Haystack 2.0-Beta

[bilgeyucel]

This discussion entry will be your ultimate guide until the stable release of Haystack 2.0.

Haystack 2.0-Beta was released on December 4th. Dive deeper into this milestone here. 🚀

What does this mean for you? Since the beta release, we have been delivering two simultaneous releases: one focusing on Haystack 1.x improvements and bug fixes, and another dedicated to unveiling features in Haystack 2.0. This discussion includes the Haystack Roadmap, additional integrations available for Haystack 2.0, Haystack 2.0-Beta versions release notes, and a feature table continuously updated with every new beta release.

If you’d like to get notified about new Haystack 2.0 features, subscribe to this entry 🔔

🛣️ Stay Updated with the Roadmap:

Explore the Haystack Roadmap for insights into the future developments and enhancements we have on the horizon.

🌐 Explore New Integrations:

Discover additional integrations available for Haystack 2.0 by checking out haystack-core-integrations, which include names like Amazon, VertexAI, Elasticsearch, OpenSearch, Chroma, Pinecone, Cohere, Gradient, Jina and Ollama.

📄 Release Notes:

Find detailed release notes for Haystack 2.0-Beta:

• v2.0.0-beta.5
• v2.0.0-beta.4
• v2.0.0-beta.3
• v2.0.0-beta.2
• v2.0.0-beta.1

🎛️ Feature Table:

We're committed to refining and expanding Haystack 2.0 with each beta release. Expect frequent updates to the feature table to ensure you're always aware of the latest additions and improvements.

🪶 List of Features

✅ Ready in the latest Beta release (v2.0.0-beta.4)
🏗️ Under construction

Feature	Haystack 2.0-Beta
Document Stores
InMemoryDocumentStore	✅
ElasticsearchDocumentstore	✅
OpenSearchDocumentStore	✅
ChromaDocumentStore	✅
MarqoDocumentStore	✅
PineconeDocumentStore	✅
QdrantDocumentStore	✅
MongoDBAtlasDocumentStore	🏗️
AstraDBDocumentStore	✅
PGVectorDocumentStore	🏗️
WeaviateDocumentStore	🏗️
Generators
OpenAIGenerator	✅
OpenAIChatGenerator	✅
HuggingFaceLocalGenerator	✅
HuggingFaceTGIGenerator	✅
HuggingFaceTGIChatGenerator	✅
GradientGenerator	✅
CohereGenerator	✅
CohereChatGenerator	✅
GoogleAIGeminiGenerator	✅
GoogleAIGeminiChatGenerator	✅
VertexAIGeminiGenerator	✅
VertexAIGeminiChatGenerator	✅
VertexAITextGenerator	✅
VertexAICodeGenerator	✅
VertexAIImageCaptioner	✅
VertexAIImageGenerator	✅
VertexAIImageQA	✅
AWSBedrockGenerator	✅
Ollama Generator	✅
Anthropic - Claude	🏗️
AzureOpenAIGenerator	🏗️
AWSSageMakerGenerator	🏗️
PromptBuilder	✅
AnswerBuilder	✅
DynamicPromptBuilder	✅
DynamicChatPromptBuilder	✅
Embedders
OpenAI Embedders	✅
SentenceTransformers Embedder	✅
Cohere Embeders	✅
Gradient Embedders	✅
Jina Embedders	✅
HuggingFaceTEI Embedders	✅
AzureOpenAI Embedders	✅
AmazonBedrockEmbedder	🏗️
Retrievers
InMemoryBM25Retriever	✅
InMemoryEmbeddingRetriever	✅
ElasticsearchBM25Retriever	✅
ElasticsearchEmbeddingRetriever	✅
OpensearchBM25Retriever	✅
OpensearchEmbeddingRetriever	✅
TableTextRetriever	🏗️
Rankers
TransformersSimilarityRanker	✅
CohereRanker	🏗️
DiversityRanker	🏗️
LostInTheMiddleRanker	🏗️
RecentnessRanker	🏗️
MetaFieldRanker	✅
Readers
ExtractiveReader
(successor of both FARMReader and TransformersReader)	✅
TableReader	🏗️
Data Processing
Local + Remote WhisperTranscriber	✅
AssemblyAITranscriber	✅
UnstructuredFileConverter	✅
NamedEntityExtractor	✅
UrlCacheChecker	✅
LinkContentFetcher	✅
AzureOCRDocumentConverter	✅
HTMLToDocument	✅
PyPDFToDocument	✅
TikaDocumentConverter	✅
TextFileToDocument	✅
MarkdownToDocument	✅
DocumentCleaner	✅
DocumentSplitter	✅
DocumentLanguageClassifier	✅
ConditionalRouter	✅
TextLanguageRouter	✅
FileTypeRouter	✅
MetadataRouter	✅
DocumentWriter	✅
DocumentJoiner	✅
Multiplexer	✅
Misc
SerperDevWebSearch	✅
Evaluation	🏗️
Agents	🏗️
Conversational Agent	🏗️
TopPSampler	✅

[TuanaCelik]

Warning: Installing farm-haystack and haystack-ai in the same Python environment (virtualenv, Colab, or system) causes problems.

Installing both packages in the same environment can somehow work or fail in obscure ways. We suggest installing only one of these packages per Python environment. Make sure that you remove both packages if they are installed in the same environment, followed by installing only one of them:

pip uninstall -y farm-haystack haystack-ai
pip install haystack-ai

If you have any questions, please reach out to us on this discussion or Discord.

[sandyis]

@masci if i have to use BM25Retriever, FARMReader then i have to use Haystack-farm , right?
I started with haystack-ai , when the tutorial led me to install haystack-farm, and that lead me to the error

   from haystack.nodes import TextConverter, PreProcessor
ModuleNotFoundError: No module named 'haystack.nodes'
(haystak) hni@hniai:~/haystak$ pip install farm-haystack

ImportError: cannot import name 'send_event' from 'haystack.telemetry' (/home/hni/haystak/lib/python3.10/site-packages/haystack/telemetry/__init__.py)

which lead me here
sure i can open a dedicated discussion that might be better as haystack-ai is fast moving and changing, plus like i said i might be way off base here

[masci]

farm-haystack and haystack-ai are not compatible, they can't be installed in the same env and they are not supposed to be. Normally this shouldn't be an issue, as different versions of the same software come with the same package name and the package manager (pip in this case) would take care of this, but we took the chance to ship Haystack 2.0 under a new package name, haystack-ai to finally get rid of the FARM keyword. Unfortunately in this case the package manager doesn't know, and will happily overwrite Haystack 1.x over 2.x and vice versa.

[basharullah]

I tried the following commands:

pip uninstall -y farm-haystack haystack-ai
pip install haystack-ai

However now I can't use the libraries from farm-haystack.

Do you guys have ideas on what can I do to resolve this?

[masci]

@basharullah you're trying to run Haystack 1.x code after installing Haystack 2.0. Please refer to the 2.x docs to solve your problem. Note that Haystack 2.0 is still in beta, and a full port of your 1.x code might not be possible straight ahead.

[fserv]

@masci I am new to haystack and am reading haystack-tutorials/tree/main/tutorials
/03_Scalable_QA_System.ipynb Are these 1.0 or 2.0 ? Thanks.

[TuanaCelik]

Evaluation of RAG

Hi all, we are currently implementing evaluation for RAG in Haystack 2.0. We would love to hear your thoughts on a few things:

• What do you think are some of the most promising approaches to evaluation generative models and RAG pipelines out there today?
• What do you like about (if you've tried) the implementations out there today?
• What don't you like about current implementations out there today?
• And anything else you'd like to add to the discussion

I am hosting a Community Call: Evaluation of RAG in Haystack 2.0 this Thursday the 25th at 7PM CET on Discord. Feel free to join the discussion there, or simply add your thoughts here:

Discord event link: https://discord.com/events/993534733298450452/1194241314498293800

[sandangel]

Our team is considering https://github.com/Arize-ai/phoenix for running evaluation and trace llm responses. So having an integration with that tool would be great. We also considered Trulens, but it seems not supporting tracing yet.

Another tool we are using is chainlit. In chainlit UI, it allows you to add decoration to the tool output so chainlit can display it as traces in the UI. You can take a look at its example here: https://github.com/Chainlit/chainlit#-pure-python . Being able to integrate with frontend framework like chainlit would be great

For evaluation dataset, we use llama-index. It provides pre-made dataset and evaluation pack that we can just download and run.

[TuanaCelik]

Thanks for the input @sandangel
I just wanted to add this here: https://haystack.deepset.ai/integrations/chainlit Chainlit made an integration for Haystack a while back, but it's not been ported to 2.0 yet :)

[ananddw24]

Hi,

Great initiative. What was the outcome of this meeting ? Are there any next steps or libraries that came up from this discussion ?

[bilgeyucel]

🎉 Haystack 2.0.0-beta.5 is released

We are at version 2.0.0b5 now. The new version comes with some breaking changes but also new features and enhancements.

⚠️ Breaking Changes

🗺️ Import paths of DocumentJoiner, InMemoryDocumentStore and InMemoryEmbeddingRetriever have changed
🔨 The model_name_or_path and model_name parameters of all Embedders, Generators, and various other components are renamed to model
⚙️ GPU/CPU device management changed, introducing a ComponentDevice instead of strings as device representations

Features and Enhancements

🚀 New NamedEntityExtractor component
🌡️ EvaluationResult has a calculate_metrics function for computation of evaluation metrics
📄 Single metadata dictionaries can be provided as inputs to MarkdownToDocument, TikaDocumentConverter and AzureOCRDocumentConverter
🏬 URLCacheChecker works with any type of data in the DocumentStore and we renamed it to CacheChecker.

Check out the full release notes for more 👉 v2.0.0-beta.5

[annthurium]

Missed announcing a few 2.0 beta releases here over the past few weeks - sorry about that!

v2.0.0-beta.7 is released

⬆️ Upgrade Notes
Pipeline.connect() arguments have renamed for clarity. This is a breaking change if connect was called with keyword arguments only. connect_from and connect_to arguments have been renamed respectively to sender and receiver. The behaviour of Pipeline.connect() is not changed.

Check out the full release notes

v2.0.0-beta.6

⚠️ Breaking change: Update secret handling for components using the Secret type.

Read the full release notes here

[annthurium]

Integration updates

We have a ton of new integrations that are 2.0 compatible, as well as updating 1.x integrations to be 2.0 compatible. You can see the full list here.

New Integrations

• Astra DB
• AWS Bedrock
• AWS Sagemaker
• AssemblyAI
• DeepEval
• Mistral AI
• mixedbread AI
• Neo4j
• Ollama
• pgvector
• ragas
• uptrain

1.x integrations with updates:

• Chroma
• Cohere
• Elasticsearch
• Mastodon Fetcher

[TuanaCelik]

New Resources and Community Updates
Our documentation team has a call for feedback from the community 💜
If you're a user of the Haystack 2.0 docs, and especially, if you've also used Haystack 1.x docs before too, please consider filling our our survey 🙏

Now for the updates 👇

• The AstraDB integration has now been updated to use a simpler authentication API. https://haystack.deepset.ai/integrations/astradb

📚 New Haystack 2.0 Tutorials

• Building Fallbacks with Conditional Routers
• Evaluating RAG with Eval Framework Integrations
• Hybrid Retrieval

👩🏻‍🍳 New Cookbook

• vLLM + Haystack 2.0

🚀 New Guides
Check our our new guide to chosing between the Haystack Generators vs Chat generators here

🧡 Contributions wanted
Are you looking to contribute to Haystack? You can see all the issues marked with "Contributions wanted!" in this new public GitHub project. Here, you'll not only see issues open to contributions from the main Haystack repo, but also haystack integrations.
https://github.com/orgs/deepset-ai/projects/14

[annthurium]

Some new updates!

💥v2.0.0-beta.8 release

Highlight: Pipeline templates! Introducing a flexible and dynamic approach to creating NLP pipelines with Haystack's new PipelineTemplate class!. Details in the release notes

📓 Hypothetical Document Embeddings

Hypothetical Document Embeddings (HyDE) improves retrieval by generating “fake” hypothetical documents based on a given query, and then uses those “fake” documents embeddings to retrieve similar documents from the same embedding space. See Haystack's HyDE docs and blog article to try it in your pipelines.

🧩 New 2.0 Integrations

Weaviate
ContextAI
MongoDB
Mistral
FastEmbed
DeepEval
RAGAS
Pinecone
TitanML

📚 New 2.0 Tutorials

Simplifying Pipeline Inputs with Multiplexer
Embedding Metadata for Improved Retrieval
We updated the Model-Based Evaluation of RAG Pipelines tutorial with the examples of DeepEval and RAGAS

🧑‍🍳 New 2.0 Cookbooks

Build with Gemma + Haystack 2.0
Using Mistral AI with Haystack

📖 New Guides

Check out our new guide to learn about how to evaluate your pipelines without labeled data: Model-Based Evaluation

[TuanaCelik]

Hey everyone 👋

I'm very happy to announce the stable version of Haystack 2.0 just went live, finally🥳🎉 

Which means we can bring this discussion to a close.

You can give it a try here 👉 https://haystack.deepset.ai/overview/quick-start
Learn more about the release 👉 https://haystack.deepset.ai/blog/haystack-2-release/
And of course, the full release notes are on GitHub 👉 https://github.com/deepset-ai/haystack/releases/tag/v2.0.0