Publish completed ant RTX3090 mean-profile teacher

latency-aware/ant/small-policy/sample-factory-qwenoft-rtx3090-mean-v1/README.md

@@ -0,0 +1,5 @@
+# ant Sample Factory mean-profile teacher
+
+Completed 10006528 frames with seed 2222; budget 10000000. Selection: best_reward_after_full_training_budget. Fixed mean latency 90.56460638563993 ms from archived RTX3090 profile; training GPU was RTX PRO6000 Blackwell.
+
+`checkpoint.pth` is a tensor-only inference export with `model`, `train_step`, `env_steps`; every model tensor compared exactly with selected training checkpoint. Original recovery state remains on host. Training reward in source filename is not independent task evaluation.

latency-aware/ant/small-policy/sample-factory-qwenoft-rtx3090-mean-v1/checkpoint.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7899b32beb9b2cfec1bcff85b285dddc243a73f1c322fafefae599df0f67b3b
+size 33077

latency-aware/ant/small-policy/sample-factory-qwenoft-rtx3090-mean-v1/config.json

@@ -0,0 +1,297 @@
+{
+  "help": false,
+  "algo": "APPO",
+  "env": "latency_gymnasium",
+  "experiment": "ant_mean_3090_s2222",
+  "train_dir": "/home/ubuntu/lzj/mean-profiling/ant/teacher",
+  "restart_behavior": "overwrite",
+  "device": "gpu",
+  "seed": 2222,
+  "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": 10000,
+  "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": 1.0,
+  "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": 10,
+  "decorrelate_envs_on_one_worker": true,
+  "actor_worker_gpus": [],
+  "set_workers_cpu_affinity": true,
+  "force_envs_single_thread": true,
+  "default_niceness": 0,
+  "log_to_file": true,
+  "experiment_summaries_interval": 10,
+  "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": 600,
+  "keep_checkpoints": 3,
+  "load_checkpoint_kind": "latest",
+  "save_milestones_sec": -1,
+  "save_best_every_sec": 60,
+  "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": "dongqianyu99-zhejiang-university",
+  "wandb_project": "starvla_tasks",
+  "wandb_group": "mean-profiling-3090-blackwell",
+  "wandb_job_type": "SF",
+  "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": "ant_rgb_state",
+  "gym_env_id": "LatencyBench/AntContinuous-v0",
+  "gym_make_kwargs_json": "{\"base_env_id\": \"Ant-v4\", \"base_make_kwargs\": {\"exclude_current_positions_from_observation\": true, \"use_contact_forces\": false}, \"render_mode\": \"rgb_array\"}",
+  "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_ant\"]",
+  "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_right_hip_torque\", \"back_right_ankle_torque\", \"front_left_hip_torque\", \"front_left_ankle_torque\", \"front_right_hip_torque\", \"front_right_ankle_torque\", \"back_left_hip_torque\", \"back_left_ankle_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}",
+  "gym_action_labels_json": null,
+  "gym_action_values_json": null,
+  "gym_noop_action_id": null,
+  "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]",
+  "gym_base_prompt": "Make the Ant move forward as fast as possible without falling. Predict eight continuous torques in [-1, 1] ordered as back right hip, back right ankle, front left hip, front left ankle, front right hip, front right ankle, back left hip, and back left ankle.",
+  "gym_state_labels_json": "[\"torso_z\", \"torso_quaternion_w\", \"torso_quaternion_x\", \"torso_quaternion_y\", \"torso_quaternion_z\", \"front_left_hip_angle\", \"front_left_ankle_angle\", \"front_right_hip_angle\", \"front_right_ankle_angle\", \"back_left_hip_angle\", \"back_left_ankle_angle\", \"back_right_hip_angle\", \"back_right_ankle_angle\", \"torso_x_velocity\", \"torso_y_velocity\", \"torso_z_velocity\", \"torso_angular_velocity_x\", \"torso_angular_velocity_y\", \"torso_angular_velocity_z\", \"front_left_hip_angular_velocity\", \"front_left_ankle_angular_velocity\", \"front_right_hip_angular_velocity\", \"front_right_ankle_angular_velocity\", \"back_left_hip_angular_velocity\", \"back_left_ankle_angular_velocity\", \"back_right_hip_angular_velocity\", \"back_right_ankle_angular_velocity\"]",
+  "env_fps": 10.0,
+  "obs_fps": 10.0,
+  "frame_stack": 1,
+  "action_history_decisions": null,
+  "rollout_profile_seed_by_reset": false,
+  "simulator": "cpu",
+  "gpu_env_device": "auto",
+  "gpu_env_batch_size": 1,
+  "noop_max": null,
+  "terminal_on_life_loss": null,
+  "clip_reward": null,
+  "max_episode_steps": null,
+  "export_env_raw_rgb_frames": false,
+  "mode": "train",
+  "latency_type": "fixed",
+  "fixed_latency_ms": 90.56460638563993,
+  "mean_latency_ms": null,
+  "std_latency_ms": null,
+  "min_latency_ms": null,
+  "max_latency_ms": null,
+  "latency_seed": 0,
+  "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": 10,
+  "eval_parallel_envs": 1,
+  "eval_latency_raw_frame_values": null,
+  "eval_max_steps": 1000,
+  "eval_deterministic": true,
+  "eval_raw_reward": false,
+  "episode_metrics_path": "/home/ubuntu/lzj/mean-profiling/ant/teacher/ant_mean_3090_s2222/episode_metrics.jsonl",
+  "ppo": null,
+  "command_line": "--mode train --algo APPO --env latency_gymnasium --experiment ant_profile_20260911T112232Z --train_dir /mnt/results/latency-sensitive-bench/profile-latency-new/ant/20260911T112232Z-ant/small_models --restart_behavior overwrite --device gpu --seed 2222 --episode_metrics_path /mnt/results/latency-sensitive-bench/profile-latency-new/ant/20260911T112232Z-ant/small_models/ant_profile_20260911T112232Z/episode_metrics.jsonl --train_for_env_steps 10000000 --num_workers 8 --num_envs_per_worker 8 --num_policies 1 --batch_size 1024 --rollout 64 --recurrence 1 --num_epochs 2 --num_batches_per_epoch 4 --worker_num_splits 2 --max_policy_lag 10000 --num_batches_to_accumulate 2 --policy_workers_per_policy 1 --learning_rate 0.00295 --kl_loss_coeff 0.1 --lr_schedule_kl_threshold 0.008 --nonlinearity tanh --policy_initialization torch_default --continuous_tanh_scale 0.0 --initial_stddev 1.0 --env_framestack 1 --gamma 0.99 --gae_lambda 0.95 --ppo_clip_ratio 0.2 --ppo_clip_value 1.0 --exploration_loss entropy --exploration_loss_coeff 0.0 --reward_scale 1.0 --reward_clip 1000.0 --value_loss_coeff 1.3 --max_grad_norm 3.5 --optimizer adam --adam_eps 1e-06 --adam_beta1 0.9 --adam_beta2 0.999 --obs_subtract_mean 0.0 --obs_scale 1.0 --decorrelate_experience_max_seconds 10 --default_niceness 0 --rnn_type gru --rnn_size 512 --save_every_sec 600 --keep_checkpoints 3 --save_milestones_sec -1 --save_best_every_sec 60 --save_best_after 100000 --stats_avg 100 --experiment_summaries_interval 10 --fasttd3_replay_capacity 6553600 --fasttd3_replay_batch_size 32768 --fasttd3_transitions_per_update 64 --fasttd3_v_min -250.0 --fasttd3_v_max 250.0 --fasttd3_actor_action_l2 0.0 --async_rl False --batched_sampling False --serial_mode False --use_rnn False --normalize_returns True --normalize_input True --adaptive_stddev False --shuffle_minibatches False --value_bootstrap True --with_vtrace False --decorrelate_envs_on_one_worker True --set_workers_cpu_affinity True --force_envs_single_thread True --actor_critic_share_weights True --with_wandb False --fasttd3_compile True --encoder_mlp_layers 64 64 --latency-type iid --latency-seed 0 --latency-profile-path /mnt/results/latency-sensitive-bench/profile-latency-new/ant/20260911T112232Z-ant/profile/profile.json --latency-profile-worker-slot 0 --add-latency-info False --eval-episodes 10 --eval-parallel-envs 1 --eval-max-steps 1000 --eval-deterministic True --gym-task-name ant_rgb_state --gym-env-id LatencyBench/AntContinuous-v0 --gym-make-kwargs-json {\"base_env_id\": \"Ant-v4\", \"base_make_kwargs\": {\"exclude_current_positions_from_observation\": true, \"use_contact_forces\": false}, \"render_mode\": \"rgb_array\"} --gym-registration-imports-json [\"latency_bench.envs.gymnasium_ant\"] --gym-action-space-json {\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_right_hip_torque\", \"back_right_ankle_torque\", \"front_left_hip_torque\", \"front_left_ankle_torque\", \"front_right_hip_torque\", \"front_right_ankle_torque\", \"back_left_hip_torque\", \"back_left_ankle_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"} --gym-noop-action-json [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0] --gym-base-prompt Make the Ant move forward as fast as possible without falling. Predict eight continuous torques in [-1, 1] ordered as back right hip, back right ankle, front left hip, front left ankle, front right hip, front right ankle, back left hip, and back left ankle. --gym-state-labels-json [\"torso_z\", \"torso_quaternion_w\", \"torso_quaternion_x\", \"torso_quaternion_y\", \"torso_quaternion_z\", \"front_left_hip_angle\", \"front_left_ankle_angle\", \"front_right_hip_angle\", \"front_right_ankle_angle\", \"back_left_hip_angle\", \"back_left_ankle_angle\", \"back_right_hip_angle\", \"back_right_ankle_angle\", \"torso_x_velocity\", \"torso_y_velocity\", \"torso_z_velocity\", \"torso_angular_velocity_x\", \"torso_angular_velocity_y\", \"torso_angular_velocity_z\", \"front_left_hip_angular_velocity\", \"front_left_ankle_angular_velocity\", \"front_right_hip_angular_velocity\", \"front_right_ankle_angular_velocity\", \"back_left_hip_angular_velocity\", \"back_left_ankle_angular_velocity\", \"back_right_hip_angular_velocity\", \"back_right_ankle_angular_velocity\"] --env-fps 10.0 --obs-fps 10.0 --frame-stack 1 --hold-policy hold --ordering-policy issue_order_fifo --use_env_info_cache False",
+  "cli_args": {
+    "algo": "APPO",
+    "env": "latency_gymnasium",
+    "experiment": "ant_mean_3090_s2222",
+    "train_dir": "/home/ubuntu/lzj/mean-profiling/ant/teacher",
+    "restart_behavior": "overwrite",
+    "device": "gpu",
+    "seed": 2222,
+    "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": 10000,
+    "num_workers": 8,
+    "num_envs_per_worker": 8,
+    "batch_size": 1024,
+    "num_batches_per_epoch": 4,
+    "num_epochs": 2,
+    "rollout": 64,
+    "fasttd3_replay_capacity": 6553600,
+    "fasttd3_replay_batch_size": 32768,
+    "fasttd3_transitions_per_update": 64,
+    "fasttd3_v_min": -250.0,
+    "fasttd3_v_max": 250.0,
+    "fasttd3_actor_action_l2": 0.0,
+    "fasttd3_compile": true,
+    "recurrence": 1,
+    "shuffle_minibatches": false,
+    "gamma": 0.99,
+    "reward_scale": 1.0,
+    "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,
+    "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_kl_threshold": 0.008,
+    "obs_subtract_mean": 0.0,
+    "obs_scale": 1.0,
+    "normalize_input": true,
+    "decorrelate_experience_max_seconds": 10,
+    "decorrelate_envs_on_one_worker": true,
+    "set_workers_cpu_affinity": true,
+    "force_envs_single_thread": true,
+    "default_niceness": 0,
+    "experiment_summaries_interval": 10,
+    "stats_avg": 100,
+    "train_for_env_steps": 10000000,
+    "save_every_sec": 600,
+    "keep_checkpoints": 3,
+    "save_milestones_sec": -1,
+    "save_best_every_sec": 60,
+    "save_best_after": 100000,
+    "encoder_mlp_layers": [
+      64,
+      64
+    ],
+    "use_rnn": false,
+    "rnn_size": 512,
+    "rnn_type": "gru",
+    "nonlinearity": "tanh",
+    "policy_initialization": "torch_default",
+    "actor_critic_share_weights": true,
+    "adaptive_stddev": false,
+    "continuous_tanh_scale": 0.0,
+    "initial_stddev": 1.0,
+    "use_env_info_cache": false,
+    "env_framestack": 1,
+    "with_wandb": true,
+    "gym_task_name": "ant_rgb_state",
+    "gym_env_id": "LatencyBench/AntContinuous-v0",
+    "gym_make_kwargs_json": "{\"base_env_id\": \"Ant-v4\", \"base_make_kwargs\": {\"exclude_current_positions_from_observation\": true, \"use_contact_forces\": false}, \"render_mode\": \"rgb_array\"}",
+    "gym_registration_imports_json": "[\"latency_bench.envs.gymnasium_ant\"]",
+    "gym_action_space_json": "{\"dtype\": \"float32\", \"high\": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], \"labels\": [\"back_right_hip_torque\", \"back_right_ankle_torque\", \"front_left_hip_torque\", \"front_left_ankle_torque\", \"front_right_hip_torque\", \"front_right_ankle_torque\", \"back_left_hip_torque\", \"back_left_ankle_torque\"], \"low\": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0], \"type\": \"box\"}",
+    "gym_noop_action_json": "[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]",
+    "gym_base_prompt": "Make the Ant move forward as fast as possible without falling. Predict eight continuous torques in [-1, 1] ordered as back right hip, back right ankle, front left hip, front left ankle, front right hip, front right ankle, back left hip, and back left ankle.",
+    "gym_state_labels_json": "[\"torso_z\", \"torso_quaternion_w\", \"torso_quaternion_x\", \"torso_quaternion_y\", \"torso_quaternion_z\", \"front_left_hip_angle\", \"front_left_ankle_angle\", \"front_right_hip_angle\", \"front_right_ankle_angle\", \"back_left_hip_angle\", \"back_left_ankle_angle\", \"back_right_hip_angle\", \"back_right_ankle_angle\", \"torso_x_velocity\", \"torso_y_velocity\", \"torso_z_velocity\", \"torso_angular_velocity_x\", \"torso_angular_velocity_y\", \"torso_angular_velocity_z\", \"front_left_hip_angular_velocity\", \"front_left_ankle_angular_velocity\", \"front_right_hip_angular_velocity\", \"front_right_ankle_angular_velocity\", \"back_left_hip_angular_velocity\", \"back_left_ankle_angular_velocity\", \"back_right_hip_angular_velocity\", \"back_right_ankle_angular_velocity\"]",
+    "env_fps": 10.0,
+    "obs_fps": 10.0,
+    "frame_stack": 1,
+    "mode": "train",
+    "latency_type": "fixed",
+    "latency_seed": 0,
+    "latency_profile_path": null,
+    "latency_profile_worker_slot": null,
+    "add_latency_info": false,
+    "hold_policy": "hold",
+    "ordering_policy": "issue_order_fifo",
+    "eval_episodes": 10,
+    "eval_parallel_envs": 1,
+    "eval_max_steps": 1000,
+    "eval_deterministic": true,
+    "episode_metrics_path": "/home/ubuntu/lzj/mean-profiling/ant/teacher/ant_mean_3090_s2222/episode_metrics.jsonl",
+    "output_dir": "/home/ubuntu/lzj/mean-profiling/ant/teacher-log",
+    "wandb_unique_id": "mean3090-ant-teacher-blackwell-20260930-nice0"
+  },
+  "git_hash": "unknown",
+  "git_repo_name": "not a git repository",
+  "eval_env_frameskip": 1,
+  "initial_model_path": null,
+  "fasttd3_replay_capacity": 6553600,
+  "fasttd3_replay_batch_size": 32768,
+  "fasttd3_transitions_per_update": 64,
+  "fasttd3_v_min": -250.0,
+  "fasttd3_v_max": 250.0,
+  "fasttd3_actor_action_l2": 0.0,
+  "fasttd3_compile": true,
+  "fasttd3_sonic_decoder_path": null,
+  "output_dir": "/home/ubuntu/lzj/mean-profiling/ant/teacher-log",
+  "wandb_dir": "/home/ubuntu/lzj/mean-profiling/ant/wandb",
+  "wandb_unique_id": "mean3090-ant-teacher-blackwell-20260930-nice0"
+}

latency-aware/ant/small-policy/sample-factory-qwenoft-rtx3090-mean-v1/files.sha256.json

@@ -0,0 +1,22 @@
+{
+  "checkpoint.pth": {
+    "bytes": 33077,
+    "sha256": "a7899b32beb9b2cfec1bcff85b285dddc243a73f1c322fafefae599df0f67b3b"
+  },
+  "provenance.json": {
+    "bytes": 1197,
+    "sha256": "2b5f1bd56c5901c0ee7bbfff43cd6ed0b8c4482a7dd1fb44cefc18c2a35b4659"
+  },
+  "README.md": {
+    "bytes": 540,
+    "sha256": "07c4ca08b1592f55bb180f2bb03b0d18970acc9f284029d454dc58d011413363"
+  },
+  "profile.source.json": {
+    "bytes": 2148,
+    "sha256": "0e49f72ec05ac600a54e07bbca5af3143708444d606167f33fa21788a9fefe50"
+  },
+  "config.json": {
+    "bytes": 16688,
+    "sha256": "1c699cf792010d4fc8856ca4f6949c8c1946993b53e9031e119d263b75d30d33"
+  }
+}

latency-aware/ant/small-policy/sample-factory-qwenoft-rtx3090-mean-v1/profile.source.json

@@ -0,0 +1,66 @@
+{
+  "burst_model_path": "latency_burst_model.json",
+  "distribution_path": "latency_distribution.json",
+  "env_fps": 10,
+  "frame_ms": 100.0,
+  "gpu_class": "1x-rtx3090",
+  "instance_id": "instance_859cf1e47bca6046",
+  "latency_kind": "observation_to_action_latency",
+  "latency_method": "temporal",
+  "model_id": "qwenoft",
+  "n_admitted_observations": 7976,
+  "n_capacity_drops": 294,
+  "n_observation_attempts": 8270,
+  "per_slot_summary": {
+    "0": {
+      "admitted_count": 7976,
+      "mean_observation_to_action_latency_ms": 90.56460638563993,
+      "mean_worker_service_time_ms": 90.22241150451042,
+      "p95_observation_to_action_latency_ms": 92.13036412000656,
+      "p95_worker_service_time_ms": 91.75596672296524,
+      "p99_worker_service_time_ms": 110.92562991380692
+    }
+  },
+  "provenance": {
+    "base_config": "/workspace/tasks/20260911T023128Z-p-only4/ant/profile.yaml",
+    "checkpoint_kind": "best",
+    "model_artifact": {
+      "checkpoint": "checkpoints/steps_5000_pytorch_model.pt",
+      "model_config": "config.full.yaml",
+      "path_in_repo": "OpenVLA/zero-latency/ant_rgb_state_l0_return_gt3000_100ep_openvla_native_sft_5k",
+      "repo_id": "latency-sensitive-bench/extra-envs-checkpoints",
+      "source": "local"
+    },
+    "session_ids": [
+      0,
+      1,
+      2,
+      3,
+      4
+    ]
+  },
+  "sample_model_type": "hidden_regime",
+  "source_run_id": "20260911T033037730561Z",
+  "summary": {
+    "frame_ms": 100.0,
+    "max_ms": 130.29119229316711,
+    "mean_effective_frames": 0.9056460638563993,
+    "mean_ms": 90.56460638563993,
+    "min_ms": 87.98952293395996,
+    "n_samples": 7976,
+    "p50_frames": 0.8970676946640015,
+    "p50_ms": 89.70676946640015,
+    "p90_frames": 0.9074710392951966,
+    "p90_ms": 90.74710392951965,
+    "p95_frames": 0.9213036412000656,
+    "p95_ms": 92.13036412000656,
+    "p99_frames": 1.1217999053001404,
+    "p99_ms": 112.17999053001404,
+    "prob_latency_gt_1_frame": 0.03686058174523571,
+    "prob_latency_gt_2_frames": 0.0,
+    "prob_latency_gt_3_frames": 0.0,
+    "std_ms": 4.164609685850939
+  },
+  "visualization_path": "latency_profile.png",
+  "workload_id": "ant"
+}

latency-aware/ant/small-policy/sample-factory-qwenoft-rtx3090-mean-v1/provenance.json

@@ -0,0 +1,26 @@
+{
+  "task": "ant",
+  "algorithm": "APPO",
+  "selection_rule": "best_reward_after_full_training_budget",
+  "budget_env_steps": 10000000,
+  "completed_env_steps": 10006528,
+  "selected_train_step": 19472,
+  "selected_env_steps": 9969664,
+  "selected_source": "/home/ubuntu/lzj/mean-profiling/ant/teacher/ant_mean_3090_s2222/checkpoint_p0/best_000019472_9969664_reward_3371.900.pth",
+  "selected_source_sha256": "868669c201eb0a42a136e724f5f8038304e10f9bc9045297995963261a2f9368",
+  "checkpoint_sha256": "a7899b32beb9b2cfec1bcff85b285dddc243a73f1c322fafefae599df0f67b3b",
+  "tensor_equality_verified": true,
+  "weights_only_export": true,
+  "seed": 2222,
+  "env_fps": 10.0,
+  "obs_fps": 10.0,
+  "fixed_latency_ms": 90.56460638563993,
+  "profile_source": {
+    "mean_ms": 90.56460638563993,
+    "profile": "/home/ubuntu/lzj/mean-profiling/reference/checkpoint-configs/latency-aware/ant/small-policy/sample-factory-qwenoft-rtx3090-v1/profile/profile.json",
+    "sha256": "0e49f72ec05ac600a54e07bbca5af3143708444d606167f33fa21788a9fefe50"
+  },
+  "training_hardware": "RTX PRO6000 Blackwell physicalGPU1",
+  "training_status": "complete",
+  "closed_loop_evaluation": "not performed by this publication"
+}