sync : llama.cpp

[slaren]

The moved files weren't removed, they are duplicated now.

[mudler]

seems part of this PR ( https://github.com/ggerganov/ggml/commit/538487872e43030458b23d5dbc87bc30db9cc170 ) have broken building here with sycl, not sure if am I missing something?

2024-11-19T09:13:54.2414646Z #56 97.42 /build/backend/cpp/llama-avx2/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp:1240:55: warning: cast from 'const void *' to 'unsigned char *' drops const qualifier [-Wcast-qual]
2024-11-19T09:13:54.2417134Z #56 97.42  1240 |                 auto it = m_map.upper_bound((byte_t *)ptr);
2024-11-19T09:13:54.2418252Z #56 97.42       |                                                       ^
2024-11-19T09:13:54.2420526Z #56 97.42 /build/backend/cpp/llama-avx2/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp:1837:16: error: no member named 'dp4a' in namespace 'syclcompat'; did you mean simply 'dp4a'?
2024-11-19T09:13:54.2422486Z #56 97.42  1837 |         return syclcompat::dp4a(a, b, c);
2024-11-19T09:13:54.2423252Z #56 97.42       |                ^~~~~~~~~~~~~~~~
2024-11-19T09:13:54.2424107Z #56 97.42       |                dp4a
2024-11-19T09:13:54.2425546Z #56 97.42 /build/backend/cpp/llama-avx2/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp:1835:17: note: 'dp4a' declared here
2024-11-19T09:13:54.2427095Z #56 97.42  1835 |     inline auto dp4a(T1 a, T2 b, T3 c)

Any help would be appreciated, thank you! :pray:

[slaren]

You probably need to update the oneAPI etc SYCL libraries. See https://github.com/ggerganov/llama.cpp/pull/10267 for more details.

[mudler]

You probably need to update the oneAPI etc SYCL libraries. See ggerganov/llama.cpp#10267 for more details.

Gonna check in a bit. Thank you @slaren!

[mudler]

that was it indeed! just for reference if anyone else is facing this, I had to update the Docker base image from intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04 to intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04