feat(tracing): add tracing to llm and llm-base crates

[radu-matei]

Hi, everyone!
First, thanks for the project!

This commit begins adding tracing and some instrumentation to this project.

As folks are beginning to look at using and optimising this project, I suspect this will be very useful.
Below is an example of sending the data to an open telemetry collector and visualising it:

This is a draft PR for now, as we understand whether this is something we would like to add to the project.
If it is, this can be expanded to include more robust logging and instrumentation.

[LLukas22]

See #331. I would love to have some better tracing/performance stats ❤️

[philpax]

No objections from me! Would you like to do more or should I merge this PR as-is?

We may also want to switch over llm-cli and other applications to use tracing, so that we don't have duplicate loggers.

[radu-matei]

I can update the CLI to use tracing as part of this PR as well, yeah.

[philpax]

Cool, go ahead and switch things over to tracing. Should be a pretty straightforward PR :)

[radu-matei]

Besides consuming the tracing data in an OTEL environment which is demonstrated above, you can now get basic logging with timestamps when using the llm CLI as well:

$ RUST_LOG=llm=trace ./target/release/llm ...
⣾ Loading model...2023-07-16T11:56:04.390067Z TRACE infer: llm_base::loader: Read model file from "/Users/radu/models/llama/open-llama-13b-open-instruct.ggmlv3.q3_K_L.bin"    
⣾ Loaded hyperparameters2023-07-16T11:56:04.538068Z TRACE infer: llm_base::loader: Loaded GGML model from reader    
2023-07-16T11:56:04.538102Z TRACE infer: llm_base::loader: Determined quantization version of model as 2    
2023-07-16T11:56:04.538163Z TRACE infer: llm_base::loader: Context size: 92928    
2023-07-16T11:56:04.538175Z DEBUG infer: llm::cli_args: ggml ctx size = 92.9 KB    
✓ Loaded 363 tensors (6.9 GB) after 150ms
2023-07-16T11:56:04.555690Z TRACE infer: llm_base::loader: Loaded model    
2023-07-16T11:56:04.559279Z TRACE infer:infer:infer: llm_base::inference_session: Starting inference request with max_token_count: 18446744073709551615    
2023-07-16T11:56:12.273385Z TRACE infer:infer:infer:feed_prompt: llm_base::inference_session: Finished feed prompt

[philpax]

Excellent, thank you :)

[yangyaofei]

After this PR, run the following command will get nothing :

cargo run --release -- info -a bloom --model-path models/ggml-model-f16.bin

I had to add RUST_LOG=llm=trace for all command, is that good?