Add files using upload-large-folder tool

README.md

@@ -0,0 +1,31 @@
+---
+tags:
+- robotics
+- human-robot-interaction
+- flow-matching
+- robot-learning
+---
+
+# HRI-diff large, H50, 100k steps
+
+This is the 768-wide HRI-diff joint checkpoint trained for 100,000 steps on
+HITBench tomato-to-bowl. `model.pt` includes the conditioning encoder, trust
+flow, action decoder, frozen trust-tube tokenizer, and latent statistics.
+
+The policy takes RGB-D at causal offsets `[-9, -6, -3, 0]`, current 8-D robot
+state, and separate regular and interaction prompts. It predicts 50 continuous
+actions. In the closed-loop benchmark it executes one action and replans at
+every step using 20 flow integration steps.
+
+Evaluation artifacts: https://huggingface.co/datasets/nhatcm/hri_diff_large_eval
+
+```python
+from huggingface_hub import snapshot_download
+from hitbench.hitbench.models.hri_diff.training import HRIDiffPhaseTwo
+
+path = snapshot_download("nhatcm/hri_diff_large")
+model = HRIDiffPhaseTwo.from_pretrained(path, map_location="cpu")
+```
+
+The flow-matching dependency is licensed CC BY-NC; check its terms before
+commercial use.

config.json

@@ -0,0 +1,41 @@
+{
+  "hri": {
+    "observation_dim": 768,
+    "motion_dim": 768,
+    "instruction_dim": 768,
+    "proprio_dim": 8,
+    "action_dim": 7,
+    "trust_dim": 512,
+    "hidden_dim": 768,
+    "trust_tokens": 50,
+    "action_horizon": 50,
+    "num_heads": 12,
+    "context_layers": 3,
+    "flow_layers": 12,
+    "decoder_layers": 8,
+    "mlp_ratio": 4.0,
+    "dropout": 0.1,
+    "smooth_l1_beta": 1.0,
+    "bound_actions": false
+  },
+  "conditioning": {
+    "feature_dim": 768,
+    "spatial_grid": 8,
+    "motion_grid": 4,
+    "max_prompt_bytes": 256,
+    "prompt_heads": 12,
+    "dropout": 0.1
+  },
+  "tokenizer": {
+    "horizon": 50,
+    "mask_size": 64,
+    "trust_dim": 512,
+    "hidden_dim": 1024,
+    "depth_scale_m": 3.0
+  },
+  "training_contract": {
+    "trust_tail": "forward-fill-last-causal-target",
+    "decoder_trust": "linear-flow-clean-estimate-curriculum",
+    "action_history": "two-pass-self-conditioning-curriculum"
+  }
+}

model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c4fe040382609645a65b7b4967b162ac7a570597f1fa201a5c5cc123611c220
+size 1192855103