Integrate HuggingFace MCP for automated model uploads

[suparious]

Overview

This repo contains scripts for quantizing models. Once the HuggingFace MCP is configured in srt-concierge, we can automate the upload process.

Current Workflow (Manual)

1. Run quantization script
2. Manually upload to HuggingFace
3. Manually update model card
4. Manually announce on social media

Proposed Workflow (Automated)

1. Run quantization script
2. Ask concierge: "Upload [model] to HuggingFace and announce"
3. Concierge uses HuggingFace MCP to:
   • Create model repo (if needed)
   • Upload model files
   • Generate/update model card
4. Concierge uses Twitter MCP to announce

Model Card Template

Standard template for AWQ quantized models:

# {model_name}-AWQ

AWQ quantized version of [{original_model}]({original_url})

## Quantization Details
- Method: AWQ (Activation-aware Weight Quantization)
- Bits: 4-bit
- Group Size: 128
- Tool: [AutoAWQ](https://github.com/casper-hansen/AutoAWQ)

## Usage
Compatible with vLLM, TGI, and other AWQ-supporting frameworks.

## Deployment
Currently deployed on [SolidRusT AI](https://solidrust.ai) inference platform.

## Credits
Quantized by [SolidRusT Networks](https://huggingface.co/solidrust)

Success Criteria

• Documented workflow for Claude-assisted uploads
• Model card template in repo
• README updated with upload instructions

Depends On

• srt-concierge #70: Add Social Media MCP Servers