Update index.html

index.html

@@ -99,8 +99,9 @@
     static model = null;
     static processor = null;
     static model_id = null;
-    static async getInstance(modelId, dtypeSettings, device) {
+    static async getInstance(modelId, dtypeSettings, device, revision) {
       if (this.model_id !== modelId) {
+        await this.model?.dispose();
         this.model = null;
         this.processor = null;
         this.model_id = modelId;
@@ -116,6 +117,7 @@
             decoder_model_merged: dtypeSettings.decoder,
           },
           device: device,
+          revision,
         });
       }
       return [this.processor, this.model];
@@ -128,11 +130,11 @@
     const resultsDiv = document.getElementById("results");
     resultsDiv.innerHTML = "";
 
-    const modelIds = [
-      "HuggingFaceTB/SmolVLM-256M-Instruct",
-      "HuggingFaceTB/SmolVLM-500M-Instruct",
-      "HuggingFaceTB/SmolVLM-Instruct"
-    ];
+    const modelIds = {
+      "HuggingFaceTB/SmolVLM-256M-Instruct": "refs/pr/11",
+      "HuggingFaceTB/SmolVLM-500M-Instruct": "refs/pr/9",
+      "HuggingFaceTB/SmolVLM-Instruct": "main"
+    };
 
     const decoder_dtype = document.getElementById("decoder-dtype").value || "q4";
     const embed_dtype = document.getElementById("embed-dtype").value || "q4";
@@ -146,7 +148,7 @@
     const dtypeSettings = { decoder: decoder_dtype, embed: embed_dtype, vision: vision_dtype };
     const image = await load_image(imageUrl);
 
-    for (const modelId of modelIds) {
+    for (const [modelId, revision] of Object.entries(modelIds)) {
       const modelShortName = modelId.split("/").pop();
       const modelSection = document.createElement("div");
       modelSection.className = "model-results";
@@ -158,7 +160,7 @@
 
       try {
         status.innerText = "Loading processor and model...";
-        const [processor, model] = await SmolVLM.getInstance(modelId, dtypeSettings, device);
+        const [processor, model] = await SmolVLM.getInstance(modelId, dtypeSettings, device, revision);
 
         status.innerText = "Warming up...";
         const messages = [{
@@ -180,7 +182,7 @@
           status.innerText = `Running benchmark... (${i + 1}/${numRuns})`;
           bar.innerText = createProgressBar(i + 1, numRuns);
           const start = performance.now();
-          const [processor, model] = await SmolVLM.getInstance(modelId, dtypeSettings, device);
+          // const [processor, model] = await SmolVLM.getInstance(modelId, dtypeSettings, device, revision);
           const text = processor.apply_chat_template(messages, { add_generation_prompt: true });
           const inputs = await processor(text, [image], { do_image_splitting: doImageSplitting });