CANN : Optimize mul_mat_id quantization#17782
Closed
jjjxp03 wants to merge 1 commit intoggml-org:masterfrom
Closed
CANN : Optimize mul_mat_id quantization#17782jjjxp03 wants to merge 1 commit intoggml-org:masterfrom
jjjxp03 wants to merge 1 commit intoggml-org:masterfrom
Conversation
Contributor
|
The results of running the above-mentioned modifications on the qwen2.5:0.5b f16 model are as follows. |
Contributor
|
@TianHao324 This PR optimize mul_mat_id for quantization type, not for fp16. |
hipudding
reviewed
Dec 18, 2025
| @@ -2863,175 +2863,39 @@ void ggml_cann_step(ggml_backend_cann_context& ctx, ggml_tensor* dst){ | |||
| * MoE architectures, possibly involving sparse expert routing. | |||
| */ | |||
| static void ggml_cann_mul_mat_id_fp(ggml_backend_cann_context& ctx, ggml_tensor* dst) { | |||
Contributor
There was a problem hiding this comment.
ggml_cann_mul_mat_id_fp has updated. Could you please resolve conflicts?
Contributor
|
See #19228 |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
For CANN backend:Currently, mul_mat_id performs a device-to-host copy, which does not support full-image inference.Optimize mul_mat_id quantization by referencing the non-quantized version and removing the device_to_host copy operation.
Verify the correctness of the modification using the existing test tool:
./bin/test-backend-ops test -b CANN0 -o MUL_MAT_ID
Passed the q_8 test. Currently, only q_8 has been implemented and tested. Tensor self is not implemented for DT_INT4; to support q_4, the corresponding type implementation needs to be added.