GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
FINWHALE
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
FINWHALE
Model Tree
unsloth/Qwen2.5-VL-7B-Instruct
Input Modalities
Output Modalities
Supported Functionality
This repository currently hosts Wukong-T2-Adapter: a QLoRA adapter (rank 16, 51.5M trainable params, 206MB) on top of unsloth/Qwen2.5-VL-7B-Instruct, fine-tuned for visual-spatial question answering from egocentric images.
adapter_config.json + adapter_model.safetensors — the trained adapter (final checkpoint, step 49730)eval_results.jsonl — per-item evaluation records (question, reference, prediction, score) for full reproducibilityTraining setup: 198,919 samples (195,919 VSI-style spatial QA derived from Hypersim + 3,000 programmatic egocentric↔allocentric coordinate-transform QA), 2 epochs, total batch size 8, ~18.5 hours on a single RTX 5090 with Unsloth. Final training loss ≈ 0.19.
Roadmap stages beyond T2 (semantic waypoint prediction, embodiment-specific action adapters, world-action predictive pathway) are under active development and will be released as separate artifacts.
Held-out spatial QA split from our own data pipeline (never seen in training, n=500), scored in VSI-Bench style: exact match for multiple-choice items; MRA (max(0, 1 − |pred−ref|/ref)) for numeric items.
| Question type | n | Score |
|---|---|---|
| relative_direction_camera | 57 | 0.877 |
| relative_distance_object | 130 | 0.846 |
| relative_size_object | 23 | 0.826 |
| relative_direction_object | 108 | 0.815 |
| absolute_count | 38 | 0.805 |
| absolute_size_object | 27 |
Breakdown by answer format: multiple-choice (n=345) accuracy 0.832; numeric (n=155) MRA 0.671.
| Benchmark | What it measures | Status |
|---|---|---|
| VSI-style spatial QA (own held-out split) | Visual-spatial intelligence from egocentric images | 0.782 (see above) |
| VSI-Bench (official) | Visual-spatial intelligence from egocentric video | Planned |
| SPAR-Bench | 3D spatial perception and reasoning | Planned |
| R2R-CE / RxR-CE (VLN-CE, Habitat) | Instruction-following navigation (SR / SPL) | Planned |
| Real-robot trials (Unitree) | Sim-to-real transfer | Planned |
Note: the reported 0.782 is measured on our own held-out split, not the official VSI-Bench; official-benchmark evaluation is on the roadmap.
python
from unsloth import FastVisionModelmodel, tokenizer = FastVisionModel.from_pretrained("unsloth/Qwen2.5-VL-7B-Instruct",load_in_4bit=True,max_seq_length=4096,)model.load_adapter("FINWHALE/Wukong") # adapter files at repo rootFastVisionModel.for_inference(model)
A FastAPI serving wrapper and an evaluation script are available in the companion training pipeline (to be linked from this card).
bibtex
@misc{wukong2026,title = {Wukong: An Open Spatial-Reasoning Vision-Language-Action Model},author = {FINWHALE},year = {2026},url = {https://huggingface.co/FINWHALE/Wukong}}
Apache 2.0. Please also comply with the licenses of the underlying base models and datasets.
| 0.758 |
| relative_distance_camera | 4 | 0.750 |
| relative_count | 23 | 0.739 |
| absolute_distance_object | 56 | 0.610 |
| absolute_direction_object | 34 | 0.551 |
| TOTAL | 500 | 0.782 |