Image-Text-to-Text
Transformers
Safetensors
Korean
English
qwen3_5
vocabulary-pruning
korean
warmly
conversational
Instructions to use neureps/Qwen3.5-0.8B-enko with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use neureps/Qwen3.5-0.8B-enko with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="neureps/Qwen3.5-0.8B-enko") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("neureps/Qwen3.5-0.8B-enko") model = AutoModelForMultimodalLM.from_pretrained("neureps/Qwen3.5-0.8B-enko", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use neureps/Qwen3.5-0.8B-enko with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "neureps/Qwen3.5-0.8B-enko" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "neureps/Qwen3.5-0.8B-enko", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/neureps/Qwen3.5-0.8B-enko
- SGLang
How to use neureps/Qwen3.5-0.8B-enko 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 "neureps/Qwen3.5-0.8B-enko" \ --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": "neureps/Qwen3.5-0.8B-enko", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'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 "neureps/Qwen3.5-0.8B-enko" \ --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": "neureps/Qwen3.5-0.8B-enko", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use neureps/Qwen3.5-0.8B-enko with Docker Model Runner:
docker model run hf.co/neureps/Qwen3.5-0.8B-enko
Qwen3.5-0.8B-enko — EN/KO 무손실 어휘 프루닝 마스터
Qwen/Qwen3.5-0.8B의 어휘를 영어·한국어·이모지·기호로 축소한 무손실 프루닝 마스터 체크포인트입니다. Warmly(온기) 앱에서 이미지 캡셔너(비전) 및 저사양 기기 폴백 후보로 사용합니다.
프루닝 내용 (2026-07-13, 파일럿 — 전 항목 PASS)
- vocab 248,320 → ~148k (2B와 동일 keepset —
pruning_keepset.json) - 파라미터 0.873B → 0.771B (−11.7%), 파일 1.75GB → 1.55GB
- 비전 타워(~91M)·MTP 헤드 유지
- 무손실 검증: 로짓 diff 0.00e+00, greedy 생성 동일, NLL 3.5987 → 3.5905, EN/KO/이모지/코드 토큰화 동일
역할과 실측 근거
- 캡셔너 3파전 승자 (2026-07-13): 0.8B 비전이 2B q4_0의 환각(두부)을 내지 않음 — 작은 캡셔너가 더 충실. SmolVLM-500M은 빈약해 기각.
- 한국어 댓글 직접 생성은 기각(8~9/20): 스타일이 아니라 한국어 의미 접지 부족이 원인. 영어 댓글은 ~16/20으로 강함 → "0.8B(영어) + NMT 번역" B안 트랙의 전제.
사용 시 주의
2B 마스터(neureps/Qwen3.5-2B-enko)와 동일: GGUF 변환 시 --no-mtp + chkhsh 등록(res = "qwen35", 해시는 2B 리포 카드 참조), 추론 시 enable_thinking: false 필수.
관련 리포
- 배포 GGUF: neureps/warmly-qwen35-08b-enko-gguf
- Downloads last month
- 31