fix(glm5): support FP8 context-parallel launches - #17

Closed

JackRao123 wants to merge 7 commits into

trainers-main-20260907from

jackrao/glm-cp-bridge-20260907

Commits on Jul 12, 2026

perf(ckpt): vectorize FP8 blockwise dequant (bit-exact, ~17x/tensor)

JackRao123
and
claude
committed
f8016ce
fix(glm5): resolve raw config.json via the hub cache for HF-id launches

JackRao123
and
claude
committed
d48814a
build: pin Megatron-LM to the packed-CP q_causal_offsets fix (LM#14 head)
JackRao123
committed
8e6e368

Commits on Jul 13, 2026

build: update Megatron-LM PR pin
JackRao123
committed
31ee2e5
fix(glm5): fail fast without raw config

JackRao123
and
cursoragent
committed
ce1c285
build: advance Megatron-LM PR pin
JackRao123
committed
ab50d21
docs(glm5): clarify raw config workaround

JackRao123
and
cursoragent
committed
32d432c