| ---
|
| title: PackedLLM
|
| emoji: 🧠
|
| colorFrom: indigo
|
| colorTo: purple
|
| sdk: gradio
|
| sdk_version: 6.19.0
|
| python_version: "3.11"
|
| app_file: app.py
|
| pinned: false
|
| license: other
|
| short_description: Official demo for the PackedLLM RoE system
|
| ---
|
|
|
| # PackedLLM Demo
|
|
|
| A Hugging Face Space demo for **PackedLLM**, a custom Routing-of-Experts system with persistent memory, web search, sandboxed Python execution, persona layers, and direct expert access.
|
|
|
| This Space is intended to showcase the loaded checkpoint through a Gradio interface with multiple tabs:
|
|
|
| * **Chat** — interact with the full PackedLLM pipeline
|
| * **Experts** — call individual specialists directly
|
| * **Memory** — store and recall memory, edit profiles
|
| * **Web** — run embedded web search
|
| * **CodeBox** — execute code in the sandbox
|
| * **System** — warmup, status, expert reload/unload, checkpoint save
|
|
|
| ## What this demo loads
|
|
|
| The app expects a custom checkpoint file named:
|
|
|
| ```text
|
| PackedLLM.pt
|
| ```
|
|
|
| The checkpoint is loaded through the custom `PackedLLMRunner` wrapper, not through a standard Transformers `AutoModel` workflow.
|
|
|
| ## Notes
|
|
|
| * The demo includes potentially powerful capabilities such as web search and code execution.
|
| * The implementation exposes system controls for warmup, expert reload/unload, memory operations, and checkpoint saving.
|
| * This repository uses a custom checkpoint format and custom runtime code.
|
|
|
| ## Intended use
|
|
|
| This Space is meant for interactive testing, demos, and showcasing the current PackedLLM architecture.
|
|
|
