Packages

ex_llama

0.2.3
0.2.3 0.2.2 0.2.1 0.2.0 0.1.0 0.0.2 0.0.1

NIF Wrapper around llama.cpp allowing elixir code to load/infer against gguf format models.

HexDocs

Current section

Files

Jump to
Readme
7 Versions
5 Dependencies
Dependants
Files
Activity
ex_llama CHANGELOG.md
Raw

CHANGELOG.md 

Change Log

=====



## 0.2.3

Complete the Linux build fix started in 0.2.2:



- Pass -DCMAKE_POSITION_INDEPENDENT_CODE=ON so static archives can be

  linked into the NIF .so without "relocation ... can not be used when

  making a shared object" errors on any Linux arch.

- Add -lopenblas -lgomp -ldl to the Linux link line so sgemm_/dgemm_

  (from libggml-blas.a), OpenMP symbols (from libggml-cpu.a), and

  dlopen (from libggml.a) resolve at both link time and runtime.



Darwin link line unchanged.



## 0.2.2

Fix Linux build: enable GGML_BLAS=ON in the cmake step so libggml-blas.a

is produced, matching what the NIF link line already expects. On Darwin

this flag picks Apple Accelerate (prior default); on Linux it selects

OpenBLAS (requires libopenblas-dev at build time).



## 0.1.0

Update to use GenAI Core structs