Expose `ggml_backend_load()` and `ggml_backend_load_all()` to make use of builds with `GGML_BACKEND_DL=ON` and `GGML_CPU_ALL_VARIANTS=ON`

Open
#2,069 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
45/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Stale
Tech stack
c, cmake, python

Research direction

Look at the existing C bindings in llama-cpp-python for backend-related functions. The issue mentions ggml_backend_load() and ggml_backend_load_all() from llama.cpp; find where similar functions are exposed and add these. Check the build configuration with GGML_BACKEND_DL=ON to understand the dynamic loading mechanism. Test by compiling and loading a model to ensure the backend libraries are correctly loaded and the error is resolved.

Written by the indexing model from the issue text.

Description

I just tried compiling llama-cpp-python with GGML_BACKEND_DL=ON and GGML_CPU_ALL_VARIANTS=ON to make use of this nice feature with dynamic dispatch to a dynamically loaded backend, which e.g. made it possible to build llama.cpp once but dynamically choose the best backend for the current CPU, i.e. for x86_64 depending on whether certain instructions like AVX2 or AVX512 are available choose the best backend for the current microarchitecture level.

Compiling worked for me so far on Ubuntu 24.04 LTS and when inspecting the wheel I see the backend dynamic libraries like bin/libggml-cpu-x64.so, libggml-cpu-sse42.so, libggml-cpu-haswell.so and so on. So that is good already.

But when loading a model with llama-cpp-python I get this error:
llama_model_load_from_file_impl: no backends are loaded. hint: use ggml_backend_load() or ggml_backend_load_all() to load a backend before calling this function but these functions are not exposed yet via the bindings.

I think this would be a really great thing to add. That would make the CPU wheels for llama-cpp-python way better, because it wouldn't be stuck with base x86_64 instructions and could thus be way more performant for cases where the wheel cannot be compiled at installation time.

Dominant language
Python
Stars
10.7k
Forks
1.5k
Avg merge
1h 40m
Merged PRs (30d)
5

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from abetlen/llama-cpp-python

All issues in abetlen/llama-cpp-python

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.