Cherrypick 2 - #1035
Merged
Merged
Conversation
Add API option for setting any runtime options. Particularly, add
Terminate session option for the user, the session remains in the
terminated state until the user disables terminate_session.
To terminate session:
SetRuntimeOptionsConfig("terminate_session", "1")
To unset:
SetRuntimeOptionsConfig("terminate_session", "0")
---------
Co-authored-by: Pranav Sharma <emailpranav@gmail.com>
### Description This PR adds the `softcap` attribute to the `GroupQueryAttention` op. ### Motivation and Context This PR helps resolve the `NaN` output issue with Gemma-2 raised in [this issue](#692).
### Description This PR adds new INT4 quantization features to the model builder. 1. The model builder can now quantize the embedding layer and the language modeling head to INT4 precision by default. 2. For already-quantized PyTorch models that are passed to the model builder, any ops that are still created with `MatMul` can now be quantized to `MatMulNBits` via RTN. 3. A new optional flag in the extra options called `int4_op_types_to_quantize` has been added to allow more flexibility with INT4 quantization. ### Motivation and Context With these PR changes, the size of the ONNX models can be reduced by quantizing the embedding layer and/or the language modeling head. For the ONNX models built from already-quantized PyTorch models, one example is with using AutoAWQ. AutoAWQ does not quantize the language modeling head. The resulting ONNX model typically contains a `MatMul` op for the language modeling head. Now, that `MatMul` op will be quantized via RTN to `MatMulNBits` to reduce memory.
### Description This PR sets `past_present_share_buffer = False` by default for users who run the model builder with `--extra_options config_only=true`. ### Motivation and Context This allows ONNX models produced from other sources that don't support past-present buffer sharing to run with ONNX Runtime GenAI.
As title. This enables support for further reduced quantized model size and improved runtime efficiency, within acceptable range of accuracy degradation. Orthogonal to #940. This PR targets already quantized models in autoawq/autogptq format that **has** lmhead quantized.
Abhishek Jindal (ajindal1)
approved these changes
Nov 6, 2024
Pranav Sharma (pranavsharma)
approved these changes
Nov 6, 2024
kunal-vaishnavi
approved these changes
Nov 6, 2024
Baiju Meswani (baijumeswani)
approved these changes
Nov 6, 2024
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
6 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.
No description provided.