ui: add sync blocks so display/behavior settings can be set via --ui-config-file (#25132)

* ui: add sync blocks so display/behavior settings can be set via --ui-config-file

* ui: remove enable thinking setting

common/chat.cpp

@@ -2378,6 +2378,23 @@ static void func_args_not_string(json & messages) {
     }
 }
 
+// Trim leading/trailing whitespace from message contents before rendering. This
+// has to run on the messages (not on the rendered JSON) because templates with
+// string-only content caps concatenate typed content parts into a single string
+// during rendering, after which the per-part whitespace can no longer be reached.
+// Both the plain string content and the text of typed content parts are trimmed.
+static void trim_all_content(std::vector<common_chat_msg> & messages) {
+    for (auto & message : messages) {
+        message.content           = trim_whitespace(message.content);
+        message.reasoning_content = trim_whitespace(message.reasoning_content);
+        for (auto & part : message.content_parts) {
+            if (part.type == "text") {
+                part.text = trim_whitespace(part.text);
+            }
+        }
+    }
+}
+
 }
 
 // MiniCPM5 format:
@@ -2634,7 +2651,16 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
         params.tools.is_array() && tmpls->template_tool_use ? *tmpls->template_tool_use : *tmpls->template_default;
     const auto & src             = tmpl.source();
     const auto & caps            = tmpl.original_caps();
-    params.messages              = render_message_to_json(inputs.messages, tmpl.original_caps());
+    std::vector<common_chat_msg>        trimmed_messages;
+    const std::vector<common_chat_msg> * messages_to_render = &inputs.messages;
+    if (src.find("You have access to the following functions in JSONSchema format") != std::string::npos) {
+        // StepFun: trim message contents (including typed content parts) before rendering,
+        // otherwise leftover whitespace drives the model into reasoning loops (issue #24181)
+        trimmed_messages   = inputs.messages;
+        workaround::trim_all_content(trimmed_messages);
+        messages_to_render = &trimmed_messages;
+    }
+    params.messages              = render_message_to_json(*messages_to_render, tmpl.original_caps());
     params.tool_choice           = inputs.tool_choice;
     params.reasoning_format      = inputs.reasoning_format;
     params.enable_thinking       = inputs.enable_thinking;

common/speculative.cpp

@@ -955,10 +955,11 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {
         LOG_INF("%s: - block_size=%d, mask_token_id=%d, n_extract=%u\n", __func__, block_size, mask_token_id, target_layer_ids_n);
 
         // DFlash input is [id_last, <mask> * (block_size-1)], so it can draft at most block_size-1 tokens per step
-        if (this->params.n_max > block_size - 1) {
-            LOG_WRN("%s: requested draft size %d exceeds the trained DFlash block size %d -- clamping to %d draft tokens per step\n",
-                    __func__, this->params.n_max, block_size - 1, block_size - 1);
-            this->params.n_max = block_size - 1;
+        if (this->params.n_max > block_size - 1 || this->params.n_min > block_size - 1) {
+            LOG_WRN("%s: requested draft size (n_max=%d, n_min=%d) exceeds the trained DFlash block size %d -- clamping to %d\n",
+                    __func__, this->params.n_max, this->params.n_min, block_size, block_size - 1);
+            this->params.n_max = std::min(this->params.n_max, block_size - 1);
+            this->params.n_min = std::min(this->params.n_min, block_size - 1);
         }
 
         batch        = llama_batch_init(llama_n_batch(ctx_dft), 0,          n_seq);
@@ -968,7 +969,7 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {
         for (auto & s : smpls) {
             common_params_sampling sparams;
             sparams.no_perf  = false;
-            sparams.top_k    = 1;
+            sparams.top_k    = 10;
             sparams.samplers = { COMMON_SAMPLER_TYPE_TOP_K };
             s.reset(common_sampler_init(model_dft, sparams));
         }
@@ -1173,10 +1174,18 @@ struct common_speculative_impl_draft_dflash : public common_speculative_impl {
 
                 const llama_token id = cur_p->data[0].id;
 
+                if (cur_p->data[0].p < params.p_min) {
+                    break;
+                }
+
                 common_sampler_accept(smpl, id, true);
 
                 result.push_back(id);
             }
+
+            if (result.size() < (size_t) params.n_min) {
+                result.clear();
+            }
         }
     }
 

ggml/src/ggml-cpu/ops.cpp

@@ -1913,7 +1913,11 @@ static void ggml_compute_forward_concat_any(
     GGML_ASSERT(dim >= 0 && dim < 4);
 
     int64_t o[4] = {0, 0, 0, 0};
-    o[dim] = src0->ne[dim];
+    if (dim == 0) {
+        o[dim] = src0->ne[dim]/ggml_blck_size(src0->type);
+    } else {
+        o[dim] = src0->ne[dim];
+    }
 
     const char * x;
 
@@ -1921,8 +1925,8 @@ static void ggml_compute_forward_concat_any(
     for (int i3 = 0; i3 < ne3; i3++) {
         for (int i2 = ith; i2 < ne2; i2 += nth) {
             for (int i1 = 0; i1 < ne1; i1++) {
-                for (int i0 = 0; i0 < ne0; i0++) {
-                    if (i0 < ne00 && i1 < ne01 && i2 < ne02 && i3 < ne03) {
+                for (int i0 = 0; i0 < ne0/ggml_blck_size(dst->type); i0++) {
+                    if (i0 < ne00/ggml_blck_size(src0->type) && i1 < ne01 && i2 < ne02 && i3 < ne03) {
                         x = (const char *)src0->data + (i0       )*nb00 + (i1       )*nb01 + (i2       )*nb02 + (i3       )*nb03;
                     } else {
                         x = (const char *)src1->data + (i0 - o[0])*nb10 + (i1 - o[1])*nb11 + (i2 - o[2])*nb12 + (i3 - o[3])*nb13;
@@ -2071,6 +2075,14 @@ void ggml_compute_forward_concat(
     ggml_tensor * dst) {
 
     const ggml_tensor * src0 = dst->src[0];
+    const ggml_tensor * src1 = dst->src[1];
+
+    if (ggml_is_quantized(src0->type)) {
+        GGML_ASSERT(ggml_is_contiguous(src0));
+        GGML_ASSERT(ggml_is_contiguous(src1));
+        GGML_ASSERT(src0->ne[0] % ggml_blck_size(src0->type) == 0);
+        GGML_ASSERT(src1->ne[0] % ggml_blck_size(src1->type) == 0);
+    }
 
     switch (src0->type) {
         case GGML_TYPE_F16:

ggml/src/ggml-cuda/topk-moe.cu

@@ -312,6 +312,10 @@ static void launch_topk_moe_cuda(ggml_backend_cuda_context & ctx,
             ggml_cuda_kernel_launch(topk_moe_cuda<256, has_bias>, launch_params,
                 logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
             break;
+        case 288: // StepFun 3.7
+            ggml_cuda_kernel_launch(topk_moe_cuda<288, has_bias>, launch_params,
+                logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
+            break;
         case 512:
             ggml_cuda_kernel_launch(topk_moe_cuda<512, has_bias>, launch_params,
                 logits, weights, ids, bias, n_rows, n_expert_used, clamp_val, scale_val, config);
@@ -377,8 +381,10 @@ bool ggml_cuda_should_use_topk_moe(const ggml_tensor * gating_op,
                                    const ggml_tensor * weights,
                                    const ggml_tensor * logits,
                                    const ggml_tensor * ids) {
+    // must match an instantiation of launch_topk_moe_cuda: a power of 2 up to 512,
+    // or one of the non-power-of-2 expert counts of supported models
     const int n_expert = ids->nb[1] / ids->nb[0];
-    if (((n_expert & (n_expert - 1)) != 0 || n_expert > 512) && n_expert != 576) {
+    if (((n_expert & (n_expert - 1)) != 0 || n_expert > 512) && n_expert != 288 && n_expert != 576) {
         return false;
     }
 

models/templates/stepfun-ai-Step-3.5-Flash.jinja

@@ -1,80 +0,0 @@
-{% macro render_content(content) %}{% if content is none %}{{- '' }}{% elif content is string %}{{- content }}{% elif content is mapping %}{{- content['value'] if 'value' in content else content['text'] }}{% elif content is iterable %}{% for item in content %}{% if item.type == 'text' %}{{- item['value'] if 'value' in item else item['text'] }}{% elif item.type == 'image' %}<im_patch>{% endif %}{% endfor %}{% endif %}{% endmacro %}
-{{bos_token}}{%- if tools %}
-    {{- '<|im_start|>system\n' }}
-    {%- if messages[0].role == 'system' %}
-        {{- render_content(messages[0].content) + '\n\n' }}
-    {%- endif %}
-    {{- "# Tools\n\nYou have access to the following functions in JSONSchema format:\n\n<tools>" }}
-    {%- for tool in tools %}
-        {{- "\n" }}
-        {{- tool | tojson(ensure_ascii=False) }}
-    {%- endfor %}
-    {{- "\n</tools>\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...>\n...\n</function> block must be nested within <tool_call>\n...\n</tool_call> XML tags\n- Required parameters MUST be specified\n</IMPORTANT><|im_end|>\n" }}
-{%- else %}
-    {%- if messages[0].role == 'system' %}
-        {{- '<|im_start|>system\n' + render_content(messages[0].content) + '<|im_end|>\n' }}
-    {%- endif %}
-{%- endif %}
-{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
-{%- for message in messages[::-1] %}
-    {%- set index = (messages|length - 1) - loop.index0 %}
-    {%- if ns.multi_step_tool and message.role == "user" and render_content(message.content) is string and not(render_content(message.content).startswith('<tool_response>') and render_content(message.content).endswith('</tool_response>')) %}
-        {%- set ns.multi_step_tool = false %}
-        {%- set ns.last_query_index = index %}
-    {%- endif %}
-{%- endfor %}
-{%- for message in messages %}
-    {%- set content = render_content(message.content) %}
-    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
-        {%- set role_name = 'observation' if (message.role == "system" and not loop.first and message.name == 'observation') else message.role %}
-        {{- '<|im_start|>' + role_name + '\n' + content + '<|im_end|>' + '\n' }}
-    {%- elif message.role == "assistant" %}
-        {%- if message.reasoning_content is string %}
-            {%- set reasoning_content = render_content(message.reasoning_content) %}
-        {%- else %}
-            {%- if '</think>' in content %}
-                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
-                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
-            {%- else %}
-                {%- set reasoning_content = '' %}
-            {%- endif %}
-        {%- endif %}
-        {%- if loop.index0 > ns.last_query_index %}
-            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n' + content }}
-        {%- else %}
-            {{- '<|im_start|>' + message.role + '\n' + content }}
-        {%- endif %}
-        {%- if message.tool_calls %}
-            {%- for tool_call in message.tool_calls %}
-                {%- if tool_call.function is defined %}
-                    {%- set tool_call = tool_call.function %}
-                {%- endif %}
-                {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
-                {%- if tool_call.arguments is defined %}
-                    {%- set arguments = tool_call.arguments %}
-                    {%- for args_name, args_value in arguments|items %}
-                        {{- '<parameter=' + args_name + '>\n' }}
-                        {%- set args_value = args_value | tojson(ensure_ascii=False) | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
-                        {{- args_value }}
-                        {{- '\n</parameter>\n' }}
-                    {%- endfor %}
-                {%- endif %}
-                {{- '</function>\n</tool_call>' }}
-            {%- endfor %}
-        {%- endif %}
-        {{- '<|im_end|>\n' }}
-    {%- elif message.role == "tool" %}
-        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
-            {{- '<|im_start|>tool_response\n' }}
-        {%- endif %}
-        {{- '<tool_response>' }}
-        {{- content }}
-        {{- '</tool_response>' }}
-        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
-            {{- '<|im_end|>\n' }}
-        {%- endif %}
-    {%- endif %}
-{%- endfor %}
-{%- if add_generation_prompt %}
-    {{- '<|im_start|>assistant\n<think>\n' }}
-{%- endif %}

tests/test-backend-ops.cpp

@@ -8918,6 +8918,12 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
         }
     }
 
+    for (ggml_type type_a : { GGML_TYPE_Q4_0, GGML_TYPE_Q4_1, GGML_TYPE_Q5_0, GGML_TYPE_Q5_1, GGML_TYPE_Q8_0 }) {
+        for (int dim : { 0, 1, 2, 3, }) {
+            test_cases.emplace_back(new test_concat(type_a, {128, 12, 13, 14}, dim == 0 ? 256 : 7, dim, 0));
+        }
+    }
+
     for (ggml_sort_order order : {GGML_SORT_ORDER_ASC, GGML_SORT_ORDER_DESC}) {
         for (uint32_t i = 4; i <= 1024*1024; i *= 2) {
             test_cases.emplace_back(new test_argsort(GGML_TYPE_F32, {i-1, 1, 1, 1}));
@@ -9219,6 +9225,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
                     test_cases.emplace_back(new test_topk_moe({128, 1, 1, 1}, 128, with_norm, bias_probs, gate, scale_w));
                     test_cases.emplace_back(new test_topk_moe({129, 1, 1, 1}, 128, with_norm, bias_probs, gate, scale_w));
                     test_cases.emplace_back(new test_topk_moe({160, 4, 1, 1}, 160, with_norm, bias_probs, gate, scale_w));
+                    test_cases.emplace_back(new test_topk_moe({288, 22, 1, 1}, 8, with_norm, bias_probs, gate, scale_w)); // Used by StepFun 3.7
                 }
             }
         }

tests/test-chat-auto-parser.cpp

@@ -1887,7 +1887,6 @@ static void test_role_markers_all_templates(testing & t) {
         { "Qwen-Qwen3-0.6B.jinja",                           "<|im_start|>user",       "<|im_start|>assistant"      },
         { "Qwen-QwQ-32B.jinja",                              "<|im_start|>user",       "<|im_start|>assistant"      },
         { "StepFun3.5-Flash.jinja",                          "<|im_start|>user",       "<|im_start|>assistant"      },
-        { "stepfun-ai-Step-3.5-Flash.jinja",                 "<|im_start|>user",       "<|im_start|>assistant"      },
 
         // DeepSeek family
         { "deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja",  "<｜User｜>",                "<｜Assistant｜>"             },

tests/test-chat.cpp

@@ -3155,6 +3155,59 @@ static void test_template_output_peg_parsers(bool detailed_debug) {
                 }
             }
         }
+
+        {
+            // StepFun trimming regression test (see https://github.com/ggml-org/llama.cpp/pull/25238)
+            auto tmpls = read_templates("models/templates/StepFun3.5-Flash.jinja");
+
+            common_chat_msg message_chatbot = simple_assist_msg("Let me check.\n\n", "I am thinking.\n\n");
+
+            {
+                common_chat_templates_inputs inputs;
+                inputs.messages              = { message_chatbot };
+                inputs.add_generation_prompt = true;
+
+                auto params = common_chat_templates_apply(tmpls.get(), inputs);
+
+                if (params.prompt.find("Let me check.\n\n") != std::string::npos) {
+                    throw std::runtime_error("StepFun 3.5: content not trimmed");
+                }
+
+                if (params.prompt.find("I am thinking.\n\n") != std::string::npos) {
+                    throw std::runtime_error("StepFun 3.5: reasoning_content not trimmed");
+                }
+            }
+
+            {
+                // Trimming must also reach typed (text) content parts, not just string content
+                // (see https://github.com/ggml-org/llama.cpp/pull/25238)
+                common_chat_msg message_parts;
+                message_parts.role          = "user";
+                message_parts.content_parts = {
+                    { /* .type = */ "text", /* .text = */ "First part.\n\n" },
+                    { /* .type = */ "media_marker", /* .text = */ "<__media__>" },
+                    { /* .type = */ "text", /* .text = */ "Second part.\n\n" },
+                };
+
+                common_chat_templates_inputs inputs;
+                inputs.messages              = { message_parts };
+                inputs.add_generation_prompt = true;
+
+                auto params = common_chat_templates_apply(tmpls.get(), inputs);
+
+                if (params.prompt.find("First part.\n\n") != std::string::npos ||
+                    params.prompt.find("Second part.\n\n") != std::string::npos) {
+                    throw std::runtime_error("StepFun 3.5: text content parts not trimmed");
+                }
+
+                // the trimmed text itself must still be present
+                if (params.prompt.find("First part.") == std::string::npos ||
+                    params.prompt.find("Second part.") == std::string::npos) {
+                    throw std::runtime_error("StepFun 3.5: text content parts missing after trim");
+                }
+            }
+        }
+
     }
 
     {

tools/ui/src/lib/components/app/chat/ChatMessages/ChatMessages.svelte

@@ -20,9 +20,9 @@
 		agenticInjectSteeringMessage
 	} from '$lib/stores/agentic.svelte';
 	import {
+		buildSiblingInfoMap,
 		copyToClipboard,
 		formatMessageForClipboard,
-		getMessageSiblings,
 		hasAgenticContent
 	} from '$lib/utils';
 
@@ -169,6 +169,8 @@
 		});
 	});
 
+	let siblingInfoByMessageId = $derived(buildSiblingInfoMap(allConversationMessages));
+
 	let displayMessages = $derived.by(() => {
 		if (!messages.length) {
 			return [];
@@ -223,18 +225,18 @@
 				}
 			}
 
-			const siblingInfo = getMessageSiblings(allConversationMessages, msg.id);
+			const siblingInfo = siblingInfoByMessageId.get(msg.id) ?? {
+				message: msg,
+				siblingIds: [msg.id],
+				currentIndex: 0,
+				totalSiblings: 1
+			};
 
 			result.push({
 				message: msg,
 				toolMessages,
 				isLastAssistantMessage: false,
-				siblingInfo: siblingInfo || {
-					message: msg,
-					siblingIds: [msg.id],
-					currentIndex: 0,
-					totalSiblings: 1
-				}
+				siblingInfo
 			});
 		}
 

tools/ui/src/lib/constants/model-id.ts

@@ -37,3 +37,8 @@ export const MODEL_ACTIVATED_PARAMS_RE = /^[Aa]\d+(\.\d+)?[BbMmKkTt]$/;
  * Container format segments to exclude from tags (every model uses these).
  */
 export const MODEL_IGNORED_SEGMENTS = new Set(['GGUF', 'GGML']);
+
+/**
+ * Matches a trailing weight file extension, e.g. `model.gguf` -> `model`.
+ */
+export const MODEL_WEIGHT_EXTENSION_RE = /\.(gguf|ggml)$/i;

tools/ui/src/lib/constants/settings-keys.ts

@@ -69,7 +69,6 @@ export const SETTINGS_KEYS = {
 	// Developer
 	DISABLE_REASONING_PARSING: 'disableReasoningParsing',
 	EXCLUDE_REASONING_FROM_CONTEXT: 'excludeReasoningFromContext',
-	ENABLE_THINKING: 'enableThinking',
 	SHOW_RAW_OUTPUT_SWITCH: 'showRawOutputSwitch',
 	// PY_INTERPRETER_ENABLED: 'pyInterpreterEnabled',
 	JS_SANDBOX_ENABLED: 'jsSandboxEnabled',

tools/ui/src/lib/constants/settings-registry.ts

@@ -185,7 +185,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				defaultValue: false,
 				type: SettingsFieldType.CHECKBOX,
 				section: SETTINGS_SECTION_SLUGS.GENERAL,
-				isExperimental: true
+				isExperimental: true,
+				sync: {
+					serverKey: SETTINGS_KEYS.TITLE_GENERATION_USE_LLM,
+					paramType: SyncableParameterType.BOOLEAN
+				}
 			},
 			{
 				key: SETTINGS_KEYS.TITLE_GENERATION_PROMPT,
@@ -193,7 +197,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				help: 'Optional template for the title generation prompt. Use {{USER}} for the user message and {{ASSISTANT}} for the assistant message.',
 				defaultValue: TITLE_GENERATION.DEFAULT_PROMPT,
 				type: SettingsFieldType.TEXTAREA,
-				section: SETTINGS_SECTION_SLUGS.GENERAL
+				section: SETTINGS_SECTION_SLUGS.GENERAL,
+				sync: {
+					serverKey: SETTINGS_KEYS.TITLE_GENERATION_PROMPT,
+					paramType: SyncableParameterType.STRING
+				}
 			},
 			{
 				key: SETTINGS_KEYS.MAX_IMAGE_RESOLUTION,
@@ -201,7 +209,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				help: 'Images larger than this will be resized before sending to server. Set to 0 to disable.',
 				defaultValue: 0,
 				type: SettingsFieldType.INPUT,
-				section: SETTINGS_SECTION_SLUGS.GENERAL
+				section: SETTINGS_SECTION_SLUGS.GENERAL,
+				sync: {
+					serverKey: SETTINGS_KEYS.MAX_IMAGE_RESOLUTION,
+					paramType: SyncableParameterType.NUMBER
+				}
 			}
 		]
 	},
@@ -385,7 +397,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				help: 'Display the current build version in the bottom-right corner of the interface.',
 				defaultValue: false,
 				type: SettingsFieldType.CHECKBOX,
-				section: SETTINGS_SECTION_SLUGS.DISPLAY
+				section: SETTINGS_SECTION_SLUGS.DISPLAY,
+				sync: {
+					serverKey: SETTINGS_KEYS.SHOW_BUILD_VERSION,
+					paramType: SyncableParameterType.BOOLEAN
+				}
 			}
 		]
 	},
@@ -669,7 +685,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				help: 'After each response, re-submit the conversation to pre-fill the server KV cache. Makes the next turn faster since the prompt is already encoded while you read the response.',
 				defaultValue: false,
 				type: SettingsFieldType.CHECKBOX,
-				section: SETTINGS_SECTION_SLUGS.DEVELOPER
+				section: SETTINGS_SECTION_SLUGS.DEVELOPER,
+				sync: {
+					serverKey: SETTINGS_KEYS.PRE_ENCODE_CONVERSATION,
+					paramType: SyncableParameterType.BOOLEAN
+				}
 			},
 			{
 				key: SETTINGS_KEYS.DISABLE_REASONING_PARSING,
@@ -677,7 +697,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				help: 'Send reasoning_format=none so the server returns thinking tokens inline instead of extracting them into a separate field.',
 				defaultValue: false,
 				type: SettingsFieldType.CHECKBOX,
-				section: SETTINGS_SECTION_SLUGS.DEVELOPER
+				section: SETTINGS_SECTION_SLUGS.DEVELOPER,
+				sync: {
+					serverKey: SETTINGS_KEYS.DISABLE_REASONING_PARSING,
+					paramType: SyncableParameterType.BOOLEAN
+				}
 			},
 			{
 				key: SETTINGS_KEYS.EXCLUDE_REASONING_FROM_CONTEXT,
@@ -691,14 +715,6 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 					paramType: SyncableParameterType.BOOLEAN
 				}
 			},
-			{
-				key: SETTINGS_KEYS.ENABLE_THINKING,
-				label: 'Enable thinking',
-				help: 'Enable model thinking/reasoning for each request. When off, the model will skip the thinking phase and go straight to the response.',
-				defaultValue: false,
-				type: SettingsFieldType.CHECKBOX,
-				section: SETTINGS_SECTION_SLUGS.DEVELOPER
-			},
 			{
 				key: SETTINGS_KEYS.SHOW_RAW_OUTPUT_SWITCH,
 				label: 'Enable raw output toggle',
@@ -717,7 +733,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				help: 'Expose a run_javascript tool to the model. Code runs in a Web Worker inside a sandboxed iframe with an opaque origin, isolated from the WebUI and its API, with a hard timeout.',
 				defaultValue: false,
 				type: SettingsFieldType.CHECKBOX,
-				section: SETTINGS_SECTION_SLUGS.DEVELOPER
+				section: SETTINGS_SECTION_SLUGS.DEVELOPER,
+				sync: {
+					serverKey: SETTINGS_KEYS.JS_SANDBOX_ENABLED,
+					paramType: SyncableParameterType.BOOLEAN
+				}
 			},
 			{
 				key: SETTINGS_KEYS.CUSTOM_JSON,
@@ -753,7 +773,11 @@ const SETTINGS_REGISTRY: Record<string, SettingsSectionEntry> = {
 				defaultValue: DEFAULT_MCP_CONFIG.requestTimeoutSeconds,
 				type: SettingsFieldType.INPUT,
 				section: SETTINGS_SECTION_SLUGS.MCP,
-				isPositiveInteger: true
+				isPositiveInteger: true,
+				sync: {
+					serverKey: SETTINGS_KEYS.MCP_REQUEST_TIMEOUT_SECONDS,
+					paramType: SyncableParameterType.NUMBER
+				}
 			}
 		]
 	}

tools/ui/src/lib/services/migration.service.ts

@@ -551,13 +551,49 @@ const mcpDefaultEnabledMigration: Migration = {
 	}
 };
 
+const CONFIG_TYPES_MIGRATION_ID = 'config-type-normalization-v1';
+
+const configTypesMigration: Migration = {
+	id: CONFIG_TYPES_MIGRATION_ID,
+	description: 'Coerce legacy string-encoded booleans in persisted config to real booleans',
+
+	async run(): Promise<void> {
+		const configRaw = localStorage.getItem(CONFIG_LOCALSTORAGE_KEY);
+		if (configRaw === null) return;
+
+		const config = JSON.parse(configRaw);
+		let changed = false;
+
+		// Pre-schema configs persisted booleans as the strings "true"/"false", which the
+		// strict server schema now rejects. Coerce those back to real booleans. No config
+		// string field holds exactly "true"/"false", so the match is unambiguous.
+		for (const key of Object.keys(config)) {
+			if (config[key] === 'true') {
+				config[key] = true;
+				changed = true;
+			} else if (config[key] === 'false') {
+				config[key] = false;
+				changed = true;
+			}
+		}
+
+		if (changed) {
+			localStorage.setItem(CONFIG_LOCALSTORAGE_KEY, JSON.stringify(config));
+		}
+
+		if (import.meta.env.DEV && import.meta.env.VITE_DEBUG)
+			console.log(`[Migration] Config types: coerced string booleans (changed=${changed})`);
+	}
+};
+
 const migrations: Migration[] = [
 	localStorageMigration,
 	idxdbMigration,
 	legacyMessageMigration,
 	themeMigration,
 	customJsonKeyMigration,
-	mcpDefaultEnabledMigration
+	mcpDefaultEnabledMigration,
+	configTypesMigration
 ];
 
 export const MigrationService = {

tools/ui/src/lib/services/models.service.ts

@@ -1,5 +1,5 @@
 import { ServerModelStatus } from '$lib/enums';
-import { apiFetch, apiPost } from '$lib/utils';
+import { apiFetch, apiPost, normalizeModelName } from '$lib/utils';
 import type { ParsedModelId } from '$lib/types/models';
 import {
 	MODEL_QUANTIZATION_SEGMENT_RE,
@@ -7,6 +7,7 @@ import {
 	MODEL_PARAMS_RE,
 	MODEL_ACTIVATED_PARAMS_RE,
 	MODEL_IGNORED_SEGMENTS,
+	MODEL_WEIGHT_EXTENSION_RE,
 	MODEL_ID_NOT_FOUND,
 	MODEL_ID_ORG_SEPARATOR,
 	MODEL_ID_SEGMENT_SEPARATOR,
@@ -139,15 +140,19 @@ export class ModelsService {
 			tags: []
 		};
 
+		// strip directory path and weight extension so a bare `-m /path/file.gguf`
+		// parses like a clean repo id; the HF `org/model` form is preserved
+		const source = normalizeModelName(modelId).replace(MODEL_WEIGHT_EXTENSION_RE, '');
+
 		// 1. Extract colon-separated quantization (e.g. `model:Q4_K_M`)
-		const colonIdx = modelId.indexOf(MODEL_ID_QUANTIZATION_SEPARATOR);
+		const colonIdx = source.indexOf(MODEL_ID_QUANTIZATION_SEPARATOR);
 		let modelPath: string;
 
 		if (colonIdx !== MODEL_ID_NOT_FOUND) {
-			result.quantization = modelId.slice(colonIdx + 1) || null;
-			modelPath = modelId.slice(0, colonIdx);
+			result.quantization = source.slice(colonIdx + 1) || null;
+			modelPath = source.slice(0, colonIdx);
 		} else {
-			modelPath = modelId;
+			modelPath = source;
 		}
 
 		// 2. Extract org name (e.g. `org/model` -> org = "org")

tools/ui/src/lib/stores/conversations.svelte.ts

@@ -114,14 +114,13 @@ class ConversationsStore {
 
 	/** Load thinking-enabled default from localStorage */
 	private static loadThinkingDefaults(): boolean {
-		if (typeof globalThis.localStorage === 'undefined') return false;
+		if (typeof globalThis.localStorage === 'undefined') return true;
 		try {
 			const raw = localStorage.getItem(THINKING_ENABLED_DEFAULT_LOCALSTORAGE_KEY);
-			if (!raw) return false;
-			const parsed = raw === 'true';
-			return typeof parsed === 'boolean' ? parsed : false;
+			if (!raw) return true;
+			return raw === 'true';
 		} catch {
-			return false;
+			return true;
 		}
 	}
 
@@ -333,7 +332,7 @@ class ConversationsStore {
 			}
 
 			this.pendingMcpServerOverrides = [];
-			this.pendingThinkingEnabled = false;
+			this.pendingThinkingEnabled = ConversationsStore.loadThinkingDefaults();
 			this.activeConversation = conversation;
 
 			if (conversation.currNode) {

tools/ui/src/lib/utils/branching.ts

@@ -92,18 +92,14 @@ export function filterByLeafNodeId(
  * Finds the leaf node (message with no children) for a given message branch.
  * Traverses down the tree following the last child until reaching a leaf.
  *
- * @param messages - All messages in the conversation
+ * @param nodeMap - Map of messages keyed by ID
  * @param messageId - Starting message ID to find leaf for
  * @returns The leaf node ID, or the original messageId if no children
  */
-export function findLeafNode(messages: readonly DatabaseMessage[], messageId: string): string {
-	const nodeMap = new Map<string, DatabaseMessage>();
-
-	// Build node map for quick lookups
-	for (const msg of messages) {
-		nodeMap.set(msg.id, msg);
-	}
-
+function findLeafNodeInMap(
+	nodeMap: ReadonlyMap<string, DatabaseMessage>,
+	messageId: string
+): string {
 	let currentNode: DatabaseMessage | undefined = nodeMap.get(messageId);
 	while (currentNode && currentNode.children.length > 0) {
 		// Follow the last child (most recent branch)
@@ -114,6 +110,22 @@ export function findLeafNode(messages: readonly DatabaseMessage[], messageId: st
 	return currentNode?.id ?? messageId;
 }
 
+/**
+ * Convenience wrapper around {@link findLeafNodeInMap} for callers that only have
+ * a flat message array.
+ *
+ * Finds the leaf node (message with no children) for a given message branch.
+ * Traverses down the tree following the last child until reaching a leaf.
+ *
+ * @param messages - All messages in the conversation
+ * @param messageId - Starting message ID to find leaf for
+ * @returns The leaf node ID, or the original messageId if no children
+ */
+export function findLeafNode(messages: readonly DatabaseMessage[], messageId: string): string {
+	const nodeMap = new Map(messages.map((msg) => [msg.id, msg] as const));
+	return findLeafNodeInMap(nodeMap, messageId);
+}
+
 /**
  * Finds all descendant messages (children, grandchildren, etc.) of a given message.
  * This is used for cascading deletion to remove all messages in a branch.
@@ -156,21 +168,14 @@ export function findDescendantMessages(
  * Gets sibling information for a message, including all sibling IDs and current position.
  * Siblings are messages that share the same parent.
  *
- * @param messages - All messages in the conversation
+ * @param nodeMap - Map of messages keyed by ID
  * @param messageId - The message to get sibling info for
  * @returns Sibling information including leaf node IDs for navigation
  */
 export function getMessageSiblings(
-	messages: readonly DatabaseMessage[],
+	nodeMap: ReadonlyMap<string, DatabaseMessage>,
 	messageId: string
 ): ChatMessageSiblingInfo | null {
-	const nodeMap = new Map<string, DatabaseMessage>();
-
-	// Build node map for quick lookups
-	for (const msg of messages) {
-		nodeMap.set(msg.id, msg);
-	}
-
 	const message = nodeMap.get(messageId);
 	if (!message) {
 		return null;
@@ -203,7 +208,9 @@ export function getMessageSiblings(
 
 	// Convert sibling message IDs to their corresponding leaf node IDs
 	// This allows navigation between different conversation branches
-	const siblingLeafIds = siblingIds.map((siblingId: string) => findLeafNode(messages, siblingId));
+	const siblingLeafIds = siblingIds.map((siblingId: string) =>
+		findLeafNodeInMap(nodeMap, siblingId)
+	);
 
 	// Find current message's position among siblings
 	const currentIndex = siblingIds.indexOf(messageId);
@@ -217,85 +224,22 @@ export function getMessageSiblings(
 }
 
 /**
- * Creates a display-ready list of messages with sibling information for UI rendering.
- * This is the main function used by chat components to render conversation branches.
+ * Builds sibling information for every message in a conversation.
+ * A single node map is shared across all lookups for O(1) access.
  *
  * @param messages - All messages in the conversation
- * @param leafNodeId - Current leaf node being viewed
- * @returns Array of messages with sibling navigation info
+ * @returns Map of message ID to its sibling information
  */
-export function getMessageDisplayList(
-	messages: readonly DatabaseMessage[],
-	leafNodeId: string
-): ChatMessageSiblingInfo[] {
-	// Get the current conversation path
-	const currentPath = filterByLeafNodeId(messages, leafNodeId, true);
-	const result: ChatMessageSiblingInfo[] = [];
-
-	// Add sibling info for each message in the current path
-	for (const message of currentPath) {
-		if (message.type === 'root') {
-			continue; // Skip root messages in display
-		}
-
-		const siblingInfo = getMessageSiblings(messages, message.id);
-		if (siblingInfo) {
-			result.push(siblingInfo);
+export function buildSiblingInfoMap(
+	messages: readonly DatabaseMessage[]
+): Map<string, ChatMessageSiblingInfo> {
+	const nodeMap = new Map(messages.map((msg) => [msg.id, msg] as const));
+	const siblingMap = new Map<string, ChatMessageSiblingInfo>();
+	for (const msg of messages) {
+		const info = getMessageSiblings(nodeMap, msg.id);
+		if (info) {
+			siblingMap.set(msg.id, info);
 		}
 	}
-
-	return result;
-}
-
-/**
- * Checks if a message has multiple siblings (indicating branching at that point).
- *
- * @param messages - All messages in the conversation
- * @param messageId - The message to check
- * @returns True if the message has siblings
- */
-export function hasMessageSiblings(
-	messages: readonly DatabaseMessage[],
-	messageId: string
-): boolean {
-	const siblingInfo = getMessageSiblings(messages, messageId);
-	return siblingInfo ? siblingInfo.totalSiblings > 1 : false;
-}
-
-/**
- * Gets the next sibling message ID for navigation.
- *
- * @param messages - All messages in the conversation
- * @param messageId - Current message ID
- * @returns Next sibling's leaf node ID, or null if at the end
- */
-export function getNextSibling(
-	messages: readonly DatabaseMessage[],
-	messageId: string
-): string | null {
-	const siblingInfo = getMessageSiblings(messages, messageId);
-	if (!siblingInfo || siblingInfo.currentIndex >= siblingInfo.totalSiblings - 1) {
-		return null;
-	}
-
-	return siblingInfo.siblingIds[siblingInfo.currentIndex + 1];
-}
-
-/**
- * Gets the previous sibling message ID for navigation.
- *
- * @param messages - All messages in the conversation
- * @param messageId - Current message ID
- * @returns Previous sibling's leaf node ID, or null if at the beginning
- */
-export function getPreviousSibling(
-	messages: readonly DatabaseMessage[],
-	messageId: string
-): string | null {
-	const siblingInfo = getMessageSiblings(messages, messageId);
-	if (!siblingInfo || siblingInfo.currentIndex <= 0) {
-		return null;
-	}
-
-	return siblingInfo.siblingIds[siblingInfo.currentIndex - 1];
+	return siblingMap;
 }

tools/ui/src/lib/utils/index.ts

@@ -26,10 +26,7 @@ export {
 	findLeafNode,
 	findDescendantMessages,
 	getMessageSiblings,
-	getMessageDisplayList,
-	hasMessageSiblings,
-	getNextSibling,
-	getPreviousSibling
+	buildSiblingInfoMap
 } from './branching';
 
 // Code