models: Qwen3.8-Flash-Next MTP - #28243

Draft

danielhanchen wants to merge 11 commits into

ggml-org:masterfrom

danielhanchen:qwen4exp/mtp

Commits on Sep 3, 2026

gguf-py : register the qwen4exp NextN tensors

ryanmonsurate
authored and
danielhanchen
committed
c8c3a5b
model : add the qwen4exp NextN/MTP draft head

ryanmonsurate
authored and
danielhanchen
committed
57672d9
convert : export the qwen4exp NextN/MTP draft head

ryanmonsurate
authored and
danielhanchen
committed
84f9558
llama: let an MTP draft borrow the target's embeddings and lm head
danielhanchen
committed
86d321a
qwen4exp: allow loading a draft-only MTP export
danielhanchen
committed
30b6537
qwen4exp mtp: trim comments
danielhanchen
committed
44c7960
convert : declare mtp_shared_embd on the Qwen MTP mixin
danielhanchen
committed
eb65412
qwen4exp: reject a draft-only export loaded without its target
danielhanchen
committed
44ff803
llama: borrow the MTP draft's shared tensors through ctx_other
danielhanchen
committed
cc2c59a
qwen4exp mtp: cut comments that restate the code
danielhanchen
committed
2c96729

Commits on Sep 5, 2026

speculative : only gemma4-assistant shares the target KV cache
danielhanchen
committed
d1a9235