Copy pinned dex_eval_data batch2/22

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/MANIFEST.json

@@ -0,0 +1,41 @@
+{
+  "schema": "dexrsi-hf-model-only-v1",
+  "repo": "AvaX1/dex_eval_data",
+  "repo_type": "dataset",
+  "prefix": "checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027",
+  "stage": 1,
+  "step": 10000,
+  "source_checkpoint": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage1_paired_v_8gpu_20261025_v1/training/models/checkpoint_epoch_7_step_10000",
+  "files": {
+    "transformer/diffusion_pytorch_model.bin": {
+      "bytes": 9045338989,
+      "sha256": "c2b4571b7742ac0767d672d30d0146ecbd72397f77bdf8afcc20029e27a6dfe0"
+    },
+    "stats.json": {
+      "bytes": 6863,
+      "sha256": "93e2e7ba9187396b9a40625d837fbc4b80167118036c718f63016f06703ca971"
+    },
+    "stage1.recorded.json": {
+      "bytes": 2961,
+      "sha256": "6cbdac518a87d9a09036337ce59436f7e29fc063f7b7ababcf0bc2ba0eba47a4"
+    },
+    "config.recorded.json": {
+      "bytes": 6072,
+      "sha256": "1e3c4e7f69b3a4b68b2a0e51faefc203e1fc7baf816de3443ec85477ecff8370"
+    },
+    "README.md": {
+      "bytes": 1606,
+      "sha256": "e7319c0e48457ab4b8123e31d644d0f6d9b9d7c5611c1d49cfde45bca9eaa254"
+    },
+    "dexrsiv2_complete.json": {
+      "bytes": 57,
+      "sha256": "b134231560cf07a1cb0d458b96df5a3846ff7d65d5837bdda2c97a5df7fdbf58"
+    },
+    "transformer/dexrsiv2_contract.json": {
+      "bytes": 4180,
+      "sha256": "4034fe55d2cace1399c391845939dc19ab43390aba24a3d62236396c92685d63"
+    }
+  },
+  "optimizer_included": false,
+  "rng_included": false
+}

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/README.md

@@ -0,0 +1,24 @@
+# DexRSI eight-task paired-V Stage1 — 10,000 steps
+
+Model-only export of the completed eight-GPU Stage1 paired-V run (2026-10-26).
+This is the Stage1 initialization used by the paired-V Stage2 10k export.
+
+- 800 expert episodes, eight tasks; action/state dimensions 44/46, three camera slots.
+- Linear MSE reward, pre-success one-sided Gaussian: sigma 16 frames, d=0..48
+  clipped at episode start, normalized to sum 1 per successful episode.
+- Stage1 ordinary mean reward MSE (no support/zero balancing), outer weight 0.1.
+- Paired ordinary-forward V TD Huber10 weight 0.1; successor Gaussian reward
+  and detached successor V; Q weight 0 with Q parameters frozen.
+- Continuation used ordinary masked BCE. The later cont-balance code change
+  was NOT applied to this checkpoint. Inference reward is raw linear, no sigmoid.
+
+Original model and contract are retained under `transformer/`; the original completion
+marker and recorded configurations are included unchanged. Some sealed contract schema
+fields have historical mixed9/binary labels; actual settings are in `reward_prediction`,
+`value_training` and the recorded configuration. Do not rewrite sealed provenance.
+`stats.json` is the verified original normalization.
+
+Code and reproduction guide: https://github.com/lhj-lhj/DexRSI/tree/69a650bc049dd7f2c04a2426cdb6e9c6f199a877
+Supply local FLUX backbone, FLUX.2-dev AE, tokenizer/text encoder and data paths explicitly;
+these external runtime assets are not included. No optimizer, scheduler, RNG, dataset
+or credentials are uploaded; this model-only export cannot resume the original Adam state.

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/config.recorded.json

@@ -0,0 +1,180 @@
+{
+    "project_dir": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage1_paired_v_8gpu_20261025_v1/training",
+    "runners": [
+        "stage1_job.FormalTrainer"
+    ],
+    "launch": {
+        "gpu_ids": [
+            0,
+            1,
+            2,
+            3,
+            4,
+            5,
+            6,
+            7
+        ],
+        "distributed_type": "DEEPSPEED",
+        "deepspeed_config": {
+            "deepspeed_config_file": "/mnt/home/liuhongjia/DexRSIv2/variants/dexjoco_eight_stage1_pairedv8_20261025_v1/third_party/FACT/fact_train/distributed/accelerate_configs/zero2.json"
+        },
+        "executable": "/mnt/workspace/liuhongjia/robonana-env/bin/python -m accelerate.commands.accelerate_cli",
+        "until_completion": false
+    },
+    "dataloaders": {
+        "train": {
+            "data_or_config": {
+                "_class_name": "PairedMixedDexJoCoDataset",
+                "data_path": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_800_terminal_v1",
+                "cache_dir": "/mnt/project/maxinsights/liuhongjia/dataset/cache/dexrsiv2/dexjoco_mixed8_20261021_v1",
+                "stats_path": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_800_terminal_v1/meta/stats.json",
+                "action_chunk": 16,
+                "max_horizon": 16,
+                "fixed_horizon": 16,
+                "context_tokens": 76,
+                "reward_smoothing": {
+                    "kind": "pre_success_gaussian",
+                    "sigma_frames": 16.0,
+                    "window_frames": 48,
+                    "normalization": "episode_sum_one"
+                }
+            },
+            "batch_size_per_gpu": 32,
+            "num_workers": 4,
+            "pin_memory": true,
+            "persistent_workers": true,
+            "prefetch_factor": 4,
+            "transform": null,
+            "sampler": {
+                "type": "PairedTerminalFrameSampler",
+                "seed": 42,
+                "global_batch": 256,
+                "terminal_fraction": 0.02
+            },
+            "collator": {
+                "is_equal": true
+            }
+        },
+        "test": {}
+    },
+    "models": {
+        "initialization": "flux_backbone",
+        "checkpoint": "/mnt/project/maxinsights/liuhongjia/model_weights/robonana-dex/FLUX.2-klein-base-4B/flux-2-klein-base-4b.safetensors",
+        "checkpoint_config": null,
+        "checkpoint_dir": "/mnt/project/maxinsights/liuhongjia/model_weights/robonana-dex/FLUX.2-klein-base-4B",
+        "ae_checkpoint": "/mnt/project/maxinsights/liuhongjia/model_weights/imagewam-g3/modelscope-cache/black-forest-labs/FLUX___2-dev/ae.safetensors",
+        "params": {
+            "in_channels": 128,
+            "context_in_dim": 7680,
+            "hidden_size": 3072,
+            "num_heads": 24,
+            "depth": 5,
+            "depth_single_blocks": 20,
+            "axes_dim": [
+                32,
+                32,
+                32,
+                32
+            ],
+            "theta": 2000,
+            "mlp_ratio": 3.0,
+            "use_guidance_embed": false
+        },
+        "action_dim": 44,
+        "state_dim": 46,
+        "chunk_horizon": 16,
+        "expert_hidden_dim": 1024,
+        "action_expert_mlp_ratio": 4.0,
+        "gradient_checkpointing": true
+    },
+    "optimizers": {
+        "type": "AdamW",
+        "lr": 0.0001,
+        "robot_lr": 0.0001,
+        "betas": [
+            "__tuple__",
+            0.9,
+            0.95
+        ],
+        "eps": 1e-08,
+        "weight_decay": 0.01,
+        "fused": true,
+        "foreach": false
+    },
+    "schedulers": {
+        "type": "WarmupCosineScheduler",
+        "warmup_steps": 1000,
+        "decay_steps": 10000,
+        "end_value": 0.01
+    },
+    "train": {
+        "max_steps": 10000,
+        "gradient_accumulation_steps": 1,
+        "mixed_precision": "bf16",
+        "checkpoint_interval": 1000,
+        "early_checkpoint_steps": [
+            "__tuple__"
+        ],
+        "checkpoint_keeps": [
+            10000
+        ],
+        "checkpoint_total_limit": 2,
+        "checkpoint_save_optimizer": true,
+        "disable_checkpointing": false,
+        "resume": false,
+        "seed": 42,
+        "log_with": "wandb",
+        "tracker_project_name": "DexRSIv2",
+        "tracker_init_kwargs": {
+            "wandb": {
+                "name": "dexrsiv2-eight-stage1-paired-v-8gpu-20261025-v1",
+                "entity": "hongjia-liu-aalto-university",
+                "id": "dexrsiv2-eight-stage1-paired-v-8gpu-20261025-v1"
+            }
+        },
+        "log_interval": 10,
+        "dataset_profile": "dexjoco_mixed8",
+        "image_profile": "dexjoco_mixed_rgb3",
+        "action_dim": 44,
+        "state_dim": 46,
+        "latent_grid_height": 48,
+        "latent_grid_width": 16,
+        "chunk_horizon": 16,
+        "flow": {
+            "action": {
+                "train_shift": 5.0,
+                "infer_shift": 5.0,
+                "num_train_timesteps": 1000,
+                "num_inference_steps": 10
+            },
+            "video": {
+                "train_shift": 5.0,
+                "infer_shift": 5.0,
+                "num_train_timesteps": 1000,
+                "num_inference_steps": 10
+            }
+        },
+        "max_grad_norm": 1.0,
+        "memory_limit_gib": 0.0,
+        "stats_path": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_800_terminal_v1/meta/stats.json",
+        "loss_weights": {
+            "image_loss": 0.5,
+            "action_loss": 1.0,
+            "future_state_loss": 0.4,
+            "reward_loss": 0.1,
+            "cont_loss": 0.1,
+            "value_loss": 0.1,
+            "q_loss": 0.0
+        },
+        "reward_balance_alpha": null,
+        "reward_prediction_mode": "linear_mse",
+        "reward_smoothing": {
+            "kind": "pre_success_gaussian",
+            "sigma_frames": 16.0,
+            "window_frames": 48,
+            "normalization": "episode_sum_one"
+        },
+        "paired_value_td": true,
+        "with_ema": false
+    }
+}

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/dexrsiv2_complete.json

@@ -0,0 +1 @@
+{"schema": "dexrsiv2-stage1-complete-v1", "step": 10000}

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/stage1.recorded.json

@@ -0,0 +1,113 @@
+{
+  "paths": {
+    "dataset": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_800_terminal_v1",
+    "cache": "/mnt/project/maxinsights/liuhongjia/dataset/cache/dexrsiv2/dexjoco_mixed8_20261021_v1",
+    "output": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage1_paired_v_8gpu_20261025_v1/training",
+    "flux_checkpoint": "/mnt/project/maxinsights/liuhongjia/model_weights/robonana-dex/FLUX.2-klein-base-4B",
+    "ae_checkpoint": "/mnt/project/maxinsights/liuhongjia/model_weights/imagewam-g3/modelscope-cache/black-forest-labs/FLUX___2-dev/ae.safetensors",
+    "play_dataset": null,
+    "play_cache": null
+  },
+  "data": {
+    "expected_episodes": 800,
+    "expected_frames": 388891,
+    "fps": 30,
+    "action_dim": 44,
+    "chunk_horizon": 16,
+    "play_fraction": 0.5,
+    "dataset_profile": "dexjoco_mixed8",
+    "state_dim": 46,
+    "image_profile": "dexjoco_mixed_rgb3",
+    "latent_grid_height": 48,
+    "latent_grid_width": 16,
+    "terminal_fraction": 0.02,
+    "context_tokens": 76
+  },
+  "model": {
+    "expert_hidden_dim": 1024,
+    "action_expert_mlp_ratio": 4.0,
+    "gradient_checkpointing": true,
+    "flux_params": {
+      "in_channels": 128,
+      "context_in_dim": 7680,
+      "hidden_size": 3072,
+      "num_heads": 24,
+      "depth": 5,
+      "depth_single_blocks": 20,
+      "axes_dim": [
+        32,
+        32,
+        32,
+        32
+      ],
+      "theta": 2000,
+      "mlp_ratio": 3.0,
+      "use_guidance_embed": false
+    }
+  },
+  "action_flow": {
+    "train_shift": 5.0,
+    "infer_shift": 5.0,
+    "num_train_timesteps": 1000,
+    "num_inference_steps": 10
+  },
+  "video_flow": {
+    "train_shift": 5.0,
+    "infer_shift": 5.0,
+    "num_train_timesteps": 1000,
+    "num_inference_steps": 10
+  },
+  "optimization": {
+    "gpus": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7
+    ],
+    "microbatch": 32,
+    "global_batch": 256,
+    "num_workers": 4,
+    "max_steps": 10000,
+    "warmup_steps": 1000,
+    "lr": 0.0001,
+    "robot_lr": 0.0001,
+    "weight_decay": 0.01,
+    "min_lr_ratio": 0.01,
+    "max_grad_norm": 1.0,
+    "checkpoint_interval": 1000,
+    "checkpoint_total_limit": 2,
+    "checkpoint_keeps": [
+      10000
+    ],
+    "seed": 42,
+    "log_interval": 10,
+    "log_with": "wandb",
+    "tracker_project": "DexRSIv2",
+    "tracker_entity": "hongjia-liu-aalto-university",
+    "run_name": "dexrsiv2-eight-stage1-paired-v-8gpu-20261025-v1"
+  },
+  "objectives": {
+    "loss_weights": {
+      "image_loss": 0.5,
+      "action_loss": 1.0,
+      "future_state_loss": 0.4,
+      "reward_loss": 0.1,
+      "cont_loss": 0.1,
+      "value_loss": 0.1,
+      "q_loss": 0.0
+    },
+    "reward_balance_alpha": null,
+    "reward_prediction_mode": "linear_mse",
+    "reward_smoothing": {
+      "kind": "pre_success_gaussian",
+      "sigma_frames": 16.0,
+      "window_frames": 48,
+      "normalization": "episode_sum_one"
+    },
+    "paired_value_td": true
+  }
+}

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/stats.json

@@ -0,0 +1,308 @@
+{
+  "schema": "dexrsiv2-dexjoco-nine900-v1",
+  "tasks": [
+    "click_mouse",
+    "fold_glasses",
+    "hammer_nail",
+    "pick_bucket",
+    "water_plant",
+    "bimanual_hanoi",
+    "bimanual_microwave_cook",
+    "bimanual_photograph",
+    "bimanual_unlock_ipad"
+  ],
+  "frame_count": 429528,
+  "normalization": {
+    "epsilon": 1e-08,
+    "clip": 5.0
+  },
+  "state_layout": "right_pose7_hand16_left_pose7_hand16",
+  "action_layout": "right22_left22",
+  "padding": "zero_after_normalization",
+  "observation.state": {
+    "count": [
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0
+    ],
+    "mean": [
+      -0.3493002712633983,
+      -0.24383577618624375,
+      1.2603687673297312,
+      0.1517542597981339,
+      -0.9337333770099981,
+      -0.10587760477420913,
+      0.0009141433366125124,
+      -0.16219812813444257,
+      0.41020635603790867,
+      0.7455911219006179,
+      1.0211846493651184,
+      -0.15613249313711938,
+      0.5624011315549742,
+      0.9441719467193502,
+      0.8570712691416592,
+      -0.05568327765343295,
+      0.5888186135623077,
+      0.9221678194374532,
+      0.8070650762007945,
+      0.8576553418368946,
+      0.31290979362444926,
+      0.355192882421598,
+      0.9043586101693796,
+      -0.4227596512198722,
+      0.24797489027005823,
+      1.2373904027764262,
+      -0.2649642872122615,
+      -0.8150058624450773,
+      0.12567968985698738,
+      -0.02905199493387793,
+      -0.089347394532432,
+      0.8076121978902643,
+      0.9667275746802793,
+      0.7435680529616412,
+      -0.035283851988420316,
+      0.6817023347685315,
+      1.0582855823158281,
+      0.5634651563170182,
+      0.09086352400377969,
+      0.5681596016455039,
+      1.0049067440621215,
+      0.9214411895733969,
+      0.5104102994237422,
+      0.37462264894932024,
+      0.6801564402153144,
+      1.0603516666219377
+    ],
+    "std": [
+      0.0976710777189606,
+      0.1952719372835867,
+      0.15067259331790764,
+      0.18248717871112488,
+      0.07836197167250177,
+      0.19306379672901472,
+      0.1310950947355482,
+      0.19796906174800036,
+      0.2726223539033021,
+      0.37386629065467164,
+      0.45022821817075365,
+      0.22660222141445202,
+      0.4625511483770478,
+      0.5434078378445846,
+      0.5659566722936298,
+      0.22369626176297747,
+      0.5267398585108616,
+      0.5804697669723157,
+      0.6068571475261179,
+      0.3755213060582709,
+      0.28827484532194236,
+      0.40821158764002463,
+      0.3020136772836792,
+      0.13983629195153852,
+      0.1693068837853726,
+      0.13384435576213743,
+      0.33525423169737834,
+      0.18544203464808875,
+      0.1705193154124842,
+      0.2702942522721506,
+      0.2615363606938942,
+      0.5865901076738012,
+      0.5606134874079405,
+      0.6198237884368814,
+      0.26441275515169077,
+      0.4960098569787773,
+      0.542133471347803,
+      0.7102157043393293,
+      0.2517520382896313,
+      0.38595157547855746,
+      0.4427830737211223,
+      0.40390413851783996,
+      0.23061253787198172,
+      0.33560200552922864,
+      0.44744843323156713,
+      0.3728570578969547
+    ]
+  },
+  "action": {
+    "count": [
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0
+    ],
+    "mean": [
+      -0.3435870966382824,
+      -0.24430605189524088,
+      1.2452858582430288,
+      -1.7333348925195584,
+      -0.29966068250460304,
+      0.034249144029548984,
+      -0.1477162510858357,
+      0.4599402880601885,
+      0.7898841267561791,
+      1.0391852162517041,
+      -0.11342036338680773,
+      0.6047677099424269,
+      0.9745625216633769,
+      0.869016145329004,
+      -0.02162241260774774,
+      0.6106144383555913,
+      0.9359940981345297,
+      0.8121438399734229,
+      0.956466014690754,
+      0.3111335685875101,
+      0.45594914449422397,
+      0.940704994787173,
+      -0.42476617414881873,
+      0.23572861781019414,
+      1.2236116867301374,
+      1.3649826008232766,
+      -0.24927689670379943,
+      -0.12882473596042107,
+      -0.11706859621050222,
+      0.8886180367886566,
+      0.9809192887237087,
+      0.7483279517200746,
+      -0.0927217782537091,
+      0.8128981271949458,
+      1.1060596197066632,
+      0.5771746810506223,
+      0.007379295683975502,
+      0.7366597582850135,
+      1.1049460020532034,
+      0.9508986995705223,
+      0.5980035988396629,
+      0.3195434146993771,
+      0.8757944494515124,
+      1.1497788602661225
+    ],
+    "std": [
+      0.10696119930955747,
+      0.20310346095335766,
+      0.16511848823024977,
+      1.9790998254555383,
+      0.5693717289568694,
+      0.424345427445306,
+      0.22237078548999079,
+      0.37058425385763893,
+      0.43593829889072194,
+      0.445343246646198,
+      0.27486858166781575,
+      0.5265142425517862,
+      0.5673159786481191,
+      0.564584887517863,
+      0.2641023860694298,
+      0.5647272244674655,
+      0.5930366365431426,
+      0.610061746949791,
+      0.36237361971879095,
+      0.2826861987608471,
+      0.4495371880384416,
+      0.29721779199136755,
+      0.15343398456210589,
+      0.17907598457684648,
+      0.13765813424656678,
+      1.7961800291267798,
+      0.5499092678682357,
+      0.8003153234497093,
+      0.25859695337673455,
+      0.5472637295014164,
+      0.5548769770037397,
+      0.6197415412790521,
+      0.3011903617293535,
+      0.5320294954083319,
+      0.5471901710667306,
+      0.7053992334527958,
+      0.3220185644122893,
+      0.46762781220435023,
+      0.46876426237603647,
+      0.39469137099202123,
+      0.2779228258185086,
+      0.23571571572304134,
+      0.4147188061380436,
+      0.3768777698981971
+    ]
+  }
+}

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/transformer/dexrsiv2_contract.json

@@ -0,0 +1,186 @@
+{
+  "action_dim": 44,
+  "action_mapping": "official_right22_left22_masked_zscore",
+  "architecture": {
+    "condition_order": [
+      "language",
+      "current_state",
+      "current_image"
+    ],
+    "condition_visibility": {
+      "current_image": [
+        "language",
+        "current_state",
+        "current_image"
+      ],
+      "current_state": [
+        "language",
+        "current_state"
+      ],
+      "language": [
+        "language"
+      ]
+    },
+    "policy_order": [
+      "V",
+      "noisy_action",
+      "clean_action",
+      "Q"
+    ],
+    "policy_visibility": {
+      "Q": [
+        "C",
+        "clean_action",
+        "Q"
+      ],
+      "V": [
+        "C",
+        "V"
+      ],
+      "clean_action": [
+        "C",
+        "clean_action"
+      ],
+      "noisy_action": [
+        "C",
+        "noisy_action"
+      ]
+    },
+    "segment_embedding": false,
+    "sft_condition_kv_detached": false,
+    "state_injection": "packed_text_proprio",
+    "world_order": [
+      "C",
+      "R",
+      "U",
+      "G",
+      "future_state",
+      "future_image"
+    ],
+    "world_visibility": {
+      "G": [
+        "C",
+        "G"
+      ],
+      "R": [
+        "C",
+        "R"
+      ],
+      "U": [
+        "C",
+        "U"
+      ],
+      "future_image": [
+        "C",
+        "G",
+        "future_image"
+      ],
+      "future_state": [
+        "C",
+        "G",
+        "future_state"
+      ]
+    }
+  },
+  "dataset_profile": "dexjoco_mixed9",
+  "flow": {
+    "action": {
+      "infer_shift": 5.0,
+      "num_inference_steps": 10,
+      "num_train_timesteps": 1000,
+      "train_shift": 5.0
+    },
+    "video": {
+      "infer_shift": 5.0,
+      "num_inference_steps": 10,
+      "num_train_timesteps": 1000,
+      "train_shift": 5.0
+    }
+  },
+  "horizon": 16,
+  "image": {
+    "ae_dtype": "bfloat16",
+    "ae_implementation": "flux2.autoencoder.AutoEncoder",
+    "ae_sha256": "868fe7b343cc8f3a19dbcfcafbc3d5f888802be3f89bd81b65b3621a066ce8f3",
+    "camera_count": 3,
+    "camera_time_offsets": [
+      0,
+      1,
+      2
+    ],
+    "canvas_hw": [
+      256,
+      256
+    ],
+    "cuda": "12.8",
+    "latent_grid_hw": [
+      48,
+      16
+    ],
+    "packages": {
+      "Pillow": "11.3.0",
+      "av": "12.3.0",
+      "safetensors": "0.8.0",
+      "torchvision": "0.23.0+cu128"
+    },
+    "per_view_latent_grid_hw": [
+      16,
+      16
+    ],
+    "pipeline": "dexrsiv2_dexjoco_mixed_rgb3_v1",
+    "pixel_transform": "Real views separately antialiased256-square and AE encoded; absent left view zero latent with attention/loss mask",
+    "storage_dtype": "bfloat16",
+    "token_layout": "primary256_right_wrist256_left_wrist256",
+    "torch": "2.8.0+cu128",
+    "video_decoder": "PyAV sequential RGB with official LeRobot v3 per-camera video offsets",
+    "view_keys": [
+      "primary",
+      "wrist",
+      "wrist_left"
+    ]
+  },
+  "image_profile": "dexjoco_mixed_rgb3",
+  "image_time_ids": {
+    "reference": 10,
+    "target": 0
+  },
+  "normalization_sha256": "93e2e7ba9187396b9a40625d837fbc4b80167118036c718f63016f06703ca971",
+  "padding": "normalized_zero; mask absent arm losses/logprob and absent camera keys/losses",
+  "phase": "world_policy",
+  "reward_prediction": {
+    "mode": "linear_mse",
+    "smoothing": {
+      "kind": "pre_success_gaussian",
+      "normalization": "episode_sum_one",
+      "sigma_frames": 16.0,
+      "window_frames": 48
+    }
+  },
+  "schema": "dexrsiv2-stage1-v1",
+  "state_dim": 46,
+  "state_mapping": "native46_to_right_pose7_hand16_left_pose7_hand16",
+  "step": 10000,
+  "targets": {
+    "cont": "current_frame_nonterminal",
+    "expert_terminal_cont": 0,
+    "play_cont": 1,
+    "reward": "current_frame_binary_success",
+    "reward_dim": 1,
+    "reward_positive_class": 1,
+    "reward_storage": [
+      0,
+      1
+    ],
+    "value_q_supervised": false
+  },
+  "test_only": false,
+  "text_prompt_template": "A video recorded from a robot's point of view executing the following instruction: {task}",
+  "value_training": {
+    "gamma": 0.99,
+    "horizon": 16,
+    "mode": "paired_td",
+    "q_weight": 0.0,
+    "value_weight": 0.1
+  },
+  "weights_sha256": "c2b4571b7742ac0767d672d30d0146ecbd72397f77bdf8afcc20029e27a6dfe0"
+}

checkpoints/dexrsiv2_eight_stage1_pairedv_10000_20261027/transformer/diffusion_pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2b4571b7742ac0767d672d30d0146ecbd72397f77bdf8afcc20029e27a6dfe0
+size 9045338989

checkpoints/dexrsiv2_eight_stage2_base_10000_20261023/COMPLETE.json

@@ -0,0 +1,4 @@
+{
+  "step": 10000,
+  "contract_sha256": "575fae0f9039b04514a1fbc4fea398a667db45679e4ccdb421e34157d6348934"
+}

checkpoints/dexrsiv2_eight_stage2_base_10000_20261023/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cadecb9248ca4151a6a16858e8941c475873ee0020134804ede9f1a944d5bf89
+size 9045275064

checkpoints/dexrsiv2_eight_stage2_base_10000_20261023/posttrain_contract.json

@@ -0,0 +1,209 @@
+{
+  "schema": "dexrsiv2-posttrain-v1",
+  "stage": 2,
+  "step": 10000,
+  "weights_sha256": "cadecb9248ca4151a6a16858e8941c475873ee0020134804ede9f1a944d5bf89",
+  "input_contract": {
+    "schema": "dexrsiv2-stage1-v1",
+    "architecture": {
+      "condition_order": [
+        "language",
+        "current_state",
+        "current_image"
+      ],
+      "condition_visibility": {
+        "language": [
+          "language"
+        ],
+        "current_state": [
+          "language",
+          "current_state"
+        ],
+        "current_image": [
+          "language",
+          "current_state",
+          "current_image"
+        ]
+      },
+      "world_order": [
+        "C",
+        "R",
+        "U",
+        "G",
+        "future_state",
+        "future_image"
+      ],
+      "world_visibility": {
+        "R": [
+          "C",
+          "R"
+        ],
+        "U": [
+          "C",
+          "U"
+        ],
+        "G": [
+          "C",
+          "G"
+        ],
+        "future_state": [
+          "C",
+          "G",
+          "future_state"
+        ],
+        "future_image": [
+          "C",
+          "G",
+          "future_image"
+        ]
+      },
+      "policy_order": [
+        "V",
+        "noisy_action",
+        "clean_action",
+        "Q"
+      ],
+      "policy_visibility": {
+        "V": [
+          "C",
+          "V"
+        ],
+        "noisy_action": [
+          "C",
+          "noisy_action"
+        ],
+        "clean_action": [
+          "C",
+          "clean_action"
+        ],
+        "Q": [
+          "C",
+          "clean_action",
+          "Q"
+        ]
+      },
+      "sft_condition_kv_detached": false,
+      "state_injection": "packed_text_proprio",
+      "segment_embedding": false
+    },
+    "targets": {
+      "reward": "current_frame_binary_success",
+      "reward_storage": [
+        0,
+        1
+      ],
+      "reward_positive_class": 1,
+      "reward_dim": 1,
+      "cont": "current_frame_nonterminal",
+      "expert_terminal_cont": 0,
+      "play_cont": 1,
+      "value_q_supervised": false
+    },
+    "horizon": 16,
+    "flow": {
+      "action": {
+        "train_shift": 5.0,
+        "infer_shift": 5.0,
+        "num_train_timesteps": 1000,
+        "num_inference_steps": 10
+      },
+      "video": {
+        "train_shift": 5.0,
+        "infer_shift": 5.0,
+        "num_train_timesteps": 1000,
+        "num_inference_steps": 10
+      }
+    },
+    "text_prompt_template": "A video recorded from a robot's point of view executing the following instruction: {task}",
+    "image_time_ids": {
+      "reference": 10,
+      "target": 0
+    },
+    "image": {
+      "pipeline": "dexrsiv2_dexjoco_mixed_rgb3_v1",
+      "ae_sha256": "868fe7b343cc8f3a19dbcfcafbc3d5f888802be3f89bd81b65b3621a066ce8f3",
+      "view_keys": [
+        "primary",
+        "wrist",
+        "wrist_left"
+      ],
+      "pixel_transform": "Real views separately antialiased256-square and AE encoded; absent left view zero latent with attention/loss mask",
+      "canvas_hw": [
+        256,
+        256
+      ],
+      "latent_grid_hw": [
+        48,
+        16
+      ],
+      "ae_implementation": "flux2.autoencoder.AutoEncoder",
+      "ae_dtype": "bfloat16",
+      "storage_dtype": "bfloat16",
+      "torch": "2.8.0+cu128",
+      "cuda": "12.8",
+      "packages": {
+        "torchvision": "0.23.0+cu128",
+        "Pillow": "11.3.0",
+        "av": "12.3.0",
+        "safetensors": "0.8.0"
+      },
+      "per_view_latent_grid_hw": [
+        16,
+        16
+      ],
+      "camera_count": 3,
+      "token_layout": "primary256_right_wrist256_left_wrist256",
+      "camera_time_offsets": [
+        0,
+        1,
+        2
+      ],
+      "video_decoder": "PyAV sequential RGB with official LeRobot v3 per-camera video offsets"
+    },
+    "normalization_sha256": "93e2e7ba9187396b9a40625d837fbc4b80167118036c718f63016f06703ca971",
+    "action_mapping": "official_right22_left22_masked_zscore",
+    "dataset_profile": "dexjoco_mixed9",
+    "image_profile": "dexjoco_mixed_rgb3",
+    "action_dim": 44,
+    "state_dim": 46,
+    "state_mapping": "native46_to_right_pose7_hand16_left_pose7_hand16",
+    "padding": "normalized_zero; mask absent arm losses/logprob and absent camera keys/losses"
+  },
+  "config": {
+    "world_size": 8,
+    "max_steps": 10000,
+    "checkpoint_interval": 1000,
+    "seed": 42,
+    "wandb": true,
+    "stage": 2,
+    "name": "dexrsiv2-eight-stage2-base-held8-noac-r3000-20261022-v1",
+    "microbatch": 32,
+    "accumulation": 1,
+    "global_batch": 256,
+    "include_tasks": [
+      "click_mouse",
+      "fold_glasses",
+      "hammer_nail",
+      "pick_bucket",
+      "water_plant",
+      "bimanual_hanoi",
+      "bimanual_microwave_cook",
+      "bimanual_photograph"
+    ],
+    "critic_mode": "td_huber",
+    "weights": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_mixed9_stage1_10k_20261020_v1/models/checkpoint_epoch_6_step_10000/transformer/diffusion_pytorch_model.bin",
+    "eval_dataset": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_stage1_10k_eval400_randobj_seed0_uniform3_20261021_v1",
+    "output": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage2_base_held8_acc1_20261022_v1/training",
+    "gradient_checkpointing": true,
+    "migration_rng_fork": {
+      "step": 3000,
+      "source_world_size": 4,
+      "seeds": {
+        "4": 5389180079862263966,
+        "5": 3508557990337216216,
+        "6": 2828642300766811049,
+        "7": 8192082010321363905
+      }
+    }
+  }
+}

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/COMPLETE.json

@@ -0,0 +1,4 @@
+{
+  "step": 10000,
+  "contract_sha256": "073ac3a767b4af437cd0e036a121fe638932ce51847ca59820021a80e732b94d"
+}

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/MANIFEST.json

@@ -0,0 +1,41 @@
+{
+  "schema": "dexrsi-hf-model-only-v1",
+  "repo": "AvaX1/dex_eval_data",
+  "repo_type": "dataset",
+  "prefix": "checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027",
+  "stage": 2,
+  "step": 10000,
+  "source_checkpoint": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage2_pairedv_held8_20261026_v1/training/checkpoints/step_010000",
+  "files": {
+    "model.safetensors": {
+      "bytes": 9045275064,
+      "sha256": "2c4f839b5caa09c002faa357e6f2c37b4912807343d88f4ded61a5f4a144d66f"
+    },
+    "stats.json": {
+      "bytes": 6863,
+      "sha256": "93e2e7ba9187396b9a40625d837fbc4b80167118036c718f63016f06703ca971"
+    },
+    "stage2.recorded.json": {
+      "bytes": 1368,
+      "sha256": "77872b456bcad405193c8369f1fa9480ebc508742b5b04172d3a46c8f132217d"
+    },
+    "stage1.recorded.json": {
+      "bytes": 2961,
+      "sha256": "6cbdac518a87d9a09036337ce59436f7e29fc063f7b7ababcf0bc2ba0eba47a4"
+    },
+    "COMPLETE.json": {
+      "bytes": 109,
+      "sha256": "80ecf218f595eff8729dd4d2408478192ad4cf61b47b0b4d7e802375061871c5"
+    },
+    "README.md": {
+      "bytes": 1565,
+      "sha256": "027ad4473c23570bd69b9d55f44632c977aa38b79f3315cadbcb45e9f0ee384f"
+    },
+    "posttrain_contract.json": {
+      "bytes": 5645,
+      "sha256": "073ac3a767b4af437cd0e036a121fe638932ce51847ca59820021a80e732b94d"
+    }
+  },
+  "optimizer_included": false,
+  "rng_included": false
+}

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/README.md

@@ -0,0 +1,24 @@
+# DexRSI eight-task paired-V Stage2 — 10,000 steps
+
+Inference/model-only export of the completed 2026-10-26/27 eight-GPU Stage2 run.
+Source: paired-V eight-task Stage1 10k; fresh Stage2 Adam; 10k updates.
+
+- Linear MSE reward with pre-success one-sided Gaussian (sigma 16 frames,
+  d=0..48, clipped at episode start, per-episode sum 1).
+- Reward support/zero class means weighted 0.5/0.5; outer weight 0.1.
+- Paired V TD weight 0.1, Q loss 0; expert-only BC (no eval-success BC).
+- Continuation used ordinary masked BCE. The later cont-balance code change
+  was NOT applied to these weights. RL must consume raw linear reward without sigmoid.
+- 44 action dimensions, 46 state dimensions, 3 camera slots, masked absent limbs/views.
+
+`model.safetensors`, `posttrain_contract.json` and `COMPLETE.json` retain the original
+checkpoint names and bytes. The original input contract contains historical binary-label
+and mixed9 schema fields; the actual eight-task Gaussian training semantics are in its
+Stage2 `config` and the recorded configurations. Do not rewrite sealed contracts.
+
+Use https://github.com/lhj-lhj/DexRSI/tree/69a650bc049dd7f2c04a2426cdb6e9c6f199a877
+and its reproduction guide. The config files retain original path provenance: supply
+local model/backbone/AE/data locations explicitly. `stats.json` is hash-verified original
+normalization. FLUX backbone/tokenizer/text encoder and the FLUX.2-dev AE are separate
+assets. No optimizer, RNG, dataset or private credentials are included. This model-only
+export cannot resume the original optimizer state.

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c4f839b5caa09c002faa357e6f2c37b4912807343d88f4ded61a5f4a144d66f
+size 9045275064

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/posttrain_contract.json

@@ -0,0 +1,212 @@
+{
+  "schema": "dexrsiv2-posttrain-v1",
+  "stage": 2,
+  "step": 10000,
+  "weights_sha256": "2c4f839b5caa09c002faa357e6f2c37b4912807343d88f4ded61a5f4a144d66f",
+  "input_contract": {
+    "schema": "dexrsiv2-stage1-v1",
+    "architecture": {
+      "condition_order": [
+        "language",
+        "current_state",
+        "current_image"
+      ],
+      "condition_visibility": {
+        "language": [
+          "language"
+        ],
+        "current_state": [
+          "language",
+          "current_state"
+        ],
+        "current_image": [
+          "language",
+          "current_state",
+          "current_image"
+        ]
+      },
+      "world_order": [
+        "C",
+        "R",
+        "U",
+        "G",
+        "future_state",
+        "future_image"
+      ],
+      "world_visibility": {
+        "R": [
+          "C",
+          "R"
+        ],
+        "U": [
+          "C",
+          "U"
+        ],
+        "G": [
+          "C",
+          "G"
+        ],
+        "future_state": [
+          "C",
+          "G",
+          "future_state"
+        ],
+        "future_image": [
+          "C",
+          "G",
+          "future_image"
+        ]
+      },
+      "policy_order": [
+        "V",
+        "noisy_action",
+        "clean_action",
+        "Q"
+      ],
+      "policy_visibility": {
+        "V": [
+          "C",
+          "V"
+        ],
+        "noisy_action": [
+          "C",
+          "noisy_action"
+        ],
+        "clean_action": [
+          "C",
+          "clean_action"
+        ],
+        "Q": [
+          "C",
+          "clean_action",
+          "Q"
+        ]
+      },
+      "sft_condition_kv_detached": false,
+      "state_injection": "packed_text_proprio",
+      "segment_embedding": false
+    },
+    "targets": {
+      "reward": "current_frame_binary_success",
+      "reward_storage": [
+        0,
+        1
+      ],
+      "reward_positive_class": 1,
+      "reward_dim": 1,
+      "cont": "current_frame_nonterminal",
+      "expert_terminal_cont": 0,
+      "play_cont": 1,
+      "value_q_supervised": false
+    },
+    "horizon": 16,
+    "flow": {
+      "action": {
+        "train_shift": 5.0,
+        "infer_shift": 5.0,
+        "num_train_timesteps": 1000,
+        "num_inference_steps": 10
+      },
+      "video": {
+        "train_shift": 5.0,
+        "infer_shift": 5.0,
+        "num_train_timesteps": 1000,
+        "num_inference_steps": 10
+      }
+    },
+    "text_prompt_template": "A video recorded from a robot's point of view executing the following instruction: {task}",
+    "image_time_ids": {
+      "reference": 10,
+      "target": 0
+    },
+    "image": {
+      "pipeline": "dexrsiv2_dexjoco_mixed_rgb3_v1",
+      "ae_sha256": "868fe7b343cc8f3a19dbcfcafbc3d5f888802be3f89bd81b65b3621a066ce8f3",
+      "view_keys": [
+        "primary",
+        "wrist",
+        "wrist_left"
+      ],
+      "pixel_transform": "Real views separately antialiased256-square and AE encoded; absent left view zero latent with attention/loss mask",
+      "canvas_hw": [
+        256,
+        256
+      ],
+      "latent_grid_hw": [
+        48,
+        16
+      ],
+      "ae_implementation": "flux2.autoencoder.AutoEncoder",
+      "ae_dtype": "bfloat16",
+      "storage_dtype": "bfloat16",
+      "torch": "2.8.0+cu128",
+      "cuda": "12.8",
+      "packages": {
+        "torchvision": "0.23.0+cu128",
+        "Pillow": "11.3.0",
+        "av": "12.3.0",
+        "safetensors": "0.8.0"
+      },
+      "per_view_latent_grid_hw": [
+        16,
+        16
+      ],
+      "camera_count": 3,
+      "token_layout": "primary256_right_wrist256_left_wrist256",
+      "camera_time_offsets": [
+        0,
+        1,
+        2
+      ],
+      "video_decoder": "PyAV sequential RGB with official LeRobot v3 per-camera video offsets"
+    },
+    "normalization_sha256": "93e2e7ba9187396b9a40625d837fbc4b80167118036c718f63016f06703ca971",
+    "action_mapping": "official_right22_left22_masked_zscore",
+    "dataset_profile": "dexjoco_mixed9",
+    "image_profile": "dexjoco_mixed_rgb3",
+    "action_dim": 44,
+    "state_dim": 46,
+    "state_mapping": "native46_to_right_pose7_hand16_left_pose7_hand16",
+    "padding": "normalized_zero; mask absent arm losses/logprob and absent camera keys/losses"
+  },
+  "config": {
+    "world_size": 8,
+    "max_steps": 10000,
+    "checkpoint_interval": 1000,
+    "seed": 42,
+    "wandb": true,
+    "stage": 2,
+    "name": "dexrsiv2-eight-stage2-pairedv-held8-20261026-v1",
+    "microbatch": 32,
+    "accumulation": 1,
+    "global_batch": 256,
+    "include_tasks": [
+      "click_mouse",
+      "fold_glasses",
+      "hammer_nail",
+      "pick_bucket",
+      "water_plant",
+      "bimanual_hanoi",
+      "bimanual_microwave_cook",
+      "bimanual_photograph"
+    ],
+    "critic_mode": "td_huber",
+    "reward_balance_alpha": 0.5,
+    "reward_prediction_mode": "linear_mse",
+    "reward_head_init": "preserve",
+    "reward_smoothing": {
+      "kind": "pre_success_gaussian",
+      "sigma_frames": 16.0,
+      "window_frames": 48,
+      "normalization": "episode_sum_one"
+    },
+    "eval_success_bc": false,
+    "value_training": "paired_td",
+    "value_loss_weight": 0.1,
+    "q_loss_weight": 0.0,
+    "weights": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage1_paired_v_8gpu_20261025_v1/training/models/checkpoint_epoch_7_step_10000/transformer/diffusion_pytorch_model.bin",
+    "eval_dataset": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_pairedv_stage1_10k_eval400_randobj_seed0_uniform3_20261026_av1_v1",
+    "output": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage2_pairedv_held8_20261026_v1/training",
+    "gradient_checkpointing": true
+  }
+}

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/stage1.recorded.json

@@ -0,0 +1,113 @@
+{
+  "paths": {
+    "dataset": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_800_terminal_v1",
+    "cache": "/mnt/project/maxinsights/liuhongjia/dataset/cache/dexrsiv2/dexjoco_mixed8_20261021_v1",
+    "output": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage1_paired_v_8gpu_20261025_v1/training",
+    "flux_checkpoint": "/mnt/project/maxinsights/liuhongjia/model_weights/robonana-dex/FLUX.2-klein-base-4B",
+    "ae_checkpoint": "/mnt/project/maxinsights/liuhongjia/model_weights/imagewam-g3/modelscope-cache/black-forest-labs/FLUX___2-dev/ae.safetensors",
+    "play_dataset": null,
+    "play_cache": null
+  },
+  "data": {
+    "expected_episodes": 800,
+    "expected_frames": 388891,
+    "fps": 30,
+    "action_dim": 44,
+    "chunk_horizon": 16,
+    "play_fraction": 0.5,
+    "dataset_profile": "dexjoco_mixed8",
+    "state_dim": 46,
+    "image_profile": "dexjoco_mixed_rgb3",
+    "latent_grid_height": 48,
+    "latent_grid_width": 16,
+    "terminal_fraction": 0.02,
+    "context_tokens": 76
+  },
+  "model": {
+    "expert_hidden_dim": 1024,
+    "action_expert_mlp_ratio": 4.0,
+    "gradient_checkpointing": true,
+    "flux_params": {
+      "in_channels": 128,
+      "context_in_dim": 7680,
+      "hidden_size": 3072,
+      "num_heads": 24,
+      "depth": 5,
+      "depth_single_blocks": 20,
+      "axes_dim": [
+        32,
+        32,
+        32,
+        32
+      ],
+      "theta": 2000,
+      "mlp_ratio": 3.0,
+      "use_guidance_embed": false
+    }
+  },
+  "action_flow": {
+    "train_shift": 5.0,
+    "infer_shift": 5.0,
+    "num_train_timesteps": 1000,
+    "num_inference_steps": 10
+  },
+  "video_flow": {
+    "train_shift": 5.0,
+    "infer_shift": 5.0,
+    "num_train_timesteps": 1000,
+    "num_inference_steps": 10
+  },
+  "optimization": {
+    "gpus": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7
+    ],
+    "microbatch": 32,
+    "global_batch": 256,
+    "num_workers": 4,
+    "max_steps": 10000,
+    "warmup_steps": 1000,
+    "lr": 0.0001,
+    "robot_lr": 0.0001,
+    "weight_decay": 0.01,
+    "min_lr_ratio": 0.01,
+    "max_grad_norm": 1.0,
+    "checkpoint_interval": 1000,
+    "checkpoint_total_limit": 2,
+    "checkpoint_keeps": [
+      10000
+    ],
+    "seed": 42,
+    "log_interval": 10,
+    "log_with": "wandb",
+    "tracker_project": "DexRSIv2",
+    "tracker_entity": "hongjia-liu-aalto-university",
+    "run_name": "dexrsiv2-eight-stage1-paired-v-8gpu-20261025-v1"
+  },
+  "objectives": {
+    "loss_weights": {
+      "image_loss": 0.5,
+      "action_loss": 1.0,
+      "future_state_loss": 0.4,
+      "reward_loss": 0.1,
+      "cont_loss": 0.1,
+      "value_loss": 0.1,
+      "q_loss": 0.0
+    },
+    "reward_balance_alpha": null,
+    "reward_prediction_mode": "linear_mse",
+    "reward_smoothing": {
+      "kind": "pre_success_gaussian",
+      "sigma_frames": 16.0,
+      "window_frames": 48,
+      "normalization": "episode_sum_one"
+    },
+    "paired_value_td": true
+  }
+}

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/stage2.recorded.json

@@ -0,0 +1,40 @@
+{
+  "world_size": 8,
+  "max_steps": 10000,
+  "checkpoint_interval": 1000,
+  "seed": 42,
+  "wandb": true,
+  "stage": 2,
+  "name": "dexrsiv2-eight-stage2-pairedv-held8-20261026-v1",
+  "microbatch": 32,
+  "accumulation": 1,
+  "global_batch": 256,
+  "include_tasks": [
+    "click_mouse",
+    "fold_glasses",
+    "hammer_nail",
+    "pick_bucket",
+    "water_plant",
+    "bimanual_hanoi",
+    "bimanual_microwave_cook",
+    "bimanual_photograph"
+  ],
+  "critic_mode": "td_huber",
+  "reward_balance_alpha": 0.5,
+  "reward_prediction_mode": "linear_mse",
+  "reward_head_init": "preserve",
+  "reward_smoothing": {
+    "kind": "pre_success_gaussian",
+    "sigma_frames": 16.0,
+    "window_frames": 48,
+    "normalization": "episode_sum_one"
+  },
+  "eval_success_bc": false,
+  "value_training": "paired_td",
+  "value_loss_weight": 0.1,
+  "q_loss_weight": 0.0,
+  "weights": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage1_paired_v_8gpu_20261025_v1/training/models/checkpoint_epoch_7_step_10000/transformer/diffusion_pytorch_model.bin",
+  "eval_dataset": "/mnt/project/maxinsights/liuhongjia/dataset/dexjoco/mixed8_pairedv_stage1_10k_eval400_randobj_seed0_uniform3_20261026_av1_v1",
+  "output": "/mnt/project/maxinsights/liuhongjia/training_runs/DexRSIv2/dexjoco_eight_stage2_pairedv_held8_20261026_v1/training",
+  "gradient_checkpointing": true
+}

checkpoints/dexrsiv2_eight_stage2_pairedv_10000_20261027/stats.json

@@ -0,0 +1,308 @@
+{
+  "schema": "dexrsiv2-dexjoco-nine900-v1",
+  "tasks": [
+    "click_mouse",
+    "fold_glasses",
+    "hammer_nail",
+    "pick_bucket",
+    "water_plant",
+    "bimanual_hanoi",
+    "bimanual_microwave_cook",
+    "bimanual_photograph",
+    "bimanual_unlock_ipad"
+  ],
+  "frame_count": 429528,
+  "normalization": {
+    "epsilon": 1e-08,
+    "clip": 5.0
+  },
+  "state_layout": "right_pose7_hand16_left_pose7_hand16",
+  "action_layout": "right22_left22",
+  "padding": "zero_after_normalization",
+  "observation.state": {
+    "count": [
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0
+    ],
+    "mean": [
+      -0.3493002712633983,
+      -0.24383577618624375,
+      1.2603687673297312,
+      0.1517542597981339,
+      -0.9337333770099981,
+      -0.10587760477420913,
+      0.0009141433366125124,
+      -0.16219812813444257,
+      0.41020635603790867,
+      0.7455911219006179,
+      1.0211846493651184,
+      -0.15613249313711938,
+      0.5624011315549742,
+      0.9441719467193502,
+      0.8570712691416592,
+      -0.05568327765343295,
+      0.5888186135623077,
+      0.9221678194374532,
+      0.8070650762007945,
+      0.8576553418368946,
+      0.31290979362444926,
+      0.355192882421598,
+      0.9043586101693796,
+      -0.4227596512198722,
+      0.24797489027005823,
+      1.2373904027764262,
+      -0.2649642872122615,
+      -0.8150058624450773,
+      0.12567968985698738,
+      -0.02905199493387793,
+      -0.089347394532432,
+      0.8076121978902643,
+      0.9667275746802793,
+      0.7435680529616412,
+      -0.035283851988420316,
+      0.6817023347685315,
+      1.0582855823158281,
+      0.5634651563170182,
+      0.09086352400377969,
+      0.5681596016455039,
+      1.0049067440621215,
+      0.9214411895733969,
+      0.5104102994237422,
+      0.37462264894932024,
+      0.6801564402153144,
+      1.0603516666219377
+    ],
+    "std": [
+      0.0976710777189606,
+      0.1952719372835867,
+      0.15067259331790764,
+      0.18248717871112488,
+      0.07836197167250177,
+      0.19306379672901472,
+      0.1310950947355482,
+      0.19796906174800036,
+      0.2726223539033021,
+      0.37386629065467164,
+      0.45022821817075365,
+      0.22660222141445202,
+      0.4625511483770478,
+      0.5434078378445846,
+      0.5659566722936298,
+      0.22369626176297747,
+      0.5267398585108616,
+      0.5804697669723157,
+      0.6068571475261179,
+      0.3755213060582709,
+      0.28827484532194236,
+      0.40821158764002463,
+      0.3020136772836792,
+      0.13983629195153852,
+      0.1693068837853726,
+      0.13384435576213743,
+      0.33525423169737834,
+      0.18544203464808875,
+      0.1705193154124842,
+      0.2702942522721506,
+      0.2615363606938942,
+      0.5865901076738012,
+      0.5606134874079405,
+      0.6198237884368814,
+      0.26441275515169077,
+      0.4960098569787773,
+      0.542133471347803,
+      0.7102157043393293,
+      0.2517520382896313,
+      0.38595157547855746,
+      0.4427830737211223,
+      0.40390413851783996,
+      0.23061253787198172,
+      0.33560200552922864,
+      0.44744843323156713,
+      0.3728570578969547
+    ]
+  },
+  "action": {
+    "count": [
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      429528.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0,
+      250600.0
+    ],
+    "mean": [
+      -0.3435870966382824,
+      -0.24430605189524088,
+      1.2452858582430288,
+      -1.7333348925195584,
+      -0.29966068250460304,
+      0.034249144029548984,
+      -0.1477162510858357,
+      0.4599402880601885,
+      0.7898841267561791,
+      1.0391852162517041,
+      -0.11342036338680773,
+      0.6047677099424269,
+      0.9745625216633769,
+      0.869016145329004,
+      -0.02162241260774774,
+      0.6106144383555913,
+      0.9359940981345297,
+      0.8121438399734229,
+      0.956466014690754,
+      0.3111335685875101,
+      0.45594914449422397,
+      0.940704994787173,
+      -0.42476617414881873,
+      0.23572861781019414,
+      1.2236116867301374,
+      1.3649826008232766,
+      -0.24927689670379943,
+      -0.12882473596042107,
+      -0.11706859621050222,
+      0.8886180367886566,
+      0.9809192887237087,
+      0.7483279517200746,
+      -0.0927217782537091,
+      0.8128981271949458,
+      1.1060596197066632,
+      0.5771746810506223,
+      0.007379295683975502,
+      0.7366597582850135,
+      1.1049460020532034,
+      0.9508986995705223,
+      0.5980035988396629,
+      0.3195434146993771,
+      0.8757944494515124,
+      1.1497788602661225
+    ],
+    "std": [
+      0.10696119930955747,
+      0.20310346095335766,
+      0.16511848823024977,
+      1.9790998254555383,
+      0.5693717289568694,
+      0.424345427445306,
+      0.22237078548999079,
+      0.37058425385763893,
+      0.43593829889072194,
+      0.445343246646198,
+      0.27486858166781575,
+      0.5265142425517862,
+      0.5673159786481191,
+      0.564584887517863,
+      0.2641023860694298,
+      0.5647272244674655,
+      0.5930366365431426,
+      0.610061746949791,
+      0.36237361971879095,
+      0.2826861987608471,
+      0.4495371880384416,
+      0.29721779199136755,
+      0.15343398456210589,
+      0.17907598457684648,
+      0.13765813424656678,
+      1.7961800291267798,
+      0.5499092678682357,
+      0.8003153234497093,
+      0.25859695337673455,
+      0.5472637295014164,
+      0.5548769770037397,
+      0.6197415412790521,
+      0.3011903617293535,
+      0.5320294954083319,
+      0.5471901710667306,
+      0.7053992334527958,
+      0.3220185644122893,
+      0.46762781220435023,
+      0.46876426237603647,
+      0.39469137099202123,
+      0.2779228258185086,
+      0.23571571572304134,
+      0.4147188061380436,
+      0.3768777698981971
+    ]
+  }
+}

checkpoints/dexrsiv2_mixed9_stage1_10k_20261020_v1/transformer/diffusion_pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffad455de1a7cd21161ae09328177e4a7744acef0125a79f7fbeeb265cd4db0d
+size 9045338989

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/MANIFEST.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc0e603c920a61b2727638e1ce2098cb912cb845da8105be1a1629a362005b9b
+size 32453944

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__00.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9d96a6c1699bc587262d2b8158a776e220255c0310ad8bc6c779d6c3a503153
+size 158597120

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__01.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44b46c182a6b3725a258bedd67211ccc41edbd9ecc329bcddfee1074029c86e7
+size 196761600

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__02.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:550dbb2e63e1371314878911e260b2655fa284d2dffb2b8d6959df72a57ba9e9
+size 201523200

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__03.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6009a4cbe736cdb42449705cec43cd1c3304c694883db9f74f13f84cd36a7e96
+size 146565120

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__04.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c340427b595f7e55ec5f6749b8062589d0585f3938f1007347c61a4d6461c5f2
+size 186429440

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__05.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22dbaf23e1fa904102583f87c20cb1447c4a3b0dc9c8f1912e02a4f301089084
+size 182241280

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__06.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc8ebaa063df047ade4d4aca3f0fcdd6364fdeae07c16d965aac9f1c498ddf09
+size 197693440

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__07.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff7d769bb8b3911dc624ce7f554adb2078f0fceb2a9b657ef43547f12e5ca215
+size 178380800

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__08.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:677d6298b08862a046188817df731d513c7088cd781519f0ae3ccb551d0c4555
+size 196915200

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__09.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36a56d926c372eab2a92519f481dd54ae4d17e37ca0ec1dbbee0aec45cae3dc5
+size 185917440

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__10.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4d25b6439f5ed7e803744102a5e0e6d2b6c7fa9226df8efa5a461332d5bc795
+size 161249280

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__11.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b20b6b407f075636a007dfad00e8eda3088ba801a3f0a3a5db297be372ae6df
+size 196597760

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__12.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ad8134de0b284e689f7a5884a9e2b142dbfb8e3f4d3a2a4bc09ac7515e5a9e2
+size 194252800

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__13.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2915f45b3a090b4b3e3548a0cd434c879c01a76a6d06cfe9e0911ba29de9e47
+size 95959040

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__14.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3ad9b5cf45e59cfb252e0bd06adbda0a03599f93b362826eedc1570db53fa05
+size 110376960

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__15.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a0c2f58a3a3cef66bac4375e5623dae4b66b5e040518afc27d869fe483c7fca
+size 128716800

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__16.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e0ca9d359419d04d11f5423d01226c910021d12320e2d63c4acb87c812eeb00
+size 117985280

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__17.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:296fc23cdaf6c9cc2c701cd52f478da73884c779a1967db63b1b53d3e3de0ac5
+size 202588160

dexrsiv2_mixed8_eval400_randobj_seed0_uniform3_256_20261021/cases/bimanual_hanoi__seed0__18.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:947778569eaabb26ca3bcbbd82e15cc1aad48bcf9d862269ccb3d284cc8def5f
+size 157829120