[CI][UT]enable multimodal ut #158
Conversation
Signed-off-by: zouyida <zouyida@huawei.com>
|
Please add [CI][UT] label before your PR title. |
|
need a rebase to let CI happy |
Signed-off-by: zouyida <zouyida@huawei.com>
Signed-off-by: zouyida <zouyida@huawei.com>
vllm_ascend/model_runner.py
Outdated
| vision_end_token_id=hf_config.vision_end_token_id, | ||
| spatial_merge_size=hf_config.vision_config. | ||
| spatial_merge_size, | ||
| second_per_grid_ts=None, |
There was a problem hiding this comment.
Why we directly pass None as second_per_grid_ts?
There was a problem hiding this comment.
We'd better get second_per_grid_ts from mm_kwargs at line 714, Outside the for loop. like this:
second_per_grid_ts = mm_kwargs.get("second_per_grid_ts", None)There was a problem hiding this comment.
Since this is a new interface changed by vllm, we will not use it for now. But we want to reserve it for future use to facilitate calling.
There was a problem hiding this comment.
If this change is uesless currently, just keep as it is. Otherwise, we'd better make it work as expect.
There was a problem hiding this comment.
I have changed my interface because I think it will be used soon. If there is something inappropriate, please let me know.
Signed-off-by: zouyida <zouyida@huawei.com>
|
LGTM |
enable multimodal ut