(WIP) bitnet and t-mac #23540
Open
(WIP) bitnet and t-mac #23540
Conversation
This file contains 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
Signed-off-by: Liqun Fu <[email protected]>
|
Can you add description/context ? |
Signed-off-by: Liqun Fu <[email protected]>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| @@ -402,7 +402,8 @@ | |||
| struct BlockwiseQuantizer { | |||
| // To support other qbits, need to add bit packing code for | |||
| // storing to dst and zero points | |||
| static_assert(qbits == 4, "Only 4b block quantization is supported!"); | |||
| static_assert(qbits == 4 || qbits == 2, "Only 4b block quantization is supported!"); | |||
| //static_assert(qbits != 2 || Columnwise, "Only support Columnwise in qbits == 2 case."); | |||
Check notice
Code scanning / CodeQL
Commented-out code Note
Signed-off-by: Liqun Fu <[email protected]>
Comment on lines
+63
to
+74
| switch (ComputeType) { | ||
| case SQNBIT_CompInt8: { | ||
| // workspace buffer is used for block quantization of A to int8 | ||
| const size_t BlockCountK = MlasDivRoundup(K, BlkLen); | ||
| // QuantData + Scale | ||
| const size_t PerGemmWorkspaceSize = M * BlockCountK * Q8BlkSize(BlkLen); | ||
| return PerGemmWorkspaceSize; | ||
| } | ||
| default: { | ||
| return 0; | ||
| } | ||
| } |
Check notice
Code scanning / CodeQL
No trivial switch statements Note
Signed-off-by: Liqun Fu <[email protected]>
Signed-off-by: Liqun Fu <[email protected]>
Signed-off-by: Liqun Fu <[email protected]>
fajin-corp
reviewed
Feb 4, 2025
| } | ||
| const uint8_t vi1 = weights[j * q_rows + i / 2] >> 4; | ||
| const float v1 = (static_cast<float>(vi1) - zp1) * scale1; | ||
| dst[j * rows + (i + 1)] = static_cast<ElementT>(v1); | ||
| } | ||
| } |
There was a problem hiding this comment.
maybe separate the template specializations for qnbits so the code would be cleaner?
fajin-corp
reviewed
Feb 4, 2025
|
|
||
| template <typename Tin, bool signed_quant> | ||
| struct BlockwiseQDQQuantizer<Tin, 4, signed_quant> { | ||
| struct BlockwiseQDQQuantizer { |
There was a problem hiding this comment.
would it be better to separate the specializations for different qbits?
fajin-corp
reviewed
Feb 4, 2025
| } | ||
|
|
||
| size_t | ||
| SQ2BitGemmKernel_CompInt8_avx2( |
fajin-corp
reviewed
Feb 4, 2025
| @@ -113,6 +120,7 @@ struct MLAS_QNBIT_GEMM_DISPATCH { | |||
|
|
|||
| Q4BitGemmPackQuantBData_Fn* SQ4BitGemmPackQuantBData = nullptr; | |||
| Q4BitGemmPackQuantBData_Fn* HQ4BitGemmPackQuantBData = nullptr; | |||
| Q4BitGemmPackQuantBData_Fn* SQ2BitGemmPackQuantBData = nullptr; | |||
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.
Preparation for 2bit T-MAC and ternary bit BitNet implementation.