DeepSeek V3.1 beats Llama family in fair fine-tuning experiment

Independent researcher kalomaze ran an out-of-distribution fine-tuning benchmark with identical tokenizer, data, and batches, finding DeepSeek V3.1 achieved better cross-entropy than the entire Llama family.

2026-11-05 ~ 2026-11-05 · 2 related posts

• MoE beats dense in matched OOD finetune: DeepSeek V3.1 hits 0.804 CE vs Llama 405B — kalomaze · 2026-11-05
• kalomaze runs matched OOD finetuning test: big MoE vs dense models, 1:1 comparable — kalomaze · 2026-11-05