Merge commit '335eb04a91f481f37c0c9b302ee31b449b04c3e9' into concedo_experimental

# Conflicts:
#	.github/workflows/build.yml
#	CONTRIBUTING.md
#	Makefile
#	docs/build.md
#	examples/llama.swiftui/llama.swiftui/UI/ContentView.swift
#	examples/run/run.cpp
#	ggml/CMakeLists.txt
#	ggml/src/ggml-cpu/CMakeLists.txt
#	ggml/src/ggml-cuda/CMakeLists.txt
#	ggml/src/ggml-musa/CMakeLists.txt

examples/llava/clip.cpp

@@ -2811,9 +2811,13 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
 
             if (!ctx->has_glm_projector) {
                 struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
+                // The patches vector is used to get rows to index into the embeds with;
+                // we should skip dim 0 only if we have CLS to avoid going out of bounds
+                // when retrieving the rows.
+                int patch_offset = ctx->has_class_embedding ? 1 : 0;
                 int* patches_data = (int*)malloc(ggml_nbytes(patches));
                 for (int i = 0; i < num_patches; i++) {
-                    patches_data[i] = i + 1;
+                    patches_data[i] = i + patch_offset;
                 }
                 ggml_backend_tensor_set(patches, patches_data, 0, ggml_nbytes(patches));
                 free(patches_data);

examples/server/utils.hpp

@@ -7,6 +7,8 @@
 
 // increase max payload length to allow use of larger context size
 #define CPPHTTPLIB_FORM_URL_ENCODED_PAYLOAD_MAX_LENGTH 1048576
+// disable Nagle's algorithm
+#define CPPHTTPLIB_TCP_NODELAY true
 #include "httplib.h"
 
 // Change JSON_ASSERT from assert() to GGML_ASSERT:

examples/server/webui/src/components/ChatScreen.tsx

@@ -228,6 +228,7 @@ export default function ChatScreen() {
             value={inputMsg}
             onChange={(e) => setInputMsg(e.target.value)}
             onKeyDown={(e) => {
+              if (e.nativeEvent.isComposing || e.keyCode === 229) return;
               if (e.key === 'Enter' && e.shiftKey) return;
               if (e.key === 'Enter' && !e.shiftKey) {
                 e.preventDefault();

examples/server/webui/src/utils/llama-vscode.ts

@@ -40,7 +40,7 @@ export const useVSCodeContext = (
 
     window.addEventListener('message', handleMessage);
     return () => window.removeEventListener('message', handleMessage);
-  }, []);
+  }, [inputRef, setInputMsg]);
 
   // Add a keydown listener that sends the "escapePressed" message to the parent window
   useEffect(() => {