Feature Request: Use Weights & Biases (W&B) Traces in llm calls

[cfreyre]

Feature Request: Use Weights & Biases (W&B) Traces

I currently use Docker logs to verify the call and the costs associated with the calls (Prompts) and the latency of each of them.

It would be great to have a professional tool to track each call to the different LLMs.

I propose the use of Weights & Biases (W&B) Traces

https://wandb.ai/site/traces/

Here are some key benefits:

>> Enhanced Observability

>> Improved Debugging Capabilities
Error Identification: Tracing allows for the logging of inputs, outputs, and intermediate processes, which aids in pinpointing errors and understanding conditions leading to unexpected outcomes.

>> Performance Optimization
Resource Usage Measurement:

>> Reproducibility
Experiment Tracking: Capturing exact inputs, outputs, and metadata for each step.

It is possible to add a environment variables in docker-compose.dev

WANDB_API_KEY
WANDB_PROJECT

Optional environment variables:

https://docs.wandb.ai/guides/track/environment-variables/

Thank you very much in advance