Instructions to use incoai/Qwen3.8-27B-DFlash2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use incoai/Qwen3.8-27B-DFlash2 with Transformers:

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

pipe = pipeline("text-generation", model="incoai/Qwen3.8-27B-DFlash2")

# Load model directly
from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("incoai/Qwen3.8-27B-DFlash2")
model = AutoModel.from_pretrained("incoai/Qwen3.8-27B-DFlash2", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use incoai/Qwen3.8-27B-DFlash2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "incoai/Qwen3.8-27B-DFlash2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "incoai/Qwen3.8-27B-DFlash2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/incoai/Qwen3.8-27B-DFlash2

SGLang

How to use incoai/Qwen3.8-27B-DFlash2 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 "incoai/Qwen3.8-27B-DFlash2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "incoai/Qwen3.8-27B-DFlash2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "incoai/Qwen3.8-27B-DFlash2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "incoai/Qwen3.8-27B-DFlash2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use incoai/Qwen3.8-27B-DFlash2 with Docker Model Runner:
```
docker model run hf.co/incoai/Qwen3.8-27B-DFlash2
```

FP8

by vcerny - opened 2 days ago

Discussion

vcerny

2 days ago

Hi, great stuff, congrats! Will this works for FP8 quant too or does it require a new speculator? Thanks.

jianchen0311

2 days ago

•

edited 2 days ago

Yes it works with the FP8 quant target model. We tested it and the acceptance length is very close to the BF16 target model.

sleepyeldrazi

2 days ago

•

edited 2 days ago

I tried running DFlash2 with unsloth's NVFP4 27B and vllm, from the pr mentioned, rejected it due to the quantized LM Head. Are you aware of this issue? Are you working on this, or might something else be the issue that I should verify? Am booting FP8 as I write this, but the extra KV headroom from the model size delta is something I'd ideally like to keep, even if sacrificing a little acceptance (MTP on the nvfp4 was in the 3-4 range during normal use, so it's still high quality).
Hardware is a DGX Spark.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment