Module mistral_loader

Module mistral_loader 

Source
Expand description

Weight loader for Mistral Small 4 (MLA + MoE architecture).

GQA fallback: MLA LoRA projections are expanded to dense at load time via GPU matmul. Q = wq_b @ wq_a, K/V split from wkv_b @ wkv_a[:kv_lora]. Loses MLA’s 12.8x KV cache compression but produces coherent output.

Structs§

MistralWeightLoader