quark28/mini-llm-classic-sft
Text Generation • 20.1M • Updated • 189
mini-llm
20M-parameter masked-diffusion chat models trained from scratch: encoder–decoder, ONNX export, 0-shot benchmarks.
Note Main model. Best on benchmarks and in chat; ONNX included.
quark28/mini-llm-thinking-sft
Text Generation • 19.8M • Updated • 191Note Research variant with an experimental Thinking module.