Instructions to use hangxian/Qwen3.5-27B-FP8-PerChannel with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use hangxian/Qwen3.5-27B-FP8-PerChannel with Transformers:

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

pipe = pipeline("text-generation", model="hangxian/Qwen3.5-27B-FP8-PerChannel")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("hangxian/Qwen3.5-27B-FP8-PerChannel")
model = AutoModelForCausalLM.from_pretrained("hangxian/Qwen3.5-27B-FP8-PerChannel", 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 hangxian/Qwen3.5-27B-FP8-PerChannel with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hangxian/Qwen3.5-27B-FP8-PerChannel"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hangxian/Qwen3.5-27B-FP8-PerChannel",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/hangxian/Qwen3.5-27B-FP8-PerChannel

SGLang

How to use hangxian/Qwen3.5-27B-FP8-PerChannel 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 "hangxian/Qwen3.5-27B-FP8-PerChannel" \
    --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": "hangxian/Qwen3.5-27B-FP8-PerChannel",
		"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 "hangxian/Qwen3.5-27B-FP8-PerChannel" \
        --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": "hangxian/Qwen3.5-27B-FP8-PerChannel",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use hangxian/Qwen3.5-27B-FP8-PerChannel with Docker Model Runner:
```
docker model run hf.co/hangxian/Qwen3.5-27B-FP8-PerChannel
```

Qwen3.5-27B-FP8-PerChannel

⚠️ 重要说明

来源模型: Qwen/Qwen3.5-27B（官方 BF16 原版）
权重未做任何修改: 本模型是纯量化版本，未经过任何微调、蒸馏或去审查处理，权重数值与官方原版完全一致
量化方式: per-channel FP8（每列一个 scale）+ dynamic activation（每次推理在线量化激活）
量化工具: llmcompressor，scheme=FP8_DYNAMIC

量化配置

{
  "quant_method": "compressed-tensors",
  "config_groups": {
    "group_0": {
      "weights": {
        "strategy": "channel",
        "num_bits": 8,
        "type": "float",
        "dynamic": false
      },
      "input_activations": {
        "strategy": "token",
        "num_bits": 8,
        "type": "float",
        "dynamic": true
      }
    }
  }
}

与官方 FP8 版本的区别

	本模型	Qwen/Qwen3.5-27B-FP8（官方）
权重来源	官方原版，未修改	官方原版，未修改
权重量化粒度	per-channel	block 128×128
激活量化	dynamic	dynamic
SGLang GEMM 后端	`sgl_kernel.fp8_scaled_mm` (CUTLASS)	DeepGEMM / Triton
实测性能 (H20, 4k/4k, c=1)	~160 tok/s	~108 tok/s (DeepGEMM)

per-channel 版本在 decode 阶段（batch=1）比 block-wise 版本快约 50%，因为 per-channel 的 scale 处理更简单，GEMM kernel 开销更小。

模型文件

模型大小: 28GB
权重格式: safetensors (单文件)
量化前模型大小: 52GB (BF16)
压缩比: ~1.86x

推理使用方法

方式一: 直接使用本预量化模型

python3 -m sglang.launch_server \
    --model-path hangxian/Qwen3.5-27B-FP8-PerChannel \
    --tp 1 \
    --host 0.0.0.0 \
    --port 30000

方式二: 使用原始 BF16 模型 + 动态量化（效果相同）

python3 -m sglang.launch_server \
    --model-path Qwen/Qwen3.5-27B \
    --quantization fp8 \
    --tp 1 \
    --host 0.0.0.0 \
    --port 30000

两种方式在推理时效果完全一致，区别仅在于本模型预先将 FP8 权重存储在磁盘上，方式二在加载时在线量化。

适用场景

需要减少显存占用（从 52GB 降至 28GB）
需要提升 decode 吞吐（比 BF16 快约 50%）
对精度要求可接受 per-channel FP8 的轻微损失

License

Apache 2.0（与原模型一致）

Downloads last month: 28

Safetensors

Model size

27B params

Tensor type

BF16

F8_E4M3

Model tree for hangxian/Qwen3.5-27B-FP8-PerChannel

Base model

Qwen/Qwen3.5-27B

Quantized

(224)

this model