@blank-utils/llm 0.4.19 → 0.4.20

package/dist/{chunk-JOBJ4PY7.js → chunk-T4VK6BPC.js}

@@ -1461,7 +1461,7 @@ ${systemPrompt}` : systemPrompt;
           const { pipeline: pipeline2, env } = await import("@huggingface/transformers");
           env.allowLocalModels = false;
           env.useBrowserCache = true;
-          const captioner = await pipeline2("image-to-text", "onnx-community/Florence-2-base-ft", { device: "wasm", dtype: "q8" });
+          const captioner = await pipeline2("image-to-text", "Xenova/vit-gpt2-image-captioning", { device: "wasm", dtype: "q8" });
           for (let i = 0; i < needsCaptioning.length; i++) {
             if (abortRef.current) break;
             const img = needsCaptioning[i];
@@ -1469,9 +1469,9 @@ ${systemPrompt}` : systemPrompt;
             setStreamingText(`[System: Extracting detailed visual description for ${img.name}... ${i + 1}/${needsCaptioning.length}]
 `);
             const out = await captioner(img.dataUrl, {
-              text: "<MORE_DETAILED_CAPTION>",
-              max_new_tokens: 512,
-              num_beams: 3
+              max_new_tokens: 64,
+              num_beams: 4,
+              repetition_penalty: 1.5
             });
             let val = "";
             if (Array.isArray(out) && out[0] && out[0].generated_text) val = out[0].generated_text;

package/dist/index.js

@@ -26,7 +26,7 @@ import {
   useCompletion,
   useLLM,
   useStream
-} from "./chunk-JOBJ4PY7.js";
+} from "./chunk-T4VK6BPC.js";
 export {
   Chat,
   ChatInput,

package/dist/react/index.js

@@ -9,7 +9,7 @@ import {
   useCompletion,
   useLLM,
   useStream
-} from "../chunk-JOBJ4PY7.js";
+} from "../chunk-T4VK6BPC.js";
 export {
   Chat,
   ChatApp,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@blank-utils/llm",
-  "version": "0.4.19",
+  "version": "0.4.20",
   "description": "Run LLMs directly in your browser with WebGPU acceleration. Supports React hooks and eager background loading.",
   "type": "module",
   "main": "./dist/index.js",