Add Asterix zero-latency Sample Factory checkpoint at 15M steps

zero-latency/asterix/small-policy/sample-factory-v1/best_000029312_15007744_reward_5879.000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c828d7bf77d598ff2624dfd8d511662bb54b0c3c9b1ead03b00757850dd9641
+size 14499833

zero-latency/asterix/small-policy/sample-factory-v1/config.json

@@ -0,0 +1,243 @@
+{
+  "help": false,
+  "algo": "APPO",
+  "env": "latency_gymnasium",
+  "experiment": "asterix_sf_zero_latency",
+  "train_dir": "results/sample_factory",
+  "restart_behavior": "resume",
+  "device": "gpu",
+  "seed": 0,
+  "initial_model_path": null,
+  "num_policies": 1,
+  "async_rl": true,
+  "serial_mode": false,
+  "batched_sampling": false,
+  "num_batches_to_accumulate": 2,
+  "worker_num_splits": 1,
+  "policy_workers_per_policy": 1,
+  "max_policy_lag": 300,
+  "num_workers": 4,
+  "num_envs_per_worker": 16,
+  "batch_size": 1024,
+  "num_batches_per_epoch": 8,
+  "num_epochs": 2,
+  "rollout": 64,
+  "fasttd3_replay_capacity": 6553600,
+  "fasttd3_replay_batch_size": 32768,
+  "fasttd3_action_chunk_horizon": 1,
+  "fasttd3_transitions_per_update": 64,
+  "fasttd3_train_for_optimizer_steps": 10000000000,
+  "fasttd3_v_min": -250.0,
+  "fasttd3_v_max": 250.0,
+  "fasttd3_actor_action_l2": 0.0,
+  "fasttd3_compile": true,
+  "recurrence": 32,
+  "shuffle_minibatches": false,
+  "gamma": 0.99,
+  "reward_scale": 1.0,
+  "reward_clip": 1000.0,
+  "value_bootstrap": false,
+  "normalize_returns": true,
+  "exploration_loss_coeff": 0.003,
+  "value_loss_coeff": 0.5,
+  "kl_loss_coeff": 0.0,
+  "exploration_loss": "entropy",
+  "gae_lambda": 0.95,
+  "ppo_clip_ratio": 0.1,
+  "ppo_clip_value": 0.2,
+  "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": 0.0,
+  "learning_rate": 0.000303,
+  "lr_schedule": "constant",
+  "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": 1,
+  "flush_summaries_interval": 30,
+  "stats_avg": 100,
+  "summaries_use_frameskip": true,
+  "heartbeat_interval": 20,
+  "heartbeat_reporting_interval": 180,
+  "train_for_env_steps": 15000000,
+  "train_for_seconds": 10000000000,
+  "save_every_sec": 600,
+  "keep_checkpoints": 5,
+  "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": [
+    512,
+    512
+  ],
+  "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": "elu",
+  "policy_initialization": "orthogonal",
+  "policy_init_gain": 1.0,
+  "actor_critic_share_weights": true,
+  "adaptive_stddev": true,
+  "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": "saberrr-zju",
+  "wandb_project": "starVLA_rl_games",
+  "wandb_group": "asterix",
+  "wandb_job_type": "sample_factory",
+  "wandb_tags": null,
+  "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,
+  "gym_task_name": "asterix",
+  "gym_env_id": "LatencyBench/Asterix-v0",
+  "gym_make_kwargs_json": "{\"base_env_id\": \"ALE/Asterix-v5\", \"render_mode\": \"rgb_array\", \"screen_size\": 84, \"noop_max\": 30, \"base_make_kwargs\": {\"obs_type\": \"rgb\", \"frameskip\": 1, \"repeat_action_probability\": 0.0, \"full_action_space\": false, \"mode\": 0, \"difficulty\": 0, \"max_num_frames_per_episode\": 108000}}",
+  "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_asterix\"]",
+  "gym_action_space_json": "{\"type\": \"discrete\", \"labels\": [\"noop\", \"up\", \"right\", \"left\", \"down\", \"upright\", \"upleft\", \"downright\", \"downleft\"], \"values\": [0, 1, 2, 3, 4, 5, 6, 7, 8]}",
+  "gym_action_labels_json": "[\"noop\", \"up\", \"right\", \"left\", \"down\", \"upright\", \"upleft\", \"downright\", \"downleft\"]",
+  "gym_action_values_json": "[0, 1, 2, 3, 4, 5, 6, 7, 8]",
+  "gym_noop_action_id": 0,
+  "gym_noop_action_json": "\"noop\"",
+  "gym_base_prompt": "Collect useful objects and avoid lyres. Choose exactly one action from: noop, up, right, left, down, upright, upleft, downright, downleft.",
+  "gym_state_labels_json": "",
+  "env_fps": 60.0,
+  "obs_fps": 15.0,
+  "frame_stack": 4,
+  "export_env_raw_rgb_frames": false,
+  "mode": "train",
+  "latency_type": "zero",
+  "fixed_latency_ms": null,
+  "mean_latency_ms": null,
+  "std_latency_ms": null,
+  "min_latency_ms": null,
+  "max_latency_ms": null,
+  "latency_seed": null,
+  "latency_profile_path": null,
+  "latency_profile_worker_slot": null,
+  "simulated_worker_capacity": null,
+  "add_latency_info": false,
+  "max_pending_actions": null,
+  "hold_last_chunk_action": false,
+  "hold_policy": "hold",
+  "ordering_policy": "issue_order_fifo",
+  "eval_episodes": 5,
+  "eval_parallel_envs": 1,
+  "eval_latency_raw_frame_values": null,
+  "eval_max_steps": 5000,
+  "eval_deterministic": true,
+  "eval_raw_reward": false,
+  "episode_metrics_path": "results/sample_factory/asterix_sf_zero_latency/episode_metrics.jsonl",
+  "ppo": null,
+  "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment asterix_sf_zero_latency --train_dir results/sample_factory --restart_behavior resume --device gpu --seed 0 --episode_metrics_path results/sample_factory/asterix_sf_zero_latency/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 4 --num_envs_per_worker 16 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 32 --num_epochs 2 --num_batches_per_epoch 8 --worker_num_splits 1 --max_policy_lag 300 --learning_rate 0.000303 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.1 --ppo_clip_value 0.2 --value_loss_coeff 0.5 --max_grad_norm 0.0 --save_every_sec 600 --keep_checkpoints 5 --stats_avg 100 --experiment_summaries_interval 1 --async_rl True --use_rnn False --normalize_returns True --normalize_input True --latency-type zero --add-latency-info False --eval-episodes 5 --eval-parallel-envs 1 --eval-max-steps 5000 --eval-deterministic True --with_wandb True --wandb_project starVLA_rl_games --wandb_user saberrr-zju --wandb_group asterix --wandb_job_type sample_factory --gym-task-name asterix --gym-env-id LatencyBench/Asterix-v0 --gym-make-kwargs-json {\"base_env_id\": \"ALE/Asterix-v5\", \"render_mode\": \"rgb_array\", \"screen_size\": 84, \"noop_max\": 30, \"base_make_kwargs\": {\"obs_type\": \"rgb\", \"frameskip\": 1, \"repeat_action_probability\": 0.0, \"full_action_space\": false, \"mode\": 0, \"difficulty\": 0, \"max_num_frames_per_episode\": 108000}} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_asterix\"] --gym-action-space-json {\"type\": \"discrete\", \"labels\": [\"noop\", \"up\", \"right\", \"left\", \"down\", \"upright\", \"upleft\", \"downright\", \"downleft\"], \"values\": [0, 1, 2, 3, 4, 5, 6, 7, 8]} --gym-noop-action-json \"noop\" --gym-action-labels-json [\"noop\", \"up\", \"right\", \"left\", \"down\", \"upright\", \"upleft\", \"downright\", \"downleft\"] --gym-action-values-json [0, 1, 2, 3, 4, 5, 6, 7, 8] --gym-noop-action-id 0 --gym-base-prompt Collect useful objects and avoid lyres. Choose exactly one action from: noop, up, right, left, down, upright, upleft, downright, downleft. --env-fps 60 --obs-fps 15 --frame-stack 4 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False",
+  "cli_args": {
+    "algo": "APPO",
+    "env": "latency_gymnasium",
+    "experiment": "asterix_sf_zero_latency",
+    "train_dir": "results/sample_factory",
+    "restart_behavior": "resume",
+    "device": "gpu",
+    "seed": 0,
+    "num_policies": 1,
+    "async_rl": true,
+    "worker_num_splits": 1,
+    "max_policy_lag": 300,
+    "num_workers": 4,
+    "num_envs_per_worker": 16,
+    "batch_size": 1024,
+    "num_batches_per_epoch": 8,
+    "num_epochs": 2,
+    "rollout": 64,
+    "recurrence": 32,
+    "gamma": 0.99,
+    "normalize_returns": true,
+    "value_loss_coeff": 0.5,
+    "gae_lambda": 0.95,
+    "ppo_clip_ratio": 0.1,
+    "ppo_clip_value": 0.2,
+    "max_grad_norm": 0.0,
+    "learning_rate": 0.000303,
+    "normalize_input": true,
+    "experiment_summaries_interval": 1,
+    "stats_avg": 100,
+    "train_for_env_steps": 10000000,
+    "save_every_sec": 600,
+    "keep_checkpoints": 5,
+    "use_rnn": false,
+    "use_env_info_cache": false,
+    "with_wandb": true,
+    "wandb_user": "saberrr-zju",
+    "wandb_project": "starVLA_rl_games",
+    "wandb_group": "asterix",
+    "wandb_job_type": "sample_factory",
+    "gym_task_name": "asterix",
+    "gym_env_id": "LatencyBench/Asterix-v0",
+    "gym_make_kwargs_json": "{\"base_env_id\": \"ALE/Asterix-v5\", \"render_mode\": \"rgb_array\", \"screen_size\": 84, \"noop_max\": 30, \"base_make_kwargs\": {\"obs_type\": \"rgb\", \"frameskip\": 1, \"repeat_action_probability\": 0.0, \"full_action_space\": false, \"mode\": 0, \"difficulty\": 0, \"max_num_frames_per_episode\": 108000}}",
+    "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_asterix\"]",
+    "gym_action_space_json": "{\"type\": \"discrete\", \"labels\": [\"noop\", \"up\", \"right\", \"left\", \"down\", \"upright\", \"upleft\", \"downright\", \"downleft\"], \"values\": [0, 1, 2, 3, 4, 5, 6, 7, 8]}",
+    "gym_action_labels_json": "[\"noop\", \"up\", \"right\", \"left\", \"down\", \"upright\", \"upleft\", \"downright\", \"downleft\"]",
+    "gym_action_values_json": "[0, 1, 2, 3, 4, 5, 6, 7, 8]",
+    "gym_noop_action_id": 0,
+    "gym_noop_action_json": "\"noop\"",
+    "gym_base_prompt": "Collect useful objects and avoid lyres. Choose exactly one action from: noop, up, right, left, down, upright, upleft, downright, downleft.",
+    "env_fps": 60.0,
+    "obs_fps": 15.0,
+    "frame_stack": 4,
+    "mode": "train",
+    "latency_type": "zero",
+    "add_latency_info": false,
+    "hold_policy": "hold",
+    "ordering_policy": "issue_order_fifo",
+    "eval_episodes": 5,
+    "eval_parallel_envs": 1,
+    "eval_max_steps": 5000,
+    "eval_deterministic": true,
+    "episode_metrics_path": "results/sample_factory/asterix_sf_zero_latency/episode_metrics.jsonl",
+    "output_dir": "outputs/asterix_sf_zero_latency"
+  },
+  "git_hash": "68dc7e0f7cae528a85410312ddf9bf78c6a1cf2a",
+  "git_repo_name": "/home/ubuntu/codex/atari-zero-latency-master.bundle",
+  "eval_env_frameskip": 1,
+  "output_dir": "outputs/asterix_sf_zero_latency",
+  "wandb_unique_id": "asterix_sf_zero_latency"
+}