llama.cpp
model: add Hy3 (hy_v3) support with MTP speculative decoding
#25395

Open

model: add Hy3 (hy_v3) support with MTP speculative decoding #25395

satindergrewal wants to merge 6 commits into ggml-org:master from satindergrewal:hy3-mtp

model: add Hy3 (hy_v3) architecture support

a1cc3a5e

model: add MTP/nextn speculative decoding for hy_v3

f370e03f

convert: add HYV3ForCausalLM (hy_v3) with MTP block export

6105ebb7

pwilkin approved these changes on 2026-07-07

github-actions added model

github-actions added conversion

tests : mark hy_v3 as MoE-mandatory in test-archs

97181b71

github-actions added testing

pwilkin marked this pull request as ready for review 1 day ago

pwilkin requested a review from

JohannesGaessler 1 day ago

pwilkin requested a review from

CISC 1 day ago

jinja : add str.format() support

692d171a

chat : detect separator between function name and arguments in autopa…

56142c5f

satindergrewal requested a review from

ggerganov 13 hours ago

satindergrewal requested a review 13 hours ago

github-actions added jinja parser

Reviewers

pwilkin

JohannesGaessler

CISC

ggerganov

Assignees

No one assigned

Labels

model testing jinja parser conversion

Milestone

No milestone