auto rope scale adjustments, added sched yield fix for apple, adjust warning for mirostat

ggml.c

@@ -16367,7 +16367,11 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
             // wait for other threads to finish
             const int last = node_n;
             do {
+                #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_METAL)
+                //apple does nothing
+                #else
                 sched_yield();
+                #endif
                 node_n = atomic_load(&state->shared->node_n);
             } while (node_n == last);
         }

gpttype_adapter.cpp

@@ -367,7 +367,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
         else
         {
             //approximate NTK aware ctx
-            rope_freq_base = (params.n_ctx <= 4096 ? 40880.0f : 82684.0f);
+            rope_freq_base = (params.n_ctx <= 3072 ? 26000.0f : (params.n_ctx <= 4096 ? 32000.0f : (params.n_ctx <= 6144 ? 54000.0f : 82684.0f)));
         }
 
         printf("Using automatic RoPE scaling (scale:%.3f, base:%.1f)\n",rope_freq_scale,rope_freq_base);

koboldcpp.py

@@ -248,7 +248,7 @@ def generate(prompt,max_length=20, max_context_length=512, temperature=0.8, top_
                 inputs.sampler_order[i] = sampler
             inputs.sampler_len = len(sampler_order)
             global showsamplerwarning
-            if showsamplerwarning and inputs.sampler_len>0 and (inputs.sampler_order[0]!=6 or inputs.sampler_order[inputs.sampler_len-1]!=5):
+            if showsamplerwarning and inputs.mirostat==0 and inputs.sampler_len>0 and (inputs.sampler_order[0]!=6 or inputs.sampler_order[inputs.sampler_len-1]!=5):
                 print("\n(Note: Sub-optimal sampler_order detected. You may have reduced quality. Recommended sampler values are [6,0,1,3,4,2,5]. This message will only show once per session.)")
                 showsamplerwarning = False
         except TypeError as e: