Desperately Seeking Evals
The leading edge of LLM capabilities is always advancing and although I'm regularly telling people to keep a cache of prompts that didn't work (so you can try them again…
Cosmo-1B, a Phi-1.5 Reproduction Attempt
🤗 tried to reproduce the Phi-1.5 "little Large Language Model" based on the papers that Microsoft has published. They fell short on their first attempt but have published the results and are continuing to work on it. This is what they did.