Ensured that we have default values for attributes name, temperature, llm_name, max_tokens and stream.

User can choose to alter the values that are actually used on the fly via new optional arguments in  __call__.

Also _call_ changed to __call__, in line with other LLMs.

Previously self.name was being used to identify the LLM to be called, but self.name is intended to identify the OllamaLLM instance as an instance of OllamaLLM. So it's now set to "ollama".

feat: Adds VitEncoder for visual transformers

Adding Ollama support.

Howto use:

from ollama import OllamaLLM
rl = RouteLayer(encoder = encoder, routes=routes, llm=llm)

chore: move pinecone demo

feat: Pinecone demo and hotfix

fix: Saving JSON/YAML of Layer Config

feat: separate indexes and PineconeIndex

test_from_file_invalid_config()

test_from_file_with_llm()

And some improvements where we make sure that temp files are definitely deleted after use.

Saving Layer Configs as JSON or YAML config files by representing the LayerConfig object as a dict and then turning into JSON/YAML wasn't working.

See issue here: https://github.com/aurelio-labs/semantic-router/issues/144

The issue was that, by attempting this, we were attempting to serialize all objects included in the Layer, including Routes, and the LLMs that those Routes use.

In the case of the above issue, the LLM was a Cohere one, which included a Client as one of its attributes, and this Client is not serializable.

So, instead of attempting to represent the entire LLM object a dict, to then be converted into JSON/YAML, we only keep key information about the LLM:

'module': self.llm.__module__,
'class': self.llm.__class__.__name__,
'model': self.llm.name

This is what's saved in route.py, and then sent to layer.py to be serialized (in LayerConfig.to_file()).

Then, when it comes time to load from the file via LayerConfig.from_file, the LLM is re-initialized dynamically.

Using index.describe instead of shape[].