README.md

metadata

title: AstheralPlus
emoji: ✨
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 5.49.1
python_version: '3.12'
app_file: app.py
license: apache-2.0
models:
  - SEE7/AstheralPlus
short_description: 20.9B assistant with reasoning modes and project workspaces

AstheralPlus

SEE7's fine-tuned 20.9B mixture-of-experts assistant, based on gpt-oss-20b.

Chat

Instruct uses low reasoning effort and displays the final answer. Thinking uses medium reasoning effort and displays the model's reasoning alongside its answer. Input context is capped at 8,192 tokens; generation allows up to 4,096 new tokens within a 110-second window.

Projects

Create named projects, save instructions, attach UTF-8 reference files, start and revisit chats, read transcripts and export a project as JSON. Project content is stored in the visitor's browser. Files are included as reference text during generation and are not executed.

Runtime

Real merged MXFP4 model weights, Transformers with MXFP4 kernels, ZeroGPU large (48GB) allocation. No dedicated paid inference GPU. The existing SEE7/Astheral quota service verifies the owner's included quota before each 120-second allocation, preserving a 180-second reserve. Hugging Face manages visitor quotas.

Interface attribution: LLAMA_UI_LICENSE.