Instructions to use axteam/Qwen3.5-2B-FP8-dynamic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use axteam/Qwen3.5-2B-FP8-dynamic with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="axteam/Qwen3.5-2B-FP8-dynamic")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("axteam/Qwen3.5-2B-FP8-dynamic")
model = AutoModelForMultimodalLM.from_pretrained("axteam/Qwen3.5-2B-FP8-dynamic", 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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use axteam/Qwen3.5-2B-FP8-dynamic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "axteam/Qwen3.5-2B-FP8-dynamic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "axteam/Qwen3.5-2B-FP8-dynamic",
		"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/axteam/Qwen3.5-2B-FP8-dynamic

SGLang

How to use axteam/Qwen3.5-2B-FP8-dynamic 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 "axteam/Qwen3.5-2B-FP8-dynamic" \
    --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": "axteam/Qwen3.5-2B-FP8-dynamic",
		"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 "axteam/Qwen3.5-2B-FP8-dynamic" \
        --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": "axteam/Qwen3.5-2B-FP8-dynamic",
		"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 axteam/Qwen3.5-2B-FP8-dynamic with Docker Model Runner:
```
docker model run hf.co/axteam/Qwen3.5-2B-FP8-dynamic
```

Qwen3.5-2B · FP8 (dynamic)

8-bit quantization of Qwen/Qwen3.5-2B, produced for efficient vLLM serving. Quantized by axteam.


Base model	`Qwen/Qwen3.5-2B`
Method	FP8 (dynamic)
Format	compressed-tensors `float-quantized` (scheme `FP8_DYNAMIC`)
Weights / activations	W8A8 — FP8 E4M3 weights (per-output-channel scale), dynamic per-token FP8 activations
Architecture	Hybrid (linear-attention + full-attention) · Dense · multimodal (vision) · MTP head
Size on disk	~3.2 GB
Calibration	None (data-free static weight scales)
License	Apache-2.0

Quantization details

Scheme — W8A8 — FP8 E4M3 weights (per-output-channel scale), dynamic per-token FP8 activations, stored as compressed-tensors float-quantized (scheme FP8_DYNAMIC).
Kept in BF16 — the vision tower & projector (model.visual.*), the multi-token-prediction head (mtp.*), all RMSNorms, token embeddings, the LM head, the SSM non-linear parts (linear_attn.conv1d/A_log/dt_bias). Left unquantized to preserve quality (about 40% of all parameters in this build).
Tokenizer / chat_template / generation_config are identical to the original. config.json is the original plus a quantization_config block.

This is post-training quantization (PTQ) — no retraining. FP8 dynamic is calibration-free.

Benchmarks (measured)

Every column was measured with the same harness and settings — lm-evaluation-harness on vLLM, greedy, few-shot completion (no chat template). Percentages are recovery vs. the BF16 original.

Benchmark	BF16 (original)	FP8-dynamic (this repo)	AWQ-8bit	NVFP4A16	NVFP4	AWQ-4bit
MMLU (5-shot)	53.33	48.18 (90.3%)	53.54 (100.4%)	54.27 (101.8%)	55.22 (103.5%)	60.38 (113.2%)
GSM8K (5-shot, strict)	58.23	56.63 (97.3%)	56.56 (97.1%)	54.51 (93.6%)	51.71 (88.8%)	54.06 (92.8%)
ARC-Challenge (0-shot)	41.47	41.98 (101.2%)	41.30 (99.6%)	41.04 (99.0%)	39.51 (95.3%)	39.59 (95.5%)
Winogrande (5-shot)	64.56	63.22 (97.9%)	64.48 (99.9%)	64.48 (99.9%)	61.09 (94.6%)	62.51 (96.8%)
KMMLU (5-shot, direct)	43.29	42.58 (98.4%)	43.20 (99.8%)	42.01 (97.1%)	41.14 (95.0%)	42.10 (97.3%)
HAE-RAE (0-shot)	45.65	45.00 (98.6%)	45.10 (98.8%)	44.91 (98.4%)	42.07 (92.2%)	41.06 (90.0%)
Average	51.09	49.60 (97.1%)	50.70 (99.2%)	50.21 (98.3%)	48.46 (94.8%)	49.95 (97.8%)

Evaluation (measured)

Fidelity to the BF16 original on an internal mixed English/Korean set (100 prompts, BF16 greedy answers scored under each build — same harness for every row).

Metric	This model	BF16 original
Perplexity (lower = closer)	1.8710	1.8515 (+1.1%)
Top-1 next-token agreement	94.4%	98.9%

These measure fidelity to the base model, not absolute task scores.

Requirements


Runtime	vLLM — the architecture is in vLLM ≥ 0.17.0; tested here with vLLM 0.32.0.
vLLM install	vLLM 0.32 itself needs Python 3.11–3.14, torch 2.13 (CUDA 13.0 build, so an NVIDIA driver with CUDA 13 support), transformers ≥5.16.1,<5.20 and compressed-tensors 0.17.0 (both pinned by vLLM; 0.17.0 reads this checkpoint, which was saved with 0.19.0 — do not upgrade it separately).
GPU	NVIDIA Turing (sm75, e.g. T4) or newer. FP8 math runs on Ada / Hopper / Blackwell (sm89+); on Turing and Ampere vLLM switches to weight-only FP8 automatically. V100 (sm70) is not supported. (Kernel choice per GPU from vLLM 0.32's selection code; only B200 was run here.)
Memory	Weights: 3.2 GB. Add KV cache (grows with context length × concurrent requests) and a few GB for activations / CUDA graphs; use `--tensor-parallel-size N` when the weights do not fit one GPU.
Transformers (without vLLM)	`transformers>=5.2.0` + `compressed-tensors>=0.15.0` (checked on CPU with Qwen3.5-0.8B FP8 / AWQ-4bit / NVFP4A16 / NVFP4).
trust_remote_code	not needed

Usage (vLLM)

Start an OpenAI-compatible server:

vllm serve axteam/Qwen3.5-2B-FP8-dynamic --max-model-len 32768

Query it:

curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
  "model": "axteam/Qwen3.5-2B-FP8-dynamic",
  "messages": [{"role": "user", "content": "대한민국의 수도는 어디인가요?"}],
  "max_tokens": 512
}'

Offline inference in Python:

from vllm import LLM, SamplingParams

llm = LLM(model="axteam/Qwen3.5-2B-FP8-dynamic", max_model_len=32768)
params = SamplingParams(max_tokens=1024)
out = llm.chat([{"role": "user", "content": "Explain model quantization in two sentences."}], params)
print(out[0].outputs[0].text)

# image input (URL or data: URI)
messages = [{"role": "user", "content": [
    {"type": "image_url", "image_url": {"url": "https://example.com/your-image.png"}},
    {"type": "text", "text": "Describe this image."},
]}]
out = llm.chat(messages, params)
print(out[0].outputs[0].text)

Thinking mode is off by default in this model's chat template; to turn it on pass chat_template_kwargs={"enable_thinking": True} (Python llm.chat(..., chat_template_kwargs=...), or "chat_template_kwargs": {"enable_thinking": true} in the request body). Text-only serving: add --limit-mm-per-prompt '{"image": 0, "video": 0}' to skip loading the vision encoder's memory budget.

Reducing KV-cache memory (optional, serve-time)

The KV cache is kept in BF16 by default. KV-cache precision is a runtime choice, independent of the weight quantization above; for long contexts or large batches FP8 roughly halves KV-cache memory:

vllm serve axteam/Qwen3.5-2B-FP8-dynamic --kv-cache-dtype fp8

License

Quantized derivative of Qwen/Qwen3.5-2B (Apache-2.0). All original weights, tokenizer and chat template belong to the Qwen team.

한국어 요약

Qwen/Qwen3.5-2B 를 FP8 (dynamic) 로 양자화한 서빙용(8-bit) 모델입니다. 작성: axteam.

방식 — FP8 E4M3. 가중치는 출력 채널마다 스케일 하나(정적·보정 불필요), 활성화는 추론 시 토큰마다 동적 FP8. 형식 compressed-tensors float-quantized (scheme FP8_DYNAMIC).
BF16 유지 — 비전 타워·MTP 헤드·모든 norm·임베딩/LM 헤드·SSM 비선형부(conv1d/A_log/dt_bias).
PTQ — 재학습 없음.
품질(실측) — BF16 원본 대비 perplexity 1.8710(원본 1.8515, +1.1%), top-1 일치도 94.4%(원본 98.9%). 내부 영/한 100문항, 같은 하네스.
벤치마크(실측) — MMLU · GSM8K · ARC-Challenge · Winogrande · KMMLU · HAE-RAE 평균 BF16 51.09 → 이 모델 49.60 (회복률 97.1%). lm-evaluation-harness · vLLM, 모든 형식 같은 설정.
라이선스 — 원본 Apache-2.0 파생물. 가중치·토크나이저·chat_template 권리는 Qwen 팀.
요구사항 — vLLM ≥0.17.0 (0.32.0 에서 실측); GPU Turing(sm75) 이상. FP8 연산은 Ada·Hopper·Blackwell(sm89+), 그 아래는 vLLM 이 가중치 전용 FP8 로 자동 전환; 가중치 3.2 GB + KV 캐시; Transformers 직접 사용: transformers ≥5.2.0 + compressed-tensors ≥0.15.0.
실행 — vllm serve axteam/Qwen3.5-2B-FP8-dynamic --max-model-len 32768 로 OpenAI 호환 서버(포트 8000), 또는 Python LLM(model="axteam/Qwen3.5-2B-FP8-dynamic").chat(...) (위 예시). thinking 은 기본으로 꺼져 있음 — 켜려면 chat_template_kwargs={"enable_thinking": True}. 이미지 입력은 image_url 메시지로.
KV 캐시 절감(런타임) — --kv-cache-dtype fp8.