Text Generation
Transformers
Safetensors
Japanese
English
qwen3_moe
qwen3
235b
lora
merged
reasoning
cot
japanese
english
conversational
Instructions to use weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4") model = AutoModelForCausalLM.from_pretrained("weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4
- SGLang
How to use weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4 with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4 with Docker Model Runner:
docker model run hf.co/weblab-llm-competition-2025-bridge/oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: Qwen/Qwen3-235B-A22B-Thinking-2507
|
| 4 |
+
tags:
|
| 5 |
+
- qwen3
|
| 6 |
+
- 235b
|
| 7 |
+
- lora
|
| 8 |
+
- merged
|
| 9 |
+
- reasoning
|
| 10 |
+
- cot
|
| 11 |
+
- japanese
|
| 12 |
+
- english
|
| 13 |
+
language:
|
| 14 |
+
- ja
|
| 15 |
+
- en
|
| 16 |
+
library_name: transformers
|
| 17 |
+
pipeline_tag: text-generation
|
| 18 |
+
model_type: qwen
|
| 19 |
+
datasets:
|
| 20 |
+
- oNo-1/difficult_problem_dataset_v4_500
|
| 21 |
+
---
|
| 22 |
+
|
| 23 |
+
# weblab-llm-competition-2025-bridge / oNo-1-Qwen3-235B-A22B-Thinking-difficult-problem-dataset-v4
|
| 24 |
+
|
| 25 |
+
## 概要
|
| 26 |
+
|
| 27 |
+
Qwen3-235B-A22B-Thinking-2507 をベースとして、最後層の self-attention(`q_proj` / `k_proj` / `v_proj` / `o_proj`)に限定して LoRA 学習を行い、その差分を CPU 並列でベース重みにマージ(merge)たモデルです。学習には oNo-1 による difficult_problem_dataset_v4 の 500 件サブセットを用い、SFT(ChatML)と KV 自己蒸留を適用しました。本モデルはベース同等サイズの重みを持ち、実運用では分散推論(FSDP/TP 等)を前提とします。
|
| 28 |
+
|
| 29 |
+
---
|
| 30 |
+
|
| 31 |
+
## 推論方法
|
| 32 |
+
|
| 33 |
+
詳細手順は以下の外部ドキュメントを参照してください。
|
| 34 |
+
|
| 35 |
+
Notion(推論手順): [https://www.notion.so/277e14b94af5809a88f5e7a89c707bcb?source=copy_link](https://www.notion.so/277e14b94af5809a88f5e7a89c707bcb?source=copy_link)
|
| 36 |
+
|
| 37 |
+
---
|
| 38 |
+
|
| 39 |
+
## 学習レシピ
|
| 40 |
+
|
| 41 |
+
* 方式: QLoRA(4bit NF4 / BF16 計算)
|
| 42 |
+
* 対象モジュール: `q_proj, k_proj, v_proj, o_proj`
|
| 43 |
+
* 主なハイパーパラメータ: `lora_r=8`, `lora_alpha=16`, `lora_dropout=0.1`, `lr=1e-4`, `epochs=1`
|
| 44 |
+
* その他: `bf16=True`, `gradient_checkpointing=True`, `group_by_length=True`, `lazy_preprocess=True`
|
| 45 |
+
* KV 自己蒸留: `--kv_sd --kv_sd_alpha 0.9`
|
| 46 |
+
|
| 47 |
+
---
|
| 48 |
+
|
| 49 |
+
## データセット
|
| 50 |
+
|
| 51 |
+
* 名称: `oNo-1/difficult_problem_dataset_v4_500`(difficult_problem_dataset_v4 の 500 件サブセット)
|
| 52 |
+
* 形式: `messages` に `<think>…</think>` を含む思考テキストと最終回答を格納
|
| 53 |
+
* 作成プロセス(要約): 多段生成 → キュレーション/多様性フィルタ → 進化的改良 → 推論過程付与(SDG)
|
| 54 |
+
* ライセンス: ODC-BY 1.0(データベース著作物の表示義務)
|
| 55 |
+
|
| 56 |
+
---
|
| 57 |
+
|
| 58 |
+
## ライセンスと表記
|
| 59 |
+
|
| 60 |
+
* モデル配布ライセンス: Apache-2.0
|
| 61 |
+
本モデルは `Qwen/Qwen3-235B-A22B-Thinking-2507`(Apache-2.0)をベースに、最後層の一部を LoRA 微調整・マージした改変物です。再配布時は Apache-2.0 の条件(著作権表示・LICENSE 同梱等)に従ってください。
|
| 62 |
+
* 学習データライセンス: ODC-BY 1.0
|
| 63 |
+
本モデルの学習には ODC-BY 1.0 に基づくデータベースを利用しています。再配布時はデータ提供者への適切な表示(Attribution)を README 等に明記してください。
|
| 64 |
+
|
| 65 |
+
## 制限事項
|
| 66 |
+
|
| 67 |
+
* 235B クラスの規模のため、単一 GPU での実用推論は困難です。分散推論(FSDP/TP 等)を推奨します。
|
| 68 |
+
* 公開・配布時は、モデル(Apache-2.0)とデータセット(ODC-BY 1.0)の両条件を順守してください。
|