hzxbzp/test-models

AnchorReason

Checkpoints fine-tuned on AnchorReasoning, a visual-grounding and causal-reasoning annotation layer for long-tail driving built on WOD-E2E.

Each model takes one forward panorama + 4 s of ego motion history + a navigation intent and produces a visually grounded chain of thought: scene context, traffic events, the decision-critical elements with their image locations, attributes, impact ranks and per-element implications, then a rationale, an action plan, and a 5 s future trajectory.

Checkpoints

Folder	Backbone	Params	Size
qwen2.5-vl-7b/stage2	Qwen2.5-VL-7B-Instruct	8.29 B	16.6 GB
qwen2.5-vl-7b/traj_only	Qwen2.5-VL-7B-Instruct	8.29 B	16.6 GB
qwen3-vl-8b/stage2	Qwen3-VL-8B-Instruct	8.77 B	17.5 GB
qwen3-vl-8b/traj_only	Qwen3-VL-8B-Instruct	8.77 B	17.5 GB
cosmos-reason2-2b/stage2	Cosmos-Reason2-2B	2.13 B	4.3 GB
cosmos-reason2-2b/traj_only	Cosmos-Reason2-2B	2.13 B	4.3 GB
cosmos-reason2-8b/stage2	Cosmos-Reason2-8B	8.77 B	17.5 GB
cosmos-reason2-8b/traj_only	Cosmos-Reason2-8B	8.77 B	17.5 GB
autovla-3b/stage2	AutoVLA (Qwen2.5-VL-3B)	3.76 B	7.5 GB
autovla-3b/traj_only	AutoVLA (Qwen2.5-VL-3B)	3.76 B	7.5 GB
alpamayo-1.5-10b/stage2	Alpamayo-1.5-10B	8.80 B	17.6 GB
alpamayo-1.5-10b/traj_only	Alpamayo-1.5-10B	8.80 B	17.6 GB
alpamayo-r1-10b/stage2	Alpamayo-R1-10B	8.80 B	17.6 GB
alpamayo-r1-10b/traj_only	Alpamayo-R1-10B	8.80 B	17.6 GB
impromptu-vla-7b/stage2	Impromptu-VLA-7B (Qwen2.5-VL-7B)	8.29 B	16.6 GB
impromptu-vla-7b/traj_only	Impromptu-VLA-7B (Qwen2.5-VL-7B)	8.29 B	16.6 GB

Input format

Getting this exactly right matters: these are full fine-tunes on a fixed prompt, and a prompt that differs from the training one degrades the output badly.

Image. One 2916 x 1079 forward panorama, stitched from the FRONT_LEFT, FRONT and FRONT_RIGHT cameras of a WOD-E2E frame by reprojecting all three onto a single vehicle-aligned virtual pinhole. The dataset card carries the exact reconstruction code. Images are fed at full resolution (long edge 2916).

User turn.

History (past 4 s, 16 waypoints at 4 Hz, ego frame, meters): [x, y], [x, y], ... (16 points)
Intent: GO_STRAIGHT | GO_LEFT | GO_RIGHT
<instruction>

Chat scaffold (Qwen2 family, identical for both backbone families):

<|im_start|>system
{system}<|im_end|>
<|im_start|>user
<|vision_start|>{image_pads}<|vision_end|>{user}<|im_end|>
<|im_start|>assistant

Point coordinates

<point> values follow the backbone's own convention:

Family	Checkpoints	Convention
Qwen2.5-VL	qwen2.5-vl-7b, autovla-3b, impromptu-vla-7b	pixel coords of the panorama (processed image)
Qwen3-VL	qwen3-vl-8b, cosmos-reason2-*, alpamayo-1.5-10b, alpamayo-r1-10b	(x,y) normalized to 0-1000

Each folder ships a prompt_hash.json recording the exact prompt fingerprint the weights were trained with — check it if you rebuild the prompt yourself.

Output format

stage2 emits the whole chain in a fixed tag order; traj_only emits <traj> alone.

<ego_state>lon=... lat=...</ego_state>
<context>weather=...;daytime=...;visibility=...;scenario=...;road=...</context>
<events>...</events>
<has_objects>yes|no</has_objects>
<obj type=...><point>x,y</point><rank>1</rank>...<implication>...</implication></obj>
<n_objects>N</n_objects>
<reason>...</reason>
<final_plan>...</final_plan>
<motion>lon=... lat=...</motion>
<traj>[x1,y1],[x2,y2],[x3,y3],[x4,y4],[x5,y5]</traj>

Base models and licenses

Every checkpoint is a derivative of its backbone and stays bound by that backbone's license. They differ, so this repository carries no single license — check the row for the checkpoint you use.

Folder	Base model	Base license	Base repo gated?
qwen2.5-vl-7b/*	Qwen/Qwen2.5-VL-7B-Instruct	Apache-2.0	no
qwen3-vl-8b/*	Qwen/Qwen3-VL-8B-Instruct	Apache-2.0	no
cosmos-reason2-2b/*	nvidia/Cosmos-Reason2-2B	NVIDIA Open Model License	yes
cosmos-reason2-8b/*	nvidia/Cosmos-Reason2-8B	NVIDIA Open Model License	yes
alpamayo-1.5-10b/*	nvidia/Alpamayo-1.5-10B	OpenMDW-1.1	no
alpamayo-r1-10b/*	nvidia/Alpamayo-R1-10B	OpenMDW-1.1	no
autovla-3b/*	Zewei-Zhou/AutoVLA	none declared upstream	no
impromptu-vla-7b/*	aaaaaap/ImpromptuVLAModel (7B_AD)	none declared upstream	no

Gated backbones. The two Cosmos-Reason2 base models are distributed behind a license gate. Before using cosmos-reason2-2b/* or cosmos-reason2-8b/*, accept the NVIDIA Open Model License Agreement on the base model page — Cosmos-Reason2-2B or Cosmos-Reason2-8B — where access is granted automatically once you are signed in and have agreed to the terms. These derivative weights are offered on the same terms; the gate lives on the upstream repository, not here.

The AnchorReasoning annotations used to train these models are released under CC BY-NC 4.0, and the underlying imagery comes from WOD-E2E under the Waymo Open Dataset License. Treat the checkpoints as non-commercial research artifacts.

Intended use and limitations

These are research checkpoints for studying grounded reasoning and planning in long-tail driving. They are not driving software: they run on a single forward panorama with no rear or side view, no LiDAR, no HD map and no temporal context, they predict a 5 s open-loop trajectory rather than closing any control loop, and their reasoning text is generated, not verified. Do not deploy them on a vehicle.

Citation

The accompanying paper is under review; a citation will be added once it is public.

@misc{anchorreasoning,
  title  = {AnchorReasoning: A Visual Grounding and Causal Reasoning Dataset in Long-Tail Autonomous Driving Scenarios},
  year   = {2026},
  note   = {Under review}
}