---
license: apache-2.0
base_model: unsloth/gemma-4-26B-A4B
library_name: transformers
pipeline_tag: text-generation
language:
- ja
tags:
- quiz
- buzz-quiz
- hayaoshi
- japanese
- reasoning
- gemma4
- unsloth
---

# quiz-main-gemma-merged ⚡ — 早押しクイズ 回答モデル (Answering model)

The **answering model** of a two-model Japanese competitive **buzz-quiz** (早押しクイズ) system.
Given a *partial* question (the prefix read so far at buzz time), it reasons inside
`<think>…</think>` and emits a short answer.

- 🕹️ **Live demo (HF Space):** https://huggingface.co/spaces/build-small-hackathon/quiz-buzzer-ai
- 💻 **Code (GitHub):** https://github.com/YUGOROU/quiz-ai
- 🔔 **Buzz-timing companion model:** [`YUGOROU/quiz-buzz-reg-1.2bjp-merged`](https://huggingface.co/YUGOROU/quiz-buzz-reg-1.2bjp-merged)

## Role in the system

| | Model | Job |
|---|---|---|
| 🔔 Buzz | `YUGOROU/quiz-buzz-reg-1.2bjp-merged` (LFM2.5-1.2B + regression head) | Reads the question char-by-char, buzzes when `conf ≥ θ` (~9 ms/char). |
| 🧠 **Answer (this model)** | **gemma-4-26B-A4B SFT** | From the partial question at buzz time, `<think>…</think>` reasoning → answer. |

Total ≈ 27.2B params (≤ 32B), built for the **HF Build Small Hackathon**.

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

repo = "YUGOROU/quiz-main-gemma-merged"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype=torch.bfloat16, device_map="auto")

prefix = "日本の首都は東京ですが、アメリカの首都は"  # partial question at buzz time
msgs = [{"role": "user", "content": f"早押しクイズ（{len(prefix)}文字目時点）:\n{prefix}"}]
ids = tok.apply_chat_template(msgs, enable_thinking=True, add_generation_prompt=True, return_tensors="pt").to(model.device)

out = model.generate(
    ids,
    max_new_tokens=320,
    do_sample=False,
    eos_token_id=[1, 106],   # gemma-4 closes the turn with <turn|>=106, not only <eos>=1
)
print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))
# <think> … </think>ワシントンD.C.
```

> **Important:** gemma-4 ends an assistant turn with `<turn|>` (id **106**). If you only stop on
> `<eos>` (id 1) the model will keep hallucinating new turns. Always include **106** in your stop set
> (vLLM: `--stop-token-ids 1 106`). `<think>` reasoning is **required** — disabling it collapses accuracy.

## Training

- Base: `unsloth/gemma-4-26B-A4B` (MoE, 26B total / 4B active), `gemma-4-thinking` chat template.
- SFT (Unsloth bf16 LoRA, merged to 16-bit) on a quiz-grammar corpus built from AI王 / JAQKET:
  user = partial question at the statistically-decidable buzz position (S-buzz), assistant =
  `<think>{reasoning}</think>{answer}` with adaptive think budget by difficulty.
- Full-question QA ≈ **76%**; at the buzz position ≈ **62–74%** depending on threshold θ (later buzz → higher accuracy).

## Attribution & license

This model is a fine-tune of Google **Gemma 4**, which Google releases under the
[Apache License 2.0](https://ai.google.dev/gemma/apache_2.md.txt). The model weights are therefore
distributed under Apache 2.0.

Training data derived from **AI王 (Project AIO) / JAQKET**. Quiz questions ©
abc/EQIDEN実行委員会 / 株式会社キュービック / クイズ法人カプリティオ.
**Non-commercial research use only. No dataset redistribution** — only model weights and inference
code are released.