diff --git a/common/arg.cpp b/common/arg.cpp
index f2a7cb5c0..ebee1a5d2 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -1783,7 +1783,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP}));
     add_opt(common_arg(
         {"-t", "--threads"}, "N",
-        string_format("number of threads to use during generation (default: %d)", params.cpuparams.n_threads),
+        string_format("number of CPU threads to use during generation (default: %d)", params.cpuparams.n_threads),
         [](common_params & params, int value) {
             params.cpuparams.n_threads = value;
             if (params.cpuparams.n_threads <= 0) {
diff --git a/ggml/src/ggml-cpu/ggml-cpu.c b/ggml/src/ggml-cpu/ggml-cpu.c
index 621f64f18..60a7c13b8 100644
--- a/ggml/src/ggml-cpu/ggml-cpu.c
+++ b/ggml/src/ggml-cpu/ggml-cpu.c
@@ -4433,13 +4433,17 @@ void ggml_cpu_init(void) {
 #ifdef GGML_USE_OPENMP
             //if (!getenv("OMP_WAIT_POLICY")) {
             //    // set the wait policy to active, so that OpenMP threads don't sleep
-            //    putenv("OMP_WAIT_POLICY=active");
+            //    setenv("OMP_WAIT_POLICY", "active", 0)
             //}
 
             if (!getenv("KMP_BLOCKTIME")) {
                 // set the time to wait before sleeping a thread
                 // this is less aggressive than setting the wait policy to active, but should achieve similar results in most cases
-                putenv("KMP_BLOCKTIME=200"); // 200ms
+#ifdef _WIN32
+                _putenv_s("KMP_BLOCKTIME", "200"); // 200ms
+#else
+                setenv("KMP_BLOCKTIME", "200", 0); // 200ms
+#endif
             }
 #endif
         }
diff --git a/tools/mtmd/clip-impl.h b/tools/mtmd/clip-impl.h
index 7a7523851..1669fad99 100644
--- a/tools/mtmd/clip-impl.h
+++ b/tools/mtmd/clip-impl.h
@@ -30,6 +30,7 @@
 #define KEY_LAYER_NORM_EPS      "clip.%s.attention.layer_norm_epsilon"
 
 // vision-specific
+#define KEY_VISION_PROJ_TYPE    "clip.vision.projector_type" // for models with mixed modalities
 #define KEY_IMAGE_SIZE          "clip.vision.image_size"
 #define KEY_PREPROC_IMAGE_SIZE  "clip.vision.preproc_image_size"
 #define KEY_PATCH_SIZE          "clip.vision.patch_size"
@@ -48,6 +49,7 @@
 #define KEY_MINICPMV_QUERY_NUM    "clip.minicpmv_query_num"
 
 // audio-specific
+#define KEY_AUDIO_PROJ_TYPE     "clip.audio.projector_type" // for models with mixed modalities
 #define KEY_A_NUM_MEL_BINS      "clip.audio.num_mel_bins"
 #define KEY_A_PROJ_STACK_FACTOR "clip.audio.projector.stack_factor"
 
diff --git a/tools/mtmd/clip.cpp b/tools/mtmd/clip.cpp
index 7688c7062..ff42259e8 100644
--- a/tools/mtmd/clip.cpp
+++ b/tools/mtmd/clip.cpp
@@ -2260,15 +2260,27 @@ struct clip_model_loader {
         // projector type
         std::string proj_type;
         {
+            // default key
             get_string(KEY_PROJ_TYPE, proj_type, false);
-            if (!proj_type.empty()) {
-                model.proj_type = clip_projector_type_from_string(proj_type);
+
+            // for models with mixed modalities
+            if (proj_type.empty()) {
+                if (modality == CLIP_MODALITY_VISION) {
+                    get_string(KEY_VISION_PROJ_TYPE, proj_type, false);
+                } else if (modality == CLIP_MODALITY_AUDIO) {
+                    get_string(KEY_AUDIO_PROJ_TYPE, proj_type, false);
+                } else {
+                    GGML_ABORT("unknown modality");
+                }
             }
+
+            model.proj_type = clip_projector_type_from_string(proj_type);
+
             if (model.proj_type == PROJECTOR_TYPE_UNKNOWN) {
                 throw std::runtime_error(string_format("%s: unknown projector type: %s\n", __func__, proj_type.c_str()));
             }
 
-            // correct arch for multimodal models
+            // correct arch for multimodal models (legacy method)
             if (model.proj_type == PROJECTOR_TYPE_QWEN25O) {
                 model.proj_type = modality == CLIP_MODALITY_VISION
                                     ? PROJECTOR_TYPE_QWEN25VL
diff --git a/tools/server/public/index.html.gz b/tools/server/public/index.html.gz
index 1c62ebe96..34818012c 100644
Binary files a/tools/server/public/index.html.gz and b/tools/server/public/index.html.gz differ
diff --git a/tools/server/webui/src/lib/components/app/misc/MarkdownContent.svelte b/tools/server/webui/src/lib/components/app/misc/MarkdownContent.svelte
index 24d29c2b3..1f4caa900 100644
--- a/tools/server/webui/src/lib/components/app/misc/MarkdownContent.svelte
+++ b/tools/server/webui/src/lib/components/app/misc/MarkdownContent.svelte
@@ -154,9 +154,20 @@
 		return mutated ? tempDiv.innerHTML : html;
 	}
 
+	function normalizeMathDelimiters(text: string): string {
+		return text
+			.replace(/(^|[^\\])\\\[((?:\\.|[\s\S])*?)\\\]/g, (_, prefix: string, content: string) => {
+				return `${prefix}$$${content}$$`;
+			})
+			.replace(/(^|[^\\])\\\(((?:\\.|[\s\S])*?)\\\)/g, (_, prefix: string, content: string) => {
+				return `${prefix}$${content}$`;
+			});
+	}
+
 	async function processMarkdown(text: string): Promise<string> {
 		try {
-			const result = await processor().process(text);
+			const normalized = normalizeMathDelimiters(text);
+			const result = await processor().process(normalized);
 			const html = String(result);
 			const enhancedLinks = enhanceLinks(html);