README.md

metadata

title: Exoskeleton Reasoning Demo
emoji: 🦴
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false

A Gradio Space that applies the Appendix-style prompt (Phi-3.5 instruct-style chat). The model must prioritize the given Context and answer in plain text with two sections — Analysis and Response.

Environment variables (optional)

• EXOSKELETON_MODEL_ID (default: Inpris/humains-junior)
• DEVICE_MAP (default: auto)
• MAX_NEW_TOKENS (default: 512)
• TEMPERATURE (default: 0.3)
• TOP_P (default: 0.95)

Secrets

• HF_TOKEN — required if the model is gated.

Files

• app.py — Gradio app (forces slow tokenizer using LLaMA tokenizer if needed; Phi-3.5 fallback)
• requirements.txt — dependencies (pins transformers 4.43.3, accelerate 0.32.1)
• examples/ — (optional) assets/presets