Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.github/workflows/build.yml
#	.github/workflows/server.yml
#	CMakeLists.txt
#	Makefile
#	examples/embedding/embedding.cpp
#	examples/imatrix/imatrix.cpp
#	examples/llama-bench/llama-bench.cpp
#	examples/llava/MobileVLM-README.md
#	examples/parallel/parallel.cpp
#	examples/perplexity/perplexity.cpp
#	examples/quantize/CMakeLists.txt
#	examples/server/README.md
#	examples/speculative/speculative.cpp
#	tests/test-backend-ops.cpp

common/common.h

@@ -120,6 +120,7 @@ struct gpt_sampler_params {
     float   mirostat_eta      = 0.10f; // learning rate
     bool    penalize_nl       = false; // consider newlines as a repeatable token
     bool    ignore_eos        = false;
+    bool    no_perf           = false; // disable performance metrics
 
     std::vector<enum gpt_sampler_type> samplers = {
         GPT_SAMPLER_TYPE_TOP_K,
@@ -242,6 +243,7 @@ struct gpt_params {
     bool simple_io         = false; // improves compatibility with subprocesses and limited consoles
     bool cont_batching     = true;  // insert new sequences for decoding on-the-fly
     bool flash_attn        = false; // flash attention
+    bool no_perf           = false; // disable performance metrics
 
     bool input_prefix_bos  = false; // prefix BOS to user inputs, preceding input_prefix
     bool logits_all        = false; // return logits for all tokens in the batch