Instructions to use GadflyII/Qwen3-Coder-Next-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use GadflyII/Qwen3-Coder-Next-NVFP4 with Transformers:

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

pipe = pipeline("text-generation", model="GadflyII/Qwen3-Coder-Next-NVFP4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("GadflyII/Qwen3-Coder-Next-NVFP4")
model = AutoModelForCausalLM.from_pretrained("GadflyII/Qwen3-Coder-Next-NVFP4")
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 GadflyII/Qwen3-Coder-Next-NVFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "GadflyII/Qwen3-Coder-Next-NVFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GadflyII/Qwen3-Coder-Next-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/GadflyII/Qwen3-Coder-Next-NVFP4

SGLang

How to use GadflyII/Qwen3-Coder-Next-NVFP4 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 "GadflyII/Qwen3-Coder-Next-NVFP4" \
    --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": "GadflyII/Qwen3-Coder-Next-NVFP4",
		"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 "GadflyII/Qwen3-Coder-Next-NVFP4" \
        --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": "GadflyII/Qwen3-Coder-Next-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use GadflyII/Qwen3-Coder-Next-NVFP4 with Docker Model Runner:
```
docker model run hf.co/GadflyII/Qwen3-Coder-Next-NVFP4
```

Spark support

by everphilski - opened Apr 16

Discussion

everphilski

Apr 16

I was able to get it working

ExecStart=/usr/bin/docker run --rm --gpus all
--name vllm-coder-agent
-p 8100:8000
-v /home/user/.cache/huggingface:/root/.cache/huggingface
-v /home/user/.cache/vllm:/root/.cache/vllm
--ipc=host
vllm/vllm-openai:v0.17.1-cu130
--model GadflyII/Qwen3-Coder-Next-NVFP4
--gpu-memory-utilization 0.65
--max-model-len 65536
--kv-cache-dtype fp8
--enforce-eager
--enable-auto-tool-choice
--tool-call-parser hermes

however I can't run it without --enforce-eager so I'm capped at like 10 tok/s. Any ideas?

starstream

Apr 18

•

edited Apr 18

This is from a working docker compose I play around with:

services:
  vllm-spark-pro:
    image: nvcr.io/nvidia/vllm:26.03-py3
    container_name: vllm-spark-pro
    runtime: nvidia
    ipc: host
    deploy:
      resources:
        limits:
          memory: 100G
    ports:
      - "8000:8000"
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - HF_TOKEN=${HF_TOKEN}
      - VLLM_USE_FLASHINFER_MOE_FP4=0
      - VLLM_NVFP4_GEMM_BACKEND=marlin  # FORCES stable, fast GEMM
      - VLLM_TEST_FORCE_FP8_MARLIN=1   # Routes FP8 KV cache through stable path
      - VLLM_MARLIN_USE_ATOMIC_ADD=1 # This specifically speeds up the "summing" of huge attention matrices during those massive 100k+ token context reads.
    command:
      - "vllm"
      - "serve"
      - "GadflyII/Qwen3-Coder-Next-NVFP4"
      - "--max-model-len"
      - "262144"
      #- "--chat-template"
      #- "/app/data/pastebin_quen_template.jinja"
      - "--max-num-batched-tokens"
      - "65536"
      - "--gpu-memory-utilization"
      - "0.85"
      - "--kv-cache-dtype"
      - "fp8"
      - "--enable-prefix-caching" # CRITICAL for coding models (fast multi-turn responses)
      - "--tool-call-parser"
      - "qwen3_xml"
      - "--enable-auto-tool-choice"
      - "--max-num-seqs"
      - "4"
    restart: unless-stopped
    volumes:
      - hf-cache:/root/.cache/huggingface
      - ./:/app/data

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