shuff57/lfm2-24b-phase1-reasoning

Partial fine-tune vs LiquidAI/LFM2-24B-A2B. Partial change: 8% of the weights changed (attention, mlp, other), in 40 of 40 layers.

Tensors changed: MLP 6/2768, Embeddings 0/2, other 60/90, Norms 0/100, Attention 40/40

2 derivatives checked