Distribute and run LLMs with a single file.
C++ cross-platformggufllama-cpplocal-ailocal-inferencelocal-llmopen-source-aisingle-file-executablespeech-to-text
Fix new tracing feature
Justine Tunneyedc3f2da2fc9
diff --git a/llama.cpp/ggml.c b/llama.cpp/ggml.cindex b20137da..fc306a8f 100644--- a/llama.cpp/ggml.c+++ b/llama.cpp/ggml.c@@ -1600,7 +1600,6 @@ int ggml_delay(int backoff) {// creates barrier and blocks until all threads call thisvoid ggml_syncthreads(struct ggml_barrier * b, int nth) {- llamafile_trace_begin("syncthreads");unsigned phase = atomic_load_explicit(&b->phase, memory_order_relaxed);if (atomic_fetch_add_explicit(&b->count, 1, memory_order_acq_rel) + 1 == nth) {atomic_store_explicit(&b->count, 0, memory_order_relaxed);@@ -1611,7 +1610,6 @@ void ggml_syncthreads(struct ggml_barrier * b, int nth) {while (atomic_load_explicit(&b->phase, memory_order_relaxed) == phase)backoff = ggml_delay(backoff);}- llamafile_trace_end("syncthreads");}//@@ -11468,7 +11466,6 @@ UseGgmlGemm1:;assert(params->wsize >= ne11*ne12*ne13*row_size);GGML_ASSERT(src1->type == GGML_TYPE_F32);- llamafile_trace_begin(ggml_type_name(vec_dot_type));unsigned chore = 0; // [jart] need for speedfor (int64_t i13 = 0; i13 < ne13; ++i13) {for (int64_t i12 = 0; i12 < ne12; ++i12) {@@ -11480,7 +11477,6 @@ UseGgmlGemm1:;}}}- llamafile_trace_end(ggml_type_name(vec_dot_type));ggml_syncthreads(params->barrier, params->nth);}