[luci/pass] Add basic quantization support for weights in GPTQuantizeWeightsWithGPTQPass #14475
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
…uantizeWeightsWithGPTQPass This commit implements basic quantization of weights in `QuantizeDequantizeWeightsWithGPTQPass`, supporting both 4-bit and 8-bit quantization. Only channel-wise quantization is supported. ONE-DCO-1.0-Signed-off-by: y01000.you <[email protected]>
seanshpark
reviewed
Dec 18, 2024
| nudged_max[i]); | ||
| } | ||
|
|
||
| quantize = [&](uint32_t *indices, loco::TensorShape &dimension, int index_channel_dim) { |
There was a problem hiding this comment.
Suggested change
| quantize = [&](uint32_t *indices, loco::TensorShape &dimension, int index_channel_dim) { | |
| auto quantize = [&](uint32_t *indices, loco::TensorShape &dimension, int index_channel_dim) { |
seanshpark
reviewed
Dec 18, 2024
Comment on lines
196
to
197
| IterFunc quantize; | ||
|
|
There was a problem hiding this comment.
Suggested change
| IterFunc quantize; |
seanshpark
reviewed
Dec 18, 2024
Comment on lines
128
to
129
| assert(min <= max); | ||
| const int32_t kMinScale = 0; |
There was a problem hiding this comment.
Suggested change
| assert(min <= max); | |
| const int32_t kMinScale = 0; | |
| assert(min <= max); | |
| const int32_t kMinScale = 0; |
seanshpark
reviewed
Dec 18, 2024
Comment on lines
+75
to
+76
|
|
||
| assert(scaling_factor[idx_channel] > 0); |
There was a problem hiding this comment.
Suggested change
| assert(scaling_factor[idx_channel] > 0); | |
| assert(scaling_factor[idx_channel] > 0); | |
seanshpark
reviewed
Dec 18, 2024
Comment on lines
79
to
80
| auto data_clipped = data < min[idx_channel] ? min[idx_channel] : data; | ||
| data_clipped = data_clipped > max[idx_channel] ? max[idx_channel] : data_clipped; |
There was a problem hiding this comment.
may not use min(max(a,b),c) form ?
|
1/ plz add link to draft |
seanshpark
reviewed
Dec 18, 2024
| } | ||
|
|
There was a problem hiding this comment.
move this format change to another PR
seanshpark
reviewed
Dec 18, 2024
Comment on lines
106
to
107
| min[idx_channel] = data < min[idx_channel] ? data : min[idx_channel]; | ||
| max[idx_channel] = data > max[idx_channel] ? data : max[idx_channel]; |
There was a problem hiding this comment.
why not use std::min, std::max ???
This commit refactors the QuantizeDequantizeWeightsWithGPTQPass.cpp file to improve its readability and maintainability. ONE-DCO-1.0-Signed-off-by: y01000.you <[email protected]>
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.
This commit implements basic quantization of weights in
QuantizeDequantizeWeightsWithGPTQPass, supporting both 4-bit and 8-bit quantization. Only channel-wise quantization is supported.ONE-DCO-1.0-Signed-off-by: y01000.you [email protected]