-0.5b-instruct --host 0.0.0.0\r\n]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot",,terminal_output
+26,275503,"TERMINAL",0,0,"cat slurm/dev/alfred/berlin/crowd_pilot/prompt_engineering/chat.py",,terminal_command
+27,275506,"TERMINAL",0,0,"]633;Cimport requests\r\n\r\n# SERVER_URL = ""http://hai005:30000/generate""\r\nSERVER_URL = ""http://hai005:30000/v1/chat/completions""\r\n\r\n\r\nSYSTEM_PROMPT = """"""\r\n\t\t'Output ONLY a JSON array. No prose, no code fences.',\r\n\t\t'Allowed actions (TypeScript-like schema):',\r\n\t\t""{ kind: 'showTextDocument' }"",\r\n\t\t""{ kind: 'setSelections', selections: Array<{ start: [number, number], end: [number, number] }> }"",\r\n\t\t""{ kind: 'editInsert', position: [number, number], text: string }"",\r\n\t\t""{ kind: 'terminalShow' }"",\r\n\t\t""{ kind: 'terminalSendText', text: string }"",\r\n\t\t'Coordinates are zero-based [line, column].'\r\n""""""\r\n\r\ndemoGoal = """"""\r\n\t\t'Create a concise demo plan that:',y\r\n\t\t'- focuses the active text document',y\r\n\t\t'- moves the cursor to (0, 0)',y\r\n\t\t""- inserts the line \""hello from model\\n\"" at (0, 0)"",y\r\n\t\t'- focuses the terminal',y\r\n\t\t'- runs the command ""echo model run""'\r\n\r\n""""""\r\n\r\n\r\n\r\ndef chat(message):\r\n\r\n    # payload = {\r\n    # ""text"": message,       # <- use ""text"" instead of ""prompt""\r\n    # }\r\n\r\n    payload = {\r\n        ""model"": ""qwen/qwen2.5-0.5b-instruct"",\r\n        ""messages"": [\r\n            {""role"": ""system"", ""content"": SYSTEM_PROMPT},\r\n            {""role"": ""user"", ""content"": message}\r\n        ],\r\n    }\r\n\r\n    response = requests.post(SERVER_URL, json=payload)\r\n    if response.status_code == 200:\r\n        return response.json().get(""choices"", [{}])[0].get(""message"", {}).get(""content"", """")\r\n    else:\r\n        return f""Error: {response.status_code} - {response.text}""\r\n\r\n\r\nif __name__ == ""__main__"":\r\n    while True:\r\n        msg = input(""You: "")\r\n        if msg.lower() in [""exit"", ""quit""]:\r\n            break\r\n        reply = chat(msg)\r\n        print(""Bot:"", reply)\r\n]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot",,terminal_output

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-1dbf94d5-e80c-4f76-8c82-ec481c7acea31762603480406-2025_11_08-13.05.09.810/source.csv

@@ -0,0 +1,5 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,1838,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"1:05:09 PM [info] Activating crowd-code\n1:05:09 PM [info] Recording started\n1:05:09 PM [info] Initializing git provider using file system watchers...\n1:05:09 PM [info] No workspace folder found\n",Log,tab
+3,2285,"extension-output-pdoom-org.crowd-code-#1-crowd-code",194,0,"1:05:11 PM [info] Retrying git provider initialization...\n1:05:11 PM [info] No workspace folder found\n",Log,content
+4,7180,"TERMINAL",0,0,"",,terminal_command
+5,14300,"TERMINAL",0,0,"",,terminal_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-20676a2c-2459-41f4-8896-bf5b1bc9afe71764494282718-2025_11_30-10.18.52.300/source.csv

@@ -0,0 +1,5 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,3,"data/eval/handcrafted/add_import_easy.md",0,0,"# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nTraceback (most recent call last):\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 57, in <module>\n    convert_to_incremental_jsonl(line, args.output_file, task_name)\n    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 14, in convert_to_incremental_jsonl\n    data = json.loads(line)\n           ^^^^\nNameError: name 'json' is not defined. Did you forget to import 'json'?\n</stdout>\n\n\n# Assistant\n```bash\ncat -n src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n     1  import tyro\n     2  import os\n     3  from dataclasses import dataclass\n     4\n     5\n     6  @dataclass\n     7  class Args:\n     8      input_file: str = ""data/eval/val/validation.jsonl""\n     9      output_file: str = ""data/eval/val/validation_testcases.jsonl""\n    10      task_name: str = ""validation_set""\n    11\n    12\n    13  def convert_to_incremental_jsonl(line, output_file, task_name):\n    14      data = json.loads(line)\n    15\n    16      system_prompt = data.get(""system_prompt"", """")\n    17      conversations = data.get(""conversations"", [])\n    18\n    19      output_lines = []\n    20      context = []\n    21\n    22      for i, conv in enumerate(conversations):\n    23          # Convert ""Assistant""/""User"" to ""assistant""/""user""\n    24          role = conv[""from""].lower()\n    25          content = conv[""value""]\n    26\n    27          # Add current message to context\n    28          context.append({""role"": role, ""content"": content})\n    29\n    30          if i > 0 and role == ""assistant"":\n    31              task_entry = {\n    32                  ""system_prompt"": system_prompt,\n    33                  ""task_id"": f""{task_name}/{i-1}"",\n    34                  ""context"": context[:-1].copy(),\n    35                  ""expected_final_response"": content,\n    36              }\n    37              output_lines.append(task_entry)\n    38\n    39      with open(output_file, ""a"") as f:\n    40          for line in output_lines:\n    41              f.write(json.dumps(line) + ""\n"")\n    42\n    43      print(f""Converted {len(output_lines)} entries to {output_file}"")\n    44\n    45\n    46  if __name__ == ""__main__"":\n    47      args = tyro.cli(Args)\n    48\n    49      if os.path.exists(args.output_file):\n    50          os.remove(args.output_file)\n    51\n    52      # iterate over jsonl file and convert to incremental jsonl\n    53      conversation_id = 0\n    54      with open(args.input_file, ""r"") as f:\n    55          for line in f:\n    56              task_name = f""conversation_{conversation_id}/{args.task_name}""\n    57              convert_to_incremental_jsonl(line, args.output_file, task_name)\n    58              conversation_id += 1\n</stdout>\n\n# Assistant\n```bash\nsed -i '1i\import json' src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n</stdout>\n\n# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 2 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 6 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\n</stdout>\n",markdown,tab
+2,14760,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"10:18:52 AM [info] Activating crowd-code\n10:18:52 AM [info] Recording started\n10:18:52 AM [info] Initializing git provider using file system watchers...\n10:18:54 AM [info] Retrying git provider initialization...\n10:18:56 AM [info] Git repository found\n10:18:56 AM [info] Git provider initialized successfully\n10:18:56 AM [info] Initial git state: [object Object]\n10:18:56 AM [info] Git repository found\n10:18:56 AM [info] Git provider initialized successfully\n10:18:56 AM [info] Initial git state: [object Object]\n",Log,tab
+3,67181,"data/eval/handcrafted/add_import_easy.md",0,0,"",markdown,tab
+4,77316,"TERMINAL",0,0,"",,terminal_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-2165ae8d-ebb7-4134-a4e8-9043ba00c6551762612659284-2025_11_08-15.38.17.91/source.csv

@@ -0,0 +1,25 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,3,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",0,0,"import requests\n\n# SERVER_URL = ""http://hai005:30000/generate""\nSERVER_URL = ""http://hai006:30000/v1/chat/completions""\n\n\nSYSTEM_PROMPT = """"""\n\t\t'Output ONLY a JSON array. No prose, no code fences.',\n\t\t'Allowed actions (TypeScript-like schema):',\n\t\t""{ kind: 'showTextDocument' }"",\n\t\t""{ kind: 'setSelections', selections: Array<{ start: [number, number], end: [number, number] }> }"",\n\t\t""{ kind: 'editInsert', position: [number, number], text: string }"",\n\t\t""{ kind: 'terminalShow' }"",\n\t\t""{ kind: 'terminalSendText', text: string }"",\n\t\t'Coordinates are zero-based [line, column].'\n""""""\n\ndemoGoal = """"""\n\t\t'Create a concise demo plan that:',y\n\t\t'- focuses the active text document',y\n\t\t'- moves the cursor to (0, 0)',y\n\t\t""- inserts the line \""hello from model\\n\"" at (0, 0)"",y\n\t\t'- focuses the terminal',y\n\t\t'- runs the command ""echo model run""'\n\n""""""\n\n\n\ndef chat():\n\n    payload = {\n        ""model"": ""Qwen/Qwen3-0.6B"",\n        ""messages"": [\n            {""role"": ""system"", ""content"": SYSTEM_PROMPT},\n            {""role"": ""user"", ""content"": demoGoal}\n        ],\n    }\n\n    response = requests.post(SERVER_URL, json=payload)\n    if response.status_code == 200:\n        return response.json().get(""choices"", [{}])[0].get(""message"", {}).get(""content"", """")\n    else:\n        return f""Error: {response.status_code} - {response.text}""\n\n\nif __name__ == ""__main__"":\n    reply = chat()\n    print(reply)\n",python,tab
+2,3540,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"3:38:17 PM [info] Activating crowd-code\n3:38:17 PM [info] Recording started\n3:38:17 PM [info] Initializing git provider using file system watchers...\n3:38:17 PM [info] Git repository found\n3:38:17 PM [info] Git provider initialized successfully\n3:38:17 PM [info] Initial git state: [object Object]\n",Log,tab
+3,9553,"TERMINAL",0,0,"",,terminal_command
+4,20322,"TERMINAL",0,0,"",,terminal_command
+5,92697,"TERMINAL",0,0,"",,terminal_command
+6,186522,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",0,0,"",python,tab
+7,295932,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab
+8,298717,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",0,0,"",python,tab
+9,357980,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",842,0,"",python,selection_mouse
+10,358012,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",841,0,"",python,selection_command
+11,401294,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",665,0,"",python,selection_mouse
+12,401731,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",627,0,"",python,selection_mouse
+13,402461,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",595,38,"\t\t'Create a concise demo plan that:',y",python,selection_command
+14,402631,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",595,79,"\t\t'Create a concise demo plan that:',y\n\t\t'- focuses the active text document',y",python,selection_command
+15,402781,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",595,114,"\t\t'Create a concise demo plan that:',y\n\t\t'- focuses the active text document',y\n\t\t'- moves the cursor to (0, 0)',y",python,selection_command
+16,402891,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",595,173,"\t\t'Create a concise demo plan that:',y\n\t\t'- focuses the active text document',y\n\t\t'- moves the cursor to (0, 0)',y\n\t\t""- inserts the line \""hello from model\\n\"" at (0, 0)"",y",python,selection_command
+17,403546,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",595,202,"\t\t'Create a concise demo plan that:',y\n\t\t'- focuses the active text document',y\n\t\t'- moves the cursor to (0, 0)',y\n\t\t""- inserts the line \""hello from model\\n\"" at (0, 0)"",y\n\t\t'- focuses the terminal',y",python,selection_command
+18,403780,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",633,0,"",python,selection_command
+19,404441,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",796,1,"",python,content
+20,404441,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",767,1,"",python,content
+21,404441,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",708,1,"",python,content
+22,404441,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",673,1,"",python,content
+23,404441,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",632,1,"",python,content
+24,404811,"slurm/dev/mihir/berlin/crowd_pilot/prompt_engineering/chat.py",631,0,"",python,selection_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-23300773-7107-48dc-95de-2ef6fdc0e97c1760710518542-2025_10_17-16.15.55.319/source.csv

@@ -0,0 +1,16 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,2828,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"4:15:55 PM [info] Activating crowd-code\n4:15:55 PM [info] Recording started\n4:15:55 PM [info] Initializing git provider using file system watchers...\n4:15:56 PM [info] Git repository found\n4:15:56 PM [info] Git provider initialized successfully\n4:15:56 PM [info] Initial git state: [object Object]\n",Log,tab
+3,8847,"TERMINAL",0,0,"",,terminal_command
+4,15626,"TERMINAL",0,0,"",,terminal_command
+5,89719,"TERMINAL",0,0,"",,terminal_command
+6,277515,"slurm/utils/mihir/ali_runs/merge_and_plot.sh",0,0,"python slurm/utils/mihir/run-renamer.py slurm/utils/mihir/jasmine.csv slurm/utils/mihir/jasmine_renamed.csv\n\npython slurm/utils/mihir/merge_wandb_runs.py -i slurm/utils/mihir/ali_runs/og*.csv -o slurm/utils/mihir/ali_runs/merged.csv\npython slurm/utils/mihir/run-renamer.py slurm/utils/mihir/ali_runs/merged.csv slurm/utils/mihir/ali_runs/merged_renamed.csv\n\npython slurm/utils/mihir/wandb_csv_to_matplotlib_minimal.py slurm/utils/mihir/jasmine_renamed.csv slurm/utils/mihir/ali_runs/merged_renamed.csv --x ""time"" --metric ""Train loss"" --save slurm/utils/mihir/ali_runs/train-loss.pdf",shellscript,tab
+7,282537,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",0,0,"python slurm/utils/mihir/run-renamer.py slurm/utils/mihir/jasmine.csv slurm/utils/mihir/jasmine_renamed.csv\n\npython slurm/utils/mihir/merge_wandb_runs.py -i slurm/utils/mihir/ali_runs/og*.csv -o slurm/utils/mihir/ali_runs/merged.csv\npython slurm/utils/mihir/run-renamer.py slurm/utils/mihir/ali_runs/merged.csv slurm/utils/mihir/ali_runs/merged_renamed.csv\n\npython slurm/utils/mihir/wandb_csv_to_matplotlib_minimal.py slurm/utils/mihir/jasmine_renamed.csv slurm/utils/mihir/ali_runs/merged_renamed.csv --x ""time"" --metric ""Train loss"" --save slurm/utils/mihir/ali_runs/train-loss.pdf",shellscript,tab
+8,283659,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",108,0,"",shellscript,selection_mouse
+9,287332,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",141,1,"a",shellscript,selection_command
+10,287333,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",175,2,"al",shellscript,selection_command
+11,287512,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",175,3,"ali",shellscript,selection_command
+12,288427,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",175,4,"ali_",shellscript,selection_command
+13,288844,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",175,5,"ali_r",shellscript,selection_command
+14,288920,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",175,6,"ali_ru",shellscript,selection_command
+15,288972,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",175,7,"ali_run",shellscript,selection_command
+16,289111,"slurm/utils/mihir/jafar_full_run/merge_and_plot.sh",175,8,"ali_runs",shellscript,selection_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-2b611b25-766c-40fe-9c56-c37495c49abf1764154622152-2025_11_26-11.57.21.424/source.csv

@@ -0,0 +1,78 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,3267,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"11:57:21 AM [info] Activating crowd-code\n11:57:21 AM [info] Recording started\n11:57:21 AM [info] Initializing git provider using file system watchers...\n11:57:23 AM [info] Retrying git provider initialization...\n11:57:23 AM [info] Git repository found\n11:57:23 AM [info] Git provider initialized successfully\n11:57:23 AM [info] Initial git state: [object Object]\n11:57:24 AM [info] Git repository found\n11:57:24 AM [info] Git provider initialized successfully\n11:57:24 AM [info] Initial git state: [object Object]\n",Log,tab
+3,10154,"TERMINAL",0,0,"",,terminal_command
+4,17051,"TERMINAL",0,0,"",,terminal_focus
+5,17172,"TERMINAL",0,0,"",,terminal_command
+6,17552,"TERMINAL",0,0,"source /home/mihir.mahajan/Projects/crowd-pilot/.uv_venv/bin/activate",,terminal_command
+7,17563,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot",,terminal_output
+8,19079,"TERMINAL",0,0,"watch",,terminal_focus
+9,46947,"TERMINAL",0,0,"salloc --time=02:00:00 --gres=gpu:1 --nodes=1 --ntasks-per-node=1 --mem=200G --qos=low",,terminal_command
+10,47003,"TERMINAL",0,0,"]633;Csalloc: Granted job allocation 34851\r\n",,terminal_output
+11,47089,"TERMINAL",0,0,"salloc: Nodes hai005 are ready for job\r\n",,terminal_output
+12,47370,"TERMINAL",0,0,"Running inside SLURM, Job ID 34851.\r\n",,terminal_output
+13,47459,"TERMINAL",0,0,"]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot[?2004h[mihir.mahajan@hai005.haicore.berlin:~/Projects/crowd-pilot] $ ",,terminal_output
+14,48513,"TERMINAL",0,0,"s",,terminal_output
+15,48694,"TERMINAL",0,0,"o",,terminal_output
+16,48824,"TERMINAL",0,0,"u",,terminal_output
+17,48895,"TERMINAL",0,0,"r",,terminal_output
+18,49077,"TERMINAL",0,0,"c",,terminal_output
+19,49266,"TERMINAL",0,0,"e",,terminal_output
+20,49382,"TERMINAL",0,0," .",,terminal_output
+21,49559,"TERMINAL",0,0,"v",,terminal_output
+22,49736,"TERMINAL",0,0,"env/",,terminal_output
+23,49833,"TERMINAL",0,0,"b",,terminal_output
+24,50284,"TERMINAL",0,0,"in/",,terminal_output
+25,50410,"TERMINAL",0,0,"ac",,terminal_output
+26,50766,"TERMINAL",0,0,"tivate",,terminal_output
+27,50900,"TERMINAL",0,0,"\r\n[?2004l\r]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:~/Projects/crowd-pilot] $ ",,terminal_output
+28,57713,"TERMINAL",0,0,"sc",,terminal_output
+29,57911,"TERMINAL",0,0,"a",,terminal_output
+30,58054,"TERMINAL",0,0,"n",,terminal_output
+31,58198,"TERMINAL",0,0,"c",,terminal_output
+32,58323,"TERMINAL",0,0,"e",,terminal_output
+33,58419,"TERMINAL",0,0,"l",,terminal_output
+34,58525,"TERMINAL",0,0," ",,terminal_output
+35,58640,"TERMINAL",0,0,"34849",,terminal_output
+36,59051,"TERMINAL",0,0,"34849\r\n[?2004l\r]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:~/Projects/crowd-pilot] $ ",,terminal_output
+37,82286,"slurm/dev/mihir/berlin/crowd_pilot/nemo/iterating_on_stuff/test_full_finetune_4k.py",0,0,"import os\n\nimport nemo_run as run\nfrom nemo.collections import llm\nfrom lightning.pytorch.loggers import WandbLogger\n\nos.environ[""NEMORUN_HOME""] = ""/fast/project/HFMI_SynergyUnit/tab_model/franz/nemorun_home""\nos.environ[""NEMO_CACHE_DIR""] = ""/fast/project/HFMI_SynergyUnit/tab_model/franz/nemo_cache""\nslurm_job_id = os.environ.get(""SLURM_JOB_ID"", ""local"")\n\nNAME = f""qwen3_600m_full_4k_tokens_debug_{slurm_job_id}""\nDATASET_DIR = ""/fast/project/HFMI_SynergyUnit/tab_model/data/nemo_hf_part_jsonl_4k_tokens/"" \n\ndef recipe_to_dict(recipe, depth=4):\n    if depth == 0:\n        return dict()\n    cfg_dict = dict()\n    for k, v in vars(recipe)['__arguments__'].items():\n        if isinstance(v, run.Config) or isinstance(v, dict):\n            cfg_dict[k] = recipe_to_dict(v, depth-1)\n        elif isinstance(v, (int, float, str, bool)):\n            cfg_dict[k] = v\n        else:\n            cfg_dict[k] = str(v)\n    return cfg_dict\n                    \nrecipe = llm.qwen3_600m.finetune_recipe(\n    name=NAME,\n    dir=f""/fast/project/HFMI_SynergyUnit/mihir/logs"",\n    num_nodes=1,\n    num_gpus_per_node=1,\n    peft_scheme=None,\n    # FIXME(f.srambical): check whether this is supported by ChatDataModule\n    packed_sequence=False,\n)\n\ndef local_executor_torchrun(nodes: int, devices: int) -> run.LocalExecutor:\n    # Env vars for jobs are configured here\n    env_vars = {\n        ""TORCH_NCCL_AVOID_RECORD_STREAMS"": ""1"",\n        ""NCCL_NVLS_ENABLE"": ""0"",\n        ""NVTE_DP_AMAX_REDUCE_INTERVAL"": ""0"",\n        ""NVTE_ASYNC_AMAX_REDUCTION"": ""1"",\n    }\n\n    executor = run.LocalExecutor(ntasks_per_node=devices, launcher=""torchrun"", env_vars=env_vars)\n\n    return executor\n\ndef run_finetuning():\n    recipe.resume.restore_config.path = ""/fast/project/HFMI_SynergyUnit/tab_model/data/checkpoints/nemo_converted_weights_qwen3-600m/""\n    recipe.data = run.Config(\n        llm.ChatDataModule,\n        dataset_root=DATASET_DIR,\n        seq_length=4096,\n        global_batch_size=32,\n        micro_batch_size=4,\n        num_workers=0,\n    )\n\n    MAX_STEPS = 11\n    VAL_INTERVAL = 2\n    NUM_CHECKPOINTS = 10\n    recipe.trainer.max_steps = MAX_STEPS\n    recipe.log.ckpt.every_n_train_steps = MAX_STEPS // NUM_CHECKPOINTS\n    recipe.log.ckpt.save_top_k = NUM_CHECKPOINTS\n    recipe.trainer.val_check_interval = VAL_INTERVAL\n\n    cfg_dict = recipe_to_dict(recipe)\n    recipe.log.wandb = run.Config(\n        WandbLogger,\n        name=NAME,\n        project=""crowd-pilot-nemo"",\n        save_dir=""/fast/project/HFMI_SynergyUnit/mihir/logs/wandb"",\n        entity=""instant-uv"",\n        config=cfg_dict,\n    )\n\n    executor = local_executor_torchrun(nodes=recipe.trainer.num_nodes, devices=recipe.trainer.devices)\n\n    run.run(recipe, executor=executor, name=NAME)\n\n# This condition is necessary for the script to be compatible with Python's multiprocessing module.\nif __name__ == ""__main__"":\n    run_finetuning()",python,tab
+38,82287,"slurm/dev/mihir/berlin/crowd_pilot/nemo/iterating_on_stuff/test_full_finetune_4k.py",2083,0,"",python,selection_mouse
+39,82361,"slurm/dev/mihir/berlin/crowd_pilot/nemo/iterating_on_stuff/test_full_finetune_4k.py",2082,0,"",python,selection_command
+40,83172,"TERMINAL",0,0,"srun",,terminal_focus
+41,88676,"TERMINAL",0,0,"ls",,terminal_output
+42,89998,"TERMINAL",0,0,"\r\n[?2004l\rLICENSE    crowd-pilot-extension  maxtext         slurm  uv.lock\r\nREADME.md  crowd_pilot            pyproject.toml  tests\r\n]0;mihir.mahajan@hai-login2:~/Projects/crowd-pilot[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:~/Projects/crowd-pilot] $ ",,terminal_output
+43,91865,"TERMINAL",0,0,"l",,terminal_output
+44,91984,"TERMINAL",0,0,"o",,terminal_output
+45,92163,"TERMINAL",0,0,"g",,terminal_output
+46,92341,"TERMINAL",0,0,"s\r\n[?2004l\r]0;mihir.mahajan@hai-login2:/fast/project/HFMI_SynergyUnit/mihir/logs[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir/logs] $ ",,terminal_output
+47,92474,"TERMINAL",0,0,"l",,terminal_output
+48,92652,"TERMINAL",0,0,"s\r\n[?2004l\rfinetune_lora_4k.log        finetune_lora_4k_34587.log             tb_logs\r\nfinetune_lora_4k_34569.log  finetune_lora_4k_34588.log             train\r\nfinetune_lora_4k_34572.log  nemo                                   wandb\r\nfinetune_lora_4k_34584.log  qwen3_600m_full_4k_tokens_34783\r\nfinetune_lora_4k_34585.log  qwen3_600m_full_4k_tokens_debug_34849\r\n]0;mihir.mahajan@hai-login2:/fast/project/HFMI_SynergyUnit/mihir/logs[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir/logs] $ ",,terminal_output
+49,94698,"TERMINAL",0,0,"c",,terminal_output
+50,94752,"TERMINAL",0,0,"d",,terminal_output
+51,94892,"TERMINAL",0,0," ",,terminal_output
+52,95161,"TERMINAL",0,0,"n",,terminal_output
+53,95846,"TERMINAL",0,0,"",,terminal_output
+54,95980,"TERMINAL",0,0,"",,terminal_output
+55,100656,"TERMINAL",0,0," ",,terminal_output
+56,104023,"TERMINAL",0,0,"qwen3_600m_full_4k_tokens_debug_34849",,terminal_output
+57,104466,"TERMINAL",0,0,"\rqwen3_600m_full_4k_tokens_debug_34849\r\n[?2004l\r]0;mihir.mahajan@hai-login2:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849] $ ",,terminal_output
+58,104781,"TERMINAL",0,0,"ls",,terminal_output
+59,104958,"TERMINAL",0,0,"\r\n[?2004l\r2025-11-26_11-33-01\r\n]0;mihir.mahajan@hai-login2:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849] $ ",,terminal_output
+60,107010,"TERMINAL",0,0,"cd",,terminal_output
+61,107076,"TERMINAL",0,0," ",,terminal_output
+62,107160,"TERMINAL",0,0,"2025-11-26_11-33-01/",,terminal_output
+63,107227,"TERMINAL",0,0,"",,terminal_command
+64,107484,"TERMINAL",0,0,"\r\n[?2004l\r]0;mihir.mahajan@hai-login2:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01] $ ",,terminal_output
+65,107666,"TERMINAL",0,0,"l",,terminal_output
+66,107902,"TERMINAL",0,0,"s",,terminal_output
+67,108200,"TERMINAL",0,0,"\r\n[?2004l\rcheckpoints   git-info.log        nemo_error_log.txt\r\ncmd-args.log  lightning_logs.txt  nemo_log_globalrank-0_localrank-0.txt\r\n]0;mihir.mahajan@hai-login2:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01] $ ",,terminal_output
+68,142974,"TERMINAL",0,0,"l",,terminal_output
+69,143053,"TERMINAL",0,0,"s",,terminal_output
+70,143118,"TERMINAL",0,0," ",,terminal_output
+71,143327,"TERMINAL",0,0,"c",,terminal_output
+72,143493,"TERMINAL",0,0,"h",,terminal_output
+73,143619,"TERMINAL",0,0,"e",,terminal_output
+74,143768,"TERMINAL",0,0,"c",,terminal_output
+75,143906,"TERMINAL",0,0,"kpoints/",,terminal_output
+76,144336,"TERMINAL",0,0,"\r\n[?2004l\r'model_name=0--val_loss=1.09-step=10-consumed_samples=352.0-last'\r\n'model_name=0--val_loss=1.10-step=10-consumed_samples=352.0'\r\n'model_name=0--val_loss=1.14-step=8-consumed_samples=288.0'\r\n'model_name=0--val_loss=1.14-step=9-consumed_samples=320.0'\r\n'model_name=0--val_loss=1.15-step=6-consumed_samples=224.0'\r\n'model_name=0--val_loss=1.15-step=7-consumed_samples=256.0'\r\n'model_name=0--val_loss=1.16-step=2-consumed_samples=96.0'\r\n'model_name=0--val_loss=1.16-step=3-consumed_samples=128.0'\r\n'model_name=0--val_loss=1.16-step=4-consumed_samples=160.0'\r\n'model_name=0--val_loss=1.16-step=5-consumed_samples=192.0'\r\n]0;mihir.mahajan@hai-login2:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01[?2004h(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01] $ ",,terminal_output
+77,279191,"TERMINAL",0,0,"\r(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir\r\n\r/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01] $ \r(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir\r\n\r/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01] $ ",,terminal_output
+78,608800,"TERMINAL",0,0,"\r(crowd-pilot) [mihir.mahajan@hai005.haicore.berlin:/fast/project/HFMI_SynergyUnit/mihir\r\n\r/logs/qwen3_600m_full_4k_tokens_debug_34849/2025-11-26_11-33-01] $ ",,terminal_output

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-34b6dec0-9dde-4479-9af4-cd0656225b821761477933551-2025_10_26-12.25.57.396/source.csv

@@ -0,0 +1,39 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,2312,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"12:25:57 PM [info] Activating crowd-code\n12:25:57 PM [info] Recording started\n12:25:57 PM [info] Initializing git provider using file system watchers...\n12:25:58 PM [info] Git repository found\n12:25:58 PM [info] Git provider initialized successfully\n12:25:58 PM [info] Initial git state: [object Object]\n",Log,tab
+3,8187,"TERMINAL",0,0,"",,terminal_command
+4,15085,"TERMINAL",0,0,"",,terminal_command
+5,103178,"TERMINAL",0,0,"",,terminal_command
+6,144952,"TERMINAL",0,0,"bash",,terminal_focus
+7,147182,"TERMINAL",0,0,"git branch",,terminal_command
+8,147205,"TERMINAL",0,0,"]633;C[?1h=\r  feat/diffusion-backend\r\n  feat/diffusion-forcing\r\n  feat/diffusion-jittable-sampling\r\n  feat/diffusion-ramp-weight\r\n* feat/diffusion-train-from-mae\r\n  feat/diffusion-x-loss\r\n  feat/masked-autoencoder\r\n  feat/notebook-detect-arrayrecord-duplicates\r\n  feat/shortcut-forcing\r\n  main\r\n  prepend-action-maskgit\r\n  vizdoom-dataset\r\n\r[?1l>]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+9,149188,"TERMINAL",0,0,"^C",,terminal_command
+10,153934,"TERMINAL",0,0,"git status",,terminal_command
+11,153979,"TERMINAL",0,0,"]633;COn branch feat/diffusion-train-from-mae\r\nYour branch is up to date with 'origin/feat/diffusion-train-from-mae'.\r\n\r\nUntracked files:\r\n  (use ""git add <file>..."" to include in what will be committed)\r\n\tdata/_vizdoom.ini\r\n\tdata/jasmine_data/ViZDoomPPO/\r\n\tdata/uv.lock\r\n\tjasmine/genie_bak.py\r\n\tjasmine/genie_sliding.py\r\n\tjasmine/sample_sliding.py\r\n\tjasmine/test_dit.py\r\n\tslurm/\r\n\tuv.lock\r\n\r\nnothing added to commit but untracked files present (use ""git add"" to track)\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+12,270937,"TERMINAL",0,0,"queue",,terminal_command
+13,271006,"TERMINAL",0,0,"]633;C[?1049h(B[?7hEvery 1.0s: squeue --mehai-login2.haicore.berlin: Sun Oct 26 12:30:28 2025JOBIDUSER PARTITION NODES  CPUS STSUBMIT_TIMESTART_TIME     TIME TIME_LIMIT NODELIST(REASON)",,terminal_output
+14,271756,"TERMINAL",0,0,"[?1049l\r[?1l>]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+15,954756,"TERMINAL",0,0,"git checkout feat/diffusion-backend",,terminal_command
+16,954806,"TERMINAL",0,0,"]633;C",,terminal_output
+17,954913,"TERMINAL",0,0,"Switched to branch 'feat/diffusion-backend'\r\nYour branch is up to date with 'origin/feat/diffusion-backend'.\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+18,955782,"",0,0,"Switched from branch 'feat/diffusion-train-from-mae' to 'feat/diffusion-backend'",,git_branch_checkout
+19,956407,"TERMINAL",0,0,"git pull",,terminal_command
+20,956456,"TERMINAL",0,0,"]633;C",,terminal_output
+21,958218,"TERMINAL",0,0,"remote: Enumerating objects: 15, done.\r\nremote: Counting objects:   6% (1/15)\rremote: Counting objects:  13% (2/15)\rremote: Counting objects:  20% (3/15)\rremote: Counting objects:  26% (4/15)\rremote: Counting objects:  33% (5/15)\rremote: Counting objects:  40% (6/15)\rremote: Counting objects:  46% (7/15)\rremote: Counting objects:  53% (8/15)\rremote: Counting objects:  60% (9/15)\rremote: Counting objects:  66% (10/15)\rremote: Counting objects:  73% (11/15)\rremote: Counting objects:  80% (12/15)\rremote: Counting objects:  86% (13/15)\rremote: Counting objects:  93% (14/15)\rremote: Counting objects: 100% (15/15)\rremote: Counting objects: 100% (15/15), done.\r\nremote: Compressing objects:  20% (1/5)\rremote: Compressing objects:  40% (2/5)\rremote: Compressing objects:  60% (3/5)\rremote: Compressing objects:  80% (4/5)\rremote: Compressing objects: 100% (5/5)\rremote: Compressing objects: 100% (5/5), done.\r\nremote: Total 15 (delta 10), reused 14 (delta 10), pack-reused 0 (from 0)\r\nUnpacking objects:   6% (1/15)\rUnpacking objects:  13% (2/15)\rUnpacking objects:  20% (3/15)\rUnpacking objects:  26% (4/15)\rUnpacking objects:  33% (5/15)\rUnpacking objects:  40% (6/15)\rUnpacking objects:  46% (7/15)\rUnpacking objects:  53% (8/15)\rUnpacking objects:  60% (9/15)\rUnpacking objects:  66% (10/15)\rUnpacking objects:  73% (11/15)\rUnpacking objects:  80% (12/15)\r",,terminal_output
+22,958460,"TERMINAL",0,0,"Unpacking objects:  86% (13/15)\rUnpacking objects:  93% (14/15)\rUnpacking objects: 100% (15/15)\rUnpacking objects: 100% (15/15), 2.64 KiB | 8.00 KiB/s, done.\r\n",,terminal_output
+23,958683,"TERMINAL",0,0,"From github.com:p-doom/jasmine\r\n   64e02f4..93b4c0a  feat/diffusion-backend -> origin/feat/diffusion-backend\r\n   58d8c91..41ec2f5  vizdoom-dataset        -> origin/vizdoom-dataset\r\n",,terminal_output
+24,958743,"TERMINAL",0,0,"Updating 64e02f4..93b4c0a\r\n",,terminal_output
+25,958801,"TERMINAL",0,0,"Fast-forward\r\n",,terminal_output
+26,958826,"TERMINAL",0,0," jasmine/genie.py            | 244 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++----------------------------------------------------------\r\n jasmine/models/dynamics.py  |  22 +++++++++----\r\n jasmine/models/tokenizer.py |   1 +\r\n jasmine/sample.py           |   8 +++--\r\n jasmine/train_dynamics.py   |   6 ++--\r\n jasmine/train_tokenizer.py  |   4 ++-\r\n jasmine/utils/nn.py         |  28 ++++++++--------\r\n 7 files changed, 182 insertions(+), 131 deletions(-)\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+27,1860457,"TERMINAL",0,0,"git status",,terminal_command
+28,1860517,"TERMINAL",0,0,"]633;COn branch feat/diffusion-backend\r\nYour branch is up to date with 'origin/feat/diffusion-backend'.\r\n\r\nUntracked files:\r\n  (use ""git add <file>..."" to include in what will be committed)\r\n\tdata/_vizdoom.ini\r\n\tdata/jasmine_data/ViZDoomPPO/\r\n\tdata/uv.lock\r\n\tjasmine/genie_bak.py\r\n\tjasmine/genie_sliding.py\r\n\tjasmine/sample_sliding.py\r\n\tjasmine/test_dit.py\r\n\tslurm/\r\n\tuv.lock\r\n\r\nnothing added to commit but untracked files present (use ""git add"" to track)\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+29,1868164,"TERMINAL",0,0,"git branch",,terminal_command
+30,1868204,"TERMINAL",0,0,"]633;C[?1h=\r* feat/diffusion-backend\r\n  feat/diffusion-forcing\r\n  feat/diffusion-jittable-sampling\r\n  feat/diffusion-ramp-weight\r\n  feat/diffusion-train-from-mae\r\n  feat/diffusion-x-loss\r\n  feat/masked-autoencoder\r\n  feat/notebook-detect-arrayrecord-duplicates\r\n  feat/shortcut-forcing\r\n  main\r\n  prepend-action-maskgit\r\n  vizdoom-dataset\r\n\r[?1l>]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+31,1891700,"TERMINAL",0,0,"git checkout -b ""chore/cleanup-diffusion-branch""",,terminal_command
+32,1891772,"TERMINAL",0,0,"]633;CSwitched to a new branch 'chore/cleanup-diffusion-branch'\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+33,1895871,"",0,0,"Switched from branch 'feat/diffusion-backend' to 'chore/cleanup-diffusion-branch'",,git_branch_checkout
+34,1902533,"TERMINAL",0,0,"git checkout feat/diffusion-backend",,terminal_command
+35,1902567,"TERMINAL",0,0,"]633;CSwitched to branch 'feat/diffusion-backend'\r\nYour branch is up to date with 'origin/feat/diffusion-backend'.\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+36,1905869,"",0,0,"Switched from branch 'chore/cleanup-diffusion-branch' to 'feat/diffusion-backend'",,git_branch_checkout
+37,1918003,"TERMINAL",0,0,"git checkout -b ""feat/sliding-window-sampling""",,terminal_command
+38,1918066,"TERMINAL",0,0,"]633;CSwitched to a new branch 'feat/sliding-window-sampling'\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine",,terminal_output
+39,1920871,"",0,0,"Switched from branch 'feat/diffusion-backend' to 'feat/sliding-window-sampling'",,git_branch_checkout

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-35174d2b-ac4a-4076-abc0-3de665efb30d1764851634497-2025_12_04-13.34.18.756/source.csv

@@ -0,0 +1,78 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,3781,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"1:34:18 PM [info] Activating crowd-code\n1:34:18 PM [info] Recording started\n1:34:18 PM [info] Initializing git provider using file system watchers...\n1:34:20 PM [info] Retrying git provider initialization...\n1:34:20 PM [info] Git repository found\n1:34:20 PM [info] Git provider initialized successfully\n1:34:21 PM [info] Initial git state: [object Object]\n1:34:21 PM [info] Git repository found\n1:34:21 PM [info] Git provider initialized successfully\n1:34:21 PM [info] Initial git state: [object Object]\n",Log,tab
+3,9132,"TERMINAL",0,0,"",,terminal_command
+4,16099,"TERMINAL",0,0,"",,terminal_command
+5,16109,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login1:~/Projects/tab-model-eval",,terminal_output
+6,23505,"TERMINAL",0,0,"bash",,terminal_focus
+7,25812,"TERMINAL",0,0,"queue",,terminal_command
+8,25875,"TERMINAL",0,0,"]633;C[?1049h(B[?7hEvery 1.0s: sque...  hai-login1.haicore.berlin: Thu Dec  4 13:34:44 2025JOBIDUSER PARTITION NODES  CPUS STSUBMIT_TIME\r START_TIME     TIME TIME_LIMIT NODELIST(REASON)",,terminal_output
+9,26838,"TERMINAL",0,0,"5",,terminal_output
+10,26876,"TERMINAL",0,0,"[?1049l\r[?1l>]0;mihir.mahajan@hai-login1:~/Projects/tab-model-eval",,terminal_output
+11,30522,"TERMINAL",0,0,"clear",,terminal_command
+12,30532,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login1:~/Projects/tab-model-eval",,terminal_output
+13,37004,"data/eval/unit_tests/eval_delete.md",0,0,"cat -n jasmine/baselines/diffusion/sample_diffusion.py\n\n\n\n     1\tfrom dataclasses import dataclass\n     2\timport time\n     3\timport os\n     4\timport optax\n     5\t\n     6\timport dm_pix as pix\n     7\timport einops\n     8\timport jax\n     9\timport jax.numpy as jnp\n    10\timport numpy as np\n    11\timport orbax.checkpoint as ocp\n    12\tfrom PIL import Image, ImageDraw\n    13\timport tyro\n    14\tfrom flax import nnx\n    15\t\n    16\tfrom jasmine.models.genie import GenieDiffusion\n    17\tfrom jasmine.utils.dataloader import get_dataloader\n    18\t\n    19\t\n    20\t@dataclass\n    21\tclass Args:\n    22\t    # Experiment\n    23\t    seed: int = 0\n    24\t    seq_len: int = 16\n    25\t    image_channels: int = 3\n    26\t    image_height: int = 90\n    27\t    image_width: int = 160\n    28\t    data_dir: str = ""data/coinrun_episodes""\n    29\t    checkpoint: str = """"\n    30\t    print_action_indices: bool = True\n    31\t    output_dir: str = ""gifs/""\n    32\t    # Sampling\n    33\t    batch_size: int = 1\n    34\t    start_frame: int = 1\n    35\t    diffusion_denoise_steps: int = 64\n    36\t    diffusion_corrupt_context_factor: float = 0.1\n    37\t    # Tokenizer checkpoint\n    38\t    tokenizer_dim: int = 512\n    39\t    tokenizer_ffn_dim: int = 2048\n    40\t    latent_patch_dim: int = 32\n    41\t    num_patch_latents: int = 1024\n    42\t    patch_size: int = 16\n    43\t    tokenizer_num_blocks: int = 4\n    44\t    tokenizer_num_heads: int = 8\n    45\t    # LAM checkpoint\n    46\t    lam_dim: int = 512\n    47\t    lam_ffn_dim: int = 2048\n    48\t    latent_action_dim: int = 32\n    49\t    num_actions: int = 6\n    50\t    lam_patch_size: int = 16\n    51\t    lam_num_blocks: int = 4\n    52\t    lam_num_heads: int = 8\n    53\t    use_gt_actions: bool = False\n    54\t    # Dynamics checkpoint\n    55\t    dyna_dim: int = 512\n    56\t    dyna_ffn_dim: int = 2048\n    57\t    dyna_num_blocks: int = 6\n    58\t    dyna_num_heads: int = 8\n    59\t    param_dtype = jnp.float32\n    60\t    dtype = jnp.bfloat16\n    61\t    use_flash_attention: bool = True\n    62\t\n    63\t\n    64\targs = tyro.cli(Args)\n    65\t\n    66\tif __name__ == ""__main__"":\n    67\t    """"""\n    68\t    Dimension keys:\n    69\t        B: batch size\n    70\t        T: number of input (conditioning) frames\n    71\t        N: number of patches per frame\n    72\t        S: sequence length\n    73\t        H: height\n    74\t        W: width\n    75\t        E: B * (S - 1)\n    76\t    """"""\n    77\t    jax.distributed.initialize()\n    78\t\n    79\t    rng = jax.random.key(args.seed)\n    80\t\n    81\t    # --- Load Genie checkpoint ---\n    82\t    rngs = nnx.Rngs(rng)\n    83\t    genie = GenieDiffusion(\n    84\t        # Tokenizer\n    85\t        in_dim=args.image_channels,\n    86\t        tokenizer_dim=args.tokenizer_dim,\n    87\t        tokenizer_ffn_dim=args.tokenizer_ffn_dim,\n    88\t        latent_patch_dim=args.latent_patch_dim,\n    89\t        num_patch_latents=args.num_patch_latents,\n    90\t        patch_size=args.patch_size,\n    91\t        tokenizer_num_blocks=args.tokenizer_num_blocks,\n    92\t        tokenizer_num_heads=args.tokenizer_num_heads,\n    93\t        # LAM\n    94\t        lam_dim=args.lam_dim,\n    95\t        lam_ffn_dim=args.lam_ffn_dim,\n    96\t        latent_action_dim=args.latent_action_dim,\n    97\t        num_actions=args.num_actions,\n    98\t        lam_patch_size=args.lam_patch_size,\n    99\t        lam_num_blocks=args.lam_num_blocks,\n   100\t        lam_num_heads=args.lam_num_heads,\n   101\t        lam_co_train=False,\n   102\t        use_gt_actions=args.use_gt_actions,\n   103\t        # Dynamics\n   104\t        dyna_dim=args.dyna_dim,\n   105\t        dyna_ffn_dim=args.dyna_ffn_dim,\n   106\t        dyna_num_blocks=args.dyna_num_blocks,\n   107\t        dyna_num_heads=args.dyna_num_heads,\n   108\t        param_dtype=args.param_dtype,\n   109\t        dtype=args.dtype,\n   110\t        use_flash_attention=args.use_flash_attention,\n   111\t        diffusion_denoise_steps=args.diffusion_denoise_steps,\n   112\t        # FIXME (f.srambical): implement spatiotemporal KV caching and set decode=True\n   113\t        decode=False,\n   114\t        rngs=rngs,\n   115\t    )\n   116\t\n   117\t    # Need to delete lam decoder for checkpoint loading\n   118\t    if not args.use_gt_actions:\n   119\t        assert genie.lam is not None\n   120\t        del genie.lam.decoder\n   121\t\n   122\t    handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n   123\t    handler_registry.add(\n   124\t        ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n   125\t    )\n   126\t    handler_registry.add(\n   127\t        ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n   128\t    )\n   129\t    checkpoint_options = ocp.CheckpointManagerOptions(\n   130\t        step_format_fixed_length=6,\n   131\t    )\n   132\t    checkpoint_manager = ocp.CheckpointManager(\n   133\t        args.checkpoint,\n   134\t        options=checkpoint_options,\n   135\t        handler_registry=handler_registry,\n   136\t    )\n   137\t\n   138\t    dummy_tx = optax.adamw(\n   139\t        learning_rate=optax.linear_schedule(0.0001, 0.0001, 10000),\n   140\t        b1=0.9,\n   141\t        b2=0.9,\n   142\t        weight_decay=1e-4,\n   143\t        mu_dtype=args.dtype,\n   144\t    )\n   145\t    dummy_optimizer = nnx.ModelAndOptimizer(genie, dummy_tx)\n   146\t\n   147\t    abstract_optimizer = nnx.eval_shape(lambda: dummy_optimizer)\n   148\t    abstract_optimizer_state = nnx.state(abstract_optimizer)\n   149\t    restored = checkpoint_manager.restore(\n   150\t        checkpoint_manager.latest_step(),\n   151\t        args=ocp.args.Composite(\n   152\t            model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n   153\t        ),\n   154\t    )\n   155\t    restored_optimizer_state = restored[""model_state""]\n   156\t    nnx.update(dummy_optimizer, restored_optimizer_state)\n   157\t\n   158\t    # --- Define sampling function ---\n   159\t    def _sampling_fn(model: GenieDiffusion, batch: dict) -> jax.Array:\n   160\t        """"""Runs Genie.sample with pre-defined generation hyper-parameters.""""""\n   161\t        frames = model.sample(\n   162\t            batch,\n   163\t            args.seq_len,\n   164\t            args.diffusion_denoise_steps,\n   165\t            args.diffusion_corrupt_context_factor,\n   166\t        )\n   167\t        return frames\n   168\t\n   169\t    # --- Define autoregressive sampling loop ---\n   170\t    def _autoreg_sample(\n   171\t        genie: GenieDiffusion, rng: jax.Array, batch: dict\n   172\t    ) -> jax.Array:\n   173\t        batch[""videos""] = batch[""videos""][:, : args.start_frame]\n   174\t        batch[""rng""] = rng\n   175\t        generated_vid_BSHWC = _sampling_fn(genie, batch)\n   176\t        return generated_vid_BSHWC\n   177\t\n   178\t    # --- Get video + latent actions ---\n   179\t    array_record_files = [\n   180\t        os.path.join(args.data_dir, x)\n   181\t        for x in os.listdir(args.data_dir)\n   182\t        if x.endswith("".array_record"")\n   183\t    ]\n   184\t    dataloader = get_dataloader(\n   185\t        array_record_files,\n   186\t        args.seq_len,\n   187\t        args.batch_size,\n   188\t        args.image_height,\n   189\t        args.image_width,\n   190\t        args.image_channels,\n   191\t        # We don't use workers in order to avoid grain shutdown issues (https://github.com/google/grain/issues/398)\n   192\t        num_workers=0,\n   193\t        prefetch_buffer_size=1,\n   194\t        seed=args.seed,\n   195\t    )\n   196\t    dataloader = iter(dataloader)\n   197\t    batch = next(dataloader)\n   198\t    gt_video = jnp.asarray(batch[""videos""], dtype=jnp.float32) / 255.0\n   199\t    print(f""DEBUG: gt_video shape: {gt_video.shape}"")\n   200\t    batch[""videos""] = gt_video.astype(args.dtype)\n   201\t    # Get latent actions for all videos in the batch\n   202\t    action_batch_E = None\n   203\t    if not args.use_gt_actions:\n   204\t        action_batch_E = genie.vq_encode(batch, training=False)\n   205\t        batch[""latent_actions""] = action_batch_E\n   206\t        print(f""DEBUG: action_batch_E shape: {action_batch_E.shape}"")\n   207\t\n   208\t    # --- Sample + evaluate video ---\n   209\t    recon_video_BSHWC = _autoreg_sample(genie, rng, batch)\n   210\t    print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   211\t    recon_video_BSHWC = recon_video_BSHWC.astype(jnp.float32)\n   212\t    print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   213\t\n   214\t    gt = gt_video.clip(0, 1)[:, args.start_frame :]\n   215\t    print(f""DEBUG: gt shape: {gt.shape}"")\n   216\t    recon = recon_video_BSHWC.clip(0, 1)[:, args.start_frame :]\n   217\t    print(f""DEBUG: recon shape: {recon.shape}"")\n   218\t\n   219\t    ssim_vmap = jax.vmap(pix.ssim, in_axes=(0, 0))\n   220\t    psnr_vmap = jax.vmap(pix.psnr, in_axes=(0, 0))\n   221\t    ssim = jnp.asarray(ssim_vmap(gt, recon))\n   222\t    psnr = jnp.asarray(psnr_vmap(gt, recon))\n   223\t    per_frame_ssim = ssim.mean(0)\n   224\t    per_frame_psnr = psnr.mean(0)\n   225\t    avg_ssim = ssim.mean()\n   226\t    avg_psnr = psnr.mean()\n   227\t\n   228\t    print(""Per-frame SSIM:\n"", per_frame_ssim)\n   229\t    print(""Per-frame PSNR:\n"", per_frame_psnr)\n   230\t\n   231\t    print(f""SSIM: {avg_ssim}"")\n   232\t    print(f""PSNR: {avg_psnr}"")\n   233\t\n   234\t    # --- Construct video ---\n   235\t    true_videos = (gt_video * 255).astype(np.uint8)\n   236\t    pred_videos = (recon_video_BSHWC * 255).astype(np.uint8)\n   237\t    video_comparison = np.zeros((2, *recon_video_BSHWC.shape), dtype=np.uint8)\n   238\t    video_comparison[0] = true_videos[:, : args.seq_len]\n   239\t    video_comparison[1] = pred_videos\n   240\t    frames = einops.rearrange(video_comparison, ""n b t h w c -> t (b h) (n w) c"")\n   241\t\n   242\t    # --- Save video ---\n   243\t    imgs = [Image.fromarray(img) for img in frames]\n   244\t    # Write actions on each frame, on each row (i.e., for each video in the batch, on the GT row)\n   245\t    B = batch[""videos""].shape[0]\n   246\t    if action_batch_E is not None:\n   247\t        action_batch_BSm11 = jnp.reshape(action_batch_E, (B, args.seq_len - 1, 1))\n   248\t    else:\n   249\t        action_batch_BSm11 = jnp.reshape(\n   250\t            batch[""actions""][:, :-1], (B, args.seq_len - 1, 1)\n   251\t        )\n   252\t    for t, img in enumerate(imgs[1:]):\n   253\t        d = ImageDraw.Draw(img)\n   254\t        for row in range(B):\n   255\t            if args.print_action_indices:\n   256\t                action = action_batch_BSm11[row, t, 0]\n   257\t                y_offset = row * batch[""videos""].shape[2] + 2\n   258\t                d.text((2, y_offset), f""{action}"", fill=255)\n   259\t\n   260\t    os.makedirs(args.output_dir, exist_ok=True)\n   261\t    imgs[0].save(\n   262\t        os.path.join(args.output_dir, f""generation_{time.time()}.gif""),\n   263\t        save_all=True,\n   264\t        append_images=imgs[1:],\n   265\t        duration=250,\n   266\t        loop=0,\n   267\t    )\n   268\t    # Save predicted videos as PNG image with all frames (skipping first 4) for each item in batch\n   269\t    skip = 4\n   270\t    # pred_videos: (B, T, H, W, C)\n   271\t    B, T, H, W, C = pred_videos.shape\n   272\t\n   273\t    for i in range(B):\n   274\t        # Predicted\n   275\t        pred_strip = np.concatenate(\n   276\t            [pred_videos[i, t] for t in range(skip, args.seq_len)], axis=1\n   277\t        )  # resulting shape: (H, (T-skip)*W, C)\n   278\t        pred_img = Image.fromarray(pred_strip)\n   279\t\n   280\t        # Write actions on each frame for the predicted strip\n   281\t        if args.print_action_indices:\n   282\t            d = ImageDraw.Draw(pred_img)\n   283\t            for t in range(skip, args.seq_len):\n   284\t                if action_batch_E is not None:\n   285\t                    action = int(action_batch_BSm11[i, t - 1, 0])\n   286\t                else:\n   287\t                    action = int(batch[""actions""][i, t - 1])\n   288\t                x_offset = (t - skip) * pred_videos.shape[3] + 2\n   289\t                y_offset = 2\n   290\t                d.text((x_offset, y_offset), f""{action}"", fill=255)\n   291\t\n   292\t        pred_img.save(\n   293\t            os.path.join(\n   294\t                args.output_dir, f""generation-{args.seed}-sample{i}-diffusion.png""\n   295\t            )\n   296\t        )\n   297\t        # GT\n   298\t        gt_strip = np.concatenate(\n   299\t            [true_videos[i, t] for t in range(skip, args.seq_len)], axis=1\n   300\t        )  # resulting shape: (H, (T-skip)*W, C)\n   301\t        gt_img = Image.fromarray(gt_strip)\n   302\t        gt_img.save(\n   303\t            os.path.join(args.output_dir, f""generation-{args.seed}-sample{i}-gt.png"")\n   304\t        )\n\n\n\n\n# Assistant\n```bash\ncat -n jasmine/baselines/diffusion/sample_diffusion.py | sed -n '195,206p'\n```\n\n   195      )\n   196      dataloader = iter(dataloader)\n   197      batch = next(dataloader)\n   198      gt_video = jnp.asarray(batch[""videos""], dtype=jnp.float32) / 255.0\n   199      print(f""DEBUG: gt_video shape: {gt_video.shape}"")\n   200      batch[""videos""] = gt_video.astype(args.dtype)\n   201      # Get latent actions for all videos in the batch\n   202      action_batch_E = None\n   203      if not args.use_gt_actions:\n   204          action_batch_E = genie.vq_encode(batch, training=False)\n   205          batch[""latent_actions""] = action_batch_E\n   206          print(f""DEBUG: action_batch_E shape: {action_batch_E.shape}"")\n\n\nsed -i '199,199d' jasmine/baselines/diffusion/sample_diffusion.py\n\n\nsed -i '205,205d' jasmine/baselines/diffusion/sample_diffusion.py\n\n\n\ncat -n jasmine/baselines/diffusion/sample_diffusion.py | sed -n '205,216p'\n\n\n\n   205\n   206      # --- Sample + evaluate video ---\n   207      recon_video_BSHWC = _autoreg_sample(genie, rng, batch)\n   208      print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   209      recon_video_BSHWC = recon_video_BSHWC.astype(jnp.float32)\n   210      print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   211\n   212      gt = gt_video.clip(0, 1)[:, args.start_frame :]\n   213      print(f""DEBUG: gt shape: {gt.shape}"")\n   214      recon = recon_video_BSHWC.clip(0, 1)[:, args.start_frame :]\n   215      print(f""DEBUG: recon shape: {recon.shape}"")\n   216\n\n\nsed -i '208,208d' jasmine/baselines/diffusion/sample_diffusion.py\n\nsed -i '209,209d' jasmine/baselines/diffusion/sample_diffusion.py\n\nsed -i '211,211d' jasmine/baselines/diffusion/sample_diffusion.py\n\nsed -i '212,212d' jasmine/baselines/diffusion/sample_diffusion.py",markdown,tab
+14,39521,"data/eval/unit_tests/eval_delete.md",14389,0,"",markdown,selection_mouse
+15,39522,"data/eval/unit_tests/eval_delete.md",14388,0,"",markdown,selection_command
+16,40601,"data/eval/unit_tests/eval_delete.md",14324,65,"sed -i '212,212d' jasmine/baselines/diffusion/sample_diffusion.py",markdown,selection_command
+17,42435,"data/eval/unit_tests/eval_delete.md",14388,0,"",markdown,selection_command
+18,44071,"data/eval/unit_tests/eval_delete.md",14324,65,"sed -i '212,212d' jasmine/baselines/diffusion/sample_diffusion.py",markdown,selection_command
+19,45309,"data/eval/unit_tests/eval_delete.md",14324,65,"sed -i '212,212d' jasmine/baselines/diffusion/sample_diffusion.py",markdown,selection_command
+20,46093,"data/eval/unit_tests/eval_delete.md",0,14389,"cat -n jasmine/baselines/diffusion/sample_diffusion.py\n\n\n\n     1\tfrom dataclasses import dataclass\n     2\timport time\n     3\timport os\n     4\timport optax\n     5\t\n     6\timport dm_pix as pix\n     7\timport einops\n     8\timport jax\n     9\timport jax.numpy as jnp\n    10\timport numpy as np\n    11\timport orbax.checkpoint as ocp\n    12\tfrom PIL import Image, ImageDraw\n    13\timport tyro\n    14\tfrom flax import nnx\n    15\t\n    16\tfrom jasmine.models.genie import GenieDiffusion\n    17\tfrom jasmine.utils.dataloader import get_dataloader\n    18\t\n    19\t\n    20\t@dataclass\n    21\tclass Args:\n    22\t    # Experiment\n    23\t    seed: int = 0\n    24\t    seq_len: int = 16\n    25\t    image_channels: int = 3\n    26\t    image_height: int = 90\n    27\t    image_width: int = 160\n    28\t    data_dir: str = ""data/coinrun_episodes""\n    29\t    checkpoint: str = """"\n    30\t    print_action_indices: bool = True\n    31\t    output_dir: str = ""gifs/""\n    32\t    # Sampling\n    33\t    batch_size: int = 1\n    34\t    start_frame: int = 1\n    35\t    diffusion_denoise_steps: int = 64\n    36\t    diffusion_corrupt_context_factor: float = 0.1\n    37\t    # Tokenizer checkpoint\n    38\t    tokenizer_dim: int = 512\n    39\t    tokenizer_ffn_dim: int = 2048\n    40\t    latent_patch_dim: int = 32\n    41\t    num_patch_latents: int = 1024\n    42\t    patch_size: int = 16\n    43\t    tokenizer_num_blocks: int = 4\n    44\t    tokenizer_num_heads: int = 8\n    45\t    # LAM checkpoint\n    46\t    lam_dim: int = 512\n    47\t    lam_ffn_dim: int = 2048\n    48\t    latent_action_dim: int = 32\n    49\t    num_actions: int = 6\n    50\t    lam_patch_size: int = 16\n    51\t    lam_num_blocks: int = 4\n    52\t    lam_num_heads: int = 8\n    53\t    use_gt_actions: bool = False\n    54\t    # Dynamics checkpoint\n    55\t    dyna_dim: int = 512\n    56\t    dyna_ffn_dim: int = 2048\n    57\t    dyna_num_blocks: int = 6\n    58\t    dyna_num_heads: int = 8\n    59\t    param_dtype = jnp.float32\n    60\t    dtype = jnp.bfloat16\n    61\t    use_flash_attention: bool = True\n    62\t\n    63\t\n    64\targs = tyro.cli(Args)\n    65\t\n    66\tif __name__ == ""__main__"":\n    67\t    """"""\n    68\t    Dimension keys:\n    69\t        B: batch size\n    70\t        T: number of input (conditioning) frames\n    71\t        N: number of patches per frame\n    72\t        S: sequence length\n    73\t        H: height\n    74\t        W: width\n    75\t        E: B * (S - 1)\n    76\t    """"""\n    77\t    jax.distributed.initialize()\n    78\t\n    79\t    rng = jax.random.key(args.seed)\n    80\t\n    81\t    # --- Load Genie checkpoint ---\n    82\t    rngs = nnx.Rngs(rng)\n    83\t    genie = GenieDiffusion(\n    84\t        # Tokenizer\n    85\t        in_dim=args.image_channels,\n    86\t        tokenizer_dim=args.tokenizer_dim,\n    87\t        tokenizer_ffn_dim=args.tokenizer_ffn_dim,\n    88\t        latent_patch_dim=args.latent_patch_dim,\n    89\t        num_patch_latents=args.num_patch_latents,\n    90\t        patch_size=args.patch_size,\n    91\t        tokenizer_num_blocks=args.tokenizer_num_blocks,\n    92\t        tokenizer_num_heads=args.tokenizer_num_heads,\n    93\t        # LAM\n    94\t        lam_dim=args.lam_dim,\n    95\t        lam_ffn_dim=args.lam_ffn_dim,\n    96\t        latent_action_dim=args.latent_action_dim,\n    97\t        num_actions=args.num_actions,\n    98\t        lam_patch_size=args.lam_patch_size,\n    99\t        lam_num_blocks=args.lam_num_blocks,\n   100\t        lam_num_heads=args.lam_num_heads,\n   101\t        lam_co_train=False,\n   102\t        use_gt_actions=args.use_gt_actions,\n   103\t        # Dynamics\n   104\t        dyna_dim=args.dyna_dim,\n   105\t        dyna_ffn_dim=args.dyna_ffn_dim,\n   106\t        dyna_num_blocks=args.dyna_num_blocks,\n   107\t        dyna_num_heads=args.dyna_num_heads,\n   108\t        param_dtype=args.param_dtype,\n   109\t        dtype=args.dtype,\n   110\t        use_flash_attention=args.use_flash_attention,\n   111\t        diffusion_denoise_steps=args.diffusion_denoise_steps,\n   112\t        # FIXME (f.srambical): implement spatiotemporal KV caching and set decode=True\n   113\t        decode=False,\n   114\t        rngs=rngs,\n   115\t    )\n   116\t\n   117\t    # Need to delete lam decoder for checkpoint loading\n   118\t    if not args.use_gt_actions:\n   119\t        assert genie.lam is not None\n   120\t        del genie.lam.decoder\n   121\t\n   122\t    handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n   123\t    handler_registry.add(\n   124\t        ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n   125\t    )\n   126\t    handler_registry.add(\n   127\t        ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n   128\t    )\n   129\t    checkpoint_options = ocp.CheckpointManagerOptions(\n   130\t        step_format_fixed_length=6,\n   131\t    )\n   132\t    checkpoint_manager = ocp.CheckpointManager(\n   133\t        args.checkpoint,\n   134\t        options=checkpoint_options,\n   135\t        handler_registry=handler_registry,\n   136\t    )\n   137\t\n   138\t    dummy_tx = optax.adamw(\n   139\t        learning_rate=optax.linear_schedule(0.0001, 0.0001, 10000),\n   140\t        b1=0.9,\n   141\t        b2=0.9,\n   142\t        weight_decay=1e-4,\n   143\t        mu_dtype=args.dtype,\n   144\t    )\n   145\t    dummy_optimizer = nnx.ModelAndOptimizer(genie, dummy_tx)\n   146\t\n   147\t    abstract_optimizer = nnx.eval_shape(lambda: dummy_optimizer)\n   148\t    abstract_optimizer_state = nnx.state(abstract_optimizer)\n   149\t    restored = checkpoint_manager.restore(\n   150\t        checkpoint_manager.latest_step(),\n   151\t        args=ocp.args.Composite(\n   152\t            model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n   153\t        ),\n   154\t    )\n   155\t    restored_optimizer_state = restored[""model_state""]\n   156\t    nnx.update(dummy_optimizer, restored_optimizer_state)\n   157\t\n   158\t    # --- Define sampling function ---\n   159\t    def _sampling_fn(model: GenieDiffusion, batch: dict) -> jax.Array:\n   160\t        """"""Runs Genie.sample with pre-defined generation hyper-parameters.""""""\n   161\t        frames = model.sample(\n   162\t            batch,\n   163\t            args.seq_len,\n   164\t            args.diffusion_denoise_steps,\n   165\t            args.diffusion_corrupt_context_factor,\n   166\t        )\n   167\t        return frames\n   168\t\n   169\t    # --- Define autoregressive sampling loop ---\n   170\t    def _autoreg_sample(\n   171\t        genie: GenieDiffusion, rng: jax.Array, batch: dict\n   172\t    ) -> jax.Array:\n   173\t        batch[""videos""] = batch[""videos""][:, : args.start_frame]\n   174\t        batch[""rng""] = rng\n   175\t        generated_vid_BSHWC = _sampling_fn(genie, batch)\n   176\t        return generated_vid_BSHWC\n   177\t\n   178\t    # --- Get video + latent actions ---\n   179\t    array_record_files = [\n   180\t        os.path.join(args.data_dir, x)\n   181\t        for x in os.listdir(args.data_dir)\n   182\t        if x.endswith("".array_record"")\n   183\t    ]\n   184\t    dataloader = get_dataloader(\n   185\t        array_record_files,\n   186\t        args.seq_len,\n   187\t        args.batch_size,\n   188\t        args.image_height,\n   189\t        args.image_width,\n   190\t        args.image_channels,\n   191\t        # We don't use workers in order to avoid grain shutdown issues (https://github.com/google/grain/issues/398)\n   192\t        num_workers=0,\n   193\t        prefetch_buffer_size=1,\n   194\t        seed=args.seed,\n   195\t    )\n   196\t    dataloader = iter(dataloader)\n   197\t    batch = next(dataloader)\n   198\t    gt_video = jnp.asarray(batch[""videos""], dtype=jnp.float32) / 255.0\n   199\t    print(f""DEBUG: gt_video shape: {gt_video.shape}"")\n   200\t    batch[""videos""] = gt_video.astype(args.dtype)\n   201\t    # Get latent actions for all videos in the batch\n   202\t    action_batch_E = None\n   203\t    if not args.use_gt_actions:\n   204\t        action_batch_E = genie.vq_encode(batch, training=False)\n   205\t        batch[""latent_actions""] = action_batch_E\n   206\t        print(f""DEBUG: action_batch_E shape: {action_batch_E.shape}"")\n   207\t\n   208\t    # --- Sample + evaluate video ---\n   209\t    recon_video_BSHWC = _autoreg_sample(genie, rng, batch)\n   210\t    print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   211\t    recon_video_BSHWC = recon_video_BSHWC.astype(jnp.float32)\n   212\t    print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   213\t\n   214\t    gt = gt_video.clip(0, 1)[:, args.start_frame :]\n   215\t    print(f""DEBUG: gt shape: {gt.shape}"")\n   216\t    recon = recon_video_BSHWC.clip(0, 1)[:, args.start_frame :]\n   217\t    print(f""DEBUG: recon shape: {recon.shape}"")\n   218\t\n   219\t    ssim_vmap = jax.vmap(pix.ssim, in_axes=(0, 0))\n   220\t    psnr_vmap = jax.vmap(pix.psnr, in_axes=(0, 0))\n   221\t    ssim = jnp.asarray(ssim_vmap(gt, recon))\n   222\t    psnr = jnp.asarray(psnr_vmap(gt, recon))\n   223\t    per_frame_ssim = ssim.mean(0)\n   224\t    per_frame_psnr = psnr.mean(0)\n   225\t    avg_ssim = ssim.mean()\n   226\t    avg_psnr = psnr.mean()\n   227\t\n   228\t    print(""Per-frame SSIM:\n"", per_frame_ssim)\n   229\t    print(""Per-frame PSNR:\n"", per_frame_psnr)\n   230\t\n   231\t    print(f""SSIM: {avg_ssim}"")\n   232\t    print(f""PSNR: {avg_psnr}"")\n   233\t\n   234\t    # --- Construct video ---\n   235\t    true_videos = (gt_video * 255).astype(np.uint8)\n   236\t    pred_videos = (recon_video_BSHWC * 255).astype(np.uint8)\n   237\t    video_comparison = np.zeros((2, *recon_video_BSHWC.shape), dtype=np.uint8)\n   238\t    video_comparison[0] = true_videos[:, : args.seq_len]\n   239\t    video_comparison[1] = pred_videos\n   240\t    frames = einops.rearrange(video_comparison, ""n b t h w c -> t (b h) (n w) c"")\n   241\t\n   242\t    # --- Save video ---\n   243\t    imgs = [Image.fromarray(img) for img in frames]\n   244\t    # Write actions on each frame, on each row (i.e., for each video in the batch, on the GT row)\n   245\t    B = batch[""videos""].shape[0]\n   246\t    if action_batch_E is not None:\n   247\t        action_batch_BSm11 = jnp.reshape(action_batch_E, (B, args.seq_len - 1, 1))\n   248\t    else:\n   249\t        action_batch_BSm11 = jnp.reshape(\n   250\t            batch[""actions""][:, :-1], (B, args.seq_len - 1, 1)\n   251\t        )\n   252\t    for t, img in enumerate(imgs[1:]):\n   253\t        d = ImageDraw.Draw(img)\n   254\t        for row in range(B):\n   255\t            if args.print_action_indices:\n   256\t                action = action_batch_BSm11[row, t, 0]\n   257\t                y_offset = row * batch[""videos""].shape[2] + 2\n   258\t                d.text((2, y_offset), f""{action}"", fill=255)\n   259\t\n   260\t    os.makedirs(args.output_dir, exist_ok=True)\n   261\t    imgs[0].save(\n   262\t        os.path.join(args.output_dir, f""generation_{time.time()}.gif""),\n   263\t        save_all=True,\n   264\t        append_images=imgs[1:],\n   265\t        duration=250,\n   266\t        loop=0,\n   267\t    )\n   268\t    # Save predicted videos as PNG image with all frames (skipping first 4) for each item in batch\n   269\t    skip = 4\n   270\t    # pred_videos: (B, T, H, W, C)\n   271\t    B, T, H, W, C = pred_videos.shape\n   272\t\n   273\t    for i in range(B):\n   274\t        # Predicted\n   275\t        pred_strip = np.concatenate(\n   276\t            [pred_videos[i, t] for t in range(skip, args.seq_len)], axis=1\n   277\t        )  # resulting shape: (H, (T-skip)*W, C)\n   278\t        pred_img = Image.fromarray(pred_strip)\n   279\t\n   280\t        # Write actions on each frame for the predicted strip\n   281\t        if args.print_action_indices:\n   282\t            d = ImageDraw.Draw(pred_img)\n   283\t            for t in range(skip, args.seq_len):\n   284\t                if action_batch_E is not None:\n   285\t                    action = int(action_batch_BSm11[i, t - 1, 0])\n   286\t                else:\n   287\t                    action = int(batch[""actions""][i, t - 1])\n   288\t                x_offset = (t - skip) * pred_videos.shape[3] + 2\n   289\t                y_offset = 2\n   290\t                d.text((x_offset, y_offset), f""{action}"", fill=255)\n   291\t\n   292\t        pred_img.save(\n   293\t            os.path.join(\n   294\t                args.output_dir, f""generation-{args.seed}-sample{i}-diffusion.png""\n   295\t            )\n   296\t        )\n   297\t        # GT\n   298\t        gt_strip = np.concatenate(\n   299\t            [true_videos[i, t] for t in range(skip, args.seq_len)], axis=1\n   300\t        )  # resulting shape: (H, (T-skip)*W, C)\n   301\t        gt_img = Image.fromarray(gt_strip)\n   302\t        gt_img.save(\n   303\t            os.path.join(args.output_dir, f""generation-{args.seed}-sample{i}-gt.png"")\n   304\t        )\n\n\n\n\n# Assistant\n```bash\ncat -n jasmine/baselines/diffusion/sample_diffusion.py | sed -n '195,206p'\n```\n\n   195      )\n   196      dataloader = iter(dataloader)\n   197      batch = next(dataloader)\n   198      gt_video = jnp.asarray(batch[""videos""], dtype=jnp.float32) / 255.0\n   199      print(f""DEBUG: gt_video shape: {gt_video.shape}"")\n   200      batch[""videos""] = gt_video.astype(args.dtype)\n   201      # Get latent actions for all videos in the batch\n   202      action_batch_E = None\n   203      if not args.use_gt_actions:\n   204          action_batch_E = genie.vq_encode(batch, training=False)\n   205          batch[""latent_actions""] = action_batch_E\n   206          print(f""DEBUG: action_batch_E shape: {action_batch_E.shape}"")\n\n\nsed -i '199,199d' jasmine/baselines/diffusion/sample_diffusion.py\n\n\nsed -i '205,205d' jasmine/baselines/diffusion/sample_diffusion.py\n\n\n\ncat -n jasmine/baselines/diffusion/sample_diffusion.py | sed -n '205,216p'\n\n\n\n   205\n   206      # --- Sample + evaluate video ---\n   207      recon_video_BSHWC = _autoreg_sample(genie, rng, batch)\n   208      print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   209      recon_video_BSHWC = recon_video_BSHWC.astype(jnp.float32)\n   210      print(f""DEBUG: recon_video_BSHWC shape: {recon_video_BSHWC.shape}"")\n   211\n   212      gt = gt_video.clip(0, 1)[:, args.start_frame :]\n   213      print(f""DEBUG: gt shape: {gt.shape}"")\n   214      recon = recon_video_BSHWC.clip(0, 1)[:, args.start_frame :]\n   215      print(f""DEBUG: recon shape: {recon.shape}"")\n   216\n\n\nsed -i '208,208d' jasmine/baselines/diffusion/sample_diffusion.py\n\nsed -i '209,209d' jasmine/baselines/diffusion/sample_diffusion.py\n\nsed -i '211,211d' jasmine/baselines/diffusion/sample_diffusion.py\n\nsed -i '212,212d' jasmine/baselines/diffusion/sample_diffusion.py",markdown,selection_command
+21,102756,"TERMINAL",0,0,"",,terminal_command
+22,216425,"data/eval/unit_tests/eval_delete.md",14389,0,"```\n\n# User\n<stdout>\n</stdout>\n\n",markdown,content
+23,216425,"data/eval/unit_tests/eval_delete.md",14324,0,"# Assistant\n```bash\n",markdown,content
+24,216425,"data/eval/unit_tests/eval_delete.md",14323,0,"```\n\n# User\n<stdout>\n</stdout>\n\n",markdown,content
+25,216425,"data/eval/unit_tests/eval_delete.md",14257,0,"# Assistant\n```bash\n",markdown,content
+26,216425,"data/eval/unit_tests/eval_delete.md",14256,0,"```\n\n# User\n<stdout>\n</stdout>\n\n",markdown,content
+27,216425,"data/eval/unit_tests/eval_delete.md",14190,0,"# Assistant\n```bash\n",markdown,content
+28,216425,"data/eval/unit_tests/eval_delete.md",14189,0,"```\n\n# User\n<stdout>\n</stdout>\n\n",markdown,content
+29,216425,"data/eval/unit_tests/eval_delete.md",14123,0,"# Assistant\n```bash\n",markdown,content
+30,216425,"data/eval/unit_tests/eval_delete.md",14121,0,"</stdout>\n",markdown,content
+31,216425,"data/eval/unit_tests/eval_delete.md",13517,1,"# User\n<stdout>",markdown,content
+32,216425,"data/eval/unit_tests/eval_delete.md",13516,0,"```\n",markdown,content
+33,216425,"data/eval/unit_tests/eval_delete.md",13441,0,"# Assistant\n```bash\n",markdown,content
+34,216425,"data/eval/unit_tests/eval_delete.md",13439,0,"# User\n<stdout>\n</stdout>\n",markdown,content
+35,216425,"data/eval/unit_tests/eval_delete.md",13438,0,"```\n",markdown,content
+36,216425,"data/eval/unit_tests/eval_delete.md",13372,0,"\n# Assistant\n```bash\n",markdown,content
+37,216425,"data/eval/unit_tests/eval_delete.md",13371,0,"# User\n<stdout>\n</stdout>\n",markdown,content
+38,216425,"data/eval/unit_tests/eval_delete.md",13370,0,"```\n",markdown,content
+39,216425,"data/eval/unit_tests/eval_delete.md",13304,0,"# Assistant\n```bash\n",markdown,content
+40,216425,"data/eval/unit_tests/eval_delete.md",13302,0,"</stdout>\n",markdown,content
+41,216425,"data/eval/unit_tests/eval_delete.md",12668,0,"# User\n<stdout>\n",markdown,content
+42,216425,"data/eval/unit_tests/eval_delete.md",12564,1,"</stdout>",markdown,content
+43,216425,"data/eval/unit_tests/eval_delete.md",56,1,"# User\n<stdout>",markdown,content
+44,216425,"data/eval/unit_tests/eval_delete.md",55,0,"```\n",markdown,content
+45,216425,"data/eval/unit_tests/eval_delete.md",0,0,"# Assistant\n```bash\n",markdown,content
+46,216429,"data/eval/unit_tests/eval_delete.md",14817,1,"",markdown,content
+47,216429,"data/eval/unit_tests/eval_delete.md",14721,68,"sed -i '212,212d' jasmine/baselines/diffusion/sample_diffusion.py\n```",markdown,content
+48,379094,"data/eval/unit_tests/eval_delete.md",14700,0,"",markdown,selection_mouse
+49,401814,"TERMINAL",0,0,"clear",,terminal_command
+50,401814,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login1:~/Projects/tab-model-eval",,terminal_output
+51,404526,"TERMINAL",0,0,"clear",,terminal_command
+52,464515,"data/eval/unit_tests/eval_delete.md",0,0,"",markdown,tab
+53,464516,"data/eval/unit_tests/eval_delete.md",14593,0,"",markdown,selection_mouse
+54,464570,"data/eval/unit_tests/eval_delete.md",14592,0,"",markdown,selection_command
+55,793294,"data/eval/handcrafted/add_import_hard.md",0,0,"# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nTraceback (most recent call last):\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 47, in <module>\n    args = tyro.cli(Args)\n           ^^^^\nNameError: name 'tyro' is not defined\n</stdout>\n\n# Assistant\n```bash\ncat -n src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n     1  import json\n     2  import os\n     3  from dataclasses import dataclass\n     4\n     5\n     6  @dataclass\n     7  class Args:\n     8      input_file: str = ""data/eval/val/validation.jsonl""\n     9      output_file: str = ""data/eval/val/validation_testcases.jsonl""\n    10      task_name: str = ""validation_set""\n    11\n    12\n    13  def convert_to_incremental_jsonl(line, output_file, task_name):\n    14      data = json.loads(line)\n    15\n    16      system_prompt = data.get(""system_prompt"", """")\n    17      conversations = data.get(""conversations"", [])\n    18\n    19      output_lines = []\n    20      context = []\n    21\n    22      for i, conv in enumerate(conversations):\n    23          # Convert ""Assistant""/""User"" to ""assistant""/""user""\n    24          role = conv[""from""].lower()\n    25          content = conv[""value""]\n    26\n    27          # Add current message to context\n    28          context.append({""role"": role, ""content"": content})\n    29\n    30          if i > 0 and role == ""assistant"":\n    31              task_entry = {\n    32                  ""system_prompt"": system_prompt,\n    33                  ""task_id"": f""{task_name}/{i-1}"",\n    34                  ""context"": context[:-1].copy(),\n    35                  ""expected_final_response"": content,\n    36              }\n    37              output_lines.append(task_entry)\n    38\n    39      with open(output_file, ""a"") as f:\n    40          for line in output_lines:\n    41              f.write(json.dumps(line) + ""\n"")\n    42\n    43      print(f""Converted {len(output_lines)} entries to {output_file}"")\n    44\n    45\n    46  if __name__ == ""__main__"":\n    47      args = tyro.cli(Args)\n    48\n    49      if os.path.exists(args.output_file):\n    50          os.remove(args.output_file)\n    51\n    52      # iterate over jsonl file and convert to incremental jsonl\n    53      conversation_id = 0\n    54      with open(args.input_file, ""r"") as f:\n    55          for line in f:\n    56              task_name = f""conversation_{conversation_id}/{args.task_name}""\n    57              convert_to_incremental_jsonl(line, args.output_file, task_name)\n    58              conversation_id += 1\n</stdout>\n\n# Assistant\n```bash\nsed -i '1i\import tyro' src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n</stdout>\n\n\n# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 2 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 6 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\n</stdout>\n",markdown,tab
+56,821305,"data/eval/handcrafted/add_import_easy.md",0,0,"# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nTraceback (most recent call last):\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 57, in <module>\n    convert_to_incremental_jsonl(line, args.output_file, task_name)\n    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 14, in convert_to_incremental_jsonl\n    data = json.loads(line)\n           ^^^^\nNameError: name 'json' is not defined. Did you forget to import 'json'?\n</stdout>\n\n\n# Assistant\n```bash\ncat -n src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n     1  import tyro\n     2  import os\n     3  from dataclasses import dataclass\n     4\n     5\n     6  @dataclass\n     7  class Args:\n     8      input_file: str = ""data/eval/val/validation.jsonl""\n     9      output_file: str = ""data/eval/val/validation_testcases.jsonl""\n    10      task_name: str = ""validation_set""\n    11\n    12\n    13  def convert_to_incremental_jsonl(line, output_file, task_name):\n    14      data = json.loads(line)\n    15\n    16      system_prompt = data.get(""system_prompt"", """")\n    17      conversations = data.get(""conversations"", [])\n    18\n    19      output_lines = []\n    20      context = []\n    21\n    22      for i, conv in enumerate(conversations):\n    23          # Convert ""Assistant""/""User"" to ""assistant""/""user""\n    24          role = conv[""from""].lower()\n    25          content = conv[""value""]\n    26\n    27          # Add current message to context\n    28          context.append({""role"": role, ""content"": content})\n    29\n    30          if i > 0 and role == ""assistant"":\n    31              task_entry = {\n    32                  ""system_prompt"": system_prompt,\n    33                  ""task_id"": f""{task_name}/{i-1}"",\n    34                  ""context"": context[:-1].copy(),\n    35                  ""expected_final_response"": content,\n    36              }\n    37              output_lines.append(task_entry)\n    38\n    39      with open(output_file, ""a"") as f:\n    40          for line in output_lines:\n    41              f.write(json.dumps(line) + ""\n"")\n    42\n    43      print(f""Converted {len(output_lines)} entries to {output_file}"")\n    44\n    45\n    46  if __name__ == ""__main__"":\n    47      args = tyro.cli(Args)\n    48\n    49      if os.path.exists(args.output_file):\n    50          os.remove(args.output_file)\n    51\n    52      # iterate over jsonl file and convert to incremental jsonl\n    53      conversation_id = 0\n    54      with open(args.input_file, ""r"") as f:\n    55          for line in f:\n    56              task_name = f""conversation_{conversation_id}/{args.task_name}""\n    57              convert_to_incremental_jsonl(line, args.output_file, task_name)\n    58              conversation_id += 1\n</stdout>\n\n# Assistant\n```bash\nsed -i '1i\import json' src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n</stdout>\n\n# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 2 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 6 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\n</stdout>\n",markdown,tab
+57,829092,"data/eval/unit_tests/add_import_easy.md",0,0,"# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nTraceback (most recent call last):\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 57, in <module>\n    convert_to_incremental_jsonl(line, args.output_file, task_name)\n    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 14, in convert_to_incremental_jsonl\n    data = json.loads(line)\n           ^^^^\nNameError: name 'json' is not defined. Did you forget to import 'json'?\n</stdout>\n\n\n# Assistant\n```bash\ncat -n src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n     1  import tyro\n     2  import os\n     3  from dataclasses import dataclass\n     4\n     5\n     6  @dataclass\n     7  class Args:\n     8      input_file: str = ""data/eval/val/validation.jsonl""\n     9      output_file: str = ""data/eval/val/validation_testcases.jsonl""\n    10      task_name: str = ""validation_set""\n    11\n    12\n    13  def convert_to_incremental_jsonl(line, output_file, task_name):\n    14      data = json.loads(line)\n    15\n    16      system_prompt = data.get(""system_prompt"", """")\n    17      conversations = data.get(""conversations"", [])\n    18\n    19      output_lines = []\n    20      context = []\n    21\n    22      for i, conv in enumerate(conversations):\n    23          # Convert ""Assistant""/""User"" to ""assistant""/""user""\n    24          role = conv[""from""].lower()\n    25          content = conv[""value""]\n    26\n    27          # Add current message to context\n    28          context.append({""role"": role, ""content"": content})\n    29\n    30          if i > 0 and role == ""assistant"":\n    31              task_entry = {\n    32                  ""system_prompt"": system_prompt,\n    33                  ""task_id"": f""{task_name}/{i-1}"",\n    34                  ""context"": context[:-1].copy(),\n    35                  ""expected_final_response"": content,\n    36              }\n    37              output_lines.append(task_entry)\n    38\n    39      with open(output_file, ""a"") as f:\n    40          for line in output_lines:\n    41              f.write(json.dumps(line) + ""\n"")\n    42\n    43      print(f""Converted {len(output_lines)} entries to {output_file}"")\n    44\n    45\n    46  if __name__ == ""__main__"":\n    47      args = tyro.cli(Args)\n    48\n    49      if os.path.exists(args.output_file):\n    50          os.remove(args.output_file)\n    51\n    52      # iterate over jsonl file and convert to incremental jsonl\n    53      conversation_id = 0\n    54      with open(args.input_file, ""r"") as f:\n    55          for line in f:\n    56              task_name = f""conversation_{conversation_id}/{args.task_name}""\n    57              convert_to_incremental_jsonl(line, args.output_file, task_name)\n    58              conversation_id += 1\n</stdout>\n\n# Assistant\n```bash\nsed -i '1i\import json' src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n</stdout>\n\n# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 2 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 6 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\n</stdout>\n",markdown,tab
+58,844708,"data/eval/unit_tests/eval_terminal_invocation.md",0,0,"# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nTraceback (most recent call last):\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 57, in <module>\n    convert_to_incremental_jsonl(line, args.output_file, task_name)\n    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File ""/fast/home/mihir.mahajan/Projects/tab-model-eval/src/input_pipeline/validation_to_testcases.py"", line 14, in convert_to_incremental_jsonl\n    data = json.loads(line)\n           ^^^^\nNameError: name 'json' is not defined. Did you forget to import 'json'?\n</stdout>\n\n\n# Assistant\n```bash\ncat -n src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n     1  import tyro\n     2  import os\n     3  from dataclasses import dataclass\n     4\n     5\n     6  @dataclass\n     7  class Args:\n     8      input_file: str = ""data/eval/val/validation.jsonl""\n     9      output_file: str = ""data/eval/val/validation_testcases.jsonl""\n    10      task_name: str = ""validation_set""\n    11\n    12\n    13  def convert_to_incremental_jsonl(line, output_file, task_name):\n    14      data = json.loads(line)\n    15\n    16      system_prompt = data.get(""system_prompt"", """")\n    17      conversations = data.get(""conversations"", [])\n    18\n    19      output_lines = []\n    20      context = []\n    21\n    22      for i, conv in enumerate(conversations):\n    23          # Convert ""Assistant""/""User"" to ""assistant""/""user""\n    24          role = conv[""from""].lower()\n    25          content = conv[""value""]\n    26\n    27          # Add current message to context\n    28          context.append({""role"": role, ""content"": content})\n    29\n    30          if i > 0 and role == ""assistant"":\n    31              task_entry = {\n    32                  ""system_prompt"": system_prompt,\n    33                  ""task_id"": f""{task_name}/{i-1}"",\n    34                  ""context"": context[:-1].copy(),\n    35                  ""expected_final_response"": content,\n    36              }\n    37              output_lines.append(task_entry)\n    38\n    39      with open(output_file, ""a"") as f:\n    40          for line in output_lines:\n    41              f.write(json.dumps(line) + ""\n"")\n    42\n    43      print(f""Converted {len(output_lines)} entries to {output_file}"")\n    44\n    45\n    46  if __name__ == ""__main__"":\n    47      args = tyro.cli(Args)\n    48\n    49      if os.path.exists(args.output_file):\n    50          os.remove(args.output_file)\n    51\n    52      # iterate over jsonl file and convert to incremental jsonl\n    53      conversation_id = 0\n    54      with open(args.input_file, ""r"") as f:\n    55          for line in f:\n    56              task_name = f""conversation_{conversation_id}/{args.task_name}""\n    57              convert_to_incremental_jsonl(line, args.output_file, task_name)\n    58              conversation_id += 1\n</stdout>\n\n# Assistant\n```bash\nsed -i '1i\import json' src/input_pipeline/validation_to_testcases.py\n```\n\n# User\n<stdout>\n</stdout>\n\n# Assistant\n```bash\npython src/input_pipeline/validation_to_testcases.py --input_file data/temp/validation.jsonl --output_file data/temp/validation_testcases.jsonl\n```\n\n# User\n<stdout>\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 7 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 8 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 2 entries to data/temp/validation_testcases.jsonl\nConverted 4 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 5 entries to data/temp/validation_testcases.jsonl\nConverted 6 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\nConverted 3 entries to data/temp/validation_testcases.jsonl\n</stdout>\n",markdown,tab
+59,855759,"data/eval/unit_tests/eval_terminal_invocation.md",3197,0,"",markdown,selection_mouse
+60,860607,"data/eval/unit_tests/eval_terminal_invocation.md",3066,0,"",markdown,selection_mouse
+61,861097,"data/eval/unit_tests/eval_terminal_invocation.md",3108,0,"",markdown,selection_mouse
+62,861876,"data/eval/unit_tests/eval_terminal_invocation.md",3120,0,"",markdown,selection_mouse
+63,861879,"data/eval/unit_tests/eval_terminal_invocation.md",3119,0,"",markdown,selection_command
+64,861899,"data/eval/unit_tests/eval_terminal_invocation.md",3119,1,"`",markdown,selection_mouse
+65,861905,"data/eval/unit_tests/eval_terminal_invocation.md",3120,0,"",markdown,selection_command
+66,862508,"data/eval/unit_tests/eval_terminal_invocation.md",3128,0,"",markdown,selection_mouse
+67,862519,"data/eval/unit_tests/eval_terminal_invocation.md",3127,0,"",markdown,selection_command
+68,863436,"data/eval/unit_tests/eval_terminal_invocation.md",3120,0,"",markdown,selection_mouse
+69,863437,"data/eval/unit_tests/eval_terminal_invocation.md",3119,0,"",markdown,selection_command
+70,863920,"data/eval/unit_tests/eval_terminal_invocation.md",3128,0,"",markdown,selection_mouse
+71,863921,"data/eval/unit_tests/eval_terminal_invocation.md",3127,0,"",markdown,selection_command
+72,864352,"data/eval/unit_tests/eval_terminal_invocation.md",3137,0,"",markdown,selection_mouse
+73,864352,"data/eval/unit_tests/eval_terminal_invocation.md",3136,0,"",markdown,selection_command
+74,864767,"data/eval/unit_tests/eval_terminal_invocation.md",3147,0,"",markdown,selection_mouse
+75,864767,"data/eval/unit_tests/eval_terminal_invocation.md",3146,0,"",markdown,selection_command
+76,1068619,"data/eval/unit_tests/eval_terminal_invocation.md",0,0,"",markdown,tab
+77,1068620,"data/eval/unit_tests/eval_terminal_invocation.md",3324,0,"",markdown,selection_mouse
+78,1068629,"data/eval/unit_tests/eval_terminal_invocation.md",3323,0,"",markdown,selection_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-41e637ef-086e-4e89-977d-363e70d7d3151764500068578-2025_11_30-11.55.34.47/source.csv

@@ -0,0 +1,74 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,2,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=32\n#SBATCH --ntasks-per-node=4\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=logs/logs_training/%x_%j.log\n#SBATCH --error=logs/logs_training/%x_%j.log\n#SBATCH --job-name=train_tokenizer_batch_size_scaling_32_node\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_new/open_ai_minecraft_arrayrecords/10fps_160x90\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/checkpoints/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\nenv | grep SLURM\n\nsrun python jasmine/train_tokenizer.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=1536 \\n    --min_lr=0 \\n    --max_lr=3e-4 \\n    --log_image_interval=250 \\n    --log \\n    --name=tokenizer-batch-size-scaling-32-node-sqrt-lr-$slurm_job_id \\n    --tags tokenizer batch-size-scaling 32-node sqrt-lr-scaling \\n    --entity instant-uv \\n    --project jafar \\n    --data_dir $array_records_dir\n",shellscript,tab
+2,4925,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"11:55:34 AM [info] Activating crowd-code\n11:55:34 AM [info] Recording started\n11:55:34 AM [info] Initializing git provider using file system watchers...\n11:55:35 AM [info] Git repository found\n11:55:35 AM [info] Git provider initialized successfully\n11:55:35 AM [info] Initial git state: [object Object]\n",Log,tab
+3,46263,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",0,0,"",shellscript,tab
+4,46267,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",1179,0,"",shellscript,selection_mouse
+5,49573,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=4\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\n#SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_new/open_ai_minecraft_arrayrecords/10fps_160x90\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/checkpoints/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\nenv | grep SLURM\n\nsrun python jasmine/train_tokenizer.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=48 \\n    --min_lr=0 \\n    --max_lr=1e-4 \\n    --log_image_interval=1000 \\n    --log_checkpoint_interval=1000 \\n    --log \\n    --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\n    --tags tokenizer batch-size-scaling 1-node sqrt-lr \\n    --entity instant-uv \\n    --project jafar \\n    --data_dir $array_records_dir\n",shellscript,tab
+6,51445,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",0,0,"",shellscript,tab
+7,52258,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",943,0,"",shellscript,selection_mouse
+8,52986,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",944,0,"",shellscript,selection_command
+9,53443,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",943,1,"",shellscript,content
+10,53514,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",942,1,"",shellscript,content
+11,53710,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",941,1,"",shellscript,content
+12,54083,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",941,0,"1",shellscript,content
+13,54084,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",942,0,"",shellscript,selection_keyboard
+14,54132,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",942,0,"0",shellscript,content
+15,54132,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",943,0,"",shellscript,selection_keyboard
+16,54264,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",943,0,"0",shellscript,content
+17,54264,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",944,0,"",shellscript,selection_keyboard
+18,54481,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",944,0,"0",shellscript,content
+19,54481,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",945,0,"",shellscript,selection_keyboard
+20,56715,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",947,0,"\n    --log_checkpoint_interval=1000 \",shellscript,content
+21,58155,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",944,0,"",shellscript,selection_command
+22,61500,"TERMINAL",0,0,"",,terminal_command
+23,86690,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch",0,0,"",shellscript,tab
+24,91839,"TERMINAL",0,0,"",,terminal_focus
+25,92728,"TERMINAL",0,0,"source /home/mihir.mahajan/Projects/tab-model-eval/.venv/bin/activate",,terminal_command
+26,92885,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+27,94149,"TERMINAL",0,0,"bash",,terminal_focus
+28,121987,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch",0,0,"",shellscript,tab
+29,123074,"TERMINAL",0,0,"bash",,terminal_focus
+30,140621,"TERMINAL",0,0,"diff /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",,terminal_command
+31,140656,"TERMINAL",0,0,"]633;C3c3\r\n< #SBATCH --nodes=1\r\n---\r\n> #SBATCH --nodes=32\r\n9,11c9,11\r\n< #SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\r\n< #SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\r\n< #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\r\n---\r\n> #SBATCH --output=logs/logs_training/%x_%j.log\r\n> #SBATCH --error=logs/logs_training/%x_%j.log\r\n> #SBATCH --job-name=train_tokenizer_batch_size_scaling_32_node\r\n32c32\r\n<     --batch_size=48 \\r\n---\r\n>     --batch_size=1536 \\r\n34c34\r\n<     --max_lr=1e-4 \\r\n---\r\n>     --max_lr=3e-4 \\r\n38,39c38,39\r\n<     --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\r\n<     --tags tokenizer batch-size-scaling 1-node sqrt-lr \\r\n---\r\n>     --name=tokenizer-batch-size-scaling-32-node-sqrt-lr-$slurm_job_id \\r\n>     --tags tokenizer batch-size-scaling 32-node sqrt-lr-scaling \\r\n]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+32,154994,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch",0,0,"",shellscript,tab
+33,160937,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",0,0,"",shellscript,tab
+34,163667,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch",0,0,"",shellscript,tab
+35,164483,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch",187,0,"",shellscript,selection_mouse
+36,167106,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",0,0,"",shellscript,tab
+37,167763,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",249,0,"",shellscript,selection_mouse
+38,168320,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",261,0,"\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log",shellscript,content
+39,168337,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",262,0,"",shellscript,selection_command
+40,168635,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",217,0,"",shellscript,selection_command
+41,168958,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",171,91,"",shellscript,content
+42,171940,"TERMINAL",0,0,"diff /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_32_nodes.sbatch",,terminal_command
+43,171941,"TERMINAL",0,0,"]633;C3c3\r\n< #SBATCH --nodes=1\r\n---\r\n> #SBATCH --nodes=32\r\n11c11\r\n< #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\r\n---\r\n> #SBATCH --job-name=train_tokenizer_batch_size_scaling_32_node\r\n32c32\r\n<     --batch_size=48 \\r\n---\r\n>     --batch_size=1536 \\r\n34c34\r\n<     --max_lr=1e-4 \\r\n---\r\n>     --max_lr=3e-4 \\r\n38,39c38,39\r\n<     --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\r\n<     --tags tokenizer batch-size-scaling 1-node sqrt-lr \\r\n---\r\n>     --name=tokenizer-batch-size-scaling-32-node-sqrt-lr-$slurm_job_id \\r\n>     --tags tokenizer batch-size-scaling 32-node sqrt-lr-scaling \\r\n]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+44,198176,"TERMINAL",0,0,"diff /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_16_nodes.sbatch",,terminal_command
+45,198187,"TERMINAL",0,0,"]633;C3c3\r\n< #SBATCH --nodes=1\r\n---\r\n> #SBATCH --nodes=16\r\n11c11\r\n< #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\r\n---\r\n> #SBATCH --job-name=train_tokenizer_batch_size_scaling_16_node\r\n32c32\r\n<     --batch_size=48 \\r\n---\r\n>     --batch_size=768 \\r\n34c34\r\n<     --max_lr=1e-4 \\r\n---\r\n>     --max_lr=4e-4 \\r\n38,39c38,39\r\n<     --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\r\n<     --tags tokenizer batch-size-scaling 1-node sqrt-lr \\r\n---\r\n>     --name=tokenizer-batch-size-scaling-16-node-sqrt-lr-$slurm_job_id \\r\n>     --tags tokenizer batch-size-scaling 16-node sqrt-lr-scaling \\r\n]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+46,208288,"TERMINAL",0,0,"diff /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_8_nodes.sbatch",,terminal_command
+47,208304,"TERMINAL",0,0,"]633;C3c3\r\n< #SBATCH --nodes=1\r\n---\r\n> #SBATCH --nodes=8\r\n11c11\r\n< #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\r\n---\r\n> #SBATCH --job-name=train_tokenizer_batch_size_scaling_8_node\r\n32c32\r\n<     --batch_size=48 \\r\n---\r\n>     --batch_size=384 \\r\n34c34\r\n<     --max_lr=1e-4 \\r\n---\r\n>     --max_lr=3e-4 \\r\n38,39c38,39\r\n<     --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\r\n<     --tags tokenizer batch-size-scaling 1-node sqrt-lr \\r\n---\r\n>     --name=tokenizer-batch-size-scaling-8-node-sqrt-lr-$slurm_job_id \\r\n>     --tags tokenizer batch-size-scaling 8-node sqrt-lr-scaling \\r\n]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+48,215509,"TERMINAL",0,0,"diff /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",,terminal_command
+49,215540,"TERMINAL",0,0,"]633;C3c3\r\n< #SBATCH --nodes=1\r\n---\r\n> #SBATCH --nodes=4\r\n11c11\r\n< #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\r\n---\r\n> #SBATCH --job-name=train_tokenizer_batch_size_scaling_4_node\r\n32c32\r\n<     --batch_size=48 \\r\n---\r\n>     --batch_size=192 \\r\n34c34\r\n<     --max_lr=1e-4 \\r\n---\r\n>     --max_lr=2.00e-4 \\r\n38,39c38,39\r\n<     --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\r\n<     --tags tokenizer batch-size-scaling 1-node sqrt-lr \\r\n---\r\n>     --name=tokenizer-batch-size-scaling-4-node-sqrt-lr-$slurm_job_id \\r\n>     --tags tokenizer batch-size-scaling 4-node sqrt-lr \\r\n]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+50,227354,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=4\n#SBATCH --ntasks-per-node=4\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\n#SBATCH --job-name=train_tokenizer_batch_size_scaling_4_node\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_new/open_ai_minecraft_arrayrecords/10fps_160x90\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/checkpoints/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\nenv | grep SLURM\n\nsrun python jasmine/train_tokenizer.py \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=192 \\n    --min_lr=0 \\n    --max_lr=2.00e-4 \\n    --log_image_interval=1000 \\n    --log_checkpoint_interval=1000 \\n    --log \\n    --name=tokenizer-batch-size-scaling-4-node-sqrt-lr-$slurm_job_id \\n    --tags tokenizer batch-size-scaling 4-node sqrt-lr \\n    --entity instant-uv \\n    --project jafar \\n    --data_dir $array_records_dir\n",shellscript,tab
+51,229213,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",1016,0,"",shellscript,selection_mouse
+52,230095,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",1015,1,"",shellscript,content
+53,230188,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",1014,1,"",shellscript,content
+54,230484,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",1013,1,"",shellscript,content
+55,230834,"/home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",1012,0,"",shellscript,selection_command
+56,234469,"TERMINAL",0,0,"diff /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_4_nodes.sbatch",,terminal_command
+57,234477,"TERMINAL",0,0,"]633;C3c3\r\n< #SBATCH --nodes=1\r\n---\r\n> #SBATCH --nodes=4\r\n11c11\r\n< #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\r\n---\r\n> #SBATCH --job-name=train_tokenizer_batch_size_scaling_4_node\r\n32c32\r\n<     --batch_size=48 \\r\n---\r\n>     --batch_size=192 \\r\n34c34\r\n<     --max_lr=1e-4 \\r\n---\r\n>     --max_lr=2e-4 \\r\n38,39c38,39\r\n<     --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\r\n<     --tags tokenizer batch-size-scaling 1-node sqrt-lr \\r\n---\r\n>     --name=tokenizer-batch-size-scaling-4-node-sqrt-lr-$slurm_job_id \\r\n>     --tags tokenizer batch-size-scaling 4-node sqrt-lr \\r\n]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+58,239342,"TERMINAL",0,0,"diff /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_2_nodes.sbatch",,terminal_command
+59,239359,"TERMINAL",0,0,"]633;C3c3\r\n< #SBATCH --nodes=1\r\n---\r\n> #SBATCH --nodes=2\r\n11c11\r\n< #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\r\n---\r\n> #SBATCH --job-name=train_tokenizer_batch_size_scaling_2_node\r\n32c32\r\n<     --batch_size=48 \\r\n---\r\n>     --batch_size=96 \\r\n34c34\r\n<     --max_lr=1e-4 \\r\n---\r\n>     --max_lr=1.4e-4 \\r\n38,39c38,39\r\n<     --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\r\n<     --tags tokenizer batch-size-scaling 1-node sqrt-lr \\r\n---\r\n>     --name=tokenizer-batch-size-scaling-2-node-sqrt-lr-$slurm_job_id \\r\n>     --tags tokenizer batch-size-scaling 2-node sqrt-lr-scaling \\r\n]0;mihir.mahajan@hai-login2:~/Projects/tab-model-eval",,terminal_output
+60,254499,"TERMINAL",0,0,"cd /home/mihir.mahajan/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr",,terminal_command
+61,255158,"TERMINAL",0,0,"ls",,terminal_command
+62,255158,"TERMINAL",0,0,"]633;Cstart_runs.sh                    train_tokenizer_1_nodes.sbatch   train_tokenizer_4_nodes.sbatch\r\ntester.sh                        train_tokenizer_2_nodes.sbatch   train_tokenizer_8_nodes.sbatch\r\ntrain_tokenizer_16_nodes.sbatch  train_tokenizer_32_nodes.sbatch\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr",,terminal_output
+63,256089,"TERMINAL",0,0,"cd ..",,terminal_command
+64,265480,"TERMINAL",0,0,"cp -r sqrt_lr sqrt_lr_bak/",,terminal_command
+65,265514,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer",,terminal_output
+66,267772,"TERMINAL",0,0,"ls",,terminal_command
+67,267772,"TERMINAL",0,0,"]633;Cconst_lr  linear_lr  sqrt_lr  sqrt_lr_bak\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine/slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer",,terminal_output
+68,314957,"TERMINAL",0,0,"cd ..",,terminal_command
+69,316867,"TERMINAL",0,0,"cd ..",,terminal_command
+70,317484,"TERMINAL",0,0,"cd ..",,terminal_command
+71,318175,"TERMINAL",0,0,"ls",,terminal_command
+72,320036,"TERMINAL",0,0,"cd ..",,terminal_command
+73,320555,"TERMINAL",0,0,"ls",,terminal_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-529e4bba-7d8a-4661-afaa-55c4c5c035291764770291321-2025_12_03-14.58.42.143/source.csv

@@ -0,0 +1,6 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,3,"data/eval/handcrafted/eval_file_switch.md",0,0,"# Assistant\n```bash\nls -l slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/\n```\n\n# User\n<stdout>\ntotal 144\n-rw-r--r-- 1 mihir.mahajan mihir.mahajan 1322 Dec  3 14:23 train_tokenizer_16_nodes.sbatch\n-rw-r--r-- 1 mihir.mahajan mihir.mahajan 1309 Dec  3 14:23 train_tokenizer_1_nodes.sbatch\n-rw-r--r-- 1 mihir.mahajan mihir.mahajan 1319 Dec  3 14:23 train_tokenizer_2_nodes.sbatch\n-rw-r--r-- 1 mihir.mahajan mihir.mahajan 1323 Dec  3 14:23 train_tokenizer_32_nodes.sbatch\n-rw-r--r-- 1 mihir.mahajan mihir.mahajan 1310 Dec  3 14:23 train_tokenizer_4_nodes.sbatch\n-rw-r--r-- 1 mihir.mahajan mihir.mahajan 1318 Dec  3 14:23 train_tokenizer_8_nodes.sbatch\n</stdout>\n\n# Assistant\n```bash\ncat -n slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch\n```\n\n# User\n<stdout>\n     1  #!/usr/bin/env bash\n     2\n     3  #SBATCH --nodes=1\n     4  #SBATCH --ntasks-per-node=4\n     5  #SBATCH --time=05:00:00\n     6  #SBATCH --partition=accelerated\n     7  #SBATCH --cpus-per-task=5\n     8  #SBATCH --gres=gpu:4\n     9  #SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\n    10  #SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/%x_%j.log\n    11  #SBATCH --job-name=train_tokenizer_batch_size_scaling_1_node\n    12\n    13  # Log the sbatch script\n    14  cat $0\n    15\n    16  module unload mpi/openmpi/5.0\n    17  module unload devel/cuda/12.4\n    18  source .venv/bin/activate\n    19\n    20  array_records_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_new/open_ai_minecraft_arrayrecords/10fps_160x90\n    21\n    22  job_name=$SLURM_JOB_NAME\n    23  slurm_job_id=$SLURM_JOB_ID\n    24\n    25  CHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data/checkpoints/$job_name/$slurm_job_id\n    26  mkdir -p $CHECKPOINT_DIR\n    27\n    28  env | grep SLURM\n    29\n    30  srun python jasmine/train_tokenizer.py \\n    31      --ckpt_dir $CHECKPOINT_DIR \\n    32      --batch_size=48 \\n    33      --min_lr=0 \\n    34      --max_lr=1e-4 \\n    35      --log_image_interval=1000 \\n    36      --log_checkpoint_interval=1000 \\n    37      --log \\n    38      --name=tokenizer-batch-size-scaling-1-node-sqrt-lr-$slurm_job_id \\n    39      --tags tokenizer batch-size-scaling 1-node sqrt-lr \\n    40      --entity instant-uv \\n    41      --project jafar \\n    42      --data_dir $array_records_dir\n</stdout>\n\n# Assistant\n```bash\nsed -i '5,5c#SBATCH --time=48:00:00' slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch\n```\n\n\n# Assistant\n```bash\ncat -n slurm/jobs/mihir/horeka/batchsize_scaling/tokenizer/sqrt_lr/train_tokenizer_1_nodes.sbatch\n```\n\n# User\n<stdout>\n</stdout>",markdown,tab
+2,3417,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"2:58:42 PM [info] Activating crowd-code\n2:58:42 PM [info] Recording started\n2:58:42 PM [info] Initializing git provider using file system watchers...\n2:58:43 PM [info] Git repository found\n2:58:43 PM [info] Git provider initialized successfully\n2:58:43 PM [info] Initial git state: [object Object]\n",Log,tab
+3,9859,"TERMINAL",0,0,"",,terminal_command
+4,16890,"TERMINAL",0,0,"",,terminal_command
+5,96256,"TERMINAL",0,0,"",,terminal_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-59804d2c-df27-4b41-9fe6-ffdfef8825021760703869419-2025_10_17-14.25.51.972/source.csv

@@ -0,0 +1,3 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,3471,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab
+3,3682,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"2:25:50 PM [info] Activating crowd-code\n2:25:51 PM [info] Recording started\n2:25:52 PM [info] Initializing git provider using file system watchers...\n2:25:52 PM [info] No workspace folder found\n2:25:55 PM [info] Retrying git provider initialization...\n2:25:55 PM [info] No workspace folder found\n",Log,content

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-6877ad19-9807-438f-8b51-95c34f7ec7121764157755585-2025_11_26-12.49.36.76/source.csv

@@ -0,0 +1,4 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,1418,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"12:49:36 PM [info] Activating crowd-code\n12:49:36 PM [info] Recording started\n12:49:36 PM [info] Initializing git provider using file system watchers...\n12:49:36 PM [info] Git repository found\n12:49:36 PM [info] Git provider initialized successfully\n12:49:36 PM [info] Initial git state: [object Object]\n",Log,tab
+3,4025,"TERMINAL",0,0,"",,terminal_command
+4,10769,"TERMINAL",0,0,"",,terminal_command

b1ccc08ffefaf10a6c1b392f91212b8c7d87421316664745744fabeefb9fdb16/crowd-code-698eb099-f494-4ec6-a868-8174b5764b0d1765283687620-2025_12_09-13.35.26.529/source.csv

@@ -0,0 +1,7 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,8859,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"1:35:26 PM [info] Activating crowd-code\n1:35:26 PM [info] Recording started\n1:35:26 PM [info] Initializing git provider using file system watchers...\n1:35:28 PM [info] Git repository found\n1:35:28 PM [info] Git provider initialized successfully\n",Log,tab
+3,14121,"TERMINAL",0,0,"",,terminal_focus
+4,16585,"TERMINAL",0,0,"",,terminal_command
+5,23298,"TERMINAL",0,0,"",,terminal_command
+6,28538,"pyproject.toml",0,0,"# Copyright (c) 2023, NVIDIA CORPORATION.  All rights reserved.\n\n[build-system]\nrequires = [""setuptools<80.0.0"", ""pybind11""]\nbuild-backend = ""setuptools.build_meta""\n\n[tool.setuptools]\ninclude-package-data = true\n\n[tool.setuptools.packages.find]\ninclude = [""megatron.core"", ""megatron.core.*""]\n\n[tool.setuptools.dynamic]\nversion = { attr = ""megatron.core.package_info.__version__"" }\nreadme = { file = ""README.md"", content-type = ""text/markdown"" }\n\n[project]\nname = ""megatron-core""\ndynamic = [""version"", ""readme""]\ndescription = ""Megatron Core - a library for efficient and scalable training of transformer based models""\nrequires-python = "">=3.10""\nlicense = { text = ""Apache 2.0"" }\ndependencies = [""torch"", ""numpy<2.0.0"", ""packaging""]\nauthors = [{ name = ""NVIDIA"", email = ""nemo-toolkit@nvidia.com"" }]\nmaintainers = [{ name = ""NVIDIA"", email = ""nemo-toolkit@nvidia.com"" }]\nkeywords = [\n    ""NLP"",\n    ""NLU"",\n    ""deep"",\n    ""gpu"",\n    ""language"",\n    ""learning"",\n    ""learning"",\n    ""machine"",\n    ""nvidia"",\n    ""pytorch"",\n    ""torch"",\n    ""transformer"",\n]\nclassifiers = [\n    ""Development Status :: 5 - Production/Stable"",\n    ""Environment :: Console"",\n    ""Intended Audience :: Developers"",\n    ""Intended Audience :: Information Technology"",\n    ""Intended Audience :: Science/Research"",\n    ""License :: OSI Approved :: BSD License"",\n    ""Natural Language :: English"",\n    ""Operating System :: OS Independent"",\n    ""Programming Language :: Python :: 3"",\n    ""Programming Language :: Python :: 3.8"",\n    ""Programming Language :: Python :: 3.9"",\n    ""Topic :: Scientific/Engineering :: Artificial Intelligence"",\n    ""Topic :: Scientific/Engineering :: Image Recognition"",\n    ""Topic :: Scientific/Engineering :: Mathematics"",\n    ""Topic :: Scientific/Engineering"",\n    ""Topic :: Software Development :: Libraries :: Python Modules"",\n    ""Topic :: Software Development :: Libraries"",\n    ""Topic :: Utilities"",\n]\n\n[project.urls]\nDownload = ""https://github.com/NVIDIA/Megatron-LM/releases""\nHomepage = ""https://github.com/NVIDIA/Megatron-LM/megatron/core""\n\n[project.optional-dependencies]\nmlm = [""flask-restful"", ""sentencepiece"", ""tiktoken"", ""wandb""]\n\ndev = [\n    ""tqdm"",\n    ""einops~=0.8"",\n    ""tensorstore~=0.1,!=0.1.46,!=0.1.72"",\n    ""nvtx~=0.2"",\n    ""transformers~=4.53"",\n    ""multi-storage-client~=0.20"",\n    ""opentelemetry-api~=1.33.1"",\n    ""setuptools<80.0.0"",\n    ""mamba-ssm~=2.2"",\n    ""causal-conv1d~=1.5"",\n    ""nv-grouped-gemm~=1.1"",\n    ""transformer-engine[pytorch]>=2.6.0a0,<2.8.0"",\n    ""nvidia-resiliency-ext>=0.4.0a0,<0.5.0"",\n    ""nvidia-modelopt[torch]>=0.33.0a0,<0.34.0; sys_platform != 'darwin'"",\n    ""megatron-energon[av_decode]~=6.0"",\n    ""flashinfer-python"",\n    ""onnxscript"",\n]\n\nlts = [\n    ""tqdm"",\n    ""einops"",\n    ""tensorstore!=0.1.46,!=0.1.72"",\n    ""nvtx"",\n    ""transformers"",\n    ""zarr"",\n    ""setuptools<80.0.0"",\n]\n\n[dependency-groups]\ntest = [\n    ""coverage"",\n    ""nltk"",\n    ""wrapt"",\n    ""pytest==8.3.5"",\n    ""pytest-mock"",\n    ""pytest-cov"",\n    ""pytest-random-order"",\n    ""pytest-asyncio"",\n    ""pygithub"",\n    ""pydantic"",\n    ""tensorboard"",\n    ""pyyaml"",\n]\n\nbuild = [\n    ""setuptools<80.0.0"",\n    ""packaging"",\n    ""hatchling"",\n    ""pybind11"",\n    ""Cython>=3.0.0"",\n    ""torch"",\n]\nlinting = [\n    ""ruff~=0.9.0"",\n    ""black==24.4.2"",\n    ""isort==5.13.2"",\n    ""flake8==7.1.0"",\n    ""pylint==3.2.6"",\n]\nci = [""python-gitlab"", ""slack-sdk"", ""pandas""]\nflash_mla = [""flash_mla""]\n\n[tool.uv]\ndefault-groups = [""linting"", ""build"", ""test""]\nno-build-isolation-package = [\n    ""transformer-engine"",\n    ""transformer-engine-torch"",\n    ""mamba-ssm"",\n    ""causal-conv1d"",\n    ""nv-grouped-gemm"",\n    ""flash_mla"",\n]\nlink-mode = ""copy""\nconflicts = [[{ extra = ""lts"" }, { extra = ""dev"" }]]\n# We don't want to install torch, torchvision, and triton\n# because they are already installed in the base image.\noverride-dependencies = [\n    ""torch; sys_platform == 'never'"",\n    ""torchvision; sys_platform == 'never'"",\n    ""triton; sys_platform == 'never'"",\n]\n\n[tool.uv.sources]\nflash_mla = [\n    { git = ""https://github.com/deepseek-ai/FlashMLA"", rev = ""9edee0c022cd0938148a18e334203b0aab43aa19"" },\n]\n\n[tool.isort]\nprofile = ""black""                                                          # black-compatible\nline_length = 100                                                          # should match black parameters\npy_version = 310                                                           # python 3.8 as a target version\nknown_first_party = [""megatron""]                                           # FIRSTPARTY section\nknown_third_party = [""transformer_engine""]                                 # THIRDPARTY section\nsections = [""FUTURE"", ""STDLIB"", ""THIRDPARTY"", ""FIRSTPARTY"", ""LOCALFOLDER""]\ndefault_section = ""THIRDPARTY""\nextend_skip = [""setup.py""]\n\n[tool.black]\nline_length = 100\nskip_string_normalization = true\n# recongized by future versions, disallows to reformat code with incompatible versions\n# Matches NeMO version so people working on both codebases don't need two different version of black installed\nrequired_version = ""24""\nskip_magic_trailing_comma = true\ninclude = '\.pyi?$'\nexclude = '''\n/(\n    \.git\n  | \.venv\n  | build\n)/\n'''\n\n[tool.pytest.ini_options]\naddopts = ""--durations=15 -s -rA -x""\ntestpaths = [""tests""]\npython_files = ""test_*.py""\nmarkers = [\n    ""internal: mark a test as a test to private/internal functions."",\n    ""flaky: mark flaky tests for LTS environment"",\n    ""flaky_in_dev: mark flaky tests for DEV environment"",\n]\n\n[tool.coverage.run]\ndata_file = "".coverage_$LOCAL_RANK""\nconcurrency = [""thread"", ""multiprocessing""]\nomit = [\n    ""/tmp/*"",\n    ""/workspace/tests/*"",\n    ""/usr/local/lib/python3.12/dist-packages/*"",\n]\nrelative_files = true\n\n[tool.coverage.paths]\nsource = [""/opt/megatron-lm/""]\n\n[tool.ruff.lint]\n# Enable all `pydocstyle` rules, limiting to those that adhere to the\n# Google convention via `convention = ""google""`, below.\n# select = [""D"", ""F""]\nselect = [""S506""]\n\n# - On top of the Google convention, disable `D417`, which requires\n#   documentation for every function parameter.\n# - F841: local variable assigned but never used (exluced to favor readability)\n# TODO: Remove D10 once we are about to release to get all the docstrings written\nignore = [""D417"", ""D10"", ""F841""]\n\n[tool.ruff.lint.pydocstyle]\nconvention = ""google""\n\n# Section to exclude errors for different file types\n[tool.ruff.per-file-ignores]\n# Ignore all directories named `tests`.\n""tests/**"" = [""D""]\n# Ignore all files that end in `_test.py`.\n""*_test.py"" = [""D""]\n# Ignore F401 (import but unused) in __init__.py\n""__init__.py"" = [""F401""]\n",plaintext,tab
+7,89041,"TERMINAL",0,0,"",,terminal_command