[WIP][Model] Add support for multiple audio chunks/audio URLs #7826
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. Once the PR is approved and ready to go, please make sure to run full CI as it is required to merge (or just use auto-merge). To run full CI, you can do one of these:
🚀 |
| # items that already have them. | ||
| prompt_fragments = [text_prompt] | ||
| missing_placeholders: List[str] = [] | ||
| for placeholder in placeholders: |
There was a problem hiding this comment.
This is pretty convoluted as it's a full generalization of the check that was there before -- happy to do something much simpler, I just wasn't sure which "placeholder is already present in the prompt" scenarios are important.
There was a problem hiding this comment.
(Also it needs tests if it remains in its current form :))
There was a problem hiding this comment.
I think this is fine, but definitely needs some examples to better make sense of it.
|
|
||
| tensors_ = cast(List[torch.Tensor], concatenated) | ||
| return torch.stack(tensors_) if all(t.shape == tensors_[0].shape | ||
| for t in tensors_) else tensors_ |
There was a problem hiding this comment.
Thanks for implementing my suggestion from the RFC! Would be great to add some tests for this.
There was a problem hiding this comment.
Thank you for laying out the plan!
| @@ -70,7 +72,7 @@ def vllm_to_hf_output(vllm_output: Tuple[List[int], str, | |||
| def run_test( | |||
There was a problem hiding this comment.
Let's separate the test for single-input (without list) and multi-input (with list)
| mm_data: MultiModalDataDict = {} | ||
|
|
||
| # Merge all the multi-modal items | ||
| for single_mm_data in (await asyncio.gather(*futures)): | ||
| for mm_key, mm_item in single_mm_data.items(): | ||
| existing_item = mm_data.get(mm_key) | ||
| if not existing_item: | ||
| # Clone it if it's already a list so we can freely | ||
| # mutate it later. | ||
| item_to_insert = mm_item[:] if isinstance( | ||
| mm_item, list) else mm_item | ||
| mm_data[mm_key] = item_to_insert # type: ignore | ||
| else: | ||
| if isinstance(existing_item, list): | ||
| result_list = existing_item | ||
| else: | ||
| result_list = [existing_item] | ||
| mm_data[mm_key] = result_list # type: ignore | ||
|
|
||
| if isinstance(mm_item, list): | ||
| result_list.extend(mm_item) | ||
| else: | ||
| result_list.append(mm_item) |
There was a problem hiding this comment.
Looks quite complicated. Can we simplify the code here using a defaultdict? We allow Mapping[str, object] in MultiModalDataDict so this shouldn't cause any type errors.
| # items that already have them. | ||
| prompt_fragments = [text_prompt] | ||
| missing_placeholders: List[str] = [] | ||
| for placeholder in placeholders: |
There was a problem hiding this comment.
I think this is fine, but definitely needs some examples to better make sense of it.
This adds:
NestedTensorsof arbitrary depth to represent multimodal inputs -- model implementations that support multiple multimodal inputs can return lists from their input mapper to include a dimension that represents the number of inputs in the prompt.UltravoxModel"audio_url"/"image_url"items in the OpenAI API server (up to the limits provided at initialization)This depends on #7230 (for changes to
repeat_and_pad_placeholder_tokens) which in turn is blocked on #7783.I intend to split this into three PRs (NestedTensors, Ultravox, API server) but I'd love to get an assessment from @DarkLight1337 and @ywang96 on the overall approach.
BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE
PR Checklist (Click to Expand)
Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.
PR Title and Classification
Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:
[Bugfix]for bug fixes.[CI/Build]for build or continuous integration improvements.[Doc]for documentation fixes and improvements.[Model]for adding a new model or improving an existing model. Model name should appear in the title.[Frontend]For changes on the vLLM frontend (e.g., OpenAI API server,LLMclass, etc.)[Kernel]for changes affecting CUDA kernels or other compute kernels.[Core]for changes in the core vLLM logic (e.g.,LLMEngine,AsyncLLMEngine,Scheduler, etc.)[Hardware][Vendor]for hardware-specific changes. Vendor name should appear in the prefix (e.g.,[Hardware][AMD]).[Misc]for PRs that do not fit the above categories. Please use this sparingly.Note: If the PR spans more than one category, please include all relevant prefixes.
Code Quality
The PR need to meet the following code quality standards:
format.shto format your code.docs/source/if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.Notes for Large Changes
Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with
rfc-requiredand might not go through the PR.What to Expect for the Reviews
The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:
action-requiredlabel on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.Thank You
Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!