Vivek Silimkhan
f998bf4a4f
feat: add Amazon Bedrock support ( #6226 )
...
* Add Bedrock
* Update supported models for Bedrock
* Fix supports and add extract response in Bedrock
* fix errors imports
* improve and refactor supports
* fix install
* fix mypy
* fix pylint
* fix existing tests
* Added Anthropic Bedrock
* fix tests
* fix sagemaker tests
* add default prompt handler, constructor and supports tests
* more tests
* invoke refactoring
* refactor model_kwargs
* fix mypy
* lstrip responses
* Add streaming support
* bump boto3 version
* add class docstrings, better exception names
* fix layer name
* add tests for anthropic and cohere model adapters
* update cohere params
* update ai21 args and add tests
* support cohere command light model
* add tital tests
* better class names
* support meta llama 2 model
* fix streaming support
* more future-proof model adapter selection
* fix import
* fix mypy
* fix pylint for preview
* add tests for streaming
* add release notes
* Apply suggestions from code review
Co-authored-by: Agnieszka Marzec <97166305+agnieszka-m@users.noreply.github.com>
* fix format
* fix tests after msg changes
* fix streaming for cohere
---------
Co-authored-by: tstadel <60758086+tstadel@users.noreply.github.com>
Co-authored-by: tstadel <thomas.stadelmann@deepset.ai>
Co-authored-by: Agnieszka Marzec <97166305+agnieszka-m@users.noreply.github.com>
2023-11-15 13:26:29 +01:00
Stefano Fiorucci
ccc9f010bb
fix: fix ChatGPT invocation layer (and add async support) ( #5979 )
...
* ChatGPT async
* release note
* fix tests
2023-10-05 18:43:26 +02:00
Massimiliano Pippi
ac408134f4
feat: add support for async openai calls ( #5946 )
...
* add support for async openai calls
* add actual async call
* split the async api
* ask permission
* Update haystack/utils/openai_utils.py
Co-authored-by: Silvano Cerza <3314350+silvanocerza@users.noreply.github.com>
* Fix OpenAI content moderation tests
* Fix ChatGPT invocation layer tests
---------
Co-authored-by: Silvano Cerza <3314350+silvanocerza@users.noreply.github.com>
Co-authored-by: Silvano Cerza <silvanocerza@gmail.com>
2023-10-03 10:42:21 +02:00
Malte Pietsch
aa3cc3d5ae
feat: Add support for OpenAI's gpt-3.5-turbo-instruct model ( #5837 )
...
* support gpt-3.5.-turbo-instruct
* add release note
2023-09-19 16:06:43 +02:00
Christian Clauss
1bc03ddc73
ci: Fix all ruff pyflakes errors except unused imports ( #5820 )
...
* ci: Fix all ruff pyflakes errors except unused imports
* Delete releasenotes/notes/fix-some-pyflakes-errors-69a1106efa5d0203.yaml
2023-09-15 18:30:33 +02:00
Fanli Lin
40d9f34e68
feat: enable passing use_fast to the underlying transformers' pipeline ( #5655 )
...
* copy instead of deepcopy
* fix pylint
* add use_fast
* add release note
* remove unrelevant changes
* black fix
* fix bug
* black
* bug fix
2023-08-30 10:25:18 +02:00
Fanli Lin
f6b50cfdf9
fix: StopWordsCriteria doesn't compare the stop word token ids with the input ids in a continuous and sequential order ( #5503 )
...
* bug fix
* add release note
* add unit test
* refactor
---------
Co-authored-by: Silvano Cerza <3314350+silvanocerza@users.noreply.github.com>
Co-authored-by: Vladimir Blagojevic <dovlex@gmail.com>
2023-08-08 08:35:10 +02:00
Vladimir Blagojevic
d96c963bc4
test: Convert two HFLocalInvocationLayer integration to unit tests ( #5446 )
...
* Convert two HFLocalInvocationLayer integration to unit tests
* Simplify unit test
* Improve HFLocalInvocationLayer unit tests
2023-08-03 17:41:32 +02:00
Fanli Lin
8d04f28e11
fix: hf agent outputs the prompt text while the openai agent not ( #5461 )
...
* add skil prompt
* fix formatting
* add release note
* add release note
* Update releasenotes/notes/add-skip-prompt-for-hf-model-agent-89aef2838edb907c.yaml
Co-authored-by: Daria Fokina <daria.f93@gmail.com>
* Update haystack/nodes/prompt/invocation_layer/handlers.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update haystack/nodes/prompt/invocation_layer/handlers.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update haystack/nodes/prompt/invocation_layer/hugging_face.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* add a unit test
* add a unit test2
* add skil prompt
* Revert "add skil prompt"
This reverts commit b1ba938c94b67a4fd636d321945990aabd2c5b2a.
* add unit test
---------
Co-authored-by: Daria Fokina <daria.f93@gmail.com>
Co-authored-by: bogdankostic <bogdankostic@web.de>
2023-08-02 16:34:33 +02:00
Fanli Lin
73fa796735
fix: enable passing max_length for text2text-generation task ( #5420 )
...
* bug fix
* add unit test
* reformatting
* add release note
* add release note
* Update releasenotes/notes/enable-set-max-length-during-runtime-097d65e537bf800b.yaml
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update test/prompt/invocation_layer/test_hugging_face.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update test/prompt/invocation_layer/test_hugging_face.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update test/prompt/invocation_layer/test_hugging_face.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update test/prompt/invocation_layer/test_hugging_face.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* bug fix
---------
Co-authored-by: bogdankostic <bogdankostic@web.de>
2023-08-02 14:13:30 +02:00
Fanli Lin
f7fd5eeb4f
feat: enable loading tokenizer for models that are not supported by the transformers library ( #5314 )
...
* add tokenizer load
* change import order
* move imports
* refactor code
* import lib
* remove pretrainedmodel
* fix linting
* update patch
* fix order
* remove tokenizer class
* use tokenizer class
* no copy
* add case for model is an instance
* fix optional
* add ut
* set default to None
* change models
* Update haystack/nodes/prompt/invocation_layer/hugging_face.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update haystack/nodes/prompt/invocation_layer/hugging_face.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* add unit tests
* add unit tests
* remove lib
* formatting
* formatting
* formatting
* add release note
* Update releasenotes/notes/load-tokenizer-if-not-load-by-transformers-5841cdc9ff69bcc2.yaml
Co-authored-by: bogdankostic <bogdankostic@web.de>
---------
Co-authored-by: bogdankostic <bogdankostic@web.de>
2023-08-02 11:42:23 +02:00
Stefano Fiorucci
6f534873a5
fix: restrict supports method in the OpenAI invocation layer and a similar method in the EmbeddingRetriever ( #5458 )
...
* restrict OpenAI supports method
* better note
* Update releasenotes/notes/restrict-openai-supports-method-fb126583e4beb057.yaml
Co-authored-by: Daria Fokina <daria.fokina@deepset.ai>
---------
Co-authored-by: Daria Fokina <daria.fokina@deepset.ai>
2023-07-31 13:14:22 +02:00
Vladimir Blagojevic
409e3471cb
feat: Enable Support for Meta LLama-2 Models in Amazon Sagemaker ( #5437 )
...
* Enable Support for Meta LLama-2 Models in Amazon Sagemaker
* Improve unit test for invocation layers positioning
* Small adjustment, add more unit tests
* mypy fixes
* Improve unit tests
* Update test/prompt/invocation_layer/test_sagemaker_meta.py
Co-authored-by: Stefano Fiorucci <44616784+anakin87@users.noreply.github.com>
* PR feedback
* Add pydocs for newly extracted methods
* simplify is_proper_chat_*
---------
Co-authored-by: Stefano Fiorucci <44616784+anakin87@users.noreply.github.com>
Co-authored-by: anakin87 <stefanofiorucci@gmail.com>
2023-07-26 15:26:39 +02:00
Julian Risch
5bb0a1f57a
Revert "fix: num_return_sequences should be less than num_beams, not top_k ( #5280 )" ( #5434 )
...
This reverts commit 514f93a6eb575d376b21d22e32080fac62cf785f.
2023-07-25 13:27:41 +02:00
Vladimir Blagojevic
597df1414c
feat: Update Anthropic Claude support with the latest models, new streaming API, context window sizes ( #5406 )
...
* Update Claude support with the latest models, new streaming API, context window sizes
* Use Github Anthropic SDK link for tokenizer, revert _init_tokenizer
* Change example key name to ANTHROPIC_API_KEY
2023-07-21 13:33:07 +02:00
Vladimir Blagojevic
adfabdd648
Improve token limit tests for OpenAI PromptNode layer ( #5351 )
2023-07-17 14:03:03 +02:00
Fanli Lin
9891bfeddd
fix: a small bug in StopWordsCriteria ( #5316 )
2023-07-13 15:58:06 +02:00
MichelBartels
fd350bbb8f
fix: Run HFLocalInvocationLayer.supports even if inference packages are not installed ( #5308 )
...
---------
Co-authored-by: Vladimir Blagojevic <dovlex@gmail.com>
2023-07-13 12:52:56 +02:00
Fanli Lin
514f93a6eb
fix: num_return_sequences should be less than num_beams, not top_k ( #5280 )
...
* formatting
* remove top_k variable
* add pytest
* add numbers
* string formatting
* fix formatting
* revert
* extend tests with assertions for num_return_sequences
---------
Co-authored-by: Julian Risch <julian.risch@deepset.ai>
2023-07-11 12:20:21 +02:00
Stefano Fiorucci
90ff3817e7
feat: support OpenAI-Organization for authentication ( #5292 )
...
* add openai_organization to invocation layer, generator and retriever
* added tests
2023-07-07 12:02:21 +02:00
MichelBartels
08f1865ddd
fix: Improve robustness of get_task HF pipeline invocations ( #5284 )
...
* replace get_task method and change invocation layer order
* add test for invocation layer order
* add test documentation
* make invocation layer test more robust
* fix type annotation
* change hf timeout
* simplify timeout mock and add get_task exception cause
---------
Co-authored-by: Stefano Fiorucci <44616784+anakin87@users.noreply.github.com>
2023-07-06 16:33:44 +02:00
Vladimir Blagojevic
5ee393226d
fix: Support all SageMaker HF text generation models (other than Falcon) ( #5205 )
...
* Create SageMaker base class and two implementation subclasses
---------
Co-authored-by: Stefano Fiorucci <44616784+anakin87@users.noreply.github.com>
2023-06-26 19:59:16 +02:00
Vladimir Blagojevic
eb2255c0dd
Rename SageMakerInvocationLayer -> SageMakerHFTextGenerationInvocationLayer ( #5204 )
2023-06-26 11:03:30 +02:00
Malte Pietsch
c9179ed0eb
feat: enable LLMs hosted via AWS SageMaker in PromptNode ( #5155 )
...
* Add SageMakerInvocationLayer
---------
Co-authored-by: oryx1729 <78848855+oryx1729@users.noreply.github.com>
Co-authored-by: Vladimir Blagojevic <dovlex@gmail.com>
2023-06-23 15:33:20 +02:00
Sebastian
7a140c1524
feat: add ensure token limit for direct prompting of ChatGPT ( #5166 )
...
* Add support for prompt truncation when using chatgpt if direct prompting is used
* Update tests for test token limit for prompt node
* Update warning message to be correct
* Minor cleanup
* Mark back to integration
* Update count_openai_tokens_messages to reflect changes shown in tiktoken
* Use mocking to avoid request call
* Fix test to make it comply with unit test requirements
* Move tests to respective invocation layers
* Moved fixture to one spot
2023-06-21 15:41:28 +02:00
Julian Risch
ce1c9c9ddb
fix: Relax ChatGPT model name check to support gpt-3.5-turbo-0613 ( #5142 )
...
* relax model name checking for chatgpt
* add unit tests
2023-06-14 09:53:00 +02:00
ZanSara
65cdf36d72
chore: block all HTTP requests in CI ( #5088 )
2023-06-13 14:52:24 +02:00
Vladimir Blagojevic
e3b069620b
feat: pass model parameters to HFLocalInvocationLayer via model_kwargs, enabling direct model usage ( #4956 )
...
* Simplify HFLocalInvocationLayer, move/add unit tests
* PR feedback
* Better pipeline invocation, add mocked tests
* Minor improvements
* Mock pipeline directly, unit test updates
* PR feedback, change pytest type to integration
* Mock supports unit test
* add full stop
* PR feedback, improve unit tests
* Add mock_get_task fixture
* Further improve unit tests
* Minor unit test improvement
* Add unit tests, increase coverage
* Add unit tests, increase test coverage
* Small optimization, improve _ensure_token_limit unit test
---------
Co-authored-by: Darja Fokina <daria.f93@gmail.com>
2023-06-07 13:34:45 +02:00
Silvano Cerza
a2156ee8fb
fix: Fix handling of streaming response in AnthropicClaudeInvocationLayer ( #4993 )
...
* Fix handling of streaming response in AnthropicClaudeInvocationLayer
---------
Co-authored-by: Vladimir Blagojevic <dovlex@gmail.com>
Co-authored-by: Darja Fokina <daria.f93@gmail.com>
2023-06-07 10:57:36 +02:00
Michael Feil
6ea8ae01a2
feat: Allow setting custom api_base for OpenAI nodes ( #5033 )
...
* add changes for api_base
* format retriever
* Update haystack/nodes/retriever/dense.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update haystack/nodes/audio/whisper_transcriber.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update haystack/preview/components/audio/whisper_remote.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update haystack/nodes/answer_generator/openai.py
Co-authored-by: bogdankostic <bogdankostic@web.de>
* Update test_retriever.py
* Update test_whisper_remote.py
* Update test_generator.py
* Update test_retriever.py
* reformat with black
* Update haystack/nodes/prompt/invocation_layer/chatgpt.py
Co-authored-by: Daria Fokina <daria.f93@gmail.com>
* Add unit tests
* apply docstring suggestions
---------
Co-authored-by: bogdankostic <bogdankostic@web.de>
Co-authored-by: michaelfeil <me@michaelfeil.eu>
Co-authored-by: Daria Fokina <daria.f93@gmail.com>
2023-06-05 11:32:06 +02:00
Silvano Cerza
524d2cba36
Fix CohereInvocationLayer _ensure_token_limit not returning resized ( #4978 )
...
prompt
2023-05-23 17:58:01 +02:00
Vladimir Blagojevic
068a967e5b
feat: HFInferenceEndpointInvocationLayer streaming support ( #4819 )
...
* HFInferenceEndpointInvocationLayer streaming support
* Small fixes
* Add unit test
* PR feedback
* Alphabetically sort params
* Convert PromptNode tests to HFInferenceEndpointInvocationLayer invoke tests
* Rewrite streaming with sseclient
* More PR updates
* Implement and test _ensure_token_limit
* Further optimize DefaultPromptHandler
* Fix CohereInvocationLayer mistypes
* PR feedback
* Break up unit tests, simplify
* Simplify unit tests even further
* PR feedback on unit test simplification
* Proper code identation under patch context manager
* More unit tests, slight adjustments
* Remove unrelated CohereInvocationLayer change
This reverts commit 82337151e8328d982f738e5da9129ff99350ea0c.
* Revert "Further optimize DefaultPromptHandler"
This reverts commit 606a761b6e3333f27df51a304cfbd1906c806e05.
* lg update
mostly full stops at the end of docstrings
---------
Co-authored-by: Silvano Cerza <3314350+silvanocerza@users.noreply.github.com>
Co-authored-by: Silvano Cerza <silvanocerza@gmail.com>
Co-authored-by: Darja Fokina <daria.f93@gmail.com>
2023-05-22 14:45:53 +02:00
Vladimir Blagojevic
73380b194a
feat: Add Cohere PromptNode invocation layer ( #4827 )
...
* Add CohereInvocationLayer
---------
Co-authored-by: bogdankostic <bogdankostic@web.de>
2023-05-12 17:50:09 +02:00
Silvano Cerza
98947e4c3c
feat: Add Anthropic invocation layer ( #4818 )
...
* feat: Add Anthropic Claude Invocation Layer
* feat: Add AnthropicClaude Invocation Layer
* fix: Permission changes
* fix: Permission changes
* Move anthropic utils in anthropic invocation layer file
* Rework method to post data
* Simplify invoke
* Simplify supports classmethod
* Remove unnecessary functions
* Use always same tokenizer
* Add module import
* Rename some members and kwargs
* Add tests
* Fix _post not handling HTTPError
* Fix handling of streamed response
* Fix kwargs handling
* Update tests
* Update supports to be generic
* Fix failing test
* Use correct tokenizer and fix tests
* Update lg
* Fix mypy issue
* Move requests-cache from dev to base dependencies
* Fix failing test
* Handle all stop words use cases
---------
Co-authored-by: recrudesce <recrudesce@gmail.com>
Co-authored-by: agnieszka-m <amarzec13@gmail.com>
2023-05-11 10:14:33 +02:00
