Feat/native gguf llama - #327

Open

amodhyh wants to merge 9 commits into

FlashML-org:mainfrom

amodhyh:feat/native-gguf-llama

Commits on Sep 1, 2026

feat(llama): add initial stubs for native GGUF support
amodhyh
committed
1cbf8a6
feat(llama): implement native GGUF support
amodhyh
committed
f42698e
feat(llama): add MoE config parsing for LLaMA 4 with shared experts
amodhyh
committed
429ebf9
feat(llama): add MoE and Shared Expert dual-routing to LLaMA Decoder
amodhyh
committed
ccfe28b
fix(llama): instantiate and route through MoE gate
amodhyh
committed
4e51005
feat(llama): stream MoE weights and Shared Experts from GGUF
amodhyh
committed
7076793
fix(llama): conditionally swap MoE and Shared Expert layers for GGUF kernels
amodhyh
committed
c05412d
feat(llama): add tokenizer EOT tokens for LLaMA 3 and 4
amodhyh
committed
1cabd37
feat(gguf): implement Llama-4 MoE routing and tensor filtering
amodhyh
committed
cd1ff5f