feat: multi-model support with KV cache (T5, Qwen, Mu2)#334

Merged

vortex-captain merged 38 commits into

mainfrom

reny/multi_model

Apr 23, 2026

Commits on Apr 14, 2026

re init branch
Yi Ren
committed
feat: fp16 KV cache support and session_options passthrough
Yi Ren
committed
revert: remove subfolder from config command (not part of this PR)
Yi Ren
committed

Commits on Apr 15, 2026

Commits on Apr 16, 2026

refactor: polymorphic KV cache for decoder-only prefill + gen
Yi Ren
committed
fix: remove unused _pad_inputs from decoder_only.py
Yi Ren
committed
docs: add static cache switching instructions to mu2.py
Yi Ren
committed
feat: WinMLCompositeModel.from_onnx + from_pretrained composite routing
Yi Ren
committed
feat: composite model support in run_eval.py + T5 summarization
Yi Ren
committed
feat: remove_isnan_in_attention_mask surgery + optim configs for T5/Qwen/Mu2
Yi Ren
committed
fix: enable DepthPro ONNX registration + update timeout skip list
Yi Ren
committed
test: add unit tests for polymorphic KV cache and composite from_onnx
Yi Ren
committed
Merge remote-tracking branch 'origin/main' into reny/multi_model
Yi Ren
committed
Potential fix for pull request finding 'CodeQL / Cyclic import'

vortex-captain
and
github-advanced-security[bot]
authored
fix: resolve ruff F821/UP037 for WinMLCompositeModel type annotation
Yi Ren
committed

Commits on Apr 17, 2026

Commits on Apr 20, 2026

Commits on Apr 22, 2026

fix(models): defensive fixes from PR #334 review

tezheng
authored and
Zheng Te
committed
fix(models): Phase 1 follow-up — 6 review findings + 2 critic fixes

tezheng
authored and
Zheng Te
committed
fix(models): Phase 2 follow-up — 9 review findings + critic regression

tezheng
authored and
Zheng Te
committed

Commits on Apr 23, 2026

Merge branch 'main' into reny/multi_model
Yi Ren
committed
fix(models): I8 — widen WinMLCompositeModel.from_onnx onnx_path type

tezheng
authored and
Zheng Te
committed