KeyLM: 75M parameter LLM beats SmolLM-135M in IFEval

View on Reddit

Trained from scratch on 18B tokens, the 75M-parameter KeyLM instruct model scores higher than SmolLM-135M-Instruct on IFEval. Available as base, instruct, and GGUF versions.