[Abhishekpandey123-git]

Related to #702.

Adds an explicit llama.cpp provider configuration, registry coverage, and benchmark tooling for reproducible Ollama-versus-llama.cpp comparisons.

This PR is stacked on #733 and targets 502-ollama-model-install so it can be reviewed with the provider/model configuration work it depends on.

Note: comparative benchmark runs and the final performance recommendation still need to be captured before #702 is complete.

[Abhishekpandey123-git]

Benchmark limitation: hardware unavailable

The implementation and benchmark tooling are ready for review, and the local backend validation passed (420 tests; Ruff passed). However, I cannot run a fair Ollama-versus-llama.cpp comparison on my current hardware due to resource constraints.

I have not measured or claimed a performance improvement. Latency, throughput, CPU, and memory comparisons remain unmeasured, so the comparison report should be treated as pending rather than as evidence that either backend is faster.

Would a maintainer or contributor with suitable hardware be able to run the documented benchmark workloads, or advise on an available CI/benchmark runner? I can help interpret the output and update the report once genuine measurements are available.

[marcvergees]

Hi @Abhishekpandey123-git, I'm a little bit out this week, lots of midterms in uni and other stuff. I'll be back with this asap. Thanks for you comprehension.
M