Upload folder using huggingface_hub

tinyvla2/pretrain_C_scaled/README.md

@@ -0,0 +1,25 @@
+# TinyVLA-2 — C-scaled (BEST CHECKPOINT — use this one)
+
+C setup (canonical base-frame EE actions + numeric morphology descriptor) trained
+with a real budget: full Bridge (53K eps) + full RT-1 (87K eps) — 9.1M frame pool —
+for 60k steps with the LM unfrozen at 0.1x lr.
+
+## Results (endpoint error / per-robot zero-prediction floor; <1.0 beats baseline)
+| robot | before scaling | after |
+|---|---|---|
+| jaco (arm) | 1.10 | **0.69** |
+| xarm | 0.95 | **0.64** |
+| dlr_edan | 0.96 | **0.69** |
+| stretch (mobile) | 0.79 | **0.69** |
+| ur5 (fastest robot) | 1.25 | 1.22 |
+| LeKiwi (HELD-OUT mobile) | 1.23 | **1.07** |
+| mean in-training | 0.99 | **0.81** |
+| mean held-out | 1.17 | **0.88** |
+
+Scaling data + steps + unfreezing the LM is what made this model competent; five
+different conditioning mechanisms (text prompts, Qwen-encoded morphology, three
+demo-conditioning architectures) moved nothing. In-training saturates at 40k;
+held-out and especially LeKiwi were still improving at 60k.
+
+Load: `TinyVLAPolicy.from_pretrained(...)`, `conditioning="morph"`,
+`action_space="canonical"`. Full history: tinyvla2/RESULTS.md

tinyvla2/pretrain_C_scaled/config.json

@@ -0,0 +1,107 @@
+{
+    "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",
+    "morph_tokens": 2,
+    "morph_to_slow": false,
+    "num_morph_readout": 4,
+    "morph_text_max_len": 32,
+    "use_demo_conditioning": false,
+    "n_support": 3,
+    "support_other_task": false,
+    "demo_tokens_per_example": 4,
+    "demo_visual_groups": 4,
+    "demo_action_keys": 8,
+    "demo_hidden_mult": 4,
+    "vlm_native": false,
+    "num_action_readout": 16,
+    "demo_action_keys_lm": 4,
+    "demo_only": false,
+    "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_scaled/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e8ea641f556eba437b5d41255f833eb4f39db320e1dda52e9a318bb03ec2936
+size 2575766712