ORT 1.25.1 release: version bump and cherry-pick #27907#28148
Closed
vraspar wants to merge 2 commits intomicrosoft:rel-1.25.1from
Closed
ORT 1.25.1 release: version bump and cherry-pick #27907#28148vraspar wants to merge 2 commits intomicrosoft:rel-1.25.1from
vraspar wants to merge 2 commits intomicrosoft:rel-1.25.1from
Conversation
) Adds custom CUDA and CPU kernels for linear attention and causal 1D convolution with state, enabling efficient inference of Qwen3.5 hybrid decoder models in ONNX Runtime. ### New Operators **`LinearAttention`** — Implements the GatedDeltaNet recurrent linear attention mechanism: - Fused kernel computing gated delta-rule update of a recurrent state matrix - Supports both prefill (multi-token) and decode (single-token) paths - Inputs: Q, K, V, decay (alpha), beta gating, optional initial recurrent state - Outputs: attention output, updated recurrent state - CUDA implementation with per-head parallelism; CPU implementation with Eigen **`CausalConvWithState`** — Implements causal 1D convolution with persistent state for autoregressive decoding: - Supports prefill (full convolution) and decode (state-based sliding window) - Inputs: input tensor, conv weights, optional bias, optional initial conv state - Outputs: convolution output, updated conv state ### Op Definitions - Registered in `com.microsoft` domain (opset 1) - Full shape inference and type constraints in `bert_defs.cc` ### Testing - Parity test (`test_parity_linear_attention_causal_conv.py`) validates CUDA and CPU kernels against PyTorch reference implementations from the FLA (Flash Linear Attention) library --------- Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Contributor
Author
|
Closing - recreating with branch from microsoft/onnxruntime instead of fork |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Version bump to 1.25.1 and cherry-pick of #27907 (Add LinearAttention and CausalConvState ops for Qwen3.5) for the 1.25.1 patch release.
Cherry-pick merge commit: 0fedb26