diff --git a/.github/workflows/compile.yml b/.github/workflows/compile.yml index 8f591eea9..36f385c86 100644 --- a/.github/workflows/compile.yml +++ b/.github/workflows/compile.yml @@ -370,7 +370,7 @@ jobs: cp artifacts/llava-bin-win-cublas-cu11.7.1-x64.dll/llava_shared.dll deps/cu11.7.1/llava_shared.dll cp artifacts/llama-bin-linux-cublas-cu11.7.1-x64.so/libllama.so deps/cu11.7.1/libllama.so - cp artifacts/llava-bin-linux-cublas-cu11.7.1-x64.so/libllava_shared.so deps/cu11.7.1/libllama_shared.so + cp artifacts/llava-bin-linux-cublas-cu11.7.1-x64.so/libllava_shared.so deps/cu11.7.1/libllava_shared.so cp artifacts/llama-bin-win-cublas-cu12.1.0-x64.dll/llama.dll deps/cu12.1.0/llama.dll cp artifacts/llava-bin-win-cublas-cu12.1.0-x64.dll/llava_shared.dll deps/cu12.1.0/llava_shared.dll diff --git a/LLama.Examples/Examples/LlavaInteractiveModeExecute.cs b/LLama.Examples/Examples/LlavaInteractiveModeExecute.cs index 34c7aca01..f7bf24945 100644 --- a/LLama.Examples/Examples/LlavaInteractiveModeExecute.cs +++ b/LLama.Examples/Examples/LlavaInteractiveModeExecute.cs @@ -22,6 +22,7 @@ public static async Task Run() { ContextSize = 4096, Seed = 1337, + GpuLayerCount = 10 }; using var model = LLamaWeights.LoadFromFile(parameters); using var context = model.CreateContext(parameters); diff --git a/LLama/LLamaSharp.Runtime.targets b/LLama/LLamaSharp.Runtime.targets index 35534d3fb..fa7a018c1 100644 --- a/LLama/LLamaSharp.Runtime.targets +++ b/LLama/LLamaSharp.Runtime.targets @@ -84,6 +84,15 @@ PreserveNewest runtimes/win-x64/native/avx512/llava_shared.dll + + PreserveNewest + runtimes/win-x64/native/cuda11/llava_shared.dll + + + PreserveNewest + runtimes/win-x64/native/cuda12/llava_shared.dll + + PreserveNewest @@ -101,6 +110,14 @@ PreserveNewest runtimes/linux-x64/native/avx512/libllava_shared.so + + PreserveNewest + runtimes/linux-x64/native/cuda11/libllava_shared.so + + + PreserveNewest + runtimes/linux-x64/native/cuda12/libllava_shared.so + PreserveNewest diff --git a/LLama/runtimes/deps/avx/libllama.dll b/LLama/runtimes/deps/avx/libllama.dll new file mode 100644 index 000000000..f09813ae6 Binary files /dev/null and b/LLama/runtimes/deps/avx/libllama.dll differ diff --git a/LLama/runtimes/deps/avx/llama.dll b/LLama/runtimes/deps/avx/llama.dll index f4769c71f..f09813ae6 100644 Binary files a/LLama/runtimes/deps/avx/llama.dll and b/LLama/runtimes/deps/avx/llama.dll differ diff --git a/LLama/runtimes/deps/avx/llava_shared.dll b/LLama/runtimes/deps/avx/llava_shared.dll index 4bc90fb52..546da7588 100644 Binary files a/LLama/runtimes/deps/avx/llava_shared.dll and b/LLama/runtimes/deps/avx/llava_shared.dll differ diff --git a/LLama/runtimes/deps/avx2/libllama.dll b/LLama/runtimes/deps/avx2/libllama.dll new file mode 100644 index 000000000..481be2352 Binary files /dev/null and b/LLama/runtimes/deps/avx2/libllama.dll differ diff --git a/LLama/runtimes/deps/avx2/llama.dll b/LLama/runtimes/deps/avx2/llama.dll index 71db06d82..481be2352 100644 Binary files a/LLama/runtimes/deps/avx2/llama.dll and b/LLama/runtimes/deps/avx2/llama.dll differ diff --git a/LLama/runtimes/deps/avx2/llava_shared.dll b/LLama/runtimes/deps/avx2/llava_shared.dll index b6a9b5c8d..f877c590f 100644 Binary files a/LLama/runtimes/deps/avx2/llava_shared.dll and b/LLama/runtimes/deps/avx2/llava_shared.dll differ diff --git a/LLama/runtimes/deps/avx512/libllama.dll b/LLama/runtimes/deps/avx512/libllama.dll new file mode 100644 index 000000000..9f3030289 Binary files /dev/null and b/LLama/runtimes/deps/avx512/libllama.dll differ diff --git a/LLama/runtimes/deps/avx512/llama.dll b/LLama/runtimes/deps/avx512/llama.dll index 7a484c4a7..9f3030289 100644 Binary files a/LLama/runtimes/deps/avx512/llama.dll and b/LLama/runtimes/deps/avx512/llama.dll differ diff --git a/LLama/runtimes/deps/avx512/llava_shared.dll b/LLama/runtimes/deps/avx512/llava_shared.dll index 3ff583286..e0cfbe44a 100644 Binary files a/LLama/runtimes/deps/avx512/llava_shared.dll and b/LLama/runtimes/deps/avx512/llava_shared.dll differ diff --git a/LLama/runtimes/deps/clblast/libllama.so b/LLama/runtimes/deps/clblast/libllama.so index 5eab24e8b..9b5f87900 100644 Binary files a/LLama/runtimes/deps/clblast/libllama.so and b/LLama/runtimes/deps/clblast/libllama.so differ diff --git a/LLama/runtimes/deps/clblast/libllava_shared.so b/LLama/runtimes/deps/clblast/libllava_shared.so new file mode 100644 index 000000000..764e7266d Binary files /dev/null and b/LLama/runtimes/deps/clblast/libllava_shared.so differ diff --git a/LLama/runtimes/deps/clblast/llama.dll b/LLama/runtimes/deps/clblast/llama.dll index 8660b110d..a08951358 100644 Binary files a/LLama/runtimes/deps/clblast/llama.dll and b/LLama/runtimes/deps/clblast/llama.dll differ diff --git a/LLama/runtimes/deps/clblast/llava_shared.dll b/LLama/runtimes/deps/clblast/llava_shared.dll new file mode 100644 index 000000000..e4a51d0ba Binary files /dev/null and b/LLama/runtimes/deps/clblast/llava_shared.dll differ diff --git a/LLama/runtimes/deps/cu11.7.1/libllama.so b/LLama/runtimes/deps/cu11.7.1/libllama.so index b76d2d0e4..ef9baa519 100644 Binary files a/LLama/runtimes/deps/cu11.7.1/libllama.so and b/LLama/runtimes/deps/cu11.7.1/libllama.so differ diff --git a/LLama/runtimes/deps/cu11.7.1/libllava_shared.so b/LLama/runtimes/deps/cu11.7.1/libllava_shared.so new file mode 100644 index 000000000..7ad6a066e Binary files /dev/null and b/LLama/runtimes/deps/cu11.7.1/libllava_shared.so differ diff --git a/LLama/runtimes/deps/cu11.7.1/llama.dll b/LLama/runtimes/deps/cu11.7.1/llama.dll index 0498c5b44..22cd79574 100644 Binary files a/LLama/runtimes/deps/cu11.7.1/llama.dll and b/LLama/runtimes/deps/cu11.7.1/llama.dll differ diff --git a/LLama/runtimes/deps/cu11.7.1/llava_shared.dll b/LLama/runtimes/deps/cu11.7.1/llava_shared.dll new file mode 100644 index 000000000..a5d1c514a Binary files /dev/null and b/LLama/runtimes/deps/cu11.7.1/llava_shared.dll differ diff --git a/LLama/runtimes/deps/cu12.1.0/libllama.so b/LLama/runtimes/deps/cu12.1.0/libllama.so index f039bacfd..ac66c69f8 100644 Binary files a/LLama/runtimes/deps/cu12.1.0/libllama.so and b/LLama/runtimes/deps/cu12.1.0/libllama.so differ diff --git a/LLama/runtimes/deps/cu12.1.0/libllava_shared.so b/LLama/runtimes/deps/cu12.1.0/libllava_shared.so new file mode 100644 index 000000000..166633a80 Binary files /dev/null and b/LLama/runtimes/deps/cu12.1.0/libllava_shared.so differ diff --git a/LLama/runtimes/deps/cu12.1.0/llama.dll b/LLama/runtimes/deps/cu12.1.0/llama.dll index 4b1ecd9c1..b12c7776d 100644 Binary files a/LLama/runtimes/deps/cu12.1.0/llama.dll and b/LLama/runtimes/deps/cu12.1.0/llama.dll differ diff --git a/LLama/runtimes/deps/cu12.1.0/llava_shared.dll b/LLama/runtimes/deps/cu12.1.0/llava_shared.dll new file mode 100644 index 000000000..fdef226c3 Binary files /dev/null and b/LLama/runtimes/deps/cu12.1.0/llava_shared.dll differ diff --git a/LLama/runtimes/deps/libllama.dll b/LLama/runtimes/deps/libllama.dll new file mode 100644 index 000000000..bd256c0a8 Binary files /dev/null and b/LLama/runtimes/deps/libllama.dll differ diff --git a/LLama/runtimes/deps/llama.dll b/LLama/runtimes/deps/llama.dll index 6c6938359..bd256c0a8 100644 Binary files a/LLama/runtimes/deps/llama.dll and b/LLama/runtimes/deps/llama.dll differ diff --git a/LLama/runtimes/deps/llava_shared.dll b/LLama/runtimes/deps/llava_shared.dll index 11186175e..d1aafcad9 100644 Binary files a/LLama/runtimes/deps/llava_shared.dll and b/LLama/runtimes/deps/llava_shared.dll differ diff --git a/LLama/runtimes/deps/osx-arm64/libllama.dylib b/LLama/runtimes/deps/osx-arm64/libllama.dylib index 89eee34b9..87295f843 100644 Binary files a/LLama/runtimes/deps/osx-arm64/libllama.dylib and b/LLama/runtimes/deps/osx-arm64/libllama.dylib differ diff --git a/LLama/runtimes/deps/osx-arm64/libllava_shared.dylib b/LLama/runtimes/deps/osx-arm64/libllava_shared.dylib index aba2882ff..84ff71671 100644 Binary files a/LLama/runtimes/deps/osx-arm64/libllava_shared.dylib and b/LLama/runtimes/deps/osx-arm64/libllava_shared.dylib differ diff --git a/README.md b/README.md index a3e50e4c4..25185b47e 100644 --- a/README.md +++ b/README.md @@ -242,7 +242,7 @@ If you want to compile llama.cpp yourself you **must** use the exact commit ID l | v0.8.1 | | [`e937066`](https://github.com/ggerganov/llama.cpp/commit/e937066420b79a757bf80e9836eb12b88420a218) | | v0.9.0, v0.9.1 | [Mixtral-8x7B](https://huggingface.co/TheBloke/Mixtral-8x7B-v0.1-GGUF) | [`9fb13f9`](https://github.com/ggerganov/llama.cpp/blob/9fb13f95840c722ad419f390dc8a9c86080a3700) | | v0.10.0 | [Phi2](https://huggingface.co/TheBloke/phi-2-GGUF) | [`d71ac90`](https://github.com/ggerganov/llama.cpp/tree/d71ac90985854b0905e1abba778e407e17f9f887) | -| v0.11.0 | [LLaVA-v1.6](https://huggingface.co/ShadowBeast/llava-v1.6-mistral-7b-Q5_K_S-GGUF), [Phi2](https://huggingface.co/TheBloke/phi-2-GGUF)| [`3ab8b3a`](https://github.com/ggerganov/llama.cpp/tree/3ab8b3a92ede46df88bc5a2dfca3777de4a2b2b6) | +| v0.11.0 | [LLaVA-v1.5](https://hf-mirror.com/jartine/llava-v1.5-7B-GGUF/blob/main/llava-v1.5-7b-mmproj-Q4_0.gguf), [Phi2](https://huggingface.co/TheBloke/phi-2-GGUF)| [`3ab8b3a`](https://github.com/ggerganov/llama.cpp/tree/3ab8b3a92ede46df88bc5a2dfca3777de4a2b2b6) | ## License