Publish Demon Attack Pi05 temporal H8 profiled small checkpoint

latency-aware/demon-attack/small-policy/sample-factory-pi05-rtx3090-temporal-hist8-capacity1-s42-v1/README.md

@@ -0,0 +1,10 @@
+# demon attack — pi05 temporal H8 profiled small policy
+
+Sample Factory APPO small policy trained with an RTX 3090 temporal latency profile, one simulated inference worker, and an eight-decision causal history. This folder contains the training-selected best checkpoint and its exact Sample Factory config.
+
+- W&B run ID: `demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1-s42`
+- Checkpoint: `best_000047936_24543232_reward_2760.000.pth`
+- Training seed: 42; target: 25 million environment steps.
+- Profile SHA256: `5c2895d34b3d821c7b24a14d9acd84e416baec9eebc87083ba86b9c8954696ae`
+
+The reward in the checkpoint filename is a training selection metric, not a formal evaluation result. See `provenance.json` for exact source paths and hashes.

latency-aware/demon-attack/small-policy/sample-factory-pi05-rtx3090-temporal-hist8-capacity1-s42-v1/best_000047936_24543232_reward_2760.000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d56843071572d2872d2f3a724fa4cc08e8254bd4086d3f149346585814fd890
+size 24472102

latency-aware/demon-attack/small-policy/sample-factory-pi05-rtx3090-temporal-hist8-capacity1-s42-v1/config.json

@@ -0,0 +1,278 @@
+{
+  "help": false,
+  "algo": "APPO",
+  "env": "latency_demon_attack",
+  "experiment": "demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42",
+  "train_dir": "results/checkpoints/demon_attack/decision_history_temporal_25m_seed42",
+  "restart_behavior": "resume",
+  "device": "gpu",
+  "seed": 42,
+  "initial_model_path": null,
+  "num_policies": 1,
+  "async_rl": true,
+  "serial_mode": false,
+  "batched_sampling": true,
+  "num_batches_to_accumulate": 2,
+  "worker_num_splits": 1,
+  "policy_workers_per_policy": 1,
+  "max_policy_lag": 300,
+  "num_workers": 1,
+  "num_envs_per_worker": 1,
+  "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-05,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "max_grad_norm": 0.0,
+  "learning_rate": 0.000303,
+  "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": 255.0,
+  "normalize_input": true,
+  "normalize_input_keys": null,
+  "decorrelate_experience_max_seconds": 0,
+  "decorrelate_envs_on_one_worker": true,
+  "actor_worker_gpus": [
+    0
+  ],
+  "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": 25000000,
+  "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_atari",
+  "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": true,
+  "env_gpu_observations": true,
+  "env_frameskip": 1,
+  "env_framestack": 1,
+  "pixel_format": "CHW",
+  "use_record_episode_statistics": false,
+  "episode_counter": false,
+  "with_wandb": true,
+  "wandb_user": null,
+  "wandb_project": "starvla_tasks",
+  "wandb_group": "demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1",
+  "wandb_job_type": "demon_attack_decision_history_temporal",
+  "wandb_tags": [
+    "demon_attack_decision_history_temporal_25m_seed42",
+    "demon-attack",
+    "frame_stack",
+    "temporal",
+    "pi05_rtx3090_temporal_hist8_capacity1",
+    "fs4",
+    "seed42"
+  ],
+  "wandb_dir": "/mnt/ssd/latency-sensitive-bench/wandb",
+  "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,
+  "env_id": "ALE/DemonAttack-v5",
+  "env_fps": 60.0,
+  "obs_fps": 15.0,
+  "screen_size": 84,
+  "noop_max": 30,
+  "reset_mode": "fire",
+  "frame_stack": 4,
+  "action_history_decisions": 8,
+  "terminal_on_life_loss": true,
+  "clip_reward": true,
+  "repeat_action_probability": 0.0,
+  "full_action_space": false,
+  "max_episode_steps": 25000,
+  "simulator": "gpu",
+  "gpu_env_device": "auto",
+  "gpu_env_batch_size": 512,
+  "export_env_raw_rgb_frames": false,
+  "mode": "train",
+  "latency_type": "profile_sample",
+  "fixed_latency_ms": null,
+  "mean_latency_ms": null,
+  "std_latency_ms": null,
+  "min_latency_ms": null,
+  "max_latency_ms": null,
+  "latency_seed": 42,
+  "latency_profile_path": "results/profiling/profiles/pi05/1x-rtx3090/demon_attack/instance_a5037b165aa0cedc/profile.json",
+  "latency_profile_worker_slot": 0,
+  "simulated_worker_capacity": 1,
+  "add_latency_info": false,
+  "max_pending_actions": null,
+  "hold_policy": "hold",
+  "ordering_policy": "latest_ready",
+  "eval_episodes": 100,
+  "eval_parallel_envs": 16,
+  "eval_latency_raw_frame_values": "0,1,2,3,4,5",
+  "eval_max_steps": 14400,
+  "eval_deterministic": true,
+  "eval_raw_reward": true,
+  "episode_metrics_path": "results/checkpoints/demon_attack/decision_history_temporal_25m_seed42/demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42/episode_metrics.jsonl",
+  "command_line": "--mode train --algo APPO --env latency_demon_attack --experiment demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42 --train_dir results/checkpoints/demon_attack/decision_history_temporal_25m_seed42 --restart_behavior resume --device gpu --actor_worker_gpus 0 --env_gpu_observations True --env_gpu_actions True --gpu-env-batch-size 512 --seed 42 --episode_metrics_path results/checkpoints/demon_attack/decision_history_temporal_25m_seed42/demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42/episode_metrics.jsonl --train_for_env_steps 25000000 --num_workers 1 --num_envs_per_worker 1 --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 --lr_schedule linear_decay --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 --batched_sampling True --async_rl True --batched_sampling True --use_rnn False --normalize_returns True --normalize_input True --latency-type profile_sample --latency-seed 42 --latency-profile-path results/profiling/profiles/pi05/1x-rtx3090/demon_attack/instance_a5037b165aa0cedc/profile.json --latency-profile-worker-slot 0 --simulated-worker-capacity 1 --add-latency-info False --eval-episodes 100 --eval-parallel-envs 16 --eval-max-steps 14400 --eval-deterministic True --eval-raw-reward --with_wandb True --wandb_project starvla_tasks --wandb_group demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1 --wandb_job_type demon_attack_decision_history_temporal --wandb_tags demon_attack_decision_history_temporal_25m_seed42 demon-attack frame_stack temporal pi05_rtx3090_temporal_hist8_capacity1 fs4 seed42 --env_id ALE/DemonAttack-v5 --env-fps 60 --obs-fps 15.0 --screen_size 84 --noop_max 30 --reset-mode fire --frame_stack 4 --action-history-decisions 8 --terminal_on_life_loss True --clip_reward True --repeat_action_probability 0.0 --full_action_space False --max_episode_steps 25000 --simulator gpu --gpu-env-device auto --hold-policy hold --ordering-policy latest_ready",
+  "cli_args": {
+    "algo": "APPO",
+    "env": "latency_demon_attack",
+    "experiment": "demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42",
+    "train_dir": "results/checkpoints/demon_attack/decision_history_temporal_25m_seed42",
+    "restart_behavior": "resume",
+    "device": "gpu",
+    "seed": 42,
+    "num_policies": 1,
+    "async_rl": true,
+    "batched_sampling": true,
+    "worker_num_splits": 1,
+    "max_policy_lag": 300,
+    "num_workers": 1,
+    "num_envs_per_worker": 1,
+    "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,
+    "lr_schedule": "linear_decay",
+    "normalize_input": true,
+    "actor_worker_gpus": [
+      0
+    ],
+    "experiment_summaries_interval": 1,
+    "stats_avg": 100,
+    "train_for_env_steps": 25000000,
+    "save_every_sec": 600,
+    "keep_checkpoints": 5,
+    "use_rnn": false,
+    "env_gpu_actions": true,
+    "env_gpu_observations": true,
+    "with_wandb": true,
+    "wandb_project": "starvla_tasks",
+    "wandb_group": "demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1",
+    "wandb_job_type": "demon_attack_decision_history_temporal",
+    "wandb_tags": [
+      "demon_attack_decision_history_temporal_25m_seed42",
+      "demon-attack",
+      "frame_stack",
+      "temporal",
+      "pi05_rtx3090_temporal_hist8_capacity1",
+      "fs4",
+      "seed42"
+    ],
+    "env_id": "ALE/DemonAttack-v5",
+    "env_fps": 60.0,
+    "obs_fps": 15.0,
+    "screen_size": 84,
+    "noop_max": 30,
+    "reset_mode": "fire",
+    "frame_stack": 4,
+    "action_history_decisions": 8,
+    "terminal_on_life_loss": true,
+    "clip_reward": true,
+    "repeat_action_probability": 0.0,
+    "full_action_space": false,
+    "max_episode_steps": 25000,
+    "simulator": "gpu",
+    "gpu_env_device": "auto",
+    "gpu_env_batch_size": 512,
+    "mode": "train",
+    "latency_type": "profile_sample",
+    "latency_seed": 42,
+    "latency_profile_path": "results/profiling/profiles/pi05/1x-rtx3090/demon_attack/instance_a5037b165aa0cedc/profile.json",
+    "latency_profile_worker_slot": 0,
+    "simulated_worker_capacity": 1,
+    "add_latency_info": false,
+    "hold_policy": "hold",
+    "ordering_policy": "latest_ready",
+    "eval_episodes": 100,
+    "eval_parallel_envs": 16,
+    "eval_max_steps": 14400,
+    "eval_deterministic": true,
+    "eval_raw_reward": true,
+    "episode_metrics_path": "results/checkpoints/demon_attack/decision_history_temporal_25m_seed42/demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42/episode_metrics.jsonl",
+    "output_dir": "outputs/tasks/demon_attack/decision_history_temporal_25m_seed42/train/frame_stack/pi05_rtx3090_temporal_hist8_capacity1/fs4/seed_42",
+    "wandb_unique_id": "demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1-s42"
+  },
+  "git_hash": "unknown",
+  "git_repo_name": "not a git repository",
+  "eval_env_frameskip": 1,
+  "output_dir": "outputs/tasks/demon_attack/decision_history_temporal_25m_seed42/train/frame_stack/pi05_rtx3090_temporal_hist8_capacity1/fs4/seed_42",
+  "wandb_unique_id": "demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1-s42"
+}

latency-aware/demon-attack/small-policy/sample-factory-pi05-rtx3090-temporal-hist8-capacity1-s42-v1/provenance.json

@@ -0,0 +1,16 @@
+{
+  "task": "demon_attack",
+  "method": "Sample Factory APPO; decision_v1 history of eight completed decisions",
+  "training_seed": 42,
+  "training_target_env_steps": 25000000,
+  "wandb_run_id": "demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1-s42",
+  "wandb_source_run": "saberrr-zju/starvla_tasks/demon-attack-fs4-pi05_rtx3090_temporal_hist8_capacity1-s42",
+  "source_checkpoint": "/mnt/ssd/latency-sensitive-bench/results/checkpoints/demon_attack/decision_history_temporal_25m_seed42/demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42/checkpoint_p0/best_000047936_24543232_reward_2760.000.pth",
+  "checkpoint_sha256": "0d56843071572d2872d2f3a724fa4cc08e8254bd4086d3f149346585814fd890",
+  "checkpoint_selection": "training best",
+  "source_config": "/mnt/ssd/latency-sensitive-bench/results/checkpoints/demon_attack/decision_history_temporal_25m_seed42/demon_attack_frame_stack_pi05_rtx3090_temporal_hist8_capacity1_fs4_seed42/config.json",
+  "config_sha256": "fd985f9894cff1a8aa5a61322b8fa2af3704ad89de6d56202297a26453c14436",
+  "source_profile_path": "/mnt/ssd/latency-sensitive-bench/results/profiling/profiles/pi05/1x-rtx3090/demon_attack/instance_a5037b165aa0cedc/profile.json",
+  "profile_sha256": "5c2895d34b3d821c7b24a14d9acd84e416baec9eebc87083ba86b9c8954696ae",
+  "hf_path": "latency-aware/demon-attack/small-policy/sample-factory-pi05-rtx3090-temporal-hist8-capacity1-s42-v1"
+}