Logan M

Hmm that s rather odd I ll try swapping

Hmm that's rather odd. I'll try swapping the sentence window notebook to use weaviate and confirm, it shouuuuld be fine 🤔

19 comments

LLogan M

You ll need to start with a fresh index

You'll need to start with a fresh index if you switch embeddings, the dimensions of every embedding vector need to be the same 👍

21 comments

LLogan M

The default model that loads for the

The default model that loads for the huggingface embeddings in the docs page that i sent usually works well

For LLMs, vicuna seems to be good (but it's also non-commericial). I like camel for commercial models so far

67 comments

LLogan M

Anyone tried the new alpha release yet?

Anyone tried the new alpha release yet? Definitely open to comments on any of the change made.

My favourite new feature is the new IngestionPipeline + cache

Plain Text

client = qdrant_client.QdrantClient(location=":memory:")
vector_store = QdrantVectorStore(client=client, collection_name="test_store")

pipeline = IngestionPipeline(
    transformations=[
        SentenceSplitter(chunk_size=25, chunk_overlap=0),
        TitleExtractor(),
        OpenAIEmbedding(),
    ],
    cache=IngestionCache(cache=RedisCache(), collection="test_cache"),
    vector_store=vector_store,
)

# Ingest directly into a vector db
pipeline.run(documents=[Document.example()])

# Create your index
from llama_index import VectorStoreIndex
index = VectorStoreIndex.from_vector_store(vector_store)

19 comments

LLogan M

Is this a llama index thing or langchain

Is this a llama index thing or langchain thing? If it's langchain, I got no idea 😅

6 comments

LLogan M

Hmmm I don t think anything like that

Hmmm, I don't think anything like that exists right now. Or at least nothing that isn't super hacky lol

24 comments

LLogan M

Unless you create a graph over your

Unless you create a graph over your indexes, you'll need one tool per index.

Be careful though, you'll run out of prompt space around 30ish tools

6 comments

LLogan M

`index service context llm predictor

index._service_context.llm_predictor.last_token_usage()

index._service_context.embed_model.last_token_usage()

69 comments

LLogan M

It makes 5 llm calls total or 5 llm

It makes 5 llm calls total, or 5 llm calls to llama index? What do your settings/indexes look like?

8 comments

LLogan M

In addition to alpaca you ll also need

In addition to alpaca, you'll also need and embed_model. By default it uses openAI text-ada-002 (which is pretty cheap thankfully).

You can use any model from huggingface locally, using this guide: https://gpt-index.readthedocs.io/en/latest/how_to/customization/embeddings.html#custom-embeddings

16 comments

LLogan M

Are you setting a system prompt

Are you setting a system prompt somewhere? By default there isn't one in llama index

2 comments

LLogan M

You can actually do a `pip install

You can actually do a pip install --upgrade -e . when you are inside the top-level after a clone, this will add the package to your env, but allow you to edit files and test the changes

5 comments

LLogan M

On its own llama index is more of a

On its own, llama index is more of a search bar than a chatbot.

If you are wanting support for an actual chatbot, you can use llama index as a tool within langchain

https://gpt-index.readthedocs.io/en/latest/guides/building_a_chatbot.html

1 comment

LLogan M

Anyone know how to get the discord token

Anyone know how to get the discord token for using the DiscordReader? I tried doing the thing where you copy from the network dev tools panel, but that doesn't seem to be working 🤔

14 comments

LLogan M

togethercomputer/GPT-NeoXT-Chat-Base-20B...

New model alert! https://huggingface.co/togethercomputer/GPT-NeoXT-Chat-Base-20B

8 comments

LLogan M

Sure does Check ths guide https github

Sure does! Check ths guide: https://github.com/jerryjliu/gpt_index/blob/main/examples/vector_indices/SimpleIndexDemo-ChatGPT.ipynb

17 comments

LLogan M

llama_index/llama_index/agent/types.py a...

Extra_state should be a dictionary, not a list 🤔
https://github.com/run-llama/llama_index/blob/d68c3bbe52399e7133bcefdf02c34838b1931181/llama_index/agent/types.py#L164

11 comments

LLogan M

Smart chunks

I think so! Have "complete" ideas in each chunk (whether a paragraph, or a section, or a chapter) usually helps the embeddings better represent the text. Normally I would do this at the document level and let the actual nodes fall where they may lol

13 comments

LLogan M

I thiiiink you can load the index from

I thiiiink you can load the index from storage (as you are doing), call insert_nodes(), and then call persist again to write to disk
index.storage_context.persist(persist_dir=...)

33 comments

LLogan M

That s a good point disiok 5183 maybe

That's a good point. @disiok maybe you know if it's possible to include the cell outputs in the new embedded notebooks?

3 comments

LLogan M

Using llama hub dataloaders with llama-index

Pretty sure the document objects from the llama hub loaders will work in llama index actually 🤔

5 comments

LLogan M

Probably the LLM stopped following

Probably the LLM stopped following instructions and printed some output that langchain couldn't parse

Pretty common error with langchain tbh. The parsing code for that specific agent it here https://github.com/hwchase17/langchain/blob/master/langchain/agents/chat/output_parser.py

Langchain at some post probably needs less-brittle parsing. Not much to do about it besides making a PR or maybe improving the tool instructions
https://github.com/hwchase17/langchain/blob/master/langchain/agents/chat/prompt.py

13 comments

LLogan M

```

Plain Text

import logging
import sys

logging.basicConfig(stream=sys.stdout, level=logging.CRITICAL)
logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))

1 comment

LLogan M

If front end stuff isn t interesting you

If front end stuff isn't interesting, you can make a quick and dirty frontend using streamlit or gradio

Personally I like streamlit, as it's a bit more customizable. But gradio can work too.

You basically build a frontend with a few lines of python, and it generally looks good.

Extremely valuable library (or libraries) to master, let's you quickly make POCs to show people and demo stuff

4 comments

LLogan M

In my experience you just need to be

In my experience, you just need to be super verbose in the description. Or if you wanted, you could even do something like "If user mentions the keyword [TOOL], use this tool"

7 comments

Find answers from the community

Hmm that s rather odd I ll try swapping

You ll need to start with a fresh index

The default model that loads for the

Anyone tried the new alpha release yet?

Is this a llama index thing or langchain

Hmmm I don t think anything like that

Unless you create a graph over your

`index service context llm predictor

It makes 5 llm calls total or 5 llm

In addition to alpaca you ll also need

Are you setting a system prompt

You can actually do a `pip install

On its own llama index is more of a

Anyone know how to get the discord token

togethercomputer/GPT-NeoXT-Chat-Base-20B...

Sure does Check ths guide https github

llama_index/llama_index/agent/types.py a...

Smart chunks

I thiiiink you can load the index from

That s a good point disiok 5183 maybe

Using llama hub dataloaders with llama-index

Probably the LLM stopped following

```

If front end stuff isn t interesting you

In my experience you just need to be