@tryhamster/gerbil 1.1.1 → 1.1.3

package/README.md

@@ -123,12 +123,12 @@ For a copy-pasteable version see [`examples/benchmark.ts`](./examples/benchmark.
 
 ## React Quickstart
 
-`useEngine` (from `@tryhamster/gerbil/gpu/hooks`) owns the full engine lifecycle —
+`useEngine` (from `@tryhamster/gerbil/hooks`) owns the full engine lifecycle —
 load, unload, hot-swap on config change, and reference-counted sharing so multiple
 components never upload the same weights to the GPU twice.
 
 ```tsx
-import { useEngine } from "@tryhamster/gerbil/gpu/hooks";
+import { useEngine } from "@tryhamster/gerbil/hooks";
 
 function Chat() {
   const { complete, completion, isLoading, isGenerating, tps } = useEngine({
@@ -156,7 +156,7 @@ and `dispose`. Pass `enableVision: true` or `embedding: true` to load those moda
 until the engine is ready — and keep it warm across navigation (no GPU re-upload):
 
 ```tsx
-import { GerbilGate } from "@tryhamster/gerbil/gpu/hooks";
+import { GerbilGate } from "@tryhamster/gerbil/hooks";
 
 <GerbilGate
   model="mlx-community/Qwen3.5-0.8B-4bit"
@@ -199,10 +199,10 @@ const engine = await WebGPUEngine.create({ repo: "mlx-community/Qwen3.5-0.8B-4bi
 await engine.generateObject("…", { schema: { required: ["title"] } });
 ```
 
-In React, use `useObject` (from `@tryhamster/gerbil/gpu/hooks`):
+In React, use `useObject` (from `@tryhamster/gerbil/hooks`):
 
 ```tsx
-import { useObject } from "@tryhamster/gerbil/gpu/hooks";
+import { useObject } from "@tryhamster/gerbil/hooks";
 
 const { generate, object, isGenerating } = useObject<{ city: string }>();
 await generate("Extract the city from: I live in Paris", {
@@ -322,17 +322,19 @@ const skill = useSkill("my-skill");
 
 ## Tools & Agents
 
-Gerbil supports tool calling with Qwen3 models for agentic workflows:
+Gerbil supports tool calling with Qwen3 models for agentic workflows. A tool is a
+plain object (`AgentTool`) — give it a `name`, `description`, optional `parameters`,
+and an `execute` function:
 
 ```typescript
-import { defineTool, executeToolCall } from "@tryhamster/gerbil";
+import type { AgentTool } from "@tryhamster/gerbil/gpu";
 
-const weatherTool = defineTool({
+const weatherTool: AgentTool = {
   name: "get_weather",
   description: "Get weather for a city",
-  parameters: z.object({ city: z.string() }),
-  execute: async ({ city }) => `Weather in ${city}: 72°F, sunny`,
-});
+  parameters: { city: "string" },
+  execute: ({ city }) => `Weather in ${city}: 72°F, sunny`,
+};
 ```
 
 **Agentic loop, on-device.** `engine.generateWithTools` (and the `useAgent` React hook)
@@ -340,7 +342,7 @@ run the whole loop — generate → call a tool → feed the result back → rep
 step trace for UIs:
 
 ```tsx
-import { useAgent } from "@tryhamster/gerbil/gpu/hooks";
+import { useAgent } from "@tryhamster/gerbil/hooks";
 
 const { run, steps, answer, isRunning } = useAgent({
   model: "mlx-community/Qwen3.5-0.8B-4bit",
@@ -377,7 +379,7 @@ Inline autocomplete — `engine.autocomplete(prefix)` and the debounced `useAuto
 hook return a brief single-line continuation (low-latency defaults + cleanup):
 
 ```tsx
-import { useAutocomplete } from "@tryhamster/gerbil/gpu/hooks";
+import { useAutocomplete } from "@tryhamster/gerbil/hooks";
 
 const { suggestion, onInput, accept, dismiss } = useAutocomplete({
   model: "mlx-community/Qwen3.5-0.8B-4bit",
@@ -418,10 +420,10 @@ gerbil update                                 # Update to latest version
 ## Browser Usage
 
 Run LLMs directly in the browser with WebGPU — no server required. The React hooks
-live at `@tryhamster/gerbil/gpu/hooks` and run pure WebGPU compute:
+live at `@tryhamster/gerbil/hooks` and run pure WebGPU compute:
 
 ```tsx
-import { useChat } from "@tryhamster/gerbil/gpu/hooks";
+import { useChat } from "@tryhamster/gerbil/hooks";
 
 function Chat() {
   const { messages, send, isLoading, isGenerating } = useChat();
@@ -454,7 +456,7 @@ function Chat() {
 | **LangChain** | `@tryhamster/gerbil/langchain` | [📖 LangChain](https://gerbilsdk.com/docs/frameworks/langchain) |
 | **MCP Server** | `npx @tryhamster/gerbil serve --mcp` | [📖 MCP](https://gerbilsdk.com/docs/mcp) |
 
-**Native engine:** `import { WebGPUEngine } from "@tryhamster/gerbil/gpu"` (or `useEngine` from `@tryhamster/gerbil/gpu/hooks` for React) is the primary surface for text, vision, embeddings, and speech.
+**Native engine:** `import { WebGPUEngine } from "@tryhamster/gerbil/gpu"` (or `useEngine` from `@tryhamster/gerbil/hooks` for React) is the primary surface for text, vision, embeddings, and speech.
 
 ## Supported Models
 

package/dist/{architectures-C1I5V3Dt.mjs → architectures-wan0swXu.mjs}

@@ -1216,17 +1216,10 @@ function generateKaniTtsGraph(rawConfig, dtype, _groupSize, kvDtype) {
 		dtype: "f32",
 		storage: "activation"
 	});
-	addTensor({
-		name: CANONICAL_KEYS.LM_HEAD,
-		shape: [vocab_size, hidden_size],
-		dtype: "f32",
-		storage: "constant",
-		safetensorsKey: CANONICAL_KEYS.EMBED
-	});
 	addNode({
 		id: "lm_head",
 		opType: "MatMul",
-		inputs: ["final_norm_last", CANONICAL_KEYS.LM_HEAD],
+		inputs: ["final_norm_last", CANONICAL_KEYS.EMBED],
 		outputs: ["logits"],
 		attributes: {
 			M_tensor: "final_norm_last",
@@ -6067,4 +6060,4 @@ function isArchitectureSupported(name) {
 
 //#endregion
 export { parseKaniConfig as C, createDefaultHFKeyMapper as D, GEMMA4_VIS_KEYS as E, nanoCodecWeightMap as S, DTYPE_BYTES as T, generateNanoCodecDecoderGraph as _, generateMoonshineEncoderGraph as a, kaniLayerAlpha as b, DEFAULT_GROUP_SIZE as c, KANI_START_OF_HUMAN as d, audioTokensToCodes as f, generateKaniTtsGraph as g, foldNanoCodecWeightNorm as h, generateMoonshineDecoderGraph as i, quantizeInt4 as l, computeKaniPositions as m, isArchitectureSupported as n, moonshineEncoderFrames as o, buildKaniLayerCosSin as p, MOONSHINE_REMAINING_WORK as r, parseMoonshineConfig as s, generateGraph as t, KANI_END_OF_HUMAN as u, kaniAttentionLayerIndices as v, CANONICAL_KEYS as w, kaniSinTensor as x, kaniCosTensor as y };
-//# sourceMappingURL=architectures-C1I5V3Dt.mjs.map
+//# sourceMappingURL=architectures-wan0swXu.mjs.map