[experimental] Add command line tool for llama_index

[iaalm]

Based on discussion #1054, add command line with following features:

# create a local config file in local dir
python -m experimental.cli init

# add file to index
python -m experimental.cli add ../data/

# query
python -m experimental.cli query "Some question?"

The idea is quite straightforward, there're two files put in current directory

• config.ini stores embedding/predicter model setup along with its parameters
• index.json the index file

There is still a lot of work to be done, such as supporting various embedding models (even a local LLM maybe?), and removing/updating the current index. However, I am considering merging this fundamental part first so that I can tackle the project step by step.

Open to suggestions/changes.

[jerryjliu]

amazing! will take a look soon

[jerryjliu]

high-level looks good! main comment is about documenting usage

[jerryjliu]

i like it - i can envision this being an easy-to-use CLI tool so that you can ask questions over any directory on your file system

[hophanms]

Got this error:

Error while finding module specification for 'experimental.cli' (ModuleNotFoundError: No module named 'experimental')

[iaalm]

Hi @hophanms, "experimental" is not included in the package yet (I think it's why it called "experimental"). For now, you need to git clone this repo and run these command in it.
Or you can use PYTHONPATH=/path/to/your/llama_index python -m experimental.cli init to run in any directory.

For long term, when this part of code is stable enough, we can move to src. At that time user will be able to call it directly with something like python -m llama_index.cli init

Hope it helps.

[hophanms]

I was in the wrong folder when I ran that CLI.

I haven't digged further but tried this
python -m experimental.cli query "Some question?"

And got this:

None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
INFO:llama_index.token_counter.token_counter:> [query] Total LLM token usage: 0 tokens
INFO:llama_index.token_counter.token_counter:> [query] Total embedding token usage: 6 tokens
None

There is no further way to debug.

[iaalm]

Hi @hophanms , could you provider some more information about your run? There're many reason can cause this output.

Maybe I can describe the work flow clearly (maybe update the doc too, later).

python -m experimental.cli init
this creates a config file "config.ini". If default config is not enough for you, you need to update it manually (for now).

python -m experimental.cli add ../data/
This use OpenAI/Azure OpenAI API to "analysis" files under "../data/" and store it in a file named "index.json" in current directory.

python -m experimental.cli query "Some question?" this checks the local "index.json" and send some more query to OpenAI/Azure OpenAI for the answer to your question.

Because we still call OpenAI API for index and question answering, environments like "OPENAI_API_KEY" still need to be set.

Your output looks ok. There're no error message just no result too. For me, I can repro same result without "add" any file. I don't know if it's your case.

[hophanms]

I found out it silently failed on generating the .json so that's why the query returned None
If I fixed the add then it works.