Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.github/workflows/winget.yml

common/chat.cpp

@@ -1662,11 +1662,12 @@ static common_chat_params common_chat_params_init_lfm2(const common_chat_templat
     data.thinking_start_tag = THINK_START;
     data.thinking_end_tag   = THINK_END;
 
-    auto has_tools         = inputs.tools.is_array() && !inputs.tools.empty();
+    auto has_tools           = inputs.tools.is_array() && !inputs.tools.empty();
+    auto has_response_format = !inputs.json_schema.is_null() && inputs.json_schema.is_object();
     // Gate by reasoning format and whether the template supports <think>
     auto extract_reasoning = inputs.reasoning_format != COMMON_REASONING_FORMAT_NONE &&
                              tmpl.source().find(THINK_START) != std::string::npos;
-    auto include_grammar   = has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE;
+    auto include_grammar   = has_response_format || (has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE);
 
     if (inputs.has_continuation()) {
         const auto & msg = inputs.continue_msg;
@@ -1689,6 +1690,10 @@ static common_chat_params common_chat_params_init_lfm2(const common_chat_templat
         }
 
         if (!has_tools || inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_NONE) {
+            if (has_response_format) {
+                auto response_format = p.content(p.schema(p.json(), "response-format-schema", inputs.json_schema));
+                return generation_prompt + reasoning + response_format + end;
+            }
             return generation_prompt + reasoning + p.content(p.rest()) + end;
         }
         auto tool_calls = p.rule("tool-calls",
@@ -1707,13 +1712,17 @@ static common_chat_params common_chat_params_init_lfm2(const common_chat_templat
     data.parser = parser.save();
 
     if (include_grammar) {
-        data.grammar_lazy = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_AUTO;
+        data.grammar_lazy = !(has_response_format || (has_tools && inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED));
         data.grammar      = build_grammar([&](const common_grammar_builder & builder) {
             foreach_function(inputs.tools, [&](const json & tool) {
                 const auto & function = tool.at("function");
                 auto         schema   = function.at("parameters");
                 builder.resolve_refs(schema);
             });
+            if (has_response_format) {
+                auto schema = inputs.json_schema;
+                builder.resolve_refs(schema);
+            }
             parser.build_grammar(builder, data.grammar_lazy);
         });
 

common/speculative.cpp

@@ -843,7 +843,8 @@ struct common_speculative_impl_ngram_map_k : public common_speculative_impl {
     common_speculative_impl_ngram_map_k(
             const common_ngram_map & config,
             uint32_t n_seq)
-        : common_speculative_impl(COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K, n_seq)
+        : common_speculative_impl(config.key_only ? COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K
+            : COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K4V, n_seq)
     {
         for (uint32_t i = 0; i < n_seq; i++) {
             this->config.push_back(config);

ggml/src/ggml-cuda/ssm-scan.cu

@@ -67,6 +67,7 @@ __global__ void __launch_bounds__(splitD, 1)
     __shared__ CubTempStorage cub_temp_storage;
 
     BlockLoad(cub_temp_storage.load_temp).Load(A_block, regA);
+    __syncthreads();
     BlockLoad(cub_temp_storage.load_temp).Load(s0_block, regs0);
 #else
     const int stride_s0 = src0_nb2 / sizeof(float);
@@ -105,6 +106,7 @@ __global__ void __launch_bounds__(splitD, 1)
             regs0[n] = state;
         }
         y_block[i * stride_y + threadIdx.x] = sumf;
+        __syncthreads();
     }
 
 #ifdef USE_CUB
@@ -249,9 +251,8 @@ static void ssm_scan_f32_cuda(const float * src0, const float * src1, const floa
         GGML_ASSERT(head_dim == 1);
         GGML_ASSERT(n_group == 1);
         const dim3 blocks(n_seq, (n_head + threads - 1) / threads, 1);
-        const int  smem_size = (threads * (d_state + 1) * 2) * sizeof(float);
         if (d_state == 16) {
-            const ggml_cuda_kernel_launch_params launch_params = ggml_cuda_kernel_launch_params(blocks, threads, smem_size, stream);
+            const ggml_cuda_kernel_launch_params launch_params = ggml_cuda_kernel_launch_params(blocks, threads, 0, stream);
             switch (n_tok)
             {
             case 1:

src/llama-graph.cpp

@@ -1873,9 +1873,9 @@ ggml_tensor * llm_graph_context::build_inp_embd(ggml_tensor * tok_embd) const {
     res->t_inp_embd = cur;
 
     // For Granite architecture
-    // NOTE: Only apply scale to token inputs. Raw embeddings are assumed to be
-    //  multimodal inputs that should not be scaled.
-    if (ubatch.token && hparams.f_embedding_scale != 0.0f) {
+    // NOTE: For deepstack models, only apply scale to token inputs (ie text-only input).
+    //  Raw embeddings are assumed to be multimodal inputs that should not be scaled.
+    if (hparams.f_embedding_scale != 0.0f && (ubatch.token || hparams.n_deepstack_layers == 0)) {
         if (!ggml_is_contiguous(cur)) {
             cur = ggml_cont(ctx0, cur);
         }

tools/mtmd/tests.sh

@@ -91,7 +91,6 @@ add_test_vision "ggml-org/LightOnOCR-1B-1025-GGUF:Q8_0"
 add_test_vision "ggml-org/DeepSeek-OCR-GGUF:Q8_0" -p "Free OCR." --chat-template deepseek-ocr
 add_test_vision "ggml-org/dots.ocr-GGUF:Q8_0" -p "OCR"
 add_test_vision "ggml-org/HunyuanOCR-GGUF:Q8_0" -p "OCR"
-add_test_vision "ggml-org/HunyuanVL-4B-GGUF:Q8_0"
 add_test_vision "ggml-org/gemma-4-E2B-it-GGUF:Q8_0" --jinja
 
 add_test_audio  "ggml-org/ultravox-v0_5-llama-3_2-1b-GGUF:Q8_0"

tools/ui/src/lib/components/app/navigation/SidebarNavigation/SidebarNavigation.svelte

@@ -1,7 +1,7 @@
 <script lang="ts">
 	import { goto } from '$app/navigation';
 	import { page } from '$app/state';
-	import { Trash2, Pencil, X } from '@lucide/svelte';
+	import { Trash2, Pencil, Pin, X } from '@lucide/svelte';
 	import { Button } from '$lib/components/ui/button';
 	import { DialogConfirmation } from '$lib/components/app';
 	import SidebarNavigationActions from './SidebarNavigationActions.svelte';
@@ -52,6 +52,14 @@
 
 	let conversationTree = $derived(buildConversationTree(filteredConversations));
 
+	let pinnedConversations = $derived.by(() => {
+		return conversationTree.filter(({ conversation }) => conversation.pinned);
+	});
+
+	let unpinnedConversations = $derived.by(() => {
+		return conversationTree.filter(({ conversation }) => !conversation.pinned);
+	});
+
 	let selectedConversationHasDescendants = $derived.by(() => {
 		if (!selectedConversation) return false;
 
@@ -199,6 +207,41 @@
 			/>
 		</Sidebar.Header>
 
+		{#if !isSearchModeActive && pinnedConversations.length > 0}
+			<Sidebar.Group class="p-0 px-4">
+				<Sidebar.GroupLabel>
+					<div class="flex items-center gap-1">
+						<Pin class="h-3.5 w-3.5" />
+						<span>Pinned</span>
+					</div>
+				</Sidebar.GroupLabel>
+				<Sidebar.GroupContent>
+					<Sidebar.Menu>
+						{#each pinnedConversations as { conversation, depth } (conversation.id)}
+							<Sidebar.MenuItem class="mb-1 p-0">
+								<SidebarNavigationConversationItem
+									conversation={{
+										id: conversation.id,
+										name: conversation.name,
+										lastModified: conversation.lastModified,
+										currNode: conversation.currNode,
+										forkedFromConversationId: conversation.forkedFromConversationId,
+										pinned: conversation.pinned
+									}}
+									{depth}
+									isActive={currentChatId === conversation.id}
+									onSelect={selectConversation}
+									onEdit={handleEditConversation}
+									onDelete={handleDeleteConversation}
+									onStop={handleStopGeneration}
+								/>
+							</Sidebar.MenuItem>
+						{/each}
+					</Sidebar.Menu>
+				</Sidebar.GroupContent>
+			</Sidebar.Group>
+		{/if}
+
 		<Sidebar.Group class="mt-2 h-[calc(100vh-21rem)] space-y-2 p-0 px-3">
 			{#if (filteredConversations.length > 0 && isSearchModeActive) || !isSearchModeActive}
 				<Sidebar.GroupLabel>
@@ -208,7 +251,7 @@
 
 			<Sidebar.GroupContent>
 				<Sidebar.Menu>
-					{#each conversationTree as { conversation, depth } (conversation.id)}
+					{#each isSearchModeActive ? conversationTree : unpinnedConversations as { conversation, depth } (conversation.id)}
 						<Sidebar.MenuItem class="mb-1 p-0">
 							<SidebarNavigationConversationItem
 								conversation={{
@@ -216,7 +259,8 @@
 									name: conversation.name,
 									lastModified: conversation.lastModified,
 									currNode: conversation.currNode,
-									forkedFromConversationId: conversation.forkedFromConversationId
+									forkedFromConversationId: conversation.forkedFromConversationId,
+									pinned: conversation.pinned
 								}}
 								{depth}
 								isActive={currentChatId === conversation.id}
@@ -228,7 +272,7 @@
 						</Sidebar.MenuItem>
 					{/each}
 
-					{#if conversationTree.length === 0}
+					{#if (isSearchModeActive ? conversationTree : unpinnedConversations).length === 0}
 						<div class="px-2 py-4 text-center">
 							<p class="mb-4 p-4 text-sm text-muted-foreground">
 								{searchQuery.length > 0

tools/ui/src/lib/components/app/navigation/SidebarNavigation/SidebarNavigationConversationItem.svelte

@@ -6,7 +6,9 @@
 		Download,
 		Loader2,
 		Square,
-		GitBranch
+		GitBranch,
+		Pin,
+		PinOff
 	} from '@lucide/svelte';
 	import { DropdownMenuActions } from '$lib/components/app';
 	import * as Tooltip from '$lib/components/ui/tooltip';
@@ -57,6 +59,10 @@
 		onStop?.(conversation.id);
 	}
 
+	function handleTogglePin() {
+		conversationsStore.toggleConversationPin(conversation.id);
+	}
+
 	function handleGlobalEditEvent(event: Event) {
 		const customEvent = event as CustomEvent<{ conversationId: string }>;
 
@@ -170,6 +176,14 @@
 				triggerTooltip="More actions"
 				bind:open={dropdownOpen}
 				actions={[
+					{
+						icon: conversation.pinned ? PinOff : Pin,
+						label: conversation.pinned ? 'Unpin' : 'Pin',
+						onclick: (e: Event) => {
+							e.stopPropagation();
+							handleTogglePin();
+						}
+					},
 					{
 						icon: Pencil,
 						label: 'Edit',

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

@@ -344,6 +344,22 @@ export class DatabaseService {
 	 *
 	 */
 
+	/**
+	 * Toggles the pinned status of a conversation.
+	 *
+	 * @param id - Conversation ID
+	 * @returns The new pinned status
+	 */
+	static async toggleConversationPin(id: string): Promise<boolean> {
+		const conversation = await db.conversations.get(id);
+		if (!conversation) {
+			throw new Error(`Conversation ${id} not found`);
+		}
+		const newPinnedState = !conversation.pinned;
+		await this.updateConversation(id, { pinned: newPinnedState });
+		return newPinnedState;
+	}
+
 	/**
 	 * Updates the conversation's current node (active branch).
 	 * This determines which conversation path is currently being viewed.

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

@@ -506,6 +506,33 @@ class ConversationsStore {
 		}
 	}
 
+	/**
+	 * Toggles the pinned status of a conversation.
+	 * @param convId - The conversation ID to toggle
+	 * @returns The new pinned status
+	 */
+	async toggleConversationPin(convId: string): Promise<boolean> {
+		try {
+			const newPinnedState = await DatabaseService.toggleConversationPin(convId);
+
+			const convIndex = this.conversations.findIndex((c) => c.id === convId);
+
+			if (convIndex !== -1) {
+				this.conversations[convIndex].pinned = newPinnedState;
+				this.conversations = [...this.conversations];
+			}
+
+			if (this.activeConversation?.id === convId) {
+				this.activeConversation = { ...this.activeConversation, pinned: newPinnedState };
+			}
+
+			return newPinnedState;
+		} catch (error) {
+			console.error('Failed to toggle conversation pin:', error);
+			return false;
+		}
+	}
+
 	/**
 	 * Updates conversation title with optional confirmation dialog based on settings
 	 * @param convId - The conversation ID to update
@@ -1057,6 +1084,14 @@ export const isConversationsInitialized = () => conversationsStore.isInitialized
  * Builds a flat tree of conversations with depth levels for nested forks.
  * Accepts a pre-filtered list so search filtering stays in the component.
  */
+
+// Pinned conversations first, then by lastModified descending
+const comparePinnedThenRecent = (a: DatabaseConversation, b: DatabaseConversation) => {
+	if (a.pinned && !b.pinned) return -1;
+	if (!a.pinned && b.pinned) return 1;
+	return b.lastModified - a.lastModified;
+};
+
 export function buildConversationTree(convs: DatabaseConversation[]): ConversationTreeItem[] {
 	const childrenByParent = new SvelteMap<string, DatabaseConversation[]>();
 	const forkIds = new SvelteSet<string>();
@@ -1081,7 +1116,7 @@ export function buildConversationTree(convs: DatabaseConversation[]): Conversati
 
 		const children = childrenByParent.get(conv.id);
 		if (children) {
-			children.sort((a, b) => b.lastModified - a.lastModified);
+			children.sort(comparePinnedThenRecent);
 
 			for (const child of children) {
 				walk(child, depth + 1);
@@ -1089,7 +1124,7 @@ export function buildConversationTree(convs: DatabaseConversation[]): Conversati
 		}
 	}
 
-	const roots = convs.filter((c) => !forkIds.has(c.id));
+	const roots = convs.filter((c) => !forkIds.has(c.id)).sort(comparePinnedThenRecent);
 	for (const root of roots) {
 		walk(root, 0);
 	}

tools/ui/src/lib/types/database.d.ts

@@ -15,6 +15,7 @@ export interface DatabaseConversation {
 	thinkingEnabled?: boolean;
 	reasoningEffort?: ReasoningEffort;
 	forkedFromConversationId?: string;
+	pinned?: boolean;
 }
 
 export interface DatabaseMessageExtraAudioFile {