[draft] feat: support GLM-5.3-Flash inference - #1525

Closed

sufubao wants to merge 29 commits into

ModelTC:mainfrom

sufubao:support-glm-5-3-flash

Commits on Aug 28, 2026

feat: support GLM-5.3 Flash inference
sufubao
committed
8a8f330
build: package GLM-5.3 H100 runtime image
sufubao
committed
2b1fbcb
ci: publish signed GLM-5.3 H100 image
sufubao
committed
3fbcbd5
build: harden GLM-5.3 runtime image
sufubao
committed
ec79db8
build: remove vulnerable packaging tools
sufubao
committed
2b68885
build: restore flattened runtime environment
sufubao
committed
624d381
build: close GLM-5.3 runtime dependencies
sufubao
committed
8e3dc81
ci: normalize GHCR image name
sufubao
committed
013da1d
ci: remove GHCR publishing
sufubao
committed
9151a7c
feat(pd): expose stage and node load metrics (#1509)
sufubao
authored
aa7cab9
perf(mtp): optimize Qwen3.5 GDN verification kernel (#1513)
sufubao
authored
3e0d9f7
perf: avoid full-vocab all-gather for draft greedy sampling (#1517)
sufubao
authored
cc25c63
feat(glm5): add multimodal and 1m deployment support
sufubao
committed
79f6713
Merge remote-tracking branch 'upstream/main' into support-glm-5-3-flash
sufubao
committed
bfdf872
build: split flattened runtime for registry upload
sufubao
committed
f4b94b9

Commits on Aug 29, 2026

docker: shard flattened GLM runtime layers
sufubao
committed
14798e4
docker: preserve metadata during layer copy-up
sufubao
committed
efff73a
docs: record validated H100 deployment results
sufubao
committed
1b408de
docs: add same-host GLM-5.3 engine benchmarks
sufubao
committed
3dc3840
perf(glm5): tune H100 concurrency profile
sufubao
committed
8efce02
build: add offline GLM-5.3 release overlay
sufubao
committed
cc6319a
docs: publish final GLM-5.3 H100 profile
sufubao
committed
c639f6b

Commits on Aug 30, 2026

perf(glm5): publish all-concurrency TP8 profile
sufubao
committed
62b8a9d
docs: record final GLM-5.3 release results
sufubao
committed
65256aa

Commits on Aug 31, 2026