2806 Commits (3984f6604f34a01dfc40116f522d39e547d7992f)

Author	SHA1	Message	Date
joelsprunger	3984f6604f	langchain: adds recursive json splitter (#17144) ... - **Description:** This adds a recursive json splitter class to the existing text_splitters as well as unit tests - **Issue:** splitting text from structured data can cause issues if you have a large nested json object and you split it as regular text you may end up losing the structure of the json. To mitigate against this you can split the nested json into large chunks and overlap them, but this causes unnecessary text processing and there will still be times where the nested json is so big that the chunks get separated from the parent keys. As an example you wouldn't want the following to be split in half: ```shell {'val0': 'DFWeNdWhapbR', 'val1': {'val10': 'QdJo', 'val11': 'FWSDVFHClW', 'val12': 'bkVnXMMlTiQh', 'val13': 'tdDMKRrOY', 'val14': 'zybPALvL', 'val15': 'JMzGMNH', 'val16': {'val160': 'qLuLKusFw', 'val161': 'DGuotLh', 'val162': 'KztlcSBropT', -----------------------------------------------------------------------split----- 'val163': 'YlHHDrN', 'val164': 'CtzsxlGBZKf', 'val165': 'bXzhcrWLmBFp', 'val166': 'zZAqC', 'val167': 'ZtyWno', 'val168': 'nQQZRsLnaBhb', 'val169': 'gSpMbJwA'}, 'val17': 'JhgiyF', 'val18': 'aJaqjUSFFrI', 'val19': 'glqNSvoyxdg'}} ``` Any llm processing the second chunk of text may not have the context of val1, and val16 reducing accuracy. Embeddings will also lack this context and this makes retrieval less accurate. Instead you want it to be split into chunks that retain the json structure. ```shell {'val0': 'DFWeNdWhapbR', 'val1': {'val10': 'QdJo', 'val11': 'FWSDVFHClW', 'val12': 'bkVnXMMlTiQh', 'val13': 'tdDMKRrOY', 'val14': 'zybPALvL', 'val15': 'JMzGMNH', 'val16': {'val160': 'qLuLKusFw', 'val161': 'DGuotLh', 'val162': 'KztlcSBropT', 'val163': 'YlHHDrN', 'val164': 'CtzsxlGBZKf'}}} ``` and ```shell {'val1':{'val16':{ 'val165': 'bXzhcrWLmBFp', 'val166': 'zZAqC', 'val167': 'ZtyWno', 'val168': 'nQQZRsLnaBhb', 'val169': 'gSpMbJwA'}, 'val17': 'JhgiyF', 'val18': 'aJaqjUSFFrI', 'val19': 'glqNSvoyxdg'}} ``` This recursive json text splitter does this. Values that contain a list can be converted to dict first by using split(... convert_lists=True) otherwise long lists will not be split and you may end up with chunks larger than the max chunk. In my testing large json objects could be split into small chunks with ✅ Increased question answering accuracy ✅ The ability to split into smaller chunks meant retrieval queries can use fewer tokens - **Dependencies:** json import added to text_splitter.py, and random added to the unit test - **Twitter handle:** @joelsprunger --------- Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>	8 months ago
Leonid Kuligin	1862900078	google-genai[patch]: added parsing of function call / response (#17245)	8 months ago
Cailin Wang	a210a8bc53	langchain[patch]: Fix create_retriever_tool missing on_retriever_end Document content (#16933) ... - **Description:** In create_retriever_tool create_tool, fix create_retriever_tool's missing Document content for on_retriever_end, caused by create_retriever_tool's missing callbacks parameter, - **Twitter handle:** @CailinWang_ --------- Co-authored-by: root <root@Bluedot-AI> Co-authored-by: Bagatur <baskaryan@gmail.com>	8 months ago
Sparsh Jain	a2167614b7	google-genai[patch]: Invoke callback prior to yielding token (#17092) ... - **Description:** Invoke callback prior to yielding token in stream and astream methods for Google-genai, - **Issue:** the issue # 16913, - **Twitter handle:** Sparsh10649446 --------- Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>	8 months ago
Liang Zhang	7306600e2f	community[patch]: Support SerDe transform functions in Databricks LLM (#16752) ... **Description:** Databricks LLM does not support SerDe the transform_input_fn and transform_output_fn. After saving and loading, the LLM will be broken. This PR serialize these functions into a hex string using pickle, and saving the hex string in the yaml file. Using pickle to serialize a function can be flaky, but this is a simple workaround that unblocks many use cases. If more sophisticated SerDe is needed, we can improve it later. Test: Added a simple unit test. I did manual test on Databricks and it works well. The saved yaml looks like: ``` llm: _type: databricks cluster_driver_port: null cluster_id: null databricks_uri: databricks endpoint_name: databricks-mixtral-8x7b-instruct extra_params: {} host: e2-dogfood.staging.cloud.databricks.com max_tokens: null model_kwargs: null n: 1 stop: null task: null temperature: 0.0 transform_input_fn: 80049520000000000000008c085f5f6d61696e5f5f948c0f7472616e73666f726d5f696e7075749493942e transform_output_fn: null ``` @baskaryan ```python from langchain_community.embeddings import DatabricksEmbeddings from langchain_community.llms import Databricks from langchain.chains import RetrievalQA from langchain.document_loaders import TextLoader from langchain.text_splitter import CharacterTextSplitter from langchain.vectorstores import FAISS import mlflow embeddings = DatabricksEmbeddings(endpoint="databricks-bge-large-en") def transform_input(**request): request["messages"] = [ { "role": "user", "content": request["prompt"] } ] del request["prompt"] return request llm = Databricks(endpoint_name="databricks-mixtral-8x7b-instruct", transform_input_fn=transform_input) persist_dir = "faiss_databricks_embedding" # Create the vector db, persist the db to a local fs folder loader = TextLoader("state_of_the_union.txt") documents = loader.load() text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0) docs = text_splitter.split_documents(documents) db = FAISS.from_documents(docs, embeddings) db.save_local(persist_dir) def load_retriever(persist_directory): embeddings = DatabricksEmbeddings(endpoint="databricks-bge-large-en") vectorstore = FAISS.load_local(persist_directory, embeddings) return vectorstore.as_retriever() retriever = load_retriever(persist_dir) retrievalQA = RetrievalQA.from_llm(llm=llm, retriever=retriever) with mlflow.start_run() as run: logged_model = mlflow.langchain.log_model( retrievalQA, artifact_path="retrieval_qa", loader_fn=load_retriever, persist_dir=persist_dir, ) # Load the retrievalQA chain loaded_model = mlflow.pyfunc.load_model(logged_model.model_uri) print(loaded_model.predict([{"query": "What did the president say about Ketanji Brown Jackson"}])) ```	8 months ago
cjpark-data	ce22e10c4b	community[patch]: Fix KeyError 'embedding' (MongoDBAtlasVectorSearch) (#17178) ... - **Description:** Embedding field name was hard-coded named "embedding". So I suggest that change `res["embedding"]` into `res[self._embedding_key]`. - **Issue:** #17177, - **Twitter handle:** [@bagcheoljun17](https://twitter.com/bagcheoljun17)	8 months ago
Neli Hateva	9bb5157a3d	langchain[patch], community[patch]: Fixes in the Ontotext GraphDB Graph and QA Chain (#17239) ... - **Description:** Fixes in the Ontotext GraphDB Graph and QA Chain related to the error handling in case of invalid SPARQL queries, for which `prepareQuery` doesn't throw an exception, but the server returns 400 and the query is indeed invalid - **Issue:** N/A - **Dependencies:** N/A - **Twitter handle:** @OntotextGraphDB	8 months ago
ByeongUk Choi	b88329e9a5	community[patch]: Implement Unique ID Enforcement in FAISS (#17244) ... **Description:** Implemented unique ID validation in the FAISS component to ensure all document IDs are distinct. This update resolves issues related to non-unique IDs, such as inconsistent behavior during deletion processes.	8 months ago
Bagatur	852973d616	langchain[minor], core[minor]: update json, pydantic parser. add openai-json structured output runnable (#16914)	8 months ago
hsuyuming	e22c4d4eb0	google-vertexai[patch]: fix _parse_response_candidate issue (#16647) ... **Description:** enable _parse_response_candidate to support complex structure format. **Issue:** currently, if Gemini response complex args format, people will get "TypeError: Object of type RepeatedComposite is not JSON serializable" error from _parse_response_candidate. response candidate example ``` content { role: "model" parts { function_call { name: "Information" args { fields { key: "people" value { list_value { values { string_value: "Joe is 30, his mom is Martha" } } } } } } } } finish_reason: STOP safety_ratings { category: HARM_CATEGORY_HARASSMENT probability: NEGLIGIBLE } safety_ratings { category: HARM_CATEGORY_HATE_SPEECH probability: NEGLIGIBLE } safety_ratings { category: HARM_CATEGORY_SEXUALLY_EXPLICIT probability: NEGLIGIBLE } safety_ratings { category: HARM_CATEGORY_DANGEROUS_CONTENT probability: NEGLIGIBLE } ``` error msg: ``` Traceback (most recent call last): File "/home/jupyter/user/abehsu/gemini_langchain_tools/example2.py", line 36, in <module> print(tagging_chain.invoke({"input": "Joe is 30, his mom is Martha"})) File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_core/runnables/base.py", line 2053, in invoke input = step.invoke( File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_core/runnables/base.py", line 3887, in invoke return self.bound.invoke( File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_core/language_models/chat_models.py", line 165, in invoke self.generate_prompt( File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_core/language_models/chat_models.py", line 543, in generate_prompt return self.generate(prompt_messages, stop=stop, callbacks=callbacks, **kwargs) File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_core/language_models/chat_models.py", line 407, in generate raise e File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_core/language_models/chat_models.py", line 397, in generate self._generate_with_cache( File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_core/language_models/chat_models.py", line 576, in _generate_with_cache return self._generate( File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_google_vertexai/chat_models.py", line 406, in _generate generations = [ File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_google_vertexai/chat_models.py", line 408, in <listcomp> message=_parse_response_candidate(c), File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/site-packages/langchain_google_vertexai/chat_models.py", line 280, in _parse_response_candidate function_call["arguments"] = json.dumps( File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/json/__init__.py", line 231, in dumps return _default_encoder.encode(obj) File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/json/encoder.py", line 199, in encode chunks = self.iterencode(o, _one_shot=True) File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/json/encoder.py", line 257, in iterencode return _iterencode(o, 0) File "/opt/conda/envs/gemini_langchain_tools/lib/python3.10/json/encoder.py", line 179, in default raise TypeError(f'Object of type {o.__class__.__name__} ' TypeError: Object of type RepeatedComposite is not JSON serializable ``` **Twitter handle:** @abehsu1992626	8 months ago
Erick Friis	d77bb7b4e9	google-vertexai[patch]: integration test fix, release 0.0.5 (#17258)	8 months ago
Aditya	98176ac982	langchain_google_vertexai : added logic to override get_num_tokens_from_messages() for ChatVertexAI (#16784) ... <!-- Thank you for contributing to LangChain! Replace this entire comment with: - **Description: added logic to override get_num_tokens_from_messages() for ChatVertexAI. Currently ChatVertexAI was inheriting get_num_tokens_from_messages() from BaseChatModel which in-turn was calling GPT-2 tokenizer - **Issue: NA - **Dependencies: NA - **Twitter handle:@aditya_rane @lkuligin for review --------- Co-authored-by: adityarane@google.com <adityarane@google.com> Co-authored-by: Leonid Kuligin <lkuligin@yandex.ru>	8 months ago
Bassem Yacoube	4e3ed7f043	community[patch]: octoai embeddings bug fix (#17216) ... fixes a bug in octoa_embeddings provider	8 months ago
Eugene Yurtsev	780e84ae79	community[minor]: SQLDatabase Add fetch mode `cursor`, query parameters, query by selectable, expose execution options, and documentation (#17191) ... - **Description:** Improve `SQLDatabase` adapter component to promote code re-use, see [suggestion](https://github.com/langchain-ai/langchain/pull/16246#pullrequestreview-1846590962). - **Needed by:** GH-16246 - **Addressed to:** @baskaryan, @cbornet ## Details - Add `cursor` fetch mode - Accept SQL query parameters - Accept both `str` and SQLAlchemy selectables as query expression - Expose `execution_options` - Documentation page (notebook) about `SQLDatabase` [^1] See [About SQLDatabase](https://github.com/langchain-ai/langchain/blob/c1c7b763/docs/docs/integrations/tools/sql_database.ipynb). [^1]: Apparently there hasn't been any yet? --------- Co-authored-by: Andreas Motl <andreas.motl@crate.io>	8 months ago
Tomaz Bratanic	7e4b676d53	community[patch]: Better error propagation for neo4jgraph (#17190) ... There are other errors that could happen when refreshing the schema, so we want to propagate specific errors for more clarity	8 months ago
Luiz Ferreira	34d2daffb3	community[patch]: Fix chat openai unit test (#17124) ... - **Description:** Actually the test named `test_openai_apredict` isn't testing the apredict method from ChatOpenAI. - **Twitter handle:** https://twitter.com/OAlmofadas	8 months ago
Dmitry Kankalovich	f92738a6f6	langchain[minor], community[minor], core[minor]: Async Cache support and AsyncRedisCache (#15817) ... * This PR adds async methods to the LLM cache. * Adds an implementation using Redis called AsyncRedisCache. * Adds a docker compose file at the /docker to help spin up docker * Updates redis tests to use a context manager so flushing always happens by default	8 months ago
Erick Friis	4153837502	google-genai[patch]: release 0.0.7 (#17193)	8 months ago
Erick Friis	927ab77d6e	google-genai[patch]: no error for FunctionMessage (#17215) ... Both should eventually match this: https://github.com/langchain-ai/langchain/blob/master/libs/partners/google-vertexai/langchain_google_vertexai/chat_models.py#L179 But seems undocumented / can't find types in genai package	8 months ago
Erick Friis	2ecf318218	google-genai[patch]: match function call interface (#17213) ... should match vertex	8 months ago
Erick Friis	e17173c403	google-vertexai[patch]: function calling integration test (#17209)	8 months ago
Erick Friis	52be84a603	google-vertexai[patch]: serializable citation metadata, release 0.0.4 (#17145) ... was breaking in langserve before	8 months ago
Nuno Campos	19ff81e74f	Fix stream events/log with some kinds of non addable output (#17205) ... <!-- Thank you for contributing to LangChain! Please title your PR "<package>: <description>", where <package> is whichever of langchain, community, core, experimental, etc. is being modified. Replace this entire comment with: - **Description:** a description of the change, - **Issue:** the issue # it fixes if applicable, - **Dependencies:** any dependencies required for this change, - **Twitter handle:** we announce bigger features on Twitter. If your PR gets announced, and you'd like a mention, we'll gladly shout you out! Please make sure your PR is passing linting and testing before submitting. Run `make format`, `make lint` and `make test` from the root of the package you've modified to check this locally. See contribution guidelines for more information on how to write/run tests, lint, etc: https://python.langchain.com/docs/contributing/ If you're adding a new integration, please include: 1. a test for the integration, preferably unit tests that do not rely on network access, 2. an example notebook showing its use. It lives in `docs/docs/integrations` directory. If no one reviews your PR within a few days, please @-mention one of @baskaryan, @eyurtsev, @hwchase17. -->	8 months ago
Bagatur	6f1403b9b6	community[patch]: Release 0.0.19 (#17207) ... Co-authored-by: Erick Friis <erick@langchain.dev>	8 months ago
Erick Friis	a13dc47a08	cli[patch]: copyright 2024 default (#17204)	8 months ago
Bagatur	00757567ba	core[patch]: Release 0.1.21 (#17202)	8 months ago
Bagatur	af74301ab9	core[patch], community[patch]: link extraction continue on failure (#17200)	8 months ago
Henry	2281f00198	langchain: Standardize `output_parser.py` across all agent types for custom `FORMAT_INSTRUCTIONS` (#17168) ... - **Description:** This PR standardizes the `output_parser.py` file across all agent types to ensure a uniform parsing mechanism is implemented. It introduces a cohesive structure and common interface for output parsing, facilitating easier modifications and extensions by users. The standardized approach enhances maintainability and scalability of the codebase by providing a consistent pattern for output parsing, which can be easily understood and utilized across different agent types. This PR builds upon the foundation set by a previously merged PR, which focused exclusively on standardizing the `output_parser.py` for the `conversational_agent` ([PR #16945](https://github.com/langchain-ai/langchain/pull/16945)). With this new update, I extend the standardization efforts to encompass `output_parser.py` files across all agent types. This enhancement not only unifies the parsing mechanism across the board but also introduces the flexibility for users to incorporate custom `FORMAT_INSTRUCTIONS`. - **Issue:** https://github.com/langchain-ai/langchain/issues/10721 https://github.com/langchain-ai/langchain/issues/4044 - **Dependencies:** No new dependencies required for this change - **Twitter handle:** With my github user is enough. Thanks I hope you accept my PR.	8 months ago
Bagatur	78409634fe	core[patch]: Release 0.1.20 (#17194)	8 months ago
Nuno Campos	65798289a4	core[minor]: Use batched tracing in sdk (#16305) ... Remove threadpool executor usage in langchain tracer, this is now handled by sdk	8 months ago
chyroc	f87b38a559	google-genai[minor]: support functions call (#15146) ... Co-authored-by: Erick Friis <erick@langchain.dev>	8 months ago
Tomaz Bratanic	302989a2b1	allow optional newline in the action responses of JSON Agent parser (#17186) ... Based on my experiments, the newline isn't always there, so we can make the regex slightly more robust by allowing an optional newline after the bacticks	8 months ago
William FH	9fa07076da	Add trace_as_chain_group metadata (#17187)	8 months ago
Erick Friis	3e58df43c2	mistralai[patch]: release 0.0.4 (#17139)	8 months ago
Erick Friis	22b6a03a28	infra: read min versions (#17135)	8 months ago
Erick Friis	f881a3330c	mistralai[patch]: 16k token batching logic embed (#17136)	8 months ago
Bagatur	226f376d59	community[patch]: Release 0.0.18 (#17129) ... Co-authored-by: Erick Friis <erick@langchain.dev>	8 months ago
Erick Friis	980e30c361	nvidia-ai-endpoints[patch]: release 0.0.2 (#17125)	8 months ago
Erick Friis	15bd1154a7	pinecone[patch]: integration test new namespace (#17121)	8 months ago
Mikhail Khludnev	14ff1438e6	nvidia-trt[patch]: propagate InferenceClientException to the caller. (#16936) ... - **Description:** before the change I've got 1. propagate InferenceClientException to the caller. 2. stop grpc receiver thread on exception ``` for token in result_queue: > result_str += token E TypeError: can only concatenate str (not "InferenceServerException") to str ../../langchain_nvidia_trt/llms.py:207: TypeError ``` And stream thread keeps running. after the change request thread stops correctly and caller got a root cause exception: ``` E tritonclient.utils.InferenceServerException: [request id: 4529729] expected number of inputs between 2 and 3 but got 10 inputs for model 'vllm_model' ../../langchain_nvidia_trt/llms.py:205: InferenceServerException ``` - **Issue:** the issue # it fixes if applicable, - **Dependencies:** any dependencies required for this change, - **Twitter handle:** [t.me/mkhl_spb](https://t.me/mkhl_spb) I'm not sure about test coverage. Should I setup deep mocks or there's a kind of triton stub via testcontainers or so.	8 months ago
Junyoung Park	1ed73f1992	community[minor]: Add SelfQueryRetriever support to PGVector (#16991) ... - **Description:** Add SelfQueryRetriever support to PGVector - **Issue:** - - **Dependencies:** - - **Twitter handle:** - --------- Co-authored-by: Bagatur <baskaryan@gmail.com>	8 months ago
Bagatur	cd945e3a5b	core[patch]: Release 0.1.19 (#17117)	8 months ago
Frank	ef082c77b1	community[minor]: add github file loader to load any github file content b… (#15305) ... ### Description support load any github file content based on file extension. Why not use [git loader](https://python.langchain.com/docs/integrations/document_loaders/git#load-existing-repository-from-disk) ? git loader clones the whole repo even only interested part of files, that's too heavy. This GithubFileLoader only downloads that you are interested files. ### Twitter handle my twitter: @shufanhaotop --------- Co-authored-by: Hao Fan <h_fan@apple.com> Co-authored-by: Bagatur <baskaryan@gmail.com>	8 months ago
Henry	eaeb8a5f71	langchain[patch]: `output_parser.py` in conversation_chat is customizable (#16945) ... **Description:** With this modification, users can customize the `FORMAT_INSTRUCTIONS` template, allowing them to create their own prompts As it is happening in [this](https://github.com/langchain-ai/langchain/issues/10721) issue, the `FORMAT_INSTRUCTIONS` is not customizable for the output parser, unless you create your own class `ConvoOutputParser`. To avoid this, a modification was done, creating a `format_instruction` variable that users can customize with ease after initialize the agent. For example: ``` agent = initialize_agent( agent = AgentType.CHAT_CONVERSATIONAL_REACT_DESCRIPTION, tools = tools, llm = llm_agent, verbose = True, max_iterations = 3, early_stopping_method = 'generate', memory = b_w_memory, handle_parsing_errors = True, agent_kwargs={ 'system_message':PREFIX, 'human_message':SUFFIX, 'template_tool_response':TEMPLATE_TOOL_RESPONSE, } ) agent.agent.output_parser.format_instructions = "MY CUSTOM FORMAT INSTRUCTIONS" print(agent.agent.output_parser.get_format_instructions()) MY CUSTOM FORMAT INSTRUCTIONS ``` Other parameters like `system_message`, `human_message`, or `template_tool_response` are already customizable and with this PR, the last parameter `FORMAT_INSTRUCTIONS` in `langchain.agents.conversational_chat.prompt` can be modified. **Issue:** https://github.com/langchain-ai/langchain/issues/10721 **Dependencies:** No new dependencies required for this change **Twitter handle:** With my github user is enough. Thanks I hope you accept my PR. --------- Co-authored-by: Bagatur <baskaryan@gmail.com>	8 months ago
Ryan Kraus	f027696b5f	community: Added new Utility runnables for NVIDIA Riva. (#15966) ... **Please tag this issue with `nvidia_genai`** - **Description:** Added new Runnables for integration NVIDIA Riva into LCEL chains for Automatic Speech Recognition (ASR) and Text To Speech (TTS). - **Issue:** N/A - **Dependencies:** To use these runnables, the NVIDIA Riva client libraries are required. It they are not installed, an error will be raised instructing how to install them. The Runnables can be safely imported without the riva client libraries. - **Twitter handle:** N/A All of the Riva Runnables are inside a single folder in the Utilities module. In this folder are four files: - common.py - Contains all code that is common to both TTS and ASR - stream.py - Contains a class representing an audio stream that allows the end user to put data into the stream like a queue. - asr.py - Contains the RivaASR runnable - tts.py - Contains the RivaTTS runnable The following Python function is an example of creating a chain that makes use of both of these Runnables: ```python def create( config: Configuration, audio_encoding: RivaAudioEncoding, sample_rate: int, audio_channels: int = 1, ) -> Runnable[ASRInputType, TTSOutputType]: """Create a new instance of the chain.""" _LOGGER.info("Instantiating the chain.") # create the riva asr client riva_asr = RivaASR( url=str(config.riva_asr.service.url), ssl_cert=config.riva_asr.service.ssl_cert, encoding=audio_encoding, audio_channel_count=audio_channels, sample_rate_hertz=sample_rate, profanity_filter=config.riva_asr.profanity_filter, enable_automatic_punctuation=config.riva_asr.enable_automatic_punctuation, language_code=config.riva_asr.language_code, ) # create the prompt template prompt = PromptTemplate.from_template("{user_input}") # model = ChatOpenAI() model = ChatNVIDIA(model="mixtral_8x7b") # type: ignore # create the riva tts client riva_tts = RivaTTS( url=str(config.riva_asr.service.url), ssl_cert=config.riva_asr.service.ssl_cert, output_directory=config.riva_tts.output_directory, language_code=config.riva_tts.language_code, voice_name=config.riva_tts.voice_name, ) # construct and return the chain return {"user_input": riva_asr} | prompt | model | riva_tts # type: ignore ``` The following code is an example of creating a new audio stream for Riva: ```python input_stream = AudioStream(maxsize=1000) # Send bytes into the stream for chunk in audio_chunks: await input_stream.aput(chunk) input_stream.close() ``` The following code is an example of how to execute the chain with RivaASR and RivaTTS ```python output_stream = asyncio.Queue() while not input_stream.complete: async for chunk in chain.astream(input_stream): output_stream.put(chunk) ``` Everything should be async safe and thread safe. Audio data can be put into the input stream while the chain is running without interruptions. --------- Co-authored-by: Hayden Wolff <hwolff@nvidia.com> Co-authored-by: Hayden Wolff <hwolff@Haydens-Laptop.local> Co-authored-by: Hayden Wolff <haydenwolff99@gmail.com> Co-authored-by: Erick Friis <erick@langchain.dev>	8 months ago
François Paupier	929f071513	community[patch]: Fix error in `LlamaCpp` community LLM with Configurable Fields, 'grammar' custom type not available (#16995) ... - **Description:** Ensure the `LlamaGrammar` custom type is always available when instantiating a `LlamaCpp` LLM - **Issue:** #16994 - **Dependencies:** None - **Twitter handle:** @fpaupier --------- Co-authored-by: Bagatur <baskaryan@gmail.com>	8 months ago
Leonid Ganeline	563f325034	experimental[patch]: fixed import in `experimental` (#17078)	8 months ago
Eugene Yurtsev	fbab8baac5	core[patch]: Add astream events config test (#17055) ... Verify that astream events propagates config correctly --------- Co-authored-by: Bagatur <baskaryan@gmail.com>	8 months ago
Scott Nath	10bd901139	infra: add integration_tests and coverage to MAKEFILE (#17053) ... - **Description: update community MAKE file** - adds `integration_tests` - adds `coverage` - **Issue:** the issue # it fixes if applicable, - moving out of https://github.com/langchain-ai/langchain/pull/17014 - **Dependencies:** n/a - **Twitter handle:** @scottnath - **Mastodon handle:** scottnath@mastodon.social --------- Co-authored-by: Bagatur <baskaryan@gmail.com>	8 months ago
Giulio Zani	9f0b63dba0	experimental[patch]: Fixes issue #17060 (#17062) ... As described in issue #17060, in the case in which text has only one sentence the following function fails. Checking for that and adding a return case fixed the issue. ```python def split_text(self, text: str) -> List[str]: """Split text into multiple components.""" # Splitting the essay on '.', '?', and '!' single_sentences_list = re.split(r"(?<=[.?!])\s+", text) sentences = [ {"sentence": x, "index": i} for i, x in enumerate(single_sentences_list) ] sentences = combine_sentences(sentences) embeddings = self.embeddings.embed_documents( [x["combined_sentence"] for x in sentences] ) for i, sentence in enumerate(sentences): sentence["combined_sentence_embedding"] = embeddings[i] distances, sentences = calculate_cosine_distances(sentences) start_index = 0 # Create a list to hold the grouped sentences chunks = [] breakpoint_percentile_threshold = 95 breakpoint_distance_threshold = np.percentile( distances, breakpoint_percentile_threshold ) # If you want more chunks, lower the percentile cutoff indices_above_thresh = [ i for i, x in enumerate(distances) if x > breakpoint_distance_threshold ] # The indices of those breakpoints on your list # Iterate through the breakpoints to slice the sentences for index in indices_above_thresh: # The end index is the current breakpoint end_index = index # Slice the sentence_dicts from the current start index to the end index group = sentences[start_index : end_index + 1] combined_text = " ".join([d["sentence"] for d in group]) chunks.append(combined_text) # Update the start index for the next group start_index = index + 1 # The last group, if any sentences remain if start_index < len(sentences): combined_text = " ".join([d["sentence"] for d in sentences[start_index:]]) chunks.append(combined_text) return chunks ``` Co-authored-by: Giulio Zani <salamanderxing@Giulios-MBP.homenet.telecomitalia.it>	8 months ago