Make TextStreamer public & add unit-tests
#1700
Open
+220
−74
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
github-actions
bot
added
category: LLM
pavel-esir
commented
Feb 10, 2025
Wovchena
reviewed
Feb 12, 2025
| if (text.size() >= 3 && text.compare(text.size() - 3, 3, replacement) == 0) { | ||
| m_decoded_lengths[m_decoded_lengths.size() - 1] = -1; | ||
| // Don't print incomplete text | ||
| return set_streaming_status(on_finalized_subword_callback(res.str())); |
There was a problem hiding this comment.
IterableStreamer needs a similar change. Please add ChunkStreamer to tests as well.
There was a problem hiding this comment.
-
Turned out
IterableStreameralready handles this situation correctly. -
We talked with @as-suvorov, looks like he will add such unit-tests together with implementation of
TextStreamerwith vectorized write.
pavel-esir
force-pushed
the
add_text_streamer_tests
branch
3 times, most recently
from
220815c to
c30f1a8
Compare
github-actions
bot
added
category: visual language
pavel-esir
force-pushed
the
add_text_streamer_tests
branch
from
1667b9e to
8d9c781
Compare
pavel-esir
changed the title
TextStreamer public & add unit-tests
| @@ -131,6 +143,11 @@ PYBIND11_MODULE(py_openvino_genai, m) { | |||
| .def(py::init<>()) | |||
| .def("write", &StreamerBase::write, "Write is called every time new token is decoded. Returns a StreamingStatus flag to indicate whether generation should be stopped or cancelled", py::arg("token")) | |||
| .def("end", &StreamerBase::end, "End is called at the end of generation. It can be used to flush cache if your own streamer has one"); | |||
|
|
|||
| py::class_<TextStreamer, std::shared_ptr<TextStreamer>>(m, "TextStreamer", text_streamer_docstring) | |||
There was a problem hiding this comment.
I think we need to highlight that TextStreamerBase is a base class
I would also create a dedicated py_text_streamer.cpp file for all streamers on python side
| # ("black-forest-labs/FLUX.1-dev", dict(subfolder="tokenizer")), # FLUX.1-dev has tokenizer in subfolder | ||
| ] | ||
|
|
||
| # Chekc that fix for CVS-157216 works. |
There was a problem hiding this comment.
Suggested change
| # Chekc that fix for CVS-157216 works. | |
| # Check that fix for CVS-157216 works. |
| callback: User-defined callback function to process the decoded text, callback should return either boolean flag or StreamingStatus. | ||
|
|
||
| """ | ||
| def __init__(self, tokenizer: ..., callback: typing.Callable[[str], bool | StreamingStatus]) -> None: |
There was a problem hiding this comment.
type ... is unexpected
Tokenizer class should be defined before TextStreamer in pybind11 PYBIND11_MODULE(py_openvino_genai, ..)
| ... | ||
| def end(self) -> None: | ||
| ... | ||
| def write(self, arg0: int) -> StreamingStatus: |
There was a problem hiding this comment.
please, use explicit argument name via py::arg to avoid automatic arg0 names
| __all__ = ['Adapter', 'AdapterConfig', 'AggregationMode', 'AutoencoderKL', 'CLIPTextModel', 'CLIPTextModelWithProjection', 'CacheEvictionConfig', 'ChunkStreamerBase', 'ContinuousBatchingPipeline', 'CppStdGenerator', 'DecodedResults', 'EncodedResults', 'FluxTransformer2DModel', 'GenerationConfig', 'GenerationResult', 'Generator', 'Image2ImagePipeline', 'ImageGenerationConfig', 'InpaintingPipeline', 'LLMPipeline', 'PerfMetrics', 'RawPerfMetrics', 'SD3Transformer2DModel', 'Scheduler', 'SchedulerConfig', 'StopCriteria', 'StreamerBase', 'StreamingStatus', 'T5EncoderModel', 'Text2ImagePipeline', 'TokenizedInputs', 'Tokenizer', 'TorchGenerator', 'UNet2DConditionModel', 'VLMPipeline', 'WhisperGenerationConfig', 'WhisperPerfMetrics', 'WhisperPipeline', 'WhisperRawPerfMetrics', 'draft_model', 'get_version', 'openvino', 'os', 'py_openvino_genai'] | ||
| __version__: str | ||
| __all__ = ['Adapter', 'AdapterConfig', 'AggregationMode', 'AutoencoderKL', 'CLIPTextModel', 'CLIPTextModelWithProjection', 'CacheEvictionConfig', 'ChunkStreamerBase', 'ContinuousBatchingPipeline', 'CppStdGenerator', 'DecodedResults', 'EncodedResults', 'FluxTransformer2DModel', 'GenerationConfig', 'GenerationResult', 'Generator', 'Image2ImagePipeline', 'ImageGenerationConfig', 'InpaintingPipeline', 'LLMPipeline', 'PerfMetrics', 'RawPerfMetrics', 'SD3Transformer2DModel', 'Scheduler', 'SchedulerConfig', 'StopCriteria', 'StreamerBase', 'StreamingStatus', 'T5EncoderModel', 'Text2ImagePipeline', 'TextStreamer', 'TokenizedInputs', 'Tokenizer', 'TorchGenerator', 'UNet2DConditionModel', 'VLMPipeline', 'WhisperGenerationConfig', 'WhisperPerfMetrics', 'WhisperPipeline', 'WhisperRawPerfMetrics', 'draft_model', 'get_version', 'openvino', 'os', 'py_openvino_genai'] | ||
| __version__: str = '2025.1.0.0-1760-6e22001f986-add_text_streamer_tests' |
There was a problem hiding this comment.
having a version is unexpected
it's explicitly removed to avoid regeneration of this file for each commit because of different hashes
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.
m_decoded_lengthsrelevant information that text in the position of this token is incomplete (is expressed with-1).TextStreamerpublic.TextStreamerwhich use pybind to make available this object from Python for test purposes. ButTextCallbackStreameris still private and visible only for developers.Ticket: CVS-148635