RubyGems - llama_cpp - Versions diffs - 0.16.1 → 0.16.2 - Mend

llama_cpp 0.16.1 → 0.16.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (29) hide show

checksums.yaml +4 -4
data/CHANGELOG.md +7 -0
data/ext/llama_cpp/extconf.rb +1 -0
data/ext/llama_cpp/llama_cpp.cpp +12 -0
data/lib/llama_cpp/version.rb +2 -2
data/sig/llama_cpp.rbs +2 -0
data/vendor/tmp/llama.cpp/Makefile +10 -2
data/vendor/tmp/llama.cpp/ggml-backend.c +14 -3
data/vendor/tmp/llama.cpp/ggml-backend.h +3 -0
data/vendor/tmp/llama.cpp/ggml-cuda/mmq.cu +10 -10
data/vendor/tmp/llama.cpp/ggml-cuda/mmvq.cu +1 -1
data/vendor/tmp/llama.cpp/ggml-cuda/unary.cu +28 -0
data/vendor/tmp/llama.cpp/ggml-impl.h +1 -1
data/vendor/tmp/llama.cpp/ggml-metal.m +6 -0
data/vendor/tmp/llama.cpp/ggml-quants.c +982 -368
data/vendor/tmp/llama.cpp/ggml-rpc.cpp +8 -3
data/vendor/tmp/llama.cpp/ggml-sycl.cpp +2124 -13202
data/vendor/tmp/llama.cpp/ggml-sycl.h +1 -10
data/vendor/tmp/llama.cpp/ggml-vulkan-shaders.hpp +27564 -23876
data/vendor/tmp/llama.cpp/ggml-vulkan.cpp +278 -366
data/vendor/tmp/llama.cpp/ggml.c +67 -150
data/vendor/tmp/llama.cpp/ggml.h +6 -0
data/vendor/tmp/llama.cpp/llama.cpp +530 -237
data/vendor/tmp/llama.cpp/llama.h +5 -1
data/vendor/tmp/llama.cpp/sgemm.cpp +2 -0
data/vendor/tmp/llama.cpp/unicode-data.cpp +851 -801
data/vendor/tmp/llama.cpp/unicode.cpp +33 -19
data/vendor/tmp/llama.cpp/unicode.h +1 -1
metadata +2 -2

data/vendor/tmp/llama.cpp/llama.h CHANGED Viewed

@@ -174,6 +174,7 @@ extern "C" {
         LLAMA_POOLING_TYPE_NONE = 0,
         LLAMA_POOLING_TYPE_MEAN = 1,
         LLAMA_POOLING_TYPE_CLS  = 2,
+        LLAMA_POOLING_TYPE_LAST = 3,
     };
     enum llama_split_mode {
@@ -293,7 +294,6 @@ extern "C" {
         enum llama_rope_scaling_type rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
         enum llama_pooling_type      pooling_type;      // whether to pool (sum) embedding results by sequence id
-                                                        // (ignored if no pooling layer)
         // ref: https://github.com/ggerganov/llama.cpp/pull/2054
         float    rope_freq_base;   // RoPE base frequency, 0 = from model
@@ -786,6 +786,10 @@ extern "C" {
     // Get the number of threads used for prompt and batch processing (multiple token).
     LLAMA_API uint32_t llama_n_threads_batch(struct llama_context * ctx);
+    // Set whether the model is in embeddings model or not
+    // If true, embeddings will be returned but logits will not
+    LLAMA_API void llama_set_embeddings(struct llama_context * ctx, bool embeddings);
     // Set whether to use causal attention or not
     // If set to true, the model will only attend to the past tokens
     LLAMA_API void llama_set_causal_attn(struct llama_context * ctx, bool causal_attn);

data/vendor/tmp/llama.cpp/sgemm.cpp CHANGED Viewed

@@ -43,8 +43,10 @@
 // [1] J. Tunney, ‘LLaMA Now Goes Faster on CPUs’, Mar. 2024. [Online].
 //     Available: https://justine.lol/matmul/. [Accessed: 29-Mar-2024].
+#if defined(__GNUC__)
 #pragma GCC diagnostic ignored "-Wpedantic"
 #pragma GCC diagnostic ignored "-Wignored-attributes"
+#endif
 #include "sgemm.h"
 #include "ggml-impl.h"