20M timesteps

maybe diminishing returns

Uploaded Unknown by Rowan G. Last edited by Rowan G.

Training information

Algorithm: ippo_rnn
Seed: 901
Progress unit: steps

The highest uploaded checkpoint may not be the final progress of the local run. Initialization-only runs do not indicate training.

Uploaded checkpoints

20M timesteps · 19.99M steps

Related evaluations

coord ring lots of training

Checkpoint evidence

[
  {
    "checkpoint_id": "a045c9a9-dbd2-407e-9ea5-cbe7bb538568",
    "id": "a045c9a9-dbd2-407e-9ea5-cbe7bb538568",
    "label": "20M timesteps \u00b7 19.99M steps",
    "progress": 19988480,
    "progress_unit": "environment_steps",
    "run_id": "1ba3d6f5-2cc8-46cf-8e36-841aada14a44",
    "run_name": "20M timesteps",
    "sha256": "fe793fb313762382b7c2d433d83c2a610e8386386698b4b1ad3067f10c49a21b",
    "source_checkpoint_id": "run:3e36ab97-b1a8-4aaa-8359-c5f39e7c6070:sha256:fe793fb313762382b7c2d433d83c2a610e8386386698b4b1ad3067f10c49a21b",
    "training_run_name": "20M timesteps"
  }
]

Training configuration

{
  "ACTIVATION": "relu",
  "ALGORITHM": "ippo_rnn",
  "ANNEAL_LR": true,
  "CHECKPOINT_INTERVAL_UPDATES": 10,
  "CHECKPOINT_KEEP_LAST": 3,
  "CLIP_EPS": 0.2,
  "CNN_FEATURES": 32,
  "ENT_COEF": 0.01,
  "ENV_KWARGS": {
    "agent_view_size": 2,
    "indicate_successful_delivery": false,
    "layout": "coord_ring",
    "max_steps": 400,
    "negative_rewards": true,
    "observation_type": "default",
    "random_agent_positions": true,
    "random_reset": false,
    "sample_recipe_on_delivery": false,
    "start_cooking_interaction": false
  },
  "ENV_NAME": "overcooked_v2",
  "FC_DIM_SIZE": 128,
  "GAE_LAMBDA": 0.95,
  "GAMMA": 0.99,
  "GRU_HIDDEN_DIM": 128,
  "GRU_UNROLL": 16,
  "LR": 0.0005,
  "LR_WARMUP": 0.05,
  "MAX_GRAD_NORM": 0.25,
  "MINIBATCH_SIZE": 2048,
  "NUM_ENVS": 256,
  "NUM_MINIBATCHES": 64,
  "NUM_SEEDS": 1,
  "NUM_STEPS": 256,
  "REW_SHAPING_HORIZON": 5000000,
  "RUNTIME": {},
  "SCALE_CLIP_EPS": false,
  "SEED": 901,
  "TOTAL_TIMESTEPS": 20000000,
  "UPDATE_EPOCHS": 4,
  "VF_COEF": 0.5,
  "WANDB_MODE": "disabled"
}

Training code and dependency provenance

{
  "algorithm_source_hash": "5cbd7c4e2faba93004cb701e5ea8ec7778c132aea6ed7862b083fde21c22fca5",
  "backend": "gpu",
  "device": {
    "architecture": "x86_64",
    "backend": "gpu",
    "devices": [
      {
        "kind": "NVIDIA GeForce RTX 3070",
        "platform": "gpu"
      }
    ],
    "logical_cpu_count": 16,
    "os": "Linux",
    "os_release": "6.18.40.1-microsoft-standard-WSL2",
    "python": "3.12.3",
    "versions": {
      "flax": "0.10.2",
      "jax": "0.4.38",
      "jaxlib": "0.4.38",
      "numpy": "1.26.4",
      "optax": "0.2.4"
    }
  },
  "git_commit": "5632bd7dc714a85b138cfdea5ea3976bdbcd5e03",
  "git_dirty": true,
  "started_at": "2026-11-07T08:32:39.395485+00:00",
  "upstream_revision": "976aeb152cb184a5095021968bba94da96eb6394",
  "versions": {
    "flax": "0.10.2",
    "jax": "0.4.38",
    "jaxlib": "0.4.38",
    "optax": "0.2.4"
  }
}

Training device

{
  "architecture": "x86_64",
  "backend": "gpu",
  "devices": [
    {
      "kind": "NVIDIA GeForce RTX 3070",
      "platform": "gpu"
    }
  ],
  "logical_cpu_count": 16,
  "os": "Linux",
  "os_release": "6.18.40.1-microsoft-standard-WSL2",
  "python": "3.12.3",
  "versions": {
    "flax": "0.10.2",
    "jax": "0.4.38",
    "jaxlib": "0.4.38",
    "numpy": "1.26.4",
    "optax": "0.2.4"
  }
}