Spaces:
Running
Apply for community grant: Personal project (gpu)
313M Open-Source LLM with Dual Reasoning Architecture—GPU Grant Request
I've trained Sam-large-2, a 313M parameter language model from scratch with a custom dual-reasoning architecture and no-think mode optimisation. It's the next generation in the Sam series, building on the Sam-large-1 architecture and lessons from Sam-x-1 and Sam-z-1. Despite its compact size, it achieves strong performance on diverse conversational tasks. The model was built on Claude-generated training data blended with outputs from other open models and curated open-source datasets.
Current Status:
81% validation accuracy across 7 epochs
Live on Hugging Face with an interactive chat Space: https://huggingface.co/Smilyai-labs/Sam-large-2
Demonstrates novel dual-reasoning approach that enables nuanced responses while maintaining fast inference
Benchmarks coming soon against comparable models
Why I Need GPU:
Right now I'm CPU-bound and can't properly demonstrate the model's potential or handle concurrent users. An A10G Large or T4 would let me run comprehensive benchmarks, optimize latency, scale the demo, and iterate on architecture improvements.
Project Goal:
This is open-source work aimed at proving that capable, accessible conversational AI doesn't require massive parameters. The Sam series dual-reasoning design is positioned to advance how efficient open models handle complex reasoning—I want to make this a community resource for developers building on edge devices and resource-constrained environments.
I'm actively developing and this isn't a one-off; having consistent GPU access would let me continue iterating and keep the demo live for users.