Upload folder using huggingface_hub

.summary/0/events.out.tfevents.1689589249.qgallouedec-MS-7C84

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28639587c888ad0050e38fe53aa110d6c1ba6d8b8cdb5144f271ab6b81b5d2cf
+size 740627

README.md

@@ -15,7 +15,7 @@ model-index:
       type: hammer-v2
     metrics:
     - type: mean_reward
-      value: 315.73 +/- 31.40
+      value: 692.49 +/- 21.25
       name: mean_reward
       verified: false
 ---

checkpoint_p0/best_000018088_9261056_reward_697.151.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f81adf9adcdefd0897f5bd6ffc7e410e342ff7ab3ab3f1d38b7220e6b7448bdd
+size 98239

checkpoint_p0/checkpoint_000019472_9969664.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a373ab520bcb67f4f07612dcef05ab7bfea59654c75e7c8a237a6878ebbff8e7
+size 98567

checkpoint_p0/checkpoint_000019544_10006528.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1cc1f7d714b541dbd6795d814171e4325aeefa81624bf798c116302533aa8b5f
+oid sha256:7444e8b3bc6e428dc8b717589c89933585a70cef5da893f8c8257af22f3691f2
 size 98567

config.json

@@ -128,7 +128,7 @@
     "wandb_user": "qgallouedec",
     "wandb_project": "sample_facotry_metaworld"
   },
-  "git_hash": "66db1b7a27030aa65fcfa2d6e3503089a7cff207",
+  "git_hash": "998bf2be1819761361d54bcbaf7d9ab9748a770e",
   "git_repo_name": "https://github.com/huggingface/gia",
-  "wandb_unique_id": "hammer-v2_20230708_190205_433334"
+  "wandb_unique_id": "hammer-v2_20230717_122047_003477"
 }

git.diff

@@ -1,121 +1,122 @@
-diff --git a/gia/eval/callback.py b/gia/eval/callback.py
-index 5c3a080..4b6198f 100644
---- a/gia/eval/callback.py
-+++ b/gia/eval/callback.py
-@@ -2,10 +2,10 @@ import glob
- import json
- import subprocess
- 
--import wandb
- from accelerate import Accelerator
- from transformers import TrainerCallback, TrainerControl, TrainerState, TrainingArguments
- 
-+import wandb
- from gia.config import Arguments
- from gia.eval.utils import is_slurm_available
- 
-diff --git a/gia/eval/evaluator.py b/gia/eval/evaluator.py
-index 91b645c..3e2cae7 100644
---- a/gia/eval/evaluator.py
-+++ b/gia/eval/evaluator.py
-@@ -1,3 +1,5 @@
-+from typing import Optional
-+
- import torch
- 
- from gia.config.arguments import Arguments
-@@ -5,11 +7,12 @@ from gia.model import GiaModel
- 
- 
- class Evaluator:
--    def __init__(self, args: Arguments, task: str) -> None:
-+    def __init__(self, args: Arguments, task: str, mean_random: Optional[float] = None) -> None:
-         self.args = args
-         self.task = task
-+        self.mean_random = mean_random
- 
--    @torch.no_grad()
-+    @torch.inference_mode()
-     def evaluate(self, model: GiaModel) -> float:
-         return self._evaluate(model)
- 
-diff --git a/gia/eval/rl/envs/core.py b/gia/eval/rl/envs/core.py
-index ec5e5b2..eeaf7cb 100644
---- a/gia/eval/rl/envs/core.py
-+++ b/gia/eval/rl/envs/core.py
-@@ -177,7 +177,6 @@ def make(task_name: str, num_envs: int = 1):
- 
-     elif task_name.startswith("metaworld"):
-         import gymnasium as gym
--        import metaworld
- 
-         env_id = TASK_TO_ENV_MAPPING[task_name]
-         env = gym.vector.SyncVectorEnv([lambda: gym.make(env_id)] * num_envs)
-diff --git a/gia/eval/rl/gia_agent.py b/gia/eval/rl/gia_agent.py
-index f0d0b9b..39dc0d2 100644
---- a/gia/eval/rl/gia_agent.py
-+++ b/gia/eval/rl/gia_agent.py
-@@ -54,7 +54,7 @@ class GiaAgent:
-         self.action_space = action_space
-         self.deterministic = deterministic
-         self.device = next(model.parameters()).device
--        self._max_length = self.model.config.max_position_embeddings - 10
-+        self._max_length = self.model.config.max_position_embeddings - 100  # TODO: fix this
- 
-         if isinstance(observation_space, spaces.Box):
-             self._observation_key = "continuous_observations"
-@@ -75,6 +75,11 @@ class GiaAgent:
-     ) -> Tuple[Tuple[Tensor, Tensor], ...]:
-         return tuple((k[:, :, -self._max_length :], v[:, :, -self._max_length :]) for (k, v) in past_key_values)
- 
-+    def set_model(self, model: GiaModel) -> None:
-+        self.model = model
-+        self.device = next(model.parameters()).device
-+        self._max_length = self.model.config.max_position_embeddings
-+
-     def reset(self, num_envs: int = 1) -> None:
-         if self.prompter is not None:
-             prompts = self.prompter.generate_prompts(num_envs)
-diff --git a/gia/eval/rl/gym_evaluator.py b/gia/eval/rl/gym_evaluator.py
-index f8531ee..754c05d 100644
---- a/gia/eval/rl/gym_evaluator.py
-+++ b/gia/eval/rl/gym_evaluator.py
-@@ -1,7 +1,7 @@
- import gym
- from gym.vector.vector_env import VectorEnv
- 
--from gia.eval.mappings import TASK_TO_ENV_MAPPING
-+# from gia.eval.rl.envs.mappings import TASK_TO_ENV_MAPPING
- from gia.eval.rl.rl_evaluator import RLEvaluator
- 
- 
-diff --git a/gia/eval/rl/rl_evaluator.py b/gia/eval/rl/rl_evaluator.py
-index c5cc423..91189f3 100644
---- a/gia/eval/rl/rl_evaluator.py
-+++ b/gia/eval/rl/rl_evaluator.py
-@@ -8,6 +8,10 @@ from gia.eval.rl.gia_agent import GiaAgent
- 
- 
- class RLEvaluator(Evaluator):
-+    def __init__(self, args, task):
-+        super().__init__(args, task)
-+        self.agent = GiaAgent()
-+
-     def _build_env(self) -> VectorEnv:  # TODO: maybe just a gym.Env ?
-         raise NotImplementedError
- 
-diff --git a/gia/eval/rl/scores_dict.json b/gia/eval/rl/scores_dict.json
-index 1b8ebee..ff7d030 100644
---- a/gia/eval/rl/scores_dict.json
-+++ b/gia/eval/rl/scores_dict.json
-@@ -929,8 +929,8 @@
-     },
-     "metaworld-assembly": {
-         "expert": {
--            "mean": 311.29314618777823,
--            "std": 75.04282151450695
-+            "mean": 3523.81468486244,
-+            "std": 63.22745220327798
-         },
-         "random": {
-             "mean": 220.65601680730813,
+diff --git a/data/envs/metaworld/generate_dataset_all.sh b/data/envs/metaworld/generate_dataset_all.sh
+index acfe879..0185b2b 100755
+--- a/data/envs/metaworld/generate_dataset_all.sh
++++ b/data/envs/metaworld/generate_dataset_all.sh
+@@ -2,8 +2,6 @@
+ 
+ ENVS=(
+     assembly
+-    basketball
+-    bin-picking
+     box-close
+     button-press-topdown
+     button-press-topdown-wall
+@@ -11,9 +9,7 @@ ENVS=(
+     button-press-wall
+     coffee-button
+     coffee-pull
+-    coffee-push
+     dial-turn
+-    disassemble
+     door-close
+     door-lock
+     door-open
+@@ -22,29 +18,15 @@ ENVS=(
+     drawer-open
+     faucet-close
+     faucet-open
+-    hammer
+     hand-insert
+     handle-press-side
+     handle-press
+     handle-pull-side
+     handle-pull
+     lever-pull
+-    peg-insert-side
+-    peg-unplug-side
+-    pick-out-of-hole
+-    pick-place
+-    pick-place-wall
+-    plate-slide-back-side
+-    plate-slide-back
+-    plate-slide-side
+-    plate-slide
+     push-back
+     push
+     push-wall
+-    reach
+-    reach-wall
+-    shelf-place
+-    soccer
+     stick-pull
+     stick-push
+     sweep-into
+@@ -54,6 +36,6 @@ ENVS=(
+ )
+ 
+ for ENV in "${ENVS[@]}"; do
+-    python -m sample_factory.huggingface.load_from_hub -r qgallouedec/$ENV-v2
++    # python -m sample_factory.huggingface.load_from_hub -r qgallouedec/$ENV-v2
+     python generate_dataset.py --env $ENV-v2 --experiment $ENV-v2 --train_dir=./train_dir
+ done
+diff --git a/data/envs/metaworld/train_all.sh b/data/envs/metaworld/train_all.sh
+index dbf328a..7b2fb85 100755
+--- a/data/envs/metaworld/train_all.sh
++++ b/data/envs/metaworld/train_all.sh
+@@ -1,56 +1,15 @@
+ #!/bin/bash
+ 
+ ENVS=(
+-    assembly
+-    basketball
+     bin-picking
+-    box-close
+-    button-press-topdown
+-    button-press-topdown-wall
+-    button-press
+-    button-press-wall
+-    coffee-button
+-    coffee-pull
+-    coffee-push
+-    dial-turn
+     disassemble
+-    door-close
+-    door-lock
+-    door-open
+-    door-unlock
+-    drawer-close
+-    drawer-open
+-    faucet-close
+-    faucet-open
+     hammer
+-    hand-insert
+-    handle-press-side
+-    handle-press
+-    handle-pull-side
+-    handle-pull
+-    lever-pull
+     peg-insert-side
+-    peg-unplug-side
+     pick-out-of-hole
+     pick-place
+     pick-place-wall
+-    plate-slide-back-side
+-    plate-slide-back
+-    plate-slide-side
+-    plate-slide
+-    push-back
+-    push
+-    push-wall
+-    reach
+-    reach-wall
+     shelf-place
+     soccer
+-    stick-pull
+-    stick-push
+-    sweep-into
+-    sweep
+-    window-close
+-    window-open
+ )
+ 
+ for ENV in "${ENVS[@]}"; do