Support no chromadb, sentence_transformers or pypdf

[thinkall]

Why are these changes needed?

Address #531

Support no chromadb, sentence_transformers or pypdf when using customized vector database, embedding function or w/o pdf files, respectively.

Related issue number

Closes #531

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. See https://microsoft.github.io/autogen/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[codecov-commenter]

Codecov Report

Merging #556 (3f044de) into main (0dd0fc5) will increase coverage by 5.58%.
The diff coverage is 40.74%.

@@            Coverage Diff             @@
##             main     #556      +/-   ##
==========================================
+ Coverage   32.39%   37.97%   +5.58%     
==========================================
  Files          22       22              
  Lines        2772     2786      +14     
  Branches      630      632       +2     
==========================================
+ Hits          898     1058     +160     
+ Misses       1807     1640     -167     
- Partials       67       88      +21     

Flag	Coverage Δ
unittests	37.90% <40.74%> (+5.58%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
...gen/agentchat/contrib/retrieve_user_proxy_agent.py	14.21% <50.00%> (+14.21%)	⬆️
autogen/retrieve_utils.py	63.63% <36.84%> (+63.63%)	⬆️

... and 1 file with indirect coverage changes

[thinkall]

@kdcokenny could you check if this PR works for you? Thanks.

[0xmatt1]

These things can be tricky, best to procede with caution

[sonichi]

shall we decouple [retrievechat] and [chromadb]? Is there any common dependency between different retrieve backends? If not, we may create [retrievechat_chromadb] etc.
It will be good to clarify first, about what is required for all retrieve chat, what is required for each specific backend, etc.

[sonichi]

make the error a constant in the header.

[thinkall]

shall we decouple [retrievechat] and [chromadb]? Is there any common dependency between different retrieve backends? If not, we may create [retrievechat_chromadb] etc. It will be good to clarify first, about what is required for all retrieve chat, what is required for each specific backend, etc.

RetrieveChat has no must-install dependencies. The [retrievechat] extras could be treated as a default solution for running it, but chromadb, sentence_transformers and pypdf all have alternatives, ipython is not needed for pure QA. Advanced users can install different dependencies as they want.

What if we keep the current [retrievechat] extras, but improve the docs in the Installation section to guide users for different choices like qdrant for vectordb, huggingface or openai for embedding, unstructured for pdf/word... extraction? @sonichi @rickyloynd-microsoft

[kdcokenny]

What I would suggest it adding import error try catch blocks which prompt the users to pip install chromadb or qdrant-client if importing one of them results in an import error. I can show an example if you’d like

…

On Sun, Nov 5, 2023 at 21:54, Li Jiang ***@***.***(mailto:On Sun, Nov 5, 2023 at 21:54, Li Jiang <<a href=)> wrote: > shall we decouple [retrievechat] and [chromadb]? Is there any common dependency between different retrieve backends? If not, we may create [retrievechat_chromadb] etc. It will be good to clarify first, about what is required for all retrieve chat, what is required for each specific backend, etc. RetrieveChat has no must-install dependencies. The [retrievechat] extras could be treated as a default solution for running it, but chromadb, sentence_transformers and pypdf all have alternatives, ipython is not needed for pure QA. Advanced users can install different dependencies as they want. What if we keep the current [retrievechat] extras, but improve the docs in the Installation section to guide users for different choices like qdrant for vectordb, huggingface or openai for embedding, unstructured for pdf/word... extraction? — Reply to this email directly, [view it on GitHub](#556 (comment)), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/AXX7GXBKVBL7DR4VSZYJB4TYDBGORAVCNFSM6AAAAAA66GTSUWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTOOJUGAYDOMJZHE). You are receiving this because you were mentioned.Message ID: ***@***.***>

[kdcokenny]

@kdcokenny could you check if this PR works for you? Thanks.

When I was testing it I was getting weird errors like: TypeError: Completions.create() got an unexpected keyword argument 'request_timeout'

This may be due to the new openai v1 upgrade that I havent taken account for. Here's my exported py notebook:

# %%
# %pip install -q pyautogen qdrant-client fastembed
# %pip install -q chromadb pypdf # required by autogen for some reason

# %%
%pip install -q git+https://github.com/microsoft/autogen.git@support_no_chromadb
%pip install -q qdrant-client fastembed

# %%
from qdrant_client import QdrantClient

host = "qdran...ws.com"

client = QdrantClient(host=host, api_key="...")

# %%
config_list = [
    {
        'model': 'gpt-3.5-turbo',
        'api_key': '...',
    },
]

# %%
# import autogen

# autogen.ChatCompletion.start_logging()

# %%
client.set_model("BAAI/bge-base-en-v1.5")

# %%
from autogen.agentchat.contrib.qdrant_retrieve_user_proxy_agent import QdrantRetrieveUserProxyAgent

ragproxyagent = QdrantRetrieveUserProxyAgent(
    name="ragproxyagent",
    human_input_mode="NEVER",
    max_consecutive_auto_reply=10,
    retrieve_config={
        "task": "qa",
        "chunk_token_size": 2000,
        "model": config_list[0]["model"],
        "client": client,
        "embedding_model": "BAAI/bge-base-en-v1.5",
        "collection_name": "wized_test",
    },
)

# %%
from autogen.agentchat.contrib.retrieve_assistant_agent import RetrieveAssistantAgent

assistant = RetrieveAssistantAgent(
    name="assistant", 
    system_message="You are a helpful assistant.",
    llm_config={
        "request_timeout": 600,
        "seed": 42,
        "config_list": config_list,
    },
)

# %%
assistant.reset()

qa_problem = "What is Wized?"
ragproxyagent.initiate_chat(assistant, problem=qa_problem)