Upload folder using huggingface_hub

tinyvla2/pretrain_C_incontext/README.md

@@ -0,0 +1,14 @@
+# TinyVLA-2 — C-incontext (in-context demo conditioning, NEGATIVE RESULT)
+
+K=3 (obs,action) exemplar pairs from the target robot fed as conditioning tokens
+at test time, no gradient updates. Trained 20k steps from C-qwen-morph.
+
+## Result: dead channel
+Real vs garbage/noise support examples give indistinguishable results (324.7mm vs
+322.5mm) — the model learned to ignore demo content. Likely cause: during training,
+support pairs were sampled from the SAME source dataset as the query, redundant
+with the already-available embodiment identity signal (text+numeric channels), so
+gradient found no reason to depend on demo content.
+
+**Do not use this checkpoint** — C-diverse remains the reference (150mm on
+held-out jaco, 295mm zero-shot LeKiwi). Full context: tinyvla2/RESULTS.md.

tinyvla2/pretrain_C_incontext/config.json

@@ -0,0 +1,98 @@
+{
+    "type": "tinyvla",
+    "n_obs_steps": 1,
+    "input_features": {
+        "observation.images.cam0": {
+            "type": "VISUAL",
+            "shape": [
+                3,
+                256,
+                256
+            ]
+        },
+        "observation.images.cam1": {
+            "type": "VISUAL",
+            "shape": [
+                3,
+                256,
+                256
+            ]
+        },
+        "observation.state": {
+            "type": "STATE",
+            "shape": [
+                16
+            ]
+        }
+    },
+    "output_features": {
+        "action": {
+            "type": "ACTION",
+            "shape": [
+                8
+            ]
+        }
+    },
+    "device": "cuda",
+    "use_amp": false,
+    "use_peft": false,
+    "push_to_hub": true,
+    "repo_id": null,
+    "private": null,
+    "tags": null,
+    "license": null,
+    "pretrained_path": null,
+    "pretrained_revision": null,
+    "chunk_size": 50,
+    "n_action_steps": 50,
+    "normalization_mapping": {
+        "VISUAL": "IDENTITY",
+        "STATE": "MEAN_STD",
+        "ACTION": "MEAN_STD"
+    },
+    "max_state_dim": 16,
+    "max_action_dim": 8,
+    "num_embodiments": 16,
+    "max_cameras": 3,
+    "conditioning": "morph_qwen",
+    "morph_tokens": 2,
+    "morph_to_slow": false,
+    "num_morph_readout": 4,
+    "morph_text_max_len": 32,
+    "use_demo_conditioning": true,
+    "n_support": 3,
+    "action_space": "canonical",
+    "image_size": 256,
+    "freeze_vision_encoder": true,
+    "lm_model_name": "Qwen/Qwen3.5-0.8B",
+    "lm_num_layers": 12,
+    "num_readout_tokens": 8,
+    "freeze_lm": false,
+    "tokenizer_max_length": 48,
+    "pad_language_to": "longest",
+    "expert_dim": 512,
+    "expert_layers": 12,
+    "expert_heads": 8,
+    "use_semantic_latent": true,
+    "use_spatial_tokens": true,
+    "spatial_vocab": 1024,
+    "spatial_loss_weight": 0.0,
+    "num_denoise_steps": 10,
+    "flow_beta_alpha": 1.5,
+    "flow_beta_beta": 1.0,
+    "min_period": 0.004,
+    "max_period": 4.0,
+    "staleness_prob": 0.0,
+    "staleness_max_s": 2.0,
+    "optimizer_lr": 0.0001,
+    "optimizer_betas": [
+        0.9,
+        0.95
+    ],
+    "optimizer_eps": 1e-08,
+    "optimizer_weight_decay": 1e-10,
+    "optimizer_grad_clip_norm": 10.0,
+    "scheduler_warmup_steps": 1000,
+    "scheduler_decay_steps": 40000,
+    "scheduler_decay_lr": 2.5e-06
+}

tinyvla2/pretrain_C_incontext/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4f01fdbff1f33230680fb3da16050c79fc43317f372895f95796d0f6dd172ec
+size 2578929032