WOD-E2E Fast-dDrive SASD TPU-ready Parquet (400-sample subset, private)

README.md

@@ -0,0 +1,82 @@
+---
+pretty_name: WOD-E2E Fast-dDrive SASD (TPU-ready)
+license: other
+license_name: waymo-open-dataset-license
+license_link: https://waymo.com/open/terms/
+tags:
+- autonomous-driving
+- waymo-open-dataset
+- vision-language-action
+- diffusion-language-model
+- tpu
+size_categories:
+- n<1K
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: train-*.parquet
+---
+
+# WOD-E2E Fast-dDrive SASD — TPU-ready training shards
+
+Pre-tokenized **Section-Aware Structured Diffusion (SASD)** training samples for the
+Fast-dDrive Qwen2.5-VL-3B block-diffusion driving model, packaged as sharded Apache
+Parquet for **multi-host TPU training** (grain / `datasets` / MaxText `hf` data path).
+
+Each row is one Waymo Open Dataset End-to-End (WOD-E2E) front-camera frame, already run
+through the Fast-dDrive prep (chat template + Qwen2.5-VL processor + deep-JSON scaffold +
+3D M-RoPE positions + per-section block indices), so the TPU side needs **no tokenizer,
+no processor, no torch** — it just loads arrays and trains.
+
+> ⚠️ **PRIVATE / license-restricted.** This data is derived from the Waymo Open Dataset and
+> is shared **privately** under the [WOD License](https://waymo.com/open/terms/), which
+> prohibits redistribution. Do not make this dataset public or share it outside your
+> licensed use.
+
+## Honest limitations (read before using)
+- **Pseudo text labels.** WOD-E2E ships **no native text targets**. `trajectory` and
+  `future_meta_behavior` derive from real GT / ego-intent; `critical_objects` and
+  `explanation` are **heuristic pseudo-labels**. Train/eval accordingly.
+- **Subset.** This is a ~400-frame representative subset. The full 263-shard WOD-E2E train
+  split runs through the identical 3-stage pipeline (see "Provenance").
+- **Not TPU-verified by the author.** The packaging + decode is bit-exact verified on CPU/GPU;
+  end-to-end TPU training is the intended downstream use, validated on emulation + a single GPU.
+
+## Schema (per row)
+Scalars: `sample_id (str)`, `L (int32)`, `n_blocks (int32)`.
+Arrays — each a little-endian binary blob + a `<name>_shape` int list:
+
+| column | dtype | shape | meaning |
+|---|---|---|---|
+| `input_ids` | int64 | [L] | token ids (mask-padded to multiple of bd_size=32) |
+| `labels` | int64 | [L] | -100 except assistant-response tokens |
+| `rbi` | int32 | [L] | response-block index per token (-1 = none) |
+| `turn` | int32 | [L] | turn index for the hybrid block-causal mask |
+| `scaffold` | bool | [L] | fixed JSON-scaffold positions (never noised) |
+| `weight_vec` | float32 | [L] | per-token section-importance weight |
+| `block_alpha` / `block_beta` | float32 | [n_blocks] | per-block Beta noise schedule |
+| `position_ids` | int32 | [3, L] | 3D M-RoPE positions |
+| `vision_mask` | bool | [L] | image/vision-start token positions |
+| `pixel_values` | float16 | [N, 1176] | Qwen2.5-VL patch features (3 front cams) |
+| `image_grid_thw` | int64 | [n_img, 3] | per-image temporal/height/width grid |
+
+## Reconstruct (numpy, framework-free)
+```python
+import numpy as np, pyarrow.parquet as pq
+ARRAY_DTYPES = {  # see dataset_info_train.json
+  "input_ids":"int64","labels":"int64","rbi":"int32","turn":"int32","scaffold":"bool",
+  "weight_vec":"float32","block_alpha":"float32","block_beta":"float32",
+  "position_ids":"int32","vision_mask":"bool","pixel_values":"float16","image_grid_thw":"int64"}
+row = pq.read_table("train-00000-of-00007.parquet").to_pylist()[0]
+def dec(name): return np.frombuffer(row[name], np.dtype(ARRAY_DTYPES[name])).reshape(row[name+"_shape"])
+input_ids = dec("input_ids"); pixel_values = dec("pixel_values")
+```
+See `load_example.py` for a runnable version.
+
+## Provenance (3-stage pipeline, reproducible at full scale)
+1. `fast_ddrive/data/convert_wod_e2e.py` (autovla env): tfrecord → train JSON + JPEGs (`--with_target`).
+2. `jax_ddrive/eval/prep_train_jax.py` (ddrive env): JSON+JPEGs → per-sample SASD npz.
+3. `jax_ddrive/ddrive_jax/convert/prep_to_parquet.py` → these Parquet shards.
+
+Model: `Efficient-Large-Model/Fast-dDrive` (Qwen2.5-VL-3B block-diffusion VLA).

dataset_info_train.json

@@ -0,0 +1,32 @@
+{
+  "split": "train",
+  "num_samples": 400,
+  "num_shards": 7,
+  "shard_size": 64,
+  "array_dtypes": {
+    "input_ids": "int64",
+    "labels": "int64",
+    "rbi": "int32",
+    "turn": "int32",
+    "scaffold": "bool",
+    "weight_vec": "float32",
+    "block_alpha": "float32",
+    "block_beta": "float32",
+    "position_ids": "int32",
+    "vision_mask": "bool",
+    "pixel_values": "float16",
+    "image_grid_thw": "int64"
+  },
+  "shape_suffix": "_shape",
+  "byte_order": "little",
+  "reconstruct": "np.frombuffer(row[name], dtype=ARRAY_DTYPES[name]).reshape(row[name+'_shape'])",
+  "files": [
+    "train-00000-of-00007.parquet",
+    "train-00001-of-00007.parquet",
+    "train-00002-of-00007.parquet",
+    "train-00003-of-00007.parquet",
+    "train-00004-of-00007.parquet",
+    "train-00005-of-00007.parquet",
+    "train-00006-of-00007.parquet"
+  ]
+}

load_example.py

@@ -0,0 +1,26 @@
+"""Runnable example: load the WOD-E2E Fast-dDrive SASD Parquet shards and reconstruct one
+sample, framework-free (numpy + pyarrow only). Works on a TPU host, GPU box, or CPU.
+
+  python load_example.py            # reads local ./train-*.parquet
+  python load_example.py <dir>      # reads <dir>/train-*.parquet
+"""
+import glob, json, os, sys
+import numpy as np
+import pyarrow.parquet as pq
+
+d = sys.argv[1] if len(sys.argv) > 1 else "."
+info = json.load(open(os.path.join(d, "dataset_info_train.json")))
+ADT = info["array_dtypes"]
+
+paths = sorted(glob.glob(os.path.join(d, "train-*.parquet")))
+row = pq.read_table(paths[0]).to_pylist()[0]
+
+def dec(name):
+    return np.frombuffer(row[name], np.dtype(ADT[name])).reshape(tuple(row[name + "_shape"]))
+
+print(f"shards={len(paths)}  num_samples={info['num_samples']}  sample_id={row['sample_id']}")
+print(f"L={row['L']}  n_blocks={row['n_blocks']}")
+for name in ADT:
+    a = dec(name)
+    print(f"  {name:16s} {str(a.shape):14s} {a.dtype}")
+print("input_ids[:12] =", dec("input_ids")[:12].tolist())

train-00000-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd406d05a1296b251c79c14b1d58464b2c59d8b4cd27332ca2d287aba53951b2
+size 30241610

train-00001-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0594e5dd26b09e7cceaf9d9754daaafb23ef5eabb01b300fce1a2f5bf9e7fb68
+size 28643718

train-00002-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f45d1951e398072b13254749b19a8349678d728cb0fc3d90745fc59fa0b48a67
+size 28514041

train-00003-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b806f980f7bcea3d2cc7b3fae745324b2a34b504dcfaed9479316e17ca6da4da
+size 28859901

train-00004-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecab0eb5739c8e7cbdae4f2ce5419a05948ecb2437d8ffb1e205cc3d2c68b98f
+size 30087414

train-00005-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cfb38fba68cb3e0877b1edbf7e101d72045dc9f537b0baf228d5d3659337b1d
+size 29171254

train-00006-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7f73f7f7a611dee2d1c2cf4ad5c37e1d1dc4401523a28cf192506fbc167545a
+size 7620718