CoolFace
Apppublic

bored10/qwen3-8-flash-next-demo

sourceHugging Faceupdated 1mo agoView on Hugging Face
2likes
App README

Qwen3.8-Flash-Next Demo

A streaming text-and-image chat demo for `Qwen/Qwen3.8-Flash-Next`. The 125B-parameter checkpoint is called through an OpenAI-compatible inference endpoint, so this Space does not download or expose model weights or API credentials.

Availability notice: The model weights are openly available, but this Space cannot provide free public inference for this very large model. To run the chat, its owner must configure an OpenAI-compatible endpoint and API key, or host the model on suitable hardware. The Space does not collect visitor API keys.

Configuration

In the Space Settings → Variables and secrets, create these secrets:

SecretRequiredValue
OPENAI_API_KEYYesAPI key for the inference endpoint
OPENAI_BASE_URLYesOpenAI-compatible API base URL, including /v1

Optional public variable:

VariableDefaultPurpose
MODEL_IDQwen/Qwen3.8-Flash-NextModel identifier accepted by the chosen endpoint

For QwenCloud, obtain the regional API host from its API-key page and use its OpenAI-compatible /compatible-mode/v1 endpoint. Confirm that the endpoint serves the exact preview model identifier; QwenCloud may use a commercial model alias instead.

Publish with the Hugging Face CLI

powershell
# Authenticate once in an interactive terminal.
hf auth login

# Replace YOUR_USERNAME with your Hugging Face namespace.
hf repos create YOUR_USERNAME/qwen3-8-flash-next-demo --type space --space-sdk gradio --flavor zero-a10g --public

# Explicit paths avoid a Windows CLI path-normalisation issue.
hf upload YOUR_USERNAME/qwen3-8-flash-next-demo app.py app.py --repo-type space
hf upload YOUR_USERNAME/qwen3-8-flash-next-demo requirements.txt requirements.txt --repo-type space
hf upload YOUR_USERNAME/qwen3-8-flash-next-demo .gitignore .gitignore --repo-type space
hf upload YOUR_USERNAME/qwen3-8-flash-next-demo README.md README.md --repo-type space

Then add the two secrets in the Space settings before making a request. Never commit an API key to this repository.

Behaviour

  • —Supports a text prompt and one optional image (max 10 MB).
  • —Streams the model's final response to the chat.
  • —Lets visitors choose Qwen thinking mode and effort while withholding internal reasoning traces from the interface.
  • —Enables the Gradio MCP server on launch.