onnxruntime
[CUDA] Extend MoEPacked for top-k, dispatch and fused MoE calculation
#33963
Merged

[CUDA] Extend MoEPacked for top-k, dispatch and fused MoE calculation #33963

rui-ren
rui-ren rui-ren requested a review from tianleiwu tianleiwu 10 days ago
rui-ren rui-ren force pushed from 520f375e to b9b776b5 10 days ago
rui-ren rui-ren changed the base branch from tlwu/qwen38-gguf to rui-ren/bolt-packed-projections-on-33953-20261107 10 days ago
rui-ren rui-ren force pushed from b9b776b5 to c8184937 10 days ago
rui-ren rui-ren force pushed from c8184937 to 736d86eb 9 days ago
rui-ren rui-ren force pushed from 736d86eb to a9bdbce3 9 days ago
rui-ren
rui-ren
rui-ren
rui-ren rui-ren marked this pull request as ready for review 7 days ago
rui-ren rui-ren requested a review from copilot-pull-request-reviewer copilot-pull-request-reviewer 7 days ago
copilot-pull-request-reviewer
copilot-pull-request-reviewer commented on 2026-11-09
rui-ren rui-ren force pushed from a9bdbce3 to 7b2b13ac 7 days ago
rui-ren rui-ren force pushed from 7b2b13ac to e50e3497 7 days ago
rui-ren
rui-ren commented on 2026-11-09
rui-ren rui-ren requested a review from baijumeswani baijumeswani 7 days ago
tianleiwu
tianleiwu commented on 2026-11-09
rui-ren rui-ren force pushed from e50e3497 to 79067e34 7 days ago
rui-ren rui-ren force pushed from 79067e34 to 97c77c1b 7 days ago
rui-ren rui-ren requested a review from tianleiwu tianleiwu 7 days ago
tianleiwu
tianleiwu commented on 2026-11-10
rui-ren rui-ren force pushed from 97c77c1b to 427252a1 7 days ago
rui-ren rui-ren force pushed from f3b1c310 to b59bba1b 7 days ago
rui-ren rui-ren force pushed from b59bba1b to 769bb7ed 7 days ago
rui-ren rui-ren requested a review from tianleiwu tianleiwu 7 days ago
rui-ren rui-ren force pushed from 769bb7ed to 0babe10d 7 days ago
rui-ren rui-ren force pushed from 0babe10d to 19447711 7 days ago
Base automatically changed from rui-ren/bolt-packed-projections-on-33953-20261107 to main 7 days ago
rui-ren rui-ren force pushed from 19447711 to 89db682c 7 days ago
rui-ren rui-ren changed the title Extend MoEPacked for Bolt ReLU-square routing and BF16 IQ2 precision [CUDA] Extend MoEPacked for top-k, dispatch and fused MoE calculation 6 days ago
rui-ren
rui-ren rui-ren force pushed from ff760d4e to 263c2266 5 days ago
rui-ren rui-ren force pushed from 87daac14 to 63aa0bf8 5 days ago
tianleiwu
tianleiwu requested changes on 2026-11-11
rui-ren rui-ren requested a review from tianleiwu tianleiwu 5 days ago
tianleiwu tianleiwu enabled auto-merge (squash) 5 days ago
tianleiwu
tianleiwu dismissed these changes on 2026-11-12
disabled auto-merge 5 days ago
Auto-merge was disabled for another pull request in the stack
rui-ren rui-ren force pushed from 2e916311 to e5655a5c 5 days ago
tianleiwu tianleiwu enabled auto-merge (squash) 5 days ago
disabled auto-merge 5 days ago
Auto-merge was disabled for another pull request in the stack
tianleiwu tianleiwu enabled auto-merge (squash) 5 days ago
disabled auto-merge 5 days ago
Auto-merge was disabled for another pull request in the stack
rui-ren rui-ren dismissed their stale review via 51b54924 4 days ago
rui-ren rui-ren force pushed from e5655a5c to 51b54924 4 days ago
rui-ren
rui-ren Support ReLU-square experts and preselected FP32 MoEPacked routes
f9341e1c
rui-ren Preserve projected sigmoid routing in packed ReLU-square MoE
780ce52e
rui-ren Preserve contiguous BF16 IQ2 ReLU-square MoE reductions
d51be3eb
rui-ren Make MoE cancellation regression portable across CUDA architectures
d421780d
rui-ren Refresh generated packed MoE operator documentation
b1fac9f1
rui-ren Format combined packed MoE rebase resolution
90d320ea
rui-ren Preserve 32-route MoEPacked modes after upstream validation rebase
c39f5f79
rui-ren Harden MoEPacked routing contracts and independent IQ2 regression tests
bc874676
rui-ren Verify malformed MoE outputs against per-element NaN expectations
2a03b16f
rui-ren Keep projected sigmoid routing outside MoEPacked
80ddfbff
rui-ren Preserve the IQ2 expert vector loader during router separation
11ccca41
rui-ren Align native MoE consumers on full-score FP32 routing
411a6431
rui-ren Keep MoEPacked routing changes independent of QMoE
331a140d
rui-ren Rename the full-score MoE routing implementation files
84fb07fa
rui-ren Keep full-score routing private to MoEPacked
4f31bea4
rui-ren Guard new MoEPacked routing modes against legacy hybrid offload
a3a46bb6
rui-ren Remove stale QMoE routing tests from MoEPacked PR
c8115c86
rui-ren Restore the legacy MoEPacked routing normalization default
98b43ec0
rui-ren Match generated MoEPacked attribute documentation
0375ed92
tianleiwu tianleiwu force pushed from 51b54924 to 0375ed92 4 days ago
tianleiwu
tianleiwu approved these changes on 2026-11-12
tianleiwu tianleiwu added release:1.32.0
tianleiwu tianleiwu enabled auto-merge (squash) 4 days ago
tianleiwu tianleiwu merged 6aa9b274 into main 4 days ago
tianleiwu tianleiwu deleted the rui-ren/bolt-packed-moe-on-33953-20261107 branch 4 days ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone