Upload three ThinkWAM benchmark checkpoints with checksums

README.md

@@ -0,0 +1,26 @@
+---
+tags:
+- robotics
+- thinkwam
+- libero
+- robotwin
+- robocasa
+---
+
+# ThinkWAM checkpoints
+
+ThinkWAM inference checkpoints exported from the selected training checkpoints.
+Each `.pt` file contains `model`, `config`, and `step`. Optimizer, scheduler, and random-number-generator states are excluded.
+Every exported model tensor was compared byte for byte against its source checkpoint, including its dtype and shape. The original BF16/FP32 storage precision and model configuration are preserved.
+
+| Benchmark | Checkpoint | Stored step |
+| --- | --- | ---: |
+| LIBERO / LIBERO-Plus | [libero_thinkwam_checkpoint.pt](libero_thinkwam_checkpoint.pt) | 75000 |
+| RoboTwin 2.0 | [robotwin2.0_thinkwam_checkpoint.pt](robotwin2.0_thinkwam_checkpoint.pt) | 155000 |
+| RoboCasa24 | [robocasa24_thinkwam_checkpoint.pt](robocasa24_thinkwam_checkpoint.pt) | 20000 |
+
+The LIBERO checkpoint is shared by LIBERO and LIBERO-Plus evaluations. The RoboCasa24 step is the counter saved in its continuation run, not a claim about total training steps across runs.
+
+Loading these checkpoints requires the compatible ThinkWAM implementation. Configurations are embedded in each checkpoint and are also provided as separate JSON files for inspection. Runtime paths and benchmark resources must be configured for the evaluation environment.
+
+File sizes and SHA-256 checksums are recorded in `checkpoint_manifest.json` and `SHA256SUMS`.

SHA256SUMS

@@ -0,0 +1,3 @@
+61d8a88069c3d0faa1b08fb9b6b2fe81d147d46ca237722f481b3d802c186eb8  libero_thinkwam_checkpoint.pt
+f37f9372a9bf748db82af6ed48e07cc6d306edaa15bd5a580bbb376a0b32c4f1  robotwin2.0_thinkwam_checkpoint.pt
+1a6700a9ec2c73f1a73ca56322305439f21d13b237296eadb37bd6d40884608c  robocasa24_thinkwam_checkpoint.pt

checkpoint_manifest.json

@@ -0,0 +1,66 @@
+{
+  "model": "ThinkWAM",
+  "format": "inference_checkpoint",
+  "files": [
+    {
+      "benchmark": "LIBERO / LIBERO-Plus",
+      "name": "libero_thinkwam_checkpoint.pt",
+      "step": 75000,
+      "bytes": 4430139219,
+      "sha256": "61d8a88069c3d0faa1b08fb9b6b2fe81d147d46ca237722f481b3d802c186eb8",
+      "source_sha256": "2ca59f2b5dcb857a920df1da0b520c6b7db8dea056514adac1626e9ee9c063e6",
+      "model_tensor_count": 1373,
+      "tensor_bytes_unchanged": true,
+      "dtypes_unchanged": true,
+      "config_unchanged": true,
+      "config_name": "libero_thinkwam_config.json",
+      "config_bytes": 5249,
+      "config_sha256": "966c826578959229c83b5f29381abac5e29c60c7bd283f795371e6fd2474a994",
+      "contents": [
+        "step",
+        "model",
+        "config"
+      ]
+    },
+    {
+      "benchmark": "RoboTwin 2.0",
+      "name": "robotwin2.0_thinkwam_checkpoint.pt",
+      "step": 155000,
+      "bytes": 4405774483,
+      "sha256": "f37f9372a9bf748db82af6ed48e07cc6d306edaa15bd5a580bbb376a0b32c4f1",
+      "source_sha256": "ce5708ee316efc8cf3ce5116fdc943fcd170e070f87ead44a433dfa34ac5012b",
+      "model_tensor_count": 1329,
+      "tensor_bytes_unchanged": true,
+      "dtypes_unchanged": true,
+      "config_unchanged": true,
+      "config_name": "robotwin2.0_thinkwam_config.json",
+      "config_bytes": 7058,
+      "config_sha256": "929e283653dd55e8f6c41d70eeb55a97ff6b2ec8676fe3e300be4426546f6440",
+      "contents": [
+        "step",
+        "model",
+        "config"
+      ]
+    },
+    {
+      "benchmark": "RoboCasa24",
+      "name": "robocasa24_thinkwam_checkpoint.pt",
+      "step": 20000,
+      "bytes": 4421635575,
+      "sha256": "1a6700a9ec2c73f1a73ca56322305439f21d13b237296eadb37bd6d40884608c",
+      "source_sha256": "b36a855ba4558a8fab9dd8870dcacb13952c5f53adddbdbe020d0813b251627b",
+      "model_tensor_count": 1365,
+      "tensor_bytes_unchanged": true,
+      "dtypes_unchanged": true,
+      "config_unchanged": true,
+      "config_name": "robocasa24_thinkwam_config.json",
+      "config_bytes": 6116,
+      "config_sha256": "02d12e747d855cb131938def09704d11ef1ff23efc42bfe09300a5855f4f07a4",
+      "contents": [
+        "step",
+        "model",
+        "config"
+      ]
+    }
+  ]
+}

libero_thinkwam_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61d8a88069c3d0faa1b08fb9b6b2fe81d147d46ca237722f481b3d802c186eb8
+size 4430139219

libero_thinkwam_config.json

@@ -0,0 +1,167 @@
+{
+  "vjepa_checkpoint": "/data/thinkwam/models/vjepa2/vjepa2_1_vitl_dist_vitG_384.pt",
+  "vjepa_dim": 1024,
+  "vjepa_image_size": 384,
+  "vjepa_patch_size": 16,
+  "vjepa_views": 2,
+  "qwen_path": "/data/thinkwam/models/Qwen2.5-0.5B",
+  "qwen_dim": 896,
+  "qwen_max_length": 512,
+  "qwen_action_tokens": 64,
+  "qwen_lora": true,
+  "qwen_lora_rank": 32,
+  "qwen_lora_alpha": 64,
+  "qwen_lora_dropout": 0.1,
+  "qwen_lora_target_modules": "all-linear",
+  "qwen_attn_implementation": "flash_attention_2",
+  "qwen_visual_tokens": 1088,
+  "qwen_hidden_only": true,
+  "action_context_dim": 1024,
+  "conditioning_mlp_hidden": 2048,
+  "online_vjepa": true,
+  "compile_vjepa": true,
+  "vjepa_compile_mode": "reduce-overhead",
+  "compile_action_dit": true,
+  "action_dit_compile_mode": "reduce-overhead",
+  "text_max_length": 128,
+  "visual_token_count": 1152,
+  "jepa_horizons": [
+    7,
+    15,
+    31
+  ],
+  "jepa_token_counts": [
+    288,
+    144,
+    72
+  ],
+  "jepa_context_dim": 1024,
+  "jepa_observation_tokens": 1152,
+  "jepa_layers": 2,
+  "jepa_heads": 8,
+  "jepa_loss_weight": 0.5,
+  "sigreg_loss_weight": 0.0,
+  "sigreg_num_projections": 128,
+  "sigreg_num_knots": 17,
+  "sigreg_t_max": 3.0,
+  "memory_dim": 1024,
+  "memory_slots": 16,
+  "memory_observation_tokens": 256,
+  "memory_action_history": 8,
+  "memory_layers": 2,
+  "memory_heads": 8,
+  "action_dim": 7,
+  "proprio_dim": 8,
+  "action_horizon": 16,
+  "action_dit_hidden": 1024,
+  "action_dit_layers": 16,
+  "action_dit_heads": 32,
+  "action_dit_inner_dim": 1536,
+  "action_dit_future_tokens": 32,
+  "flow_steps": 4,
+  "action_loss_weight": 1.0,
+  "batch_size": 24,
+  "gradient_accumulation_steps": 1,
+  "learning_rate": 5e-05,
+  "min_learning_rate": 1e-05,
+  "weight_decay": 0.0,
+  "warmup_ratio": 0.03,
+  "max_steps": 80000,
+  "num_workers": 8,
+  "dataloader_prefetch_factor": 4,
+  "dataloader_persistent_workers": true,
+  "dataloader_pin_memory": true,
+  "dataloader_worker_threads": 1,
+  "mixed_precision": "bf16",
+  "gradient_checkpointing": false,
+  "checkpoint_save_limit": 3,
+  "optimizer_beta1": 0.9,
+  "optimizer_beta2": 0.95,
+  "eval_every": 5000,
+  "eval_batches": 32,
+  "eval_fraction": 0.1,
+  "tensorboard": true,
+  "tensorboard_histogram_every": 100,
+  "open_loop_eval": true,
+  "save_best_checkpoint": true,
+  "closed_loop_after": false,
+  "closed_loop_suite": "all",
+  "closed_loop_world_size": 8,
+  "closed_loop_tasks": -1,
+  "closed_loop_episodes": 50,
+  "closed_loop_replan_steps": 20,
+  "closed_loop_max_steps": 520,
+  "closed_loop_save_videos": false,
+  "seed": 7,
+  "dataset": "libero",
+  "dataset_root": "/data/thinkwam/datasets/LIBERO",
+  "feature_cache": "",
+  "output_dir": "/data/thinkwam/runs/p1_jepawam/a0_mixed_plus_20261019_resume40k_hard_goal10_affstage_draft",
+  "vjepa_trainable": true,
+  "vjepa_adapter_enabled": true,
+  "vjepa_adapter_dim": 256,
+  "vjepa_adapter_learning_rate": 1e-05,
+  "affordance_decoder_dim": 256,
+  "affordance_decoder_layers": 2,
+  "affordance_decoder_heads": 8,
+  "affordance_channels": 1,
+  "affordance_loss_weight": 0.05,
+  "affordance_dice_weight": 0.5,
+  "affordance_loss_ramp_steps": 3000,
+  "affordance_target_key": "affordance",
+  "require_affordance_targets": true,
+  "video_model_dim": 1024,
+  "video_model_layers": 2,
+  "video_model_heads": 8,
+  "video_context_tokens": 256,
+  "video_residual_loss_weight": 0.8,
+  "action_refinement_loss_weight": 1.4,
+  "first_action_loss_weight": 1.2,
+  "video_condition_steps": 4,
+  "video_condition_train_steps": 1,
+  "action_condition_scale": 1.0,
+  "p1_stage": "staged",
+  "p0_checkpoint": "/data/thinkwam/runs/p0_jepawam/libero_qwen1152_jepa288_144_72_gr00t1536_nogc_fusedqkv_compile_b24_lr1e4_relaunch_20261013/checkpoints/best.pt",
+  "freeze_qwen_steps": 10000,
+  "freeze_vjepa_steps": 10000,
+  "freeze_memory_steps": 5000,
+  "affordance_memory_start_step": 45000,
+  "affordance_memory_ramp_steps": 5000,
+  "affordance_memory_max_scale": 0.05,
+  "affordance_confidence_threshold": 0.7,
+  "affordance_confidence_temperature": 0.1,
+  "affordance_upstream_start_step": 45000,
+  "affordance_upstream_ramp_steps": 5000,
+  "affordance_upstream_max_scale": 0.05,
+  "memory_learning_rate": 2.5e-05,
+  "qwen_lora_learning_rate": 1.5e-05,
+  "phase_num_classes": 5,
+  "phase_target_key": "phase",
+  "require_phase_targets": false,
+  "phase_loss_weight": 0.1,
+  "phase_loss_ramp_steps": 3000,
+  "phase_memory_start_step": 5000,
+  "phase_memory_ramp_steps": 5000,
+  "phase_memory_max_scale": 0.0,
+  "phase_upstream_start_step": 5000,
+  "phase_upstream_ramp_steps": 5000,
+  "phase_upstream_max_scale": 0.0,
+  "plus_dataset_root": "/data/thinkwam/datasets/libero_plus_lerobot_hard",
+  "plus_mixture_fraction": 0.35,
+  "plus_augmentations": true,
+  "plus_augmentation_probability": 0.5,
+  "plus_illumination_probability": 0.5,
+  "plus_texture_probability": 0.35,
+  "plus_camera_probability": 0.25,
+  "plus_noise_probability": 0.25,
+  "plus_object_position_probability": 0.25,
+  "plus_max_translation": 12,
+  "plus_noise_std": 0.025,
+  "plus_category_manifest": "",
+  "plus_language_category_factor": 1.25,
+  "plus_robot_initial_state_factor": 1.25,
+  "plus_initial_state_quantile": 0.7,
+  "task_sampling_enabled": true,
+  "task_sampling_goal_long_factor": 3,
+  "task_sampling_hard_factor": 3
+}

robocasa24_thinkwam_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a6700a9ec2c73f1a73ca56322305439f21d13b237296eadb37bd6d40884608c
+size 4421635575

robocasa24_thinkwam_config.json

@@ -0,0 +1,196 @@
+{
+  "vjepa_checkpoint": "/data/thinkwam/models/vjepa2/vjepa2_1_vitl_dist_vitG_384.pt",
+  "vjepa_dim": 1024,
+  "vjepa_image_size": 384,
+  "vjepa_patch_size": 16,
+  "vjepa_views": 3,
+  "qwen_path": "/data/thinkwam/models/Qwen2.5-0.5B",
+  "qwen_dim": 896,
+  "qwen_max_length": 512,
+  "qwen_action_tokens": 64,
+  "qwen_lora": true,
+  "qwen_lora_rank": 32,
+  "qwen_lora_alpha": 64,
+  "qwen_lora_dropout": 0.1,
+  "qwen_lora_target_modules": "all-linear",
+  "qwen_attn_implementation": "flash_attention_2",
+  "qwen_visual_tokens": 1152,
+  "qwen_hidden_only": true,
+  "action_context_dim": 1024,
+  "conditioning_mlp_hidden": 2048,
+  "online_vjepa": true,
+  "compile_vjepa": true,
+  "vjepa_compile_mode": "reduce-overhead",
+  "compile_action_dit": false,
+  "action_dit_compile_mode": "reduce-overhead",
+  "text_max_length": 128,
+  "visual_token_count": 1728,
+  "jepa_horizons": [
+    7,
+    15,
+    31
+  ],
+  "jepa_token_counts": [
+    288,
+    144,
+    72
+  ],
+  "jepa_context_dim": 1024,
+  "jepa_observation_tokens": 1152,
+  "jepa_layers": 2,
+  "jepa_heads": 8,
+  "jepa_loss_weight": 0.5,
+  "sigreg_loss_weight": 0.0,
+  "sigreg_num_projections": 128,
+  "sigreg_num_knots": 17,
+  "sigreg_t_max": 3.0,
+  "memory_dim": 1024,
+  "memory_slots": 16,
+  "memory_observation_tokens": 256,
+  "memory_action_history": 8,
+  "memory_layers": 2,
+  "memory_heads": 8,
+  "action_dim": 7,
+  "proprio_dim": 1,
+  "action_horizon": 16,
+  "action_dit_hidden": 1024,
+  "action_dit_layers": 16,
+  "action_dit_heads": 32,
+  "action_dit_inner_dim": 1536,
+  "action_dit_future_tokens": 32,
+  "flow_steps": 4,
+  "action_loss_weight": 1.0,
+  "batch_size": 20,
+  "gradient_accumulation_steps": 1,
+  "learning_rate": 1e-05,
+  "min_learning_rate": 5e-07,
+  "weight_decay": 0.0,
+  "warmup_ratio": 0.02,
+  "max_steps": 20000,
+  "num_workers": 8,
+  "dataloader_prefetch_factor": 2,
+  "dataloader_persistent_workers": true,
+  "dataloader_pin_memory": true,
+  "dataloader_worker_threads": 1,
+  "mixed_precision": "bf16",
+  "gradient_checkpointing": true,
+  "checkpoint_save_limit": 2,
+  "optimizer_beta1": 0.9,
+  "optimizer_beta2": 0.95,
+  "eval_every": 5000,
+  "eval_batches": 20,
+  "eval_fraction": 0.1,
+  "tensorboard": true,
+  "tensorboard_histogram_every": 100,
+  "open_loop_eval": true,
+  "save_best_checkpoint": false,
+  "closed_loop_after": false,
+  "closed_loop_suite": "all",
+  "closed_loop_world_size": 8,
+  "closed_loop_tasks": -1,
+  "closed_loop_episodes": 50,
+  "closed_loop_replan_steps": 20,
+  "closed_loop_max_steps": 520,
+  "closed_loop_save_videos": false,
+  "seed": 7,
+  "dataset": "robocasa24",
+  "dataset_root": "/data/thinkwam/robocasa24_deva",
+  "feature_cache": "",
+  "future_target_cache": "",
+  "output_dir": "/data/thinkwam/robocasa_deva_rollout60k_plus20k_20261110/output",
+  "robotwin_frame_cache": "",
+  "robotwin_manifest": "manifest.json",
+  "robotwin_stats_path": "",
+  "robocasa24_require_affordance": true,
+  "robocasa_event_sampling": true,
+  "robocasa_uniform_fraction": 0.65,
+  "robocasa_gripper_fraction": 0.2,
+  "robocasa_tail_fraction": 0.15,
+  "robocasa_transition_radius": 8,
+  "robocasa_tail_steps": 32,
+  "robocasa_task_weights": {
+    "kitchen_coffee_CoffeeSetupMug": 2.0,
+    "kitchen_pnp_PnPCounterToMicrowave": 2.0,
+    "kitchen_pnp_PnPCounterToSink": 2.0,
+    "kitchen_pnp_PnPCounterToStove": 2.0,
+    "kitchen_stove_TurnOffStove": 2.0,
+    "kitchen_pnp_PnPCabToCounter": 2.0,
+    "kitchen_pnp_PnPStoveToCounter": 2.0,
+    "kitchen_pnp_PnPMicrowaveToCounter": 2.0
+  },
+  "robocasa_split_manifest": "/data/thinkwam/robocasa24_p1_tokenmean_20261101/manifests/protected_episode_split.json",
+  "p1_selection_metric": "loss/action_dit_refined",
+  "action_dit_learning_rate": 5e-06,
+  "action_teacher_cache": "",
+  "action_distill_weight": 0.0,
+  "action_distill_ramp_steps": 0,
+  "affordance_enabled": true,
+  "phase_enabled": false,
+  "vjepa_trainable": true,
+  "vjepa_adapter_enabled": true,
+  "vjepa_adapter_dim": 256,
+  "vjepa_adapter_learning_rate": 1.5e-06,
+  "affordance_decoder_dim": 256,
+  "affordance_decoder_layers": 2,
+  "affordance_decoder_heads": 8,
+  "affordance_channels": 1,
+  "affordance_loss_weight": 0.1,
+  "affordance_dice_weight": 0.5,
+  "affordance_loss_ramp_steps": 0,
+  "affordance_target_key": "affordance",
+  "require_affordance_targets": false,
+  "video_model_dim": 1024,
+  "video_model_layers": 2,
+  "video_model_heads": 8,
+  "video_context_tokens": 256,
+  "video_residual_loss_weight": 1.0,
+  "action_refinement_loss_weight": 1.0,
+  "first_action_loss_weight": 1.0,
+  "video_condition_steps": 4,
+  "video_condition_train_steps": 1,
+  "action_condition_scale": 1.0,
+  "p1_stage": "staged",
+  "p0_checkpoint": "",
+  "freeze_qwen_steps": 0,
+  "freeze_vjepa_steps": 0,
+  "freeze_memory_steps": 0,
+  "affordance_memory_start_step": 0,
+  "affordance_memory_ramp_steps": 0,
+  "affordance_memory_max_scale": 0.05,
+  "affordance_confidence_threshold": 0.7,
+  "affordance_confidence_temperature": 0.1,
+  "affordance_upstream_start_step": 0,
+  "affordance_upstream_ramp_steps": 0,
+  "affordance_upstream_max_scale": 0.05,
+  "memory_learning_rate": 5e-06,
+  "qwen_lora_learning_rate": 1.5e-06,
+  "phase_num_classes": 5,
+  "phase_target_key": "phase",
+  "require_phase_targets": false,
+  "phase_loss_weight": 0.1,
+  "phase_loss_ramp_steps": 3000,
+  "phase_memory_start_step": 5000,
+  "phase_memory_ramp_steps": 5000,
+  "phase_memory_max_scale": 0.05,
+  "phase_upstream_start_step": 5000,
+  "phase_upstream_ramp_steps": 5000,
+  "phase_upstream_max_scale": 0.1,
+  "plus_dataset_root": "",
+  "plus_mixture_fraction": 0.25,
+  "plus_augmentations": false,
+  "plus_augmentation_probability": 0.5,
+  "plus_illumination_probability": 0.5,
+  "plus_texture_probability": 0.35,
+  "plus_camera_probability": 0.25,
+  "plus_noise_probability": 0.25,
+  "plus_object_position_probability": 0.25,
+  "plus_max_translation": 12,
+  "plus_noise_std": 0.025,
+  "plus_category_manifest": "",
+  "plus_language_category_factor": 1.0,
+  "plus_robot_initial_state_factor": 1.0,
+  "plus_initial_state_quantile": 0.7,
+  "task_sampling_enabled": true,
+  "task_sampling_goal_long_factor": 2,
+  "task_sampling_hard_factor": 2
+}

robotwin2.0_thinkwam_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f37f9372a9bf748db82af6ed48e07cc6d306edaa15bd5a580bbb376a0b32c4f1
+size 4405774483

robotwin2.0_thinkwam_config.json

@@ -0,0 +1,236 @@
+{
+  "vjepa_checkpoint": "/data/thinkwam/models/vjepa2/vjepa2_1_vitl_dist_vitG_384.pt",
+  "vjepa_dim": 1024,
+  "vjepa_image_size": 384,
+  "vjepa_patch_size": 16,
+  "vjepa_views": 3,
+  "qwen_path": "/data/thinkwam/models/Qwen2.5-0.5B",
+  "qwen_dim": 896,
+  "qwen_max_length": 512,
+  "qwen_action_tokens": 64,
+  "qwen_lora": true,
+  "qwen_lora_rank": 32,
+  "qwen_lora_alpha": 64,
+  "qwen_lora_dropout": 0.1,
+  "qwen_lora_target_modules": "all-linear",
+  "qwen_attn_implementation": "flash_attention_2",
+  "qwen_visual_tokens": 1088,
+  "qwen_hidden_only": true,
+  "action_context_dim": 1024,
+  "conditioning_mlp_hidden": 2048,
+  "online_vjepa": true,
+  "compile_vjepa": true,
+  "vjepa_compile_mode": "default",
+  "compile_action_dit": true,
+  "action_dit_compile_mode": "reduce-overhead",
+  "text_max_length": 128,
+  "visual_token_count": 1152,
+  "jepa_horizons": [
+    8,
+    16,
+    32
+  ],
+  "jepa_token_counts": [
+    288,
+    144,
+    72
+  ],
+  "jepa_context_dim": 1024,
+  "jepa_observation_tokens": 1152,
+  "jepa_layers": 2,
+  "jepa_heads": 8,
+  "jepa_loss_weight": 0.5,
+  "sigreg_loss_weight": 0.0,
+  "sigreg_num_projections": 128,
+  "sigreg_num_knots": 17,
+  "sigreg_t_max": 3.0,
+  "memory_dim": 1024,
+  "memory_slots": 16,
+  "memory_observation_tokens": 256,
+  "memory_action_history": 16,
+  "memory_layers": 2,
+  "memory_heads": 8,
+  "action_dim": 14,
+  "proprio_dim": 14,
+  "action_horizon": 32,
+  "action_dit_hidden": 1024,
+  "action_dit_layers": 16,
+  "action_dit_heads": 32,
+  "action_dit_inner_dim": 1536,
+  "action_dit_future_tokens": 32,
+  "flow_steps": 4,
+  "action_loss_weight": 1.0,
+  "batch_size": 20,
+  "gradient_accumulation_steps": 1,
+  "learning_rate": 1e-05,
+  "min_learning_rate": 2e-06,
+  "weight_decay": 0.0,
+  "warmup_ratio": 0.02,
+  "max_steps": 25000,
+  "num_workers": 8,
+  "dataloader_prefetch_factor": 2,
+  "dataloader_persistent_workers": true,
+  "dataloader_pin_memory": true,
+  "dataloader_worker_threads": 1,
+  "mixed_precision": "bf16",
+  "gradient_checkpointing": false,
+  "checkpoint_save_limit": 5,
+  "optimizer_beta1": 0.9,
+  "optimizer_beta2": 0.95,
+  "eval_every": 5000,
+  "eval_batches": 8,
+  "eval_fraction": 0.1,
+  "tensorboard": true,
+  "tensorboard_histogram_every": 100,
+  "open_loop_eval": true,
+  "save_best_checkpoint": true,
+  "closed_loop_after": false,
+  "closed_loop_suite": "all",
+  "closed_loop_world_size": 8,
+  "closed_loop_tasks": -1,
+  "closed_loop_episodes": 50,
+  "closed_loop_replan_steps": 20,
+  "closed_loop_max_steps": 520,
+  "closed_loop_save_videos": false,
+  "seed": 7,
+  "dataset": "robotwin2.0",
+  "dataset_root": "/data/thinkwam/datasets/robotwin2.0-lerobot-v3.0/clean",
+  "feature_cache": "",
+  "output_dir": "/data/thinkwam/runs/robotwin2_150k_short_25k_20261101",
+  "robotwin_frame_cache": "/data/thinkwam/datasets/thinkwam/robotwin2_clean_rgb384",
+  "robotwin_manifest": "manifest_3view.json",
+  "robotwin_stats_path": "/data/thinkwam/datasets/robotwin2.0-lerobot-v3.0/clean/dataset_stats.json",
+  "affordance_enabled": false,
+  "phase_enabled": false,
+  "vjepa_trainable": false,
+  "vjepa_adapter_enabled": true,
+  "vjepa_adapter_dim": 256,
+  "vjepa_adapter_learning_rate": 3e-06,
+  "affordance_decoder_dim": 256,
+  "affordance_decoder_layers": 2,
+  "affordance_decoder_heads": 8,
+  "affordance_channels": 1,
+  "affordance_loss_weight": 0.0,
+  "affordance_dice_weight": 0.5,
+  "affordance_loss_ramp_steps": 3000,
+  "affordance_target_key": "affordance",
+  "require_affordance_targets": false,
+  "video_model_dim": 1024,
+  "video_model_layers": 2,
+  "video_model_heads": 8,
+  "video_context_tokens": 256,
+  "video_residual_loss_weight": 0.8,
+  "action_refinement_loss_weight": 1.4,
+  "first_action_loss_weight": 1.2,
+  "video_condition_steps": 4,
+  "video_condition_train_steps": 1,
+  "action_condition_scale": 1.0,
+  "p1_stage": "joint",
+  "p0_checkpoint": "",
+  "freeze_qwen_steps": 0,
+  "freeze_vjepa_steps": 0,
+  "freeze_memory_steps": 0,
+  "affordance_memory_start_step": 5000,
+  "affordance_memory_ramp_steps": 5000,
+  "affordance_memory_max_scale": 0.05,
+  "affordance_confidence_threshold": 0.7,
+  "affordance_confidence_temperature": 0.1,
+  "affordance_upstream_start_step": 5000,
+  "affordance_upstream_ramp_steps": 5000,
+  "affordance_upstream_max_scale": 0.05,
+  "memory_learning_rate": 5e-06,
+  "qwen_lora_learning_rate": 2e-06,
+  "phase_num_classes": 5,
+  "phase_target_key": "phase",
+  "require_phase_targets": false,
+  "phase_loss_weight": 0.0,
+  "phase_loss_ramp_steps": 3000,
+  "phase_memory_start_step": 5000,
+  "phase_memory_ramp_steps": 5000,
+  "phase_memory_max_scale": 0.05,
+  "phase_upstream_start_step": 5000,
+  "phase_upstream_ramp_steps": 5000,
+  "phase_upstream_max_scale": 0.1,
+  "plus_dataset_root": "",
+  "plus_mixture_fraction": 0.25,
+  "plus_augmentations": false,
+  "plus_augmentation_probability": 0.0,
+  "plus_illumination_probability": 0.5,
+  "plus_texture_probability": 0.35,
+  "plus_camera_probability": 0.25,
+  "plus_noise_probability": 0.25,
+  "plus_object_position_probability": 0.25,
+  "plus_max_translation": 12,
+  "plus_noise_std": 0.025,
+  "plus_category_manifest": "",
+  "plus_language_category_factor": 1.0,
+  "plus_robot_initial_state_factor": 1.0,
+  "plus_initial_state_quantile": 0.7,
+  "task_sampling_enabled": true,
+  "task_sampling_goal_long_factor": 2,
+  "task_sampling_hard_factor": 2,
+  "target_manifest": "/data/thinkwam/datasets/thinkwam/robotwin2_150k_short_50c20r/manifest.json",
+  "random_dataset_root": "/data/thinkwam/datasets/robotwin2.0-lerobot-v3.0/random",
+  "base_checkpoint": "/data/thinkwam/runs/robotwin2_focused_resume_90k_to150k/checkpoints/checkpoint_step_0150000.pt",
+  "expected_global_batch": 80,
+  "base_step": 150000,
+  "transition_step": 20000,
+  "warmup_steps": 500,
+  "lr_floor_ratio": 0.2,
+  "checkpoint_every": 5000,
+  "clean_mmap_episodes": 4,
+  "label_lru_episodes": 32,
+  "decode_timeout": 30.0,
+  "robotwin_tasks": [
+    "adjust_bottle",
+    "beat_block_hammer",
+    "blocks_ranking_rgb",
+    "blocks_ranking_size",
+    "click_alarmclock",
+    "click_bell",
+    "dump_bin_bigbin",
+    "grab_roller",
+    "handover_block",
+    "handover_mic",
+    "hanging_mug",
+    "lift_pot",
+    "move_can_pot",
+    "move_pillbottle_pad",
+    "move_playingcard_away",
+    "move_stapler_pad",
+    "open_laptop",
+    "open_microwave",
+    "pick_diverse_bottles",
+    "pick_dual_bottles",
+    "place_a2b_left",
+    "place_a2b_right",
+    "place_bread_basket",
+    "place_bread_skillet",
+    "place_burger_fries",
+    "place_can_basket",
+    "place_cans_plasticbox",
+    "place_container_plate",
+    "place_dual_shoes",
+    "place_empty_cup",
+    "place_fan",
+    "place_mouse_pad",
+    "place_object_basket",
+    "place_object_scale",
+    "place_object_stand",
+    "place_phone_stand",
+    "place_shoe",
+    "press_stapler",
+    "put_bottles_dustbin",
+    "put_object_cabinet",
+    "rotate_qrcode",
+    "scan_object",
+    "shake_bottle",
+    "shake_bottle_horizontally",
+    "stack_blocks_three",
+    "stack_blocks_two",
+    "stack_bowls_three",
+    "stack_bowls_two",
+    "stamp_seal",
+    "turn_switch"
+  ]
+}