Upload folder using huggingface_hub

.summary/0/events.out.tfevents.1688782585.qgallouedec-MS-7C84

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d06d4bd297ebc0f7947212ef7c690f09ac114c72b1cb8030d83b40a64b6dd339
+size 728972

README.md

@@ -0,0 +1,56 @@
+---
+library_name: sample-factory
+tags:
+- deep-reinforcement-learning
+- reinforcement-learning
+- sample-factory
+model-index:
+- name: APPO
+  results:
+  - task:
+      type: reinforcement-learning
+      name: reinforcement-learning
+    dataset:
+      name: handle-pull-side-v2
+      type: handle-pull-side-v2
+    metrics:
+    - type: mean_reward
+      value: 107.27 +/- 101.16
+      name: mean_reward
+      verified: false
+---
+
+A(n) **APPO** model trained on the **handle-pull-side-v2** environment.
+
+This model was trained using Sample-Factory 2.0: https://github.com/alex-petrenko/sample-factory.
+Documentation for how to use Sample-Factory can be found at https://www.samplefactory.dev/
+
+
+## Downloading the model
+
+After installing Sample-Factory, download the model with:
+```
+python -m sample_factory.huggingface.load_from_hub -r qgallouedec/handle-pull-side-v2
+```
+
+    
+## Using the model
+
+To run the model after download, use the `enjoy` script corresponding to this environment:
+```
+python -m enjoy --algo=APPO --env=handle-pull-side-v2 --train_dir=./train_dir --experiment=handle-pull-side-v2
+```
+
+
+You can also upload models to the Hugging Face Hub using the same script with the `--push_to_hub` flag.
+See https://www.samplefactory.dev/10-huggingface/huggingface/ for more details
+    
+## Training with this model
+
+To continue training with this model, use the `train` script corresponding to this environment:
+```
+python -m train --algo=APPO --env=handle-pull-side-v2 --train_dir=./train_dir --experiment=handle-pull-side-v2 --restart_behavior=resume --train_for_env_steps=10000000000
+```
+
+Note, you may have to adjust `--train_for_env_steps` to a suitably high number as the experiment will resume at the number of steps it concluded at.
+    

checkpoint_p0/best_000002720_1392640_reward_120.921.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e23f6855ec735d3b3ad8108b84e015d4de173cea503d142315218cdbe12389af
+size 98239

checkpoint_p0/checkpoint_000019408_9936896.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f410a59f07499aa4d463cc94a917216ae9f6a939084a9bb9a3f2f3f7a0ca473
+size 98567

checkpoint_p0/checkpoint_000019544_10006528.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b9f062c7ae1c78d7c556c2805247f2fa65cd37b47d13ba19b705fe7cea90fa0
+size 98567

config.json

@@ -0,0 +1,134 @@
+{
+  "help": false,
+  "algo": "APPO",
+  "env": "handle-pull-side-v2",
+  "experiment": "handle-pull-side-v2",
+  "train_dir": "/home/qgallouedec/data/gia/data/envs/metaworld/train_dir",
+  "restart_behavior": "restart",
+  "device": "cpu",
+  "seed": null,
+  "num_policies": 1,
+  "async_rl": false,
+  "serial_mode": false,
+  "batched_sampling": false,
+  "num_batches_to_accumulate": 2,
+  "worker_num_splits": 2,
+  "policy_workers_per_policy": 1,
+  "max_policy_lag": 1000,
+  "num_workers": 8,
+  "num_envs_per_worker": 8,
+  "batch_size": 1024,
+  "num_batches_per_epoch": 4,
+  "num_epochs": 2,
+  "rollout": 64,
+  "recurrence": 1,
+  "shuffle_minibatches": false,
+  "gamma": 0.99,
+  "reward_scale": 0.1,
+  "reward_clip": 1000.0,
+  "value_bootstrap": true,
+  "normalize_returns": true,
+  "exploration_loss_coeff": 0.0,
+  "value_loss_coeff": 1.3,
+  "kl_loss_coeff": 0.1,
+  "exploration_loss": "entropy",
+  "gae_lambda": 0.95,
+  "ppo_clip_ratio": 0.2,
+  "ppo_clip_value": 1.0,
+  "with_vtrace": false,
+  "vtrace_rho": 1.0,
+  "vtrace_c": 1.0,
+  "optimizer": "adam",
+  "adam_eps": 1e-06,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "max_grad_norm": 3.5,
+  "learning_rate": 0.00295,
+  "lr_schedule": "linear_decay",
+  "lr_schedule_kl_threshold": 0.008,
+  "lr_adaptive_min": 1e-06,
+  "lr_adaptive_max": 0.01,
+  "obs_subtract_mean": 0.0,
+  "obs_scale": 1.0,
+  "normalize_input": true,
+  "normalize_input_keys": null,
+  "decorrelate_experience_max_seconds": 0,
+  "decorrelate_envs_on_one_worker": true,
+  "actor_worker_gpus": [],
+  "set_workers_cpu_affinity": true,
+  "force_envs_single_thread": false,
+  "default_niceness": 0,
+  "log_to_file": true,
+  "experiment_summaries_interval": 3,
+  "flush_summaries_interval": 30,
+  "stats_avg": 100,
+  "summaries_use_frameskip": true,
+  "heartbeat_interval": 20,
+  "heartbeat_reporting_interval": 180,
+  "train_for_env_steps": 10000000,
+  "train_for_seconds": 10000000000,
+  "save_every_sec": 15,
+  "keep_checkpoints": 2,
+  "load_checkpoint_kind": "latest",
+  "save_milestones_sec": -1,
+  "save_best_every_sec": 5,
+  "save_best_metric": "reward",
+  "save_best_after": 100000,
+  "benchmark": false,
+  "encoder_mlp_layers": [
+    64,
+    64
+  ],
+  "encoder_conv_architecture": "convnet_simple",
+  "encoder_conv_mlp_layers": [
+    512
+  ],
+  "use_rnn": false,
+  "rnn_size": 512,
+  "rnn_type": "gru",
+  "rnn_num_layers": 1,
+  "decoder_mlp_layers": [],
+  "nonlinearity": "tanh",
+  "policy_initialization": "torch_default",
+  "policy_init_gain": 1.0,
+  "actor_critic_share_weights": true,
+  "adaptive_stddev": false,
+  "continuous_tanh_scale": 0.0,
+  "initial_stddev": 1.0,
+  "use_env_info_cache": false,
+  "env_gpu_actions": false,
+  "env_gpu_observations": true,
+  "env_frameskip": 1,
+  "env_framestack": 1,
+  "pixel_format": "CHW",
+  "use_record_episode_statistics": false,
+  "with_wandb": true,
+  "wandb_user": "qgallouedec",
+  "wandb_project": "sample_facotry_metaworld",
+  "wandb_group": null,
+  "wandb_job_type": "SF",
+  "wandb_tags": [],
+  "with_pbt": false,
+  "pbt_mix_policies_in_one_env": true,
+  "pbt_period_env_steps": 5000000,
+  "pbt_start_mutation": 20000000,
+  "pbt_replace_fraction": 0.3,
+  "pbt_mutation_rate": 0.15,
+  "pbt_replace_reward_gap": 0.1,
+  "pbt_replace_reward_gap_absolute": 1e-06,
+  "pbt_optimize_gamma": false,
+  "pbt_target_objective": "true_objective",
+  "pbt_perturb_min": 1.1,
+  "pbt_perturb_max": 1.5,
+  "command_line": "--env handle-pull-side-v2 --experiment handle-pull-side-v2 --with_wandb True --wandb_user qgallouedec --wandb_project sample_facotry_metaworld",
+  "cli_args": {
+    "env": "handle-pull-side-v2",
+    "experiment": "handle-pull-side-v2",
+    "with_wandb": true,
+    "wandb_user": "qgallouedec",
+    "wandb_project": "sample_facotry_metaworld"
+  },
+  "git_hash": "66db1b7a27030aa65fcfa2d6e3503089a7cff207",
+  "git_repo_name": "https://github.com/huggingface/gia",
+  "wandb_unique_id": "handle-pull-side-v2_20230708_041623_261681"
+}

git.diff

@@ -0,0 +1,121 @@
+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,