onnx: add com.microsoft MatMulNBits (4-bit weight quantized matmul)#2290
Open
czoli1976 wants to merge 1 commit into
Open
onnx: add com.microsoft MatMulNBits (4-bit weight quantized matmul)#2290czoli1976 wants to merge 1 commit into
czoli1976 wants to merge 1 commit into
Conversation
Dequantizes the constant 4-bit blocked weight (B, scales, optional packed zero_points; default zero point 8) to a float [N, K] weight in Rust and emits a plain matmul (EinSum) against the activation, with an optional bias. A fused int4 kernel would be a follow-up perf optimization. Validated bit-exact against onnxruntime MatMulNBits for both symmetric (no zero_points) and asymmetric quantization. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
czoli1976
force-pushed
the
feature/onnx-matmulnbits
branch
from
2b3d915 to
1bb5e6f
Compare
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
MatMulNBits: 4-bit blocked weight-quantized matmul (the dominant quantization in edge-LLM exports). Dequantizes the constant quantized weight (B, scales, optional packed zero_points; default zero point 8) to a float [N,K] weight and emits a plain matmul (EinSum) against the activation, with optional bias. A fused int4 kernel would be a natural follow-up perf optimization. Validated BIT-EXACT vs onnxruntime across block_size 16/32/128, symmetric & asymmetric quantization, and 2D/3D activations; no node-suite regression; clippy+fmt clean. Part of com.microsoft contrib-op coverage for ORT-exported LLMs.