this is a repo I created where I can tinker w LLM360 models mostly Amber and Crystal to better understand developer experience so that I can potential imrpove their experience w LLM360. I also add my cookbooks in here too as a guide for developers.

LLM360 Financial Sentiment Prototype

Project Goal

To evaluate the inference capabilities of LLM360's Amber-7B model for specific financial domain tasks (Bitcoin Sentiment Analysis).

The Stack

• Model: LLM360/Amber-7B (Loaded via Hugging Face)
• Infrastructure: Google Colab (T4 GPU) with 4-bit quantization
• Library: Transformers & BitsAndBytes

Why This Matters

As a Community Manager, I believe in "dogfooding" our own models. I built this prototype to:

1. Test the "Time-to-Hello-World" for a Python developer.
2. Explore how Amber handles niche financial lexicon ("ETFs", "Inflows").
3. Identify friction points in the documentation flow (e.g., K2 data prep links).

Findings

• Amber loads efficiently on free-tier GPUs using 4-bit quantization.
• Inference speed is viable for real-time sentiment analysis bots.
• LLM360 K2Think data link shows 404
• Next Step: I plan to build a "Cookbook" tutorial based on this code to help Fintech developers onboard to LLM360.