xiaorui638

qwen3vl-8b-s85k-plain-grpo

README

License: apache-2.0

Results — ZoomBench/TreeBench MCQ (in-domain)

2,343 questions = 584 positives + 1,759 negative/positive twin pairs. vLLM, seed 42, T=0.7, bf16; judge = rule-match then Qwen3-30B-A3B-Instruct. Single seed.

Paired accuracy — negative and its positive twin both correct

Table with columns: category, pairs, Qwen3-VL-8B base, this model
category	pairs	Qwen3-VL-8B base	this model
needle_obj	445	20.00	54.83
needle_attr	393	12.21	48.35
ref_1_obj	269	12.27	46.84
ref_1_attr	226	6.19	39.38
ref_1_rel	202	3.47	24.26
ref_2_obj	84	5.95	42.86
ref_2_attr	74	1.35	31.08
ref_2_rel	66	0.00	13.64
micro (all pairs)	1759	11.20	43.55

Per-question accuracy

Table with columns: category, n, Qwen3-VL-8B base, this model
category	n	Qwen3-VL-8B base	this model
positive	584	44.52	58.56
needle_obj	445	53.71	92.13
needle_attr	393	27.74	87.28
ref_1_obj	269	33.09

Read the paired column, not the overall. Per-question accuracy on this benchmark rewards denial: a model that denies every premise scores well on the 1,759 negatives while failing all 584 positives. The positive half is this model's actual bottleneck — positive accuracy moves only 44.52 → 58.56, and most remaining paired error is a wrong positive twin, not a missed denial. Relational negatives (ref_*_rel) stay hard.

This benchmark is in-domain — built by the same pipeline as the training data — so it measures how well the target behaviour was learned, not general ability transfer.

Usage

Weights are bfloat16.

python
from transformers import AutoProcessor, Qwen3VLForConditionalGeneration

model = Qwen3VLForConditionalGeneration.from_pretrained(REPO_ID, dtype="auto", device_map="auto")
processor = AutoProcessor.from_pretrained(REPO_ID)

vLLM:

bash
vllm serve REPO_ID --dtype bfloat16 --limit-mm-per-prompt image=1

The model was trained with a think-then-answer format prompt; it emits reasoning followed by the answer in <answer>...</answer> tags.

Training

Table

Base	`Qwen/Qwen3-VL-8B-Instruct`
Algorithm	GRPO (KL disabled), rollout n=8 @ T=1.0
Reward	accuracy only (letter/string match, else LLM judge)
Data	85,000 false-premise twin examples, 2:1 pos:neg
Batch	rollout 384 / global 96, lr 1e-6, 1 epoch
Checkpoint	global step 221 (final)
Framework	EasyR1 (veRL), 24× GH200

Limitations

Single seed; no variance estimate. Treat <1 pt differences as noise.
Evaluated in-domain here. It trades some general MCQ ability for premise robustness; consult the fuller evaluation before using it as a general-purpose VLM.
Inherits all limitations of the base model.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

xiaorui638

Model Tree

Base

Qwen/Qwen3-VL-8B-Instruct

Fine-tuned

this model

Input Modalities

TextImage

Output Modalities

Text

Supported Functionality

Dedicated EndpointsContainer

Explore FriendliAI today

Get started Talk to an engineer