llama.cpp
14b19373 - cuda : adapt CUDA graph compat to new feature

Commit

8 hours ago

cuda : adapt CUDA graph compat to new feature

References

#18550 - ggml : add ggml_build_forward_select

#18898 - graph : utilize `ggml_build_forward_select()` to avoid reallocations

Author

ggerganov

ggerganov

Committer

ggerganov

ggerganov

Parents

Loading