The AI Toolchain — issue -385, October 31, 2023

Aider v0.15.0 adds .aiderignore support, a --commit flag, and full shell commands in /run and /git.

$ git clone --branch v0.15.0 https://github.com/Aider-AI/aider.git
# already have the repo? check out this version:
$ git checkout v0.15.0

$ aider --commit

$ /run (cd subdir; ls)

• ›Adds .aiderignore file support to instruct aider to ignore parts of the git repo during context gathering.
• ›/run and /git slash commands now accept full shell expressions, including subshells like /run (cd subdir; ls).
• ›Adds universal ctags and multi-architecture support to the aider Docker image.

Aider v0.14.2 adds Docker support, OpenRouter integration for Claude 2, and improved /add and /drop file handling.

$ git clone --branch v0.14.2 https://github.com/Aider-AI/aider.git
# already have the repo? check out this version:
$ git checkout v0.14.2

• ›Supports running Aider directly from a Docker image via the published Docker workflow.
• ›Adds Claude 2 and other LLM support through OpenRouter integration.
• ›Extends /add and /drop commands to handle absolute filenames and quoted filenames.
• ›Adds a guard in /add to verify files are within the git repo or root before adding them.
• ›Shows models supported by the API/key when the requested model is unavailable.

Zed v0.108.2 adds channel links, Save/Discard all on close, and vim Y line-wise copy.

$ git clone --branch v0.108.2 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v0.108.2

• ›Changes the mute_on_join setting default to false, so users join channels with audio unmuted by default.
• ›Adds Copy Channel Link via right-click on a channel, producing a deep link that opens Zed and auto-joins that channel.
• ›Adds Save all and Discard all options when closing Zed with unsaved buffers.
• ›Adds vim Y keybinding to copy line-wise, matching classic vim behaviour (differs from nvim).

Zed v0.107.6 adds richer channel collaboration, Vim completion bindings, and nushell venv support.

$ git clone --branch v0.107.6 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v0.107.6

{
  "terminal": {
    "detect_venv": {
      "on": {
        "activate_script": "nushell"
      }
    }
  }
}

• ›Adds nushell as a supported value for the terminal.detect_venv.on.activate_script setting.
• ›Adds Vim key bindings for Zed completion mechanisms: ctrl-x ctrl-o (completion menu), ctrl-x ctrl-l (LSP action menu), ctrl-x ctrl-c (Copilot), ctrl-x ctrl-a (inline Assistant), and ctrl-x ctrl-z (hide menus).
• ›Adds Vim ctrl-i to navigate forward in the jump list (complementing the existing ctrl-o).
• ›Adds Vim ctrl-a / ctrl-x bindings for increment/decrement.
• ›Adds Vim g <space> to open the current snippet in its own file.

Zed v0.106.2 adds project-wide find-and-replace, Next LS for Elixir, Vim command mode, and configurable diagnostics warnings.

$ git clone --branch v0.106.2 https://github.com/zed-industries/zed.git
# already have the repo? check out this version:
$ git checkout v0.106.2

{
  "elixir": {
    "lsp": "next_ls"
  }
}

{
  "diagnostics": {
    "include_warnings": false
  }
}

• ›Adds diagnostics.include_warnings setting to control whether warnings appear in project diagnostics; warnings icon in the UI can toggle them off temporarily.
• ›Adds experimental Next LS support for Elixir via the "lsp": "next_ls" field under "elixir" in Zed settings.
• ›Enables semantic_index setting by default, making Semantic mode visible in project search without manual configuration.
• ›Adds 'Replace in project' functionality for project-wide find-and-replace.
• ›Adds cmd-k shift-{left,right,up,down} keybindings to swap panes in the corresponding direction.

LangChain v0.0.326 adds Google Cloud Translation transformer, Azure Search reranking, new RAG templates, and DALL-E multi-URL support.

$ git clone --branch v0.0.326 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.326

• ›Adds rrf argument to ApproxRetrievalStrategy.__init__() to enable Reciprocal Rank Fusion in Elasticsearch approximate retrieval.
• ›Adds reranking support to the Azure Cognitive Search retriever.
• ›_dalle_image_url now returns a list of URLs when n>1, enabling multi-image generation in a single call.
• ›Adds Google Cloud Translation document transformer for translating documents as a pipeline stage.
• ›Allows astream_log to be used inside atrace_as_chain_group, enabling streaming log capture within traced chain groups.

LangChain v0.0.325 adds Google Speech-to-Text loader, JohnSnowLabs embeddings, Fireworks batching, and new RAG templates.

$ git clone --branch v0.0.325 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.325

• ›Adds AsyncHtmlLoader metadata enrichment: HTML title and page language are now extracted into document metadata.
• ›Adds JohnSnowLabs embeddings support as a new embeddings integration.
• ›Adds batch request support for the Fireworks LLM integration.
• ›New Cohere re-rank retrieval template for use with LangServe.
• ›New HyDE (Hypothetical Document Embeddings) retrieval template.

LangChain v0.0.324 adds Baidu Cloud vector search, Takeoff Pro support, Comprehend Moderation 0.2, and CohereEmbeddings retry/timeout controls.

$ git clone --branch v0.0.324 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.324

from langchain.embeddings import CohereEmbeddings

embeddings = CohereEmbeddings(
    model="embed-english-v3.0",
    max_retries=5,
    request_timeout=30,
)

• ›Adds max_retries and request_timeout parameters to CohereEmbeddings for resilience tuning.
• ›Adds allowed_operators property to QdrantTranslator for self-query filter control.
• ›Allows index name customization via environment variable in the rag-conversation template.
• ›Adds Baidu Cloud vector search as a new vectorstore integration.
• ›Adds Takeoff Pro support as a new LLM integration.

LangChain v0.0.323 integrates E2B's data analysis/code interpreter and adds serialization support for Fireworks models.

$ git clone --branch v0.0.323 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.323

• ›Integrates E2B's data analysis and code interpreter as a new tool/integration.
• ›Adds serialization properties to Fireworks and ChatFireworks model classes.

LangChain v0.0.322 adds COBOL parsing, GigaChat support, injectable boto3 client for SageMaker, and public event-handling APIs.

$ git clone --branch v0.0.322 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.322

• ›Exposes handle_event and ahandle_events as public API methods for callback/event handling.
• ›Adds injectable boto3 client support to SagemakerEndpointEmbeddings, enabling custom session and credential configurations.
• ›Adds connection args support to the pgvector vector store integration.
• ›Adds COBOL parser and splitter for ingesting COBOL source files.
• ›Adds GigaChat chat model integration.

LangChain v0.0.321 adds custom I/O schemas for runnables, optional config arg for RunnablePassthrough, and parent run ID tracking.

$ git clone --branch v0.0.321 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.321

chain = prompt | llm | parser
typed_chain = chain.with_types(input_type=MyInput, output_type=MyOutput)

• ›Adds .with_types() method to runnables, allowing custom input and output schemas to be specified explicitly.
• ›Adds optional config argument to RunnablePassthrough function argument for per-run configuration.
• ›Includes Parent Run ID in run tracking, enabling better lineage and observability across chained calls.
• ›Updates default recursion_limit for runnables (see updated docs for new value).
• ›Adds Step Back prompting notebook demonstrating the step-back question technique.

LangChain v0.0.320 adds Tencent Hunyuan chat, Tavily Search, Google Scholar tools, Neo4j env vars, and runnable factory support in .configurable_alts()

$ git clone --branch v0.0.320 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.320

• ›Adds Neo4j graph environment variables support via Add neo4j graph environment variables, allowing Neo4j connection config to be driven by env vars.
• ›Supports runnable factories in .configurable_alts(), enabling dynamic runnable construction at configuration time.
• ›Adds Tencent Hunyuan as a new chat model integration.
• ›Adds Tavily Search API as a new tool integration.
• ›Adds Google Scholar search tool via SerpAPI.

LangChain v0.0.319 adds add_embeddings support for Elasticsearch and dynamic runnable schemas from config.

$ git clone --branch v0.0.319 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.319

• ›Adds add_embeddings support for the Elasticsearch vector store integration.
• ›Adds dynamic schemas derived from config for runnables (runnable-dynamic-schemas-from-config).
• ›Changes baichuan_secret_key to use pydantic.types.SecretStr for safer credential handling.

LangChain v0.0.318 adds ERNIE-Bot-4, Weaviate multi-tenancy, Website Data Store retrieval, and configurable retry limits for output parsers.

$ git clone --branch v0.0.318 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.318

from langchain.output_parsers import RetryWithErrorOutputParser

retry_parser = RetryWithErrorOutputParser.from_llm(
    parser=base_parser,
    llm=llm,
    max_retries=2
)

• ›Adds max_retries support to RetryOutputParser and RetryWithErrorOutputParser, letting callers cap how many correction attempts are made before failing.
• ›Adds _acall async method to YandexGPT, enabling non-blocking inference calls.
• ›Adds ERNIE-Bot-4 model support to ErnieBotChat, expanding available Baidu ERNIE model options.
• ›Adds support for Website Data Stores in the Google Vertex AI Search Retriever.
• ›Updates Weaviate integration to support multi-tenancy.

LangChain v0.0.317 adds Baichuan chat model, Cohere RAG retriever, Graph interface, Hub Runnable, and Zep MMR support.

$ git clone --branch v0.0.317 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.317

• ›Adds QianfanChatEndpoint support for function_call in Qianfan ChatModels.
• ›Adds Hub Runnable to pull and run prompts/chains directly from LangChain Hub.
• ›Adds Baichuan chat model integration.
• ›Adds Cohere retrieval-augmented generation to the retrievers interface.
• ›Adds a Graph interface for graph-based data interactions.

LangChain v0.0.316 adds Together.xyz and YandexGPT LLM providers, SingleStoreDB chat history, and OutputFixingParser retry control.

$ git clone --branch v0.0.316 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.316

from langchain.output_parsers import OutputFixingParser

fixing_parser = OutputFixingParser.from_llm(parser=base_parser, llm=llm, max_retries=3)

from langchain.memory import SingleStoreDBChatMessageHistory

history = SingleStoreDBChatMessageHistory(
    session_id="user-123",
    host="singlestore-host",
    port=3306,
    user="admin",
    password="<password>",
    database="langchain"
)

• ›Adds max_retries parameter to OutputFixingParser to control how many times the parser attempts to fix malformed output.
• ›Adds SingleStoreDBChatMessageHistory class to support SingleStoreDB as a ChatMessageHistory backend.
• ›Exports merge_configs function for merging runnable configuration objects.
• ›Adds validation for configurable keys passed to .with_config(), catching invalid keys at call time.
• ›Adds together.xyz as a new LLM provider integration.

LangChain v0.0.315 adds ChatEverlyAI, the Bearly tool, candidate_count for Vertex models, and promotes Python/Pandas/Spark agents to experimental.

$ git clone --branch v0.0.315 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.315

• ›Adds candidate_count parameter support for Vertex AI models.
• ›Introduces ChatEverlyAI chat model integration.
• ›Adds the Bearly tool integration.
• ›Promotes Python, Pandas, Xorbits, and Spark agents to the experimental module.
• ›Adds get_llm_cache and set_llm_cache functions for managing LLM cache state.

LangChain v0.0.314 adds ElasticsearchChatMessageHistory, Upstash Redis integration, TrainableLLM, Alibaba Tongyi chat, RSpace loader, and Anthropic functions support.

$ git clone --branch v0.0.314 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.314

• ›Adds ElasticsearchChatMessageHistory class for storing chat message history in Elasticsearch.
• ›Adds Upstash Redis integration for caching and message history backed by Upstash Redis.
• ›Adds TrainableLLM abstract class enabling LLM fine-tuning workflows within LangChain.
• ›Adds Alibaba Tongyi chat model APIs via a new chat model integration.
• ›Adds RSpace document loader for ingesting content from RSpace electronic lab notebooks.

LangChain v0.0.313 adds configurable fields with options, Azure Cosmos DB vector store, SemaDB, and MMR for Elasticsearch retriever.

$ git clone --branch v0.0.313 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.313

• ›Adds patch_config(configurable=) argument and updates with_config(configurable=) to merge with existing configurable values, enabling runtime configuration composition.
• ›Adds configurable fields with options, allowing runnable components to expose typed, enumerable configuration surfaces.
• ›Adds allow_list support in langchain-experimental data anonymizer to whitelist terms that should not be anonymized.
• ›Adds SQLAlchemyMd5Cache implementation for MD5-keyed SQL-backed LLM response caching.
• ›Adds callback function support to RunnablePassthrough, enabling side-effects or logging within passthrough steps.

LangChain v0.0.312 adds Momento vector store, Arcee.ai integration, expanded Presidio entity support, and metadata-column control for CSV loading.

$ git clone --branch v0.0.312 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.312

• ›Adds option to specify metadata columns in the CSV loader, giving callers control over which columns are promoted to document metadata.
• ›Adds Momento Vector Index as a new vector store provider integration.
• ›Adds Arcee.ai LLM and Retriever integration.
• ›Supports all Presidio entities in the anonymizer/deanonymizer (previously a limited subset).
• ›Adds reset capability for deanonymizer mapping, allowing mappings to be cleared between runs.

LangChain v0.0.311 adds a Markdown list parser, LangSmith chat loader, autodetect encoding for CSV, and renames RunnableMap to RunnableParallel.

$ git clone --branch v0.0.311 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.311

from langchain.document_loaders.csv_loader import CSVLoader

loader = CSVLoader(file_path='data.csv', autodetect_encoding=True)
docs = loader.load()

• ›Adds autodetect_encoding option to CSVLoader to automatically detect file encoding when loading CSV documents.
• ›Adds MarkdownListParser for parsing Markdown list-formatted output from language models.
• ›Adds LangSmithRunChatLoader to load chat message history from LangSmith runs.
• ›Renames RunnableMap to RunnableParallel for clearer semantics in LCEL chains.
• ›Updates Google Document AI parser with new capabilities.

LangChain v0.0.310 adds async indexing, RL chains, streaming SageMaker LLMs, image extraction from PDFs, and new vector store filter operators.

$ git clone --branch v0.0.310 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.310

• ›Adds in and nin filter operators to Pinecone vector store queries.
• ›Adds additional filter comparators to Weaviate vector store.
• ›Adds a source constructor argument to the Vectara integration.
• ›Adds async support to SelfQueryRetriever, enabling non-blocking self-query workflows.
• ›Adds async SQL record manager and async indexing API.

LangChain v0.0.309 adds Vespa vector store, Cohere /chat integration, a project-scaffolding CLI, and optional Cypher validation tooling.

$ git clone --branch v0.0.309 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.309

from langchain.prompts import PromptTemplate

template = PromptTemplate(
    input_variables=["query"],
    input_types={"query": str},
    template="Answer the following question: {query}"
)

• ›Adds optional input_types parameter to prompt templates for stronger type hinting on template inputs.
• ›Adds a new CLI command to create a new LangChain project, with Docker Compose support included.
• ›Adds the Vespa vector store integration for similarity search via Vespa backends.
• ›Adds an optional Cypher validation tool for graph database query workflows.
• ›Adds interactive login support for the Azure Cognitive Search vector store.

LangChain v0.0.308 adds Bedrock Cohere support, custom GitHub API URLs, and default async methods.

$ git clone --branch v0.0.308 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.308

• ›Adds custom API URL support to GitHubIssuesLoader, enabling use against GitHub Enterprise or other custom endpoints.
• ›Adds Bedrock Cohere support, integrating Cohere models via AWS Bedrock into the LangChain LLM stack.
• ›Adds default async implementations across chain/runnable components via add default async.
• ›Adds _type field to the JSON functions output parser for improved schema identification.

LangChain v0.0.307 adds runtime-configurable Runnables, Tavily Search retriever, scoring chain, Kotlin splitter, and memory for SQL chains.

$ git clone --branch v0.0.307 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.307

from langchain.text_splitter import HTMLHeaderTextSplitter

splitter = HTMLHeaderTextSplitter(headers_to_split_on=[("h1", "Header 1"), ("h2", "Header 2")])
chunks = splitter.split_text(html_string)

• ›Adds .configurable_fields() and .configurable_alternatives() methods to Runnable to expose fields for runtime configuration, backed by the new RunnableSerializable base class.
• ›Adds HTMLHeaderTextSplitter for splitting HTML documents by header elements.
• ›Adds Tavily Search API retriever integration.
• ›Adds scoring chain for LLM-based evaluation.
• ›Adds Kotlin code splitter.

LangChain v0.0.306 adds a streaming JSON parser and RunnablePassthrough.assign() for inline chain composition.

$ git clone --branch v0.0.306 https://github.com/langchain-ai/langchain.git
# already have the repo? check out this version:
$ git checkout v0.0.306

from langchain.schema.runnable import RunnablePassthrough

chain = RunnablePassthrough.assign(word_count=lambda x: len(x['text'].split()))
result = chain.invoke({'text': 'Hello world from LangChain'})
# result => {'text': 'Hello world from LangChain', 'word_count': 4}

• ›Adds RunnablePassthrough.assign(...) method to attach new keys to a passthrough runnable inline, enabling richer chain composition without a separate step.
• ›Adds a streaming JSON parser for parsing partial JSON output incrementally as it streams from a model.
• ›Adds a type field to message chunks, making it easier to identify chunk provenance in streaming message flows.
• ›Updates the DeepSparse LLM integration.

Letta 0.1.6 adds local LLM support, Azure, AutoGen integration, CSV/PDF preloading, and on-the-fly embedding generation.

$ git clone --branch 0.1.6 https://github.com/letta-ai/letta.git
# already have the repo? check out this version:
$ git checkout 0.1.6

• ›Adds a flag for preloading files into archival memory at startup.
• ›Adds CSV support for preloading files into archival memory.
• ›Adds PDF support for preloading files into archival memory.
• ›Supports generating embeddings on the fly, with parallelized embedding generation.
• ›Adds local LLM support with function calling.

AutoGen v0.1.13 adds TeachableAgent for persistent long-term memory across chat sessions via vector database.

$ git clone --branch v0.1.13 https://github.com/microsoft/autogen.git
# already have the repo? check out this version:
$ git checkout v0.1.13

• ›Adds TeachableAgent class that persists user-taught facts, preferences, and skills across chat boundaries using a vector database, saving memos to disk at chat end and loading them at the next chat start.
• ›Retrieves individual memos into context as needed rather than loading the full memory store, preserving context-window space while enabling long-term recall.

AutoGen v0.1.12 adds custom text splitter support for RAG agents and function call filtering in group chat.

$ git clone --branch v0.1.12 https://github.com/microsoft/autogen.git
# already have the repo? check out this version:
$ git checkout v0.1.12

• ›Adds function call filtering in group chat, reducing failures when agents invoke tools during multi-agent conversations.
• ›Adds support for custom text splitters in RAG agents, enabling user-defined chunking logic for retrieval workflows.

AutoGen v0.1.11 adds a Langchain tool bridge, enabling agents to use Langchain tools directly.

$ git clone --branch v0.1.11 https://github.com/microsoft/autogen.git
# already have the repo? check out this version:
$ git checkout v0.1.11

• ›Adds a Langchain tool bridge so AutoGen agents can use Langchain tools natively, demonstrated in agentchat_langchain.ipynb.
• ›Improves logging in oai.completion to display token_count during model calls.
• ›Adds compatibility for custom models that do not return all fields in the response.

AutoGen v0.1.10 lets you plug in customized vector databases and embedding functions.

$ git clone --branch v0.1.10 https://github.com/microsoft/autogen.git
# already have the repo? check out this version:
$ git checkout v0.1.10

• ›Adds support for plugging in customized vector database backends and custom embedding functions.

AutoGen v0.1.7 adds .env file support for instantiating config_list in openai_utils.

$ git clone --branch v0.1.7 https://github.com/microsoft/autogen.git
# already have the repo? check out this version:
$ git checkout v0.1.7

• ›Adds .env file support to openai_utils.py for instantiating config_list, enabling credential loading from environment files without hardcoding values.

AutoGen v0.1.5 adds PDF file parsing support to RetrieveChat's retrieve_utils.

$ git clone --branch v0.1.5 https://github.com/microsoft/autogen.git
# already have the repo? check out this version:
$ git checkout v0.1.5

• ›Adds PDF file parsing to retrieve_utils.py, enabling RetrieveChat to extract and index text from PDF documents.

Jan v0.2.0 lets users fetch models directly from GitHub and improves the chat UI.

$ git clone --branch v0.2.0 https://github.com/janhq/jan.git
# already have the repo? check out this version:
$ git checkout v0.2.0

• ›Enables fetching models from GitHub directly within the app.
• ›Improved chat box UI for a cleaner conversation experience.

KoboldCpp v1.46.1 adds OpenAI Chat Completions endpoint, 64k context support, parallel horde jobs, and full API docs.

$ git clone --branch v1.46.1 https://github.com/LostRuins/koboldcpp.git
# already have the repo? check out this version:
$ git checkout v1.46.1

$ koboldcpp.exe --model my_model.gguf --contextsize 65536

• ›Adds OpenAI Chat Completions-compatible endpoint at /v1/chat/completions.
• ›Adds interactive API documentation accessible at /api (also hosted at https://lite.koboldai.net/koboldcpp_api).
• ›Expands --contextsize support up to 64k tokens (GGUF models only; remains 16k for other formats).
• ›Embedded horde worker now executes job pulls and job submits in parallel, yielding approximately 20% faster horde request handling with additional session stats.
• ›.kcpps config files can now be drag-and-dropped or opened via 'Open With' on Windows.

KoboldCpp v1.45.2 adds OpenAI-compatible /v1/completions, grammar state persistence, 32k context, and a new --foreground flag.

$ git clone --branch v1.45.2 https://github.com/LostRuins/koboldcpp.git
# already have the repo? check out this version:
$ git checkout v1.45.2

$ curl http://localhost:5001/v1/completions -H 'Content-Type: application/json' -d '{"prompt": "Once upon a time", "max_tokens": 200}'

$ koboldcpp.exe --skiplauncher my_model.kcpps --foreground

• ›Adds grammar_retain_state parameter to the grammar sampler API, allowing grammar state to persist across multiple requests.
• ›Adds --foreground flag for Windows users to bring the console terminal to the foreground on each prompt generation, avoiding idle slowdown.
• ›Increases maximum supported context via --contextsize to 32k for GGUF models (still limited to 16k for older model formats).
• ›Adds OpenAI-compatible completions API endpoint at /v1/completions.
• ›Supports launching with a .kcpps settings file via the file selector GUI combined with --skiplauncher (the settings file must already have a model selected).

Adds LLaVA v1.5 support, multi-LoRA blending, flash-attention 2 on Windows, and expanded Silero language coverage.

$ git clone --branch snapshot-2023-10-22 https://github.com/oobabooga/textgen.git
# already have the repo? check out this version:
$ git checkout snapshot-2023-10-22

• ›Supports LLaVA v1.5 and LLaVA v1.5 7B multimodal models for vision-language inference.
• ›Allows multiple loaded LoRAs to simultaneously influence generation output, enabling LoRA blending workflows.
• ›Adds option to select additional linear modules and layers as targets during LoRA training.
• ›Adds flash-attention 2 support on Windows.
• ›Adds Python 3.11 support with structured requirements.

textgen snapshot adds LLaVA-LLaMA-2 support, HTTPS for OpenAI extension, ChatML, and ExLlamaV2 LoRA.

$ git clone --branch snapshot-2023-10-15 https://github.com/oobabooga/textgen.git
# already have the repo? check out this version:
$ git checkout snapshot-2023-10-15

• ›Adds HTTPS support to the OpenAI extension, enabling secure API connections.
• ›Adds ChatML prompt format support, including Mistral-OpenOrca compatibility.
• ›Adds LoRA support for the ExLlamaV2 backend.
• ›Supports LLaVA-LLaMA-2 multimodal model loading.

v1.7 adds AutoAWQ loader support, grammar for Transformers/HF loaders, NUMA for llama.cpp, and EOS-abort for text generation.

$ git clone --branch v1.7 https://github.com/oobabooga/textgen.git
# already have the repo? check out this version:
$ git checkout v1.7

• ›Adds grammar support to the transformers and _HF loaders, enabling constrained decoding for locally loaded models.
• ›Adds initial AutoAWQ loader support, allowing AWQ-quantized models to be loaded directly.
• ›Enables NUMA feature for llama_cpp_python, improving CPU memory locality on multi-socket systems.
• ›Model downloader now downloads *.tiktoken tokenizer files alongside model weights.
• ›Adds EOS-token abort behavior in text generation — output stops immediately when the end-of-sequence token is reached.

Ollama v0.1.4 adds OpenHermes 2 Mistral, faster model switching, and unicode support for starcoder/sqlcoder/falcon.

$ git clone --branch v0.1.4 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.1.4

$ ollama pull starcoder

• ›Adds ollama pull starcoder (and sqlcoder, falcon) to get updated builds with full unicode support — existing pulls must be re-pulled.
• ›Faster model switching: models stay loaded between requests when switching parameters (e.g. temperature) or system prompts, eliminating reload overhead.
• ›Adds OpenHermes 2 Mistral to the model library (ollama pull openhermes2-mistral), a Mistral fine-tune trained on 900,000+ instructions with strong multi-turn chat performance.
• ›ollama serve now prints the current Ollama version on startup.
• ›ollama run now surfaces descriptive runtime errors (e.g. insufficient memory) instead of silent failures.

Ollama v0.1.2 adds non-streaming API responses, HTTP/HTTPS proxy support, two new models, and 25x faster pull speeds.

$ git clone --branch v0.1.2 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.1.2

$ curl -X POST http://localhost:11434/api/generate -d '{"model": "llama2", "prompt": "Why is the sky blue?", "stream": false}'

• ›Adds stream: false parameter to API endpoints (e.g. /api/generate) to return a complete, non-streaming response in a single payload instead of a token stream.
• ›Supports HTTP and HTTPS proxies via the HTTP_PROXY=http://<proxy> and HTTPS_PROXY=https://<proxy> environment variables.
• ›Adds two new models to the library: zephyr (fine-tuned Mistral 7B) and mistral-openorca (Mistral 7B fine-tuned on the OpenOrca dataset).
• ›ollama pull download speeds improved from ~60 MB/s to over 1.5 GB/s (25x) on fast network connections, enabling practical use of large models over high-bandwidth links.
• ›q8_0, q5_0, q5_1, and f32 quantization formats now use GPU acceleration on Linux.

Ollama v0.1.1 adds cancellable responses and clean session exit controls for ollama run.

$ git clone --branch v0.1.1 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.1.1

• ›Adds Ctrl+C to cancel in-progress responses during an ollama run session.
• ›Adds Ctrl+D and /bye as exit commands for ollama run sessions.

Phoenix v0.0.50 adds RAG eval relevancy conventions, protected LangChain callbacks, and a light UI theme.

$ git clone --branch v0.0.50 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout v0.0.50

• ›Adds eval relevancy conventions for RAG pipelines.
• ›Adds protection for LangChain callbacks to prevent tracing errors from interrupting application flow.
• ›Introduces a light theme option for the Phoenix UI.

Phoenix v0.0.49 adds multiclass classification support, video URL embeddings, and raises default eval retry count to 20.

$ git clone --branch v0.0.49 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout v0.0.49

• ›Raises the default retry count for evals to 20 for more resilient evaluation runs.
• ›Supports video URLs for linkToData in embedding visualizations.
• ›Adds support for multiclass classification datasets.

Phoenix v0.0.45 adds reranker span support, trace search/filtering, verbose eval mode, and gpt-3.5-turbo-instruct for evals.

$ git clone --branch v0.0.45 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout v0.0.45

• ›Adds reranker span kind to LLM Traces, enabling visibility into document reranking steps in RAG pipelines.
• ›Adds gpt-3.5-turbo-instruct as a supported model in OpenAIModel for LLM evaluations.
• ›Adds search and filtering for traces and spans, including filtering on span IO content and computed values such as latency and token counts.
• ›Adds verbose mode for LLM evaluations to surface real-time progress and aid debugging of eval runs.
• ›Adds a general-purpose LlamaIndex debug callback handler for tracing LlamaIndex applications.

Phoenix v0.0.44 adds live-updating trace UI, OpenAI and Bedrock instrumentors, Azure params, and exporter startup warnings.

$ git clone --branch v0.0.44 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout v0.0.44

from phoenix.trace.openai import OpenAIInstrumentor

OpenAIInstrumentor().instrument()

• ›Adds Azure parameters to OpenAIModel for evals, enabling Azure OpenAI endpoints as an eval backend.
• ›Adds Amazon Bedrock model support to the evals framework via a new BedrockModel integration.
• ›Adds a new OpenAI instrumentor with chat completions support for tracing OpenAI calls.
• ›Adds gql validate_span_filter_condition to the tracing GraphQL API for validating span filter conditions.
• ›Adds default UMAP parameters in launch_app() so embeddings visualizations work out of the box without manual tuning.

Chroma 0.4.14 adds gRPC segments, a distributed segment manager, and new Terraform deployment blueprints for AWS, Render, and DigitalOcean.

$ git clone --branch 0.4.14 https://github.com/chroma-core/chroma.git
# already have the repo? check out this version:
$ git checkout 0.4.14

• ›Adds gRPC-based segments and a distributed segment manager, enabling horizontally scaled Chroma deployments.
• ›Adds a Render.com Terraform blueprint for one-click cloud deployment of Chroma.
• ›Adds an improved AWS Terraform blueprint for deploying Chroma on AWS infrastructure.
• ›Adds a DigitalOcean Terraform deployment blueprint for Chroma.

LanceDB v0.3.4 adds checkout, optimize index, remap index, and dataset stats APIs for Python and Node.

$ git clone --branch v0.3.4 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.3.4

• ›Adds checkout method to table objects to reuse existing store and connections.
• ›Exposes optimize index API for managing vector indexes programmatically.
• ›Exposes remap index API for index remapping operations.
• ›Adds dataset stats APIs to both Python and Node bindings for inspecting table data statistics.
• ›Includes manifest files in mirror store, broadening mirrored-store coverage.

LanceDB python-v0.3.2 adds remote table deletion, PyArrow date/timestamp type support, and a table checkout method.

$ git clone --branch python-v0.3.2 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.3.2

• ›Adds delete operation on remote tables, enabling row deletion via the remote API for both Python and JS clients.
• ›Adds checkout method to Table to reuse an existing store and connections without re-opening.
• ›Adds PyArrow date and timestamp type conversion from Pydantic models.
• ›Adds list-table pagination support for remote table listings.
• ›Adds incremental index update and index compaction capabilities via the underlying Lance 0.8.5 upgrade.

LanceDB v0.3.3 adds PyArrow date/timestamp type conversion from Pydantic and refactors the Embeddings API.

$ git clone --branch v0.3.3 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.3.3

• ›Adds PyArrow date and timestamp type conversion from Pydantic models, enabling richer schema definitions without manual type mapping.
• ›Refactors the Embeddings API (Python) with updated embedding function support.

LanceDB v0.3.2 adds deletion operations on remote tables for both Python and JavaScript clients.

$ git clone --branch v0.3.2 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.3.2

• ›Adds deletion operation on remote tables in the Python and JavaScript clients, enabling row removal from cloud-hosted LanceDB tables.
• ›Implements remote API calls for table mutation, extending write capabilities to the remote backend.

LanceDB v0.3.1 adds GPU index creation, Cohere embeddings, mirroring object store, compaction, and new query APIs.

$ git clone --branch v0.3.1 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.3.1

df = table.search(query_vector).limit(10).to_pandas()

• ›Adds to_list and to_pandas APIs for querying table data directly into Python-native structures.
• ›Adds Cohere embedding function for generating embeddings via the Cohere API.
• ›Adds GPU support for index creation to accelerate vector index builds.
• ›Implements a mirroring object store for replicating data across storage backends.
• ›Adds cleanup and compaction support for managing table storage and reducing file fragmentation.

LanceDB v0.3.1 adds GPU index creation, Cohere embeddings, object-store mirroring, table compaction, and new query APIs.

$ git clone --branch python-v0.3.1 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.3.1

results = table.search(query_vector).limit(10).to_pandas()

• ›Adds to_list() and to_pandas() APIs for querying tables directly into Python-native result types.
• ›Adds Cohere embedding function for generating embeddings via the Cohere API.
• ›Supports GPU-accelerated index creation for faster ANN index builds.
• ›Implements object store mirroring to replicate data across storage backends.
• ›Adds table cleanup and compaction to reduce small-file overhead and reclaim storage.

LanceDB adds opt-in pre-filtering via prefilter=True on .where(), applying filters before vector search rather than after.

$ git clone --branch python-v0.2.6 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.2.6

table.search(query_vector).where("category = 'malware'", prefilter=True).limit(10).to_df()

• ›Adds prefilter=True parameter to .where() to apply filters BEFORE running KNN vector search, reducing the candidate set before similarity scoring.