Added changes to ensure mxint8 compilations of VLMs work. #336
+41
−34
Conversation
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
| @@ -1385,7 +1342,7 @@ def from_pretrained( | |||
| model, kv_offload=kv_offload | |||
| ) | |||
|
|
|||
| return cls(model, is_tlm=is_tlm, continuous_batching=continuous_batching) | |||
| return cls(model, is_tlm=is_tlm, continuous_batching=continuous_batching, enable_qnn=enable_qnn) | |||
There was a problem hiding this comment.
please inform QNN team about this change before merging, so that nothing breaks at their end due to this change.
| ctx_len: int = 150, | ||
| full_batch_size: Optional[int] = None, | ||
| kv_cache_batch_size: Optional[int] = None, | ||
| encoder_ctx_len: int = 1500, |
There was a problem hiding this comment.
always use meaningful constants rather than magic numbers in the code
quic-dhirajku
force-pushed
the
vlm_mxint8_patch
branch
from
108866f to
03f0676
Compare
quic-dhirajku
force-pushed
the
vlm_mxint8_patch
branch
from
e1a9044 to
068d3c6
Compare
Modified modelling files of InternVL and Llava to have 'vision_embeds' as the name of the image_embeddings. Modified modeling_auto file to incorporate mxint8 modifications for VLMs. LIMITATIONS: It is expected that the Processor of a model always gives vision components in 'float16'. Signed-off-by: quic-dhirajku <quic_dhirajku@quicinc.com> Signed-off-by: Dhiraj Kumar Sah <quic_dhirajku@quicinc.com>
… out older qnn based changes. Signed-off-by: quic-dhirajku <quic_dhirajku@quicinc.com> Signed-off-by: Dhiraj Kumar Sah <quic_dhirajku@quicinc.com>
quic-dhirajku
force-pushed
the
vlm_mxint8_patch
branch
from
068d3c6 to
dc3b639
Compare
Signed-off-by: Dhiraj Kumar Sah <quic_dhirajku@quicinc.com>
|
Looks good to me. Verified in vLLM. |
quic-hemagnih
approved these changes
Apr 23, 2025
# for free
to join this conversation on GitHub.
Already have an account?
# 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.
Modified modelling files of InternVL and Llava to have 'vision_embeds' as the name of the image_embeddings.
Modified modeling_auto file to incorporate mxint8 modifications for VLMs.
LIMITATIONS: It is expected that the Processor of a model always gives vision components in 'float16'.