Commit
•
06f0ef5
1
Parent(s):
f5d04e6
Upload . with huggingface_hub
Browse files- .gitattributes +1 -0
- .summary/0/events.out.tfevents.1678462938.qgallouedec-MS-7C84 +3 -0
- README.md +56 -0
- checkpoint_p0/best_000175520_89866240_reward_4737.828.pth +3 -0
- checkpoint_p0/checkpoint_000195136_99909632.pth +3 -0
- checkpoint_p0/checkpoint_000195328_100007936.pth +3 -0
- config.json +134 -0
- git.diff +41 -0
- replay.mp4 +3 -0
- sf_log.txt +0 -0
.gitattributes
CHANGED
@@ -32,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
32 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
32 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
33 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
34 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
35 |
+
replay.mp4 filter=lfs diff=lfs merge=lfs -text
|
.summary/0/events.out.tfevents.1678462938.qgallouedec-MS-7C84
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b29649bb77bddbe7dc0d0d9f53be7a71c4028adc9c1511534f3e37641551e3ce
|
3 |
+
size 5078776
|
README.md
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
library_name: sample-factory
|
3 |
+
tags:
|
4 |
+
- deep-reinforcement-learning
|
5 |
+
- reinforcement-learning
|
6 |
+
- sample-factory
|
7 |
+
model-index:
|
8 |
+
- name: APPO
|
9 |
+
results:
|
10 |
+
- task:
|
11 |
+
type: reinforcement-learning
|
12 |
+
name: reinforcement-learning
|
13 |
+
dataset:
|
14 |
+
name: push-wall-v2
|
15 |
+
type: push-wall-v2
|
16 |
+
metrics:
|
17 |
+
- type: mean_reward
|
18 |
+
value: 4317.91 +/- 1202.36
|
19 |
+
name: mean_reward
|
20 |
+
verified: false
|
21 |
+
---
|
22 |
+
|
23 |
+
A(n) **APPO** model trained on the **push-wall-v2** environment.
|
24 |
+
|
25 |
+
This model was trained using Sample-Factory 2.0: https://github.com/alex-petrenko/sample-factory.
|
26 |
+
Documentation for how to use Sample-Factory can be found at https://www.samplefactory.dev/
|
27 |
+
|
28 |
+
|
29 |
+
## Downloading the model
|
30 |
+
|
31 |
+
After installing Sample-Factory, download the model with:
|
32 |
+
```
|
33 |
+
python -m sample_factory.huggingface.load_from_hub -r qgallouedec/sample-factory-push-wall-v2
|
34 |
+
```
|
35 |
+
|
36 |
+
|
37 |
+
## Using the model
|
38 |
+
|
39 |
+
To run the model after download, use the `enjoy` script corresponding to this environment:
|
40 |
+
```
|
41 |
+
python -m enjoy --algo=APPO --env=push-wall-v2 --train_dir=./train_dir --experiment=sample-factory-push-wall-v2
|
42 |
+
```
|
43 |
+
|
44 |
+
|
45 |
+
You can also upload models to the Hugging Face Hub using the same script with the `--push_to_hub` flag.
|
46 |
+
See https://www.samplefactory.dev/10-huggingface/huggingface/ for more details
|
47 |
+
|
48 |
+
## Training with this model
|
49 |
+
|
50 |
+
To continue training with this model, use the `train` script corresponding to this environment:
|
51 |
+
```
|
52 |
+
python -m train --algo=APPO --env=push-wall-v2 --train_dir=./train_dir --experiment=sample-factory-push-wall-v2 --restart_behavior=resume --train_for_env_steps=10000000000
|
53 |
+
```
|
54 |
+
|
55 |
+
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.
|
56 |
+
|
checkpoint_p0/best_000175520_89866240_reward_4737.828.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:57837c50bf4239414d3ba622500162d62eee95639dcae1cf50b5cf0b61f1bc1c
|
3 |
+
size 98239
|
checkpoint_p0/checkpoint_000195136_99909632.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:bacdbd0fa0258bbec9dd89c295df61e5752acf75d6addf40443af994df31d245
|
3 |
+
size 98567
|
checkpoint_p0/checkpoint_000195328_100007936.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:03a14db68cf11799c790906de55f4e86fee48a83a7d6a6710bdda34483148309
|
3 |
+
size 98567
|
config.json
ADDED
@@ -0,0 +1,134 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"help": false,
|
3 |
+
"algo": "APPO",
|
4 |
+
"env": "push-wall-v2",
|
5 |
+
"experiment": "push-wall-v2",
|
6 |
+
"train_dir": "/home/qgallouedec/gia/data/envs/metaworld/train_dir",
|
7 |
+
"restart_behavior": "restart",
|
8 |
+
"device": "cpu",
|
9 |
+
"seed": null,
|
10 |
+
"num_policies": 1,
|
11 |
+
"async_rl": false,
|
12 |
+
"serial_mode": false,
|
13 |
+
"batched_sampling": false,
|
14 |
+
"num_batches_to_accumulate": 2,
|
15 |
+
"worker_num_splits": 2,
|
16 |
+
"policy_workers_per_policy": 1,
|
17 |
+
"max_policy_lag": 1000,
|
18 |
+
"num_workers": 8,
|
19 |
+
"num_envs_per_worker": 8,
|
20 |
+
"batch_size": 1024,
|
21 |
+
"num_batches_per_epoch": 4,
|
22 |
+
"num_epochs": 2,
|
23 |
+
"rollout": 64,
|
24 |
+
"recurrence": 1,
|
25 |
+
"shuffle_minibatches": false,
|
26 |
+
"gamma": 0.99,
|
27 |
+
"reward_scale": 0.1,
|
28 |
+
"reward_clip": 1000.0,
|
29 |
+
"value_bootstrap": true,
|
30 |
+
"normalize_returns": true,
|
31 |
+
"exploration_loss_coeff": 0.0,
|
32 |
+
"value_loss_coeff": 1.3,
|
33 |
+
"kl_loss_coeff": 0.1,
|
34 |
+
"exploration_loss": "entropy",
|
35 |
+
"gae_lambda": 0.95,
|
36 |
+
"ppo_clip_ratio": 0.2,
|
37 |
+
"ppo_clip_value": 1.0,
|
38 |
+
"with_vtrace": false,
|
39 |
+
"vtrace_rho": 1.0,
|
40 |
+
"vtrace_c": 1.0,
|
41 |
+
"optimizer": "adam",
|
42 |
+
"adam_eps": 1e-06,
|
43 |
+
"adam_beta1": 0.9,
|
44 |
+
"adam_beta2": 0.999,
|
45 |
+
"max_grad_norm": 3.5,
|
46 |
+
"learning_rate": 0.00295,
|
47 |
+
"lr_schedule": "linear_decay",
|
48 |
+
"lr_schedule_kl_threshold": 0.008,
|
49 |
+
"lr_adaptive_min": 1e-06,
|
50 |
+
"lr_adaptive_max": 0.01,
|
51 |
+
"obs_subtract_mean": 0.0,
|
52 |
+
"obs_scale": 1.0,
|
53 |
+
"normalize_input": true,
|
54 |
+
"normalize_input_keys": null,
|
55 |
+
"decorrelate_experience_max_seconds": 0,
|
56 |
+
"decorrelate_envs_on_one_worker": true,
|
57 |
+
"actor_worker_gpus": [],
|
58 |
+
"set_workers_cpu_affinity": true,
|
59 |
+
"force_envs_single_thread": false,
|
60 |
+
"default_niceness": 0,
|
61 |
+
"log_to_file": true,
|
62 |
+
"experiment_summaries_interval": 3,
|
63 |
+
"flush_summaries_interval": 30,
|
64 |
+
"stats_avg": 100,
|
65 |
+
"summaries_use_frameskip": true,
|
66 |
+
"heartbeat_interval": 20,
|
67 |
+
"heartbeat_reporting_interval": 180,
|
68 |
+
"train_for_env_steps": 100000000,
|
69 |
+
"train_for_seconds": 10000000000,
|
70 |
+
"save_every_sec": 15,
|
71 |
+
"keep_checkpoints": 2,
|
72 |
+
"load_checkpoint_kind": "latest",
|
73 |
+
"save_milestones_sec": -1,
|
74 |
+
"save_best_every_sec": 5,
|
75 |
+
"save_best_metric": "reward",
|
76 |
+
"save_best_after": 100000,
|
77 |
+
"benchmark": false,
|
78 |
+
"encoder_mlp_layers": [
|
79 |
+
64,
|
80 |
+
64
|
81 |
+
],
|
82 |
+
"encoder_conv_architecture": "convnet_simple",
|
83 |
+
"encoder_conv_mlp_layers": [
|
84 |
+
512
|
85 |
+
],
|
86 |
+
"use_rnn": false,
|
87 |
+
"rnn_size": 512,
|
88 |
+
"rnn_type": "gru",
|
89 |
+
"rnn_num_layers": 1,
|
90 |
+
"decoder_mlp_layers": [],
|
91 |
+
"nonlinearity": "tanh",
|
92 |
+
"policy_initialization": "torch_default",
|
93 |
+
"policy_init_gain": 1.0,
|
94 |
+
"actor_critic_share_weights": true,
|
95 |
+
"adaptive_stddev": false,
|
96 |
+
"continuous_tanh_scale": 0.0,
|
97 |
+
"initial_stddev": 1.0,
|
98 |
+
"use_env_info_cache": false,
|
99 |
+
"env_gpu_actions": false,
|
100 |
+
"env_gpu_observations": true,
|
101 |
+
"env_frameskip": 1,
|
102 |
+
"env_framestack": 1,
|
103 |
+
"pixel_format": "CHW",
|
104 |
+
"use_record_episode_statistics": false,
|
105 |
+
"with_wandb": true,
|
106 |
+
"wandb_user": "qgallouedec",
|
107 |
+
"wandb_project": "sample_facotry_metaworld",
|
108 |
+
"wandb_group": null,
|
109 |
+
"wandb_job_type": "SF",
|
110 |
+
"wandb_tags": [],
|
111 |
+
"with_pbt": false,
|
112 |
+
"pbt_mix_policies_in_one_env": true,
|
113 |
+
"pbt_period_env_steps": 5000000,
|
114 |
+
"pbt_start_mutation": 20000000,
|
115 |
+
"pbt_replace_fraction": 0.3,
|
116 |
+
"pbt_mutation_rate": 0.15,
|
117 |
+
"pbt_replace_reward_gap": 0.1,
|
118 |
+
"pbt_replace_reward_gap_absolute": 1e-06,
|
119 |
+
"pbt_optimize_gamma": false,
|
120 |
+
"pbt_target_objective": "true_objective",
|
121 |
+
"pbt_perturb_min": 1.1,
|
122 |
+
"pbt_perturb_max": 1.5,
|
123 |
+
"command_line": "--env push-wall-v2 --experiment push-wall-v2 --with_wandb True --wandb_user qgallouedec --wandb_project sample_facotry_metaworld",
|
124 |
+
"cli_args": {
|
125 |
+
"env": "push-wall-v2",
|
126 |
+
"experiment": "push-wall-v2",
|
127 |
+
"with_wandb": true,
|
128 |
+
"wandb_user": "qgallouedec",
|
129 |
+
"wandb_project": "sample_facotry_metaworld"
|
130 |
+
},
|
131 |
+
"git_hash": "2bb8d8c9cd813ffeafaad42038b1e71364092d36",
|
132 |
+
"git_repo_name": "https://github.com/huggingface/gia",
|
133 |
+
"wandb_unique_id": "push-wall-v2_20230310_164216_024863"
|
134 |
+
}
|
git.diff
ADDED
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
diff --git a/data/envs/metaworld/generate_dataset.py b/data/envs/metaworld/generate_dataset.py
|
2 |
+
index dc51b4f..b189348 100644
|
3 |
+
--- a/data/envs/metaworld/generate_dataset.py
|
4 |
+
+++ b/data/envs/metaworld/generate_dataset.py
|
5 |
+
@@ -4,7 +4,9 @@ from typing import Dict, Optional
|
6 |
+
import gym
|
7 |
+
import metaworld
|
8 |
+
import numpy as np
|
9 |
+
+import pandas as pd
|
10 |
+
import torch
|
11 |
+
+from datasets import Dataset
|
12 |
+
from huggingface_hub import HfApi, repocard, upload_folder
|
13 |
+
from sample_factory.algo.learning.learner import Learner
|
14 |
+
from sample_factory.algo.sampling.batched_sampling import preprocess_actions
|
15 |
+
@@ -12,11 +14,7 @@ from sample_factory.algo.utils.action_distributions import argmax_actions
|
16 |
+
from sample_factory.algo.utils.env_info import extract_env_info
|
17 |
+
from sample_factory.algo.utils.make_env import make_env_func_batched
|
18 |
+
from sample_factory.algo.utils.rl_utils import make_dones, prepare_and_normalize_obs
|
19 |
+
-from sample_factory.cfg.arguments import (
|
20 |
+
- load_from_checkpoint,
|
21 |
+
- parse_full_cfg,
|
22 |
+
- parse_sf_args,
|
23 |
+
-)
|
24 |
+
+from sample_factory.cfg.arguments import load_from_checkpoint, parse_full_cfg, parse_sf_args
|
25 |
+
from sample_factory.envs.env_utils import register_env
|
26 |
+
from sample_factory.model.actor_critic import create_actor_critic
|
27 |
+
from sample_factory.model.model_utils import get_rnn_size
|
28 |
+
@@ -206,6 +204,13 @@ def create_dataset(cfg: Config):
|
29 |
+
|
30 |
+
env.close()
|
31 |
+
|
32 |
+
+ # Convert dict of numpy array to pandas dataframe
|
33 |
+
+ dataset = Dataset.from_dict(dataset)
|
34 |
+
+ dataset.create_config_id
|
35 |
+
+ # Set the card of the dataset
|
36 |
+
+ dataset.card = f""""""
|
37 |
+
+ dataset.push_to_hub("qgallouedec/prj_gia_dataset_metaworld_assembly_v2_1111_demo")
|
38 |
+
+
|
39 |
+
# Save dataset
|
40 |
+
repo_path = f"{cfg.train_dir}/datasets/{cfg.experiment}"
|
41 |
+
os.makedirs(repo_path, exist_ok=True)
|
replay.mp4
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3ead79bff4c485de44e3ea6dde978fde51f2e75ed666e0329c01b7c784eb803d
|
3 |
+
size 2194215
|
sf_log.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|