[HF][streaming][4/n] Image2Text (no streaming, but lots of fixing) #855
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
This was referenced Jan 10, 2024
rossdanlm
requested review from
saqadri,
rholinshead,
suyoglastmileai,
Ankush-lastmile and
jonathanlastmileai
as code owners
added 4 commits
January 10, 2024 05:08
TSIA Adding streaming functionality to text summarization model parser ## Test Plan Rebase onto and test it with 11ace0a. Follow the README from AIConfig Editor https://github.com/lastmile-ai/aiconfig/tree/main/python/src/aiconfig/editor#dev, then run these command ```bash aiconfig_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/huggingface.aiconfig.json parsers_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/hf_model_parsers.py alias aiconfig="python3 -m 'aiconfig.scripts.aiconfig_cli'" aiconfig edit --aiconfig-path=$aiconfig_path --server-port=8080 --server-mode=debug_servers --parsers-module-path=$parsers_path ``` Then in AIConfig Editor run the prompt (it will be streaming format by default) https://github.com/lastmile-ai/aiconfig/assets/151060367/e91a1d8b-a3e9-459c-9eb1-2d8e5ec58e73
TSIA Adding streaming output support for text translation model parser. I also fixed a bug where we didn't pass in `"translation"` key into the pipeline ## Test Plan Rebase onto and test it: 5b74344. Follow the README from AIConfig Editor https://github.com/lastmile-ai/aiconfig/tree/main/python/src/aiconfig/editor#dev, then run these command ```bash aiconfig_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/huggingface.aiconfig.json parsers_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/hf_model_parsers.py alias aiconfig="python3 -m 'aiconfig.scripts.aiconfig_cli'" aiconfig edit --aiconfig-path=$aiconfig_path --server-port=8080 --server-mode=debug_servers --parsers-module-path=$parsers_path ``` With Streaming https://github.com/lastmile-ai/aiconfig/assets/151060367/d7bc9df2-2993-4709-bf9b-c5b7979fb00f Without Streaming https://github.com/lastmile-ai/aiconfig/assets/151060367/71eb6ab3-5d6f-4c5d-8b82-f3daf4c5e610
…completion params) Ok this one is weird. Today, streaming is only ever supported on text outputs in Transformers library. See `BaseStreamer` in here: https://github.com/search?q=repo%3Ahuggingface%2Ftransformers%20BaseStreamer&type=code In the future it may support other formats, but not yet. For example, OpenAI supports it: https://community.openai.com/t/streaming-from-text-to-speech-api/493784 Anyways, I basically here only did some updates to docs to clarify why completion params were null. Jonathan and I synced about this briefly ofline, but I forgot again so wanted to capture it here so no one forgets
This model parser does not support streaming (surprising!): ``` TypeError: ImageToTextPipeline._sanitize_parameters() got an unexpected keyword argument 'streamer' ``` In general, I mainly just did a lot of fixing up to make sure that this worked as expected. Things I fixed: 1. Now works for multiple images (it did before, but didn't process responses for each properly, just put the entire response) 2. Constructing responses to be in pure text output 3. Specified the completion params that are supported (only 2: https://github.com/huggingface/transformers/blob/701298d2d3d5c7bde45e71cce12736098e3f05ef/src/transformers/pipelines/image_to_text.py#L97-L102C13) Next diff I will add support for b64 encoded image format --> we need to convert this to a PIL, see https://github.com/huggingface/transformers/blob/701298d2d3d5c7bde45e71cce12736098e3f05ef/src/transformers/pipelines/image_to_text.py#L83 ## Test Plan Rebase onto and test it: 5f3b667. Follow the README from AIConfig Editor https://github.com/lastmile-ai/aiconfig/tree/main/python/src/aiconfig/editor#dev, then run these command ```bash aiconfig_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/huggingface.aiconfig.json parsers_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/hf_model_parsers.py alias aiconfig="python3 -m 'aiconfig.scripts.aiconfig_cli'" aiconfig edit --aiconfig-path=$aiconfig_path --server-port=8080 --server-mode=debug_servers --parsers-module-path=$parsers_path ``` Then in AIConfig Editor run the prompt (streaming not supported for this model so just took screenshots) These are the images I tested   Before <img width="1268" alt="Screenshot 2024-01-10 at 04 00 22" src="https://github.com/lastmile-ai/aiconfig/assets/151060367/4426f2b9-0b83-48e2-8af1-865f157ae12c"> After <img width="1277" alt="Screenshot 2024-01-10 at 04 02 01" src="https://github.com/lastmile-ai/aiconfig/assets/151060367/2ed172a8-ed26-4c1b-9a9e-5c240376a278">
saqadri
approved these changes
Jan 10, 2024
| @@ -93,10 +103,11 @@ async def deserialize( | |||
| await aiconfig.callback_manager.run_callbacks(CallbackEvent("on_deserialize_start", __name__, {"prompt": prompt, "params": params})) | |||
|
|
|||
| # Build Completion data | |||
| completion_params = self.get_model_settings(prompt, aiconfig) | |||
| model_settings = self.get_model_settings(prompt, aiconfig) | |||
| completion_params = refine_completion_params(model_settings) | |||
| prompt.outputs = [output] | ||
| await aiconfig.callback_manager.run_callbacks(CallbackEvent("on_run_complete", __name__, {"result": prompt.outputs})) | ||
| prompt.outputs = outputs | ||
| print(f"{prompt.outputs=}") |
Comment on lines
+171
to
+172
| # HuggingFace Text summarization does not support function | ||
| # calls so shouldn't get here, but just being safe |
There was a problem hiding this comment.
nit: Hugging Face image-to-text...
saqadri
added a commit
that referenced
this pull request
Jan 10, 2024
[HF][5/n] Image2Text: Allow base64 inputs for images Before we didn't allow base64, only URI (either local or http or https). This is good becuase our text2Image model parser outputs into a base64 format, so this will allow us to chain model prompts! ## Test Plan Rebase and test on 0d7ae2b. Follow the README from AIConfig Editor https://github.com/lastmile-ai/aiconfig/tree/main/python/src/aiconfig/editor#dev, then run these command ```bash aiconfig_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/huggingface.aiconfig.json parsers_path=/Users/rossdancraig/Projects/aiconfig/cookbooks/Gradio/hf_model_parsers.py alias aiconfig="python3 -m 'aiconfig.scripts.aiconfig_cli'" aiconfig edit --aiconfig-path=$aiconfig_path --server-port=8080 --server-mode=debug_servers --parsers-module-path=$parsers_path ``` Then in AIConfig Editor run the prompt (streaming not supported so just took screenshots) These are the images I tested (with bear being in base64 format)   <img width="1281" alt="Screenshot 2024-01-10 at 04 57 44" src="https://github.com/lastmile-ai/aiconfig/assets/151060367/ea60cbc5-e6ab-4bf2-82e7-17f3182fdc5c"> --- Stack created with [Sapling](https://sapling-scm.com). Best reviewed with [ReviewStack](https://reviewstack.dev/lastmile-ai/aiconfig/pull/856). * __->__ #856 * #855 * #854 * #853 * #851
rossdanlm
pushed a commit
that referenced
this pull request
Jan 10, 2024
Small fixes from comments from Sarmad + me from these diffs: - #854 - #855 - #821 Main things I did - rename `refine_chat_completion_params` --> `chat_completion_params` - edit `get_text_output` to not check for `OutputDataWithValue` - sorted the init file to be alphabetical - fixed some typos/print statements - made some error messages a bit more intuitive with prompt name - sorted some imports - fixed old class name `HuggingFaceAutomaticSpeechRecognition` --> `HuggingFaceAutomaticSpeechRecognitionTransformer` ## Test Plan These are all small nits and shouldn't change functionality
rossdanlm
added a commit
that referenced
this pull request
Jan 10, 2024
HF transformers: Small fixes nits Small fixes from comments from Sarmad + me from these diffs: - #854 - #855 - #821 Main things I did - rename `refine_chat_completion_params` --> `chat_completion_params` - edit `get_text_output` to not check for `OutputDataWithValue` - sorted the init file to be alphabetical - fixed some typos/print statements - made some error messages a bit more intuitive with prompt name - sorted some imports - fixed old class name `HuggingFaceAutomaticSpeechRecognition` --> `HuggingFaceAutomaticSpeechRecognitionTransformer` ## Test Plan These are all small nits and shouldn't change functionality
This was referenced Jan 11, 2024
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.
[HF][streaming][4/n] Image2Text (no streaming, but lots of fixing)
This model parser does not support streaming (surprising!):
In general, I mainly just did a lot of fixing up to make sure that this worked as expected. Things I fixed:
Next diff I will add support for b64 encoded image format --> we need to convert this to a PIL, see https://github.com/huggingface/transformers/blob/701298d2d3d5c7bde45e71cce12736098e3f05ef/src/transformers/pipelines/image_to_text.py#L83
Test Plan
Rebase onto and test it: 5f3b667.
Follow the README from AIConfig Editor https://github.com/lastmile-ai/aiconfig/tree/main/python/src/aiconfig/editor#dev, then run these command
Then in AIConfig Editor run the prompt (streaming not supported for this model so just took screenshots)
These are the images I tested
Before
After
Stack created with Sapling. Best reviewed with ReviewStack.