---
license: apache-2.0
base_model: Qwen/Qwen3-VL-2B-Thinking
library_name: peft
tags:
  - vision-language
  - new-yorker
  - humor
  - rlhf
  - grpo-thinking
datasets:
  - yguooo/newyorker_caption_ranking
language:
  - en
---

# humor-r1 — GRPO, with thinking (Qwen3-VL-2B-Thinking + LoRA) (E2b)

LoRA on Qwen3-VL-2B-Thinking trained via GRPO against the Bradley-Terry reward model HumorR1/rm-qwen25vl-3b-nodesc. Output format: `{thinking}</think>\n\n<caption>X</caption>`.

## Training data

- 271 New Yorker contests, top-rated caption per contest
  (`yguooo/newyorker_caption_ranking`).
- The 60k Bradley-Terry preference pairs underlying the reward model
  (separate split).
- We deliberately do NOT use the dataset's GPT-4o-generated
  Scene/Twist/Location/Entities descriptions in the prompt, since they
  hand-feed scene content to a vision-language model that can already
  see the image; this makes the policy and reward model usable on any
  single-panel cartoon, not just the curated subset.

## How it fits the project

Part of a 2x2 ablation over training method (SFT, GRPO) and output
format (no thinking, thinking) for humor caption generation. See
`HumorR1/rm-qwen25vl-3b-nodesc` for the reward model used to train (and
score) this policy.

## Inference

Backbone: `Qwen/Qwen3-VL-2B-Thinking`.
This repo is a LoRA adapter; load with `peft.PeftModel.from_pretrained`.

```python
from PIL import Image
from transformers import AutoProcessor
from vllm import LLM, SamplingParams
from vllm.lora.request import LoRARequest

processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-2B-Thinking", trust_remote_code=True)
llm = LLM(model="Qwen/Qwen3-VL-2B-Thinking", trust_remote_code=True, dtype="bfloat16",
          enable_lora=True, max_lora_rank=32, max_model_len=4096)

# Caption format: <caption>X</caption>; thinking variant prefixes <think>...</think>.
```

## Reward model used during training

- `HumorR1/rm-qwen25vl-3b-nodesc` (held-out pairwise accuracy 0.6635).