Upload 6 files

build/m.js

@@ -129,14 +129,6 @@ function handleError(f, args) {
 */
 export class Model {
 
-    static __wrap(ptr) {
-        ptr = ptr >>> 0;
-        const obj = Object.create(Model.prototype);
-        obj.__wbg_ptr = ptr;
-
-        return obj;
-    }
-
     __destroy_into_raw() {
         const ptr = this.__wbg_ptr;
         this.__wbg_ptr = 0;
@@ -170,7 +162,8 @@ export class Model {
             if (r2) {
                 throw takeObject(r1);
             }
-            return Model.__wrap(r0);
+            this.__wbg_ptr = r0 >>> 0;
+            return this;
         } finally {
             wasm.__wbindgen_add_to_stack_pointer(16);
         }
@@ -302,10 +295,10 @@ function __wbg_get_imports() {
     imports.wbg.__wbindgen_object_drop_ref = function(arg0) {
         takeObject(arg0);
     };
-    imports.wbg.__wbg_log_082f7c89404ec861 = function(arg0, arg1) {
+    imports.wbg.__wbg_log_aa756f7b1647d2ab = function(arg0, arg1) {
         console.log(getStringFromWasm0(arg0, arg1));
     };
-    imports.wbg.__wbg_crypto_c48a774b022d20ac = function(arg0) {
+    imports.wbg.__wbg_crypto_58f13aa23ffcb166 = function(arg0) {
         const ret = getObject(arg0).crypto;
         return addHeapObject(ret);
     };
@@ -314,15 +307,15 @@ function __wbg_get_imports() {
         const ret = typeof(val) === 'object' && val !== null;
         return ret;
     };
-    imports.wbg.__wbg_process_298734cf255a885d = function(arg0) {
+    imports.wbg.__wbg_process_5b786e71d465a513 = function(arg0) {
         const ret = getObject(arg0).process;
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_versions_e2e78e134e3e5d01 = function(arg0) {
+    imports.wbg.__wbg_versions_c2ab80650590b6a2 = function(arg0) {
         const ret = getObject(arg0).versions;
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_node_1cd7a5d853dbea79 = function(arg0) {
+    imports.wbg.__wbg_node_523d7bd03ef69fba = function(arg0) {
         const ret = getObject(arg0).node;
         return addHeapObject(ret);
     };
@@ -330,11 +323,11 @@ function __wbg_get_imports() {
         const ret = typeof(getObject(arg0)) === 'string';
         return ret;
     };
-    imports.wbg.__wbg_msCrypto_bcb970640f50a1e8 = function(arg0) {
+    imports.wbg.__wbg_msCrypto_abcb1295e768d1f2 = function(arg0) {
         const ret = getObject(arg0).msCrypto;
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_require_8f08ceecec0f4fee = function() { return handleError(function () {
+    imports.wbg.__wbg_require_2784e593a4674877 = function() { return handleError(function () {
         const ret = module.require;
         return addHeapObject(ret);
     }, arguments) };
@@ -346,17 +339,17 @@ function __wbg_get_imports() {
         const ret = getStringFromWasm0(arg0, arg1);
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_getRandomValues_37fa2ca9e4e07fab = function() { return handleError(function (arg0, arg1) {
-        getObject(arg0).getRandomValues(getObject(arg1));
-    }, arguments) };
-    imports.wbg.__wbg_randomFillSync_dc1e9a60c158336d = function() { return handleError(function (arg0, arg1) {
+    imports.wbg.__wbg_randomFillSync_a0d98aa11c81fe89 = function() { return handleError(function (arg0, arg1) {
         getObject(arg0).randomFillSync(takeObject(arg1));
     }, arguments) };
-    imports.wbg.__wbg_newnoargs_581967eacc0e2604 = function(arg0, arg1) {
+    imports.wbg.__wbg_getRandomValues_504510b5564925af = function() { return handleError(function (arg0, arg1) {
+        getObject(arg0).getRandomValues(getObject(arg1));
+    }, arguments) };
+    imports.wbg.__wbg_newnoargs_ccdcae30fd002262 = function(arg0, arg1) {
         const ret = new Function(getStringFromWasm0(arg0, arg1));
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_call_cb65541d95d71282 = function() { return handleError(function (arg0, arg1) {
+    imports.wbg.__wbg_call_669127b9d730c650 = function() { return handleError(function (arg0, arg1) {
         const ret = getObject(arg0).call(getObject(arg1));
         return addHeapObject(ret);
     }, arguments) };
@@ -364,19 +357,19 @@ function __wbg_get_imports() {
         const ret = getObject(arg0);
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_self_1ff1d729e9aae938 = function() { return handleError(function () {
+    imports.wbg.__wbg_self_3fad056edded10bd = function() { return handleError(function () {
         const ret = self.self;
         return addHeapObject(ret);
     }, arguments) };
-    imports.wbg.__wbg_window_5f4faef6c12b79ec = function() { return handleError(function () {
+    imports.wbg.__wbg_window_a4f46c98a61d4089 = function() { return handleError(function () {
         const ret = window.window;
         return addHeapObject(ret);
     }, arguments) };
-    imports.wbg.__wbg_globalThis_1d39714405582d3c = function() { return handleError(function () {
+    imports.wbg.__wbg_globalThis_17eff828815f7d84 = function() { return handleError(function () {
         const ret = globalThis.globalThis;
         return addHeapObject(ret);
     }, arguments) };
-    imports.wbg.__wbg_global_651f05c6a0944d1c = function() { return handleError(function () {
+    imports.wbg.__wbg_global_46f939f6541643c5 = function() { return handleError(function () {
         const ret = global.global;
         return addHeapObject(ret);
     }, arguments) };
@@ -384,34 +377,34 @@ function __wbg_get_imports() {
         const ret = getObject(arg0) === undefined;
         return ret;
     };
-    imports.wbg.__wbg_call_01734de55d61e11d = function() { return handleError(function (arg0, arg1, arg2) {
+    imports.wbg.__wbg_call_53fc3abd42e24ec8 = function() { return handleError(function (arg0, arg1, arg2) {
         const ret = getObject(arg0).call(getObject(arg1), getObject(arg2));
         return addHeapObject(ret);
     }, arguments) };
-    imports.wbg.__wbg_now_9c5990bda04c7e53 = function() {
+    imports.wbg.__wbg_now_4579335d3581594c = function() {
         const ret = Date.now();
         return ret;
     };
-    imports.wbg.__wbg_buffer_085ec1f694018c4f = function(arg0) {
+    imports.wbg.__wbg_buffer_344d9b41efe96da7 = function(arg0) {
         const ret = getObject(arg0).buffer;
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_newwithbyteoffsetandlength_6da8e527659b86aa = function(arg0, arg1, arg2) {
+    imports.wbg.__wbg_newwithbyteoffsetandlength_2dc04d99088b15e3 = function(arg0, arg1, arg2) {
         const ret = new Uint8Array(getObject(arg0), arg1 >>> 0, arg2 >>> 0);
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_new_8125e318e6245eed = function(arg0) {
+    imports.wbg.__wbg_new_d8a000788389a31e = function(arg0) {
         const ret = new Uint8Array(getObject(arg0));
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_set_5cf90238115182c3 = function(arg0, arg1, arg2) {
+    imports.wbg.__wbg_set_dcfd613a3420f908 = function(arg0, arg1, arg2) {
         getObject(arg0).set(getObject(arg1), arg2 >>> 0);
     };
-    imports.wbg.__wbg_newwithlength_e5d69174d6984cd7 = function(arg0) {
+    imports.wbg.__wbg_newwithlength_13b5319ab422dcf6 = function(arg0) {
         const ret = new Uint8Array(arg0 >>> 0);
         return addHeapObject(ret);
     };
-    imports.wbg.__wbg_subarray_13db269f57aa838d = function(arg0, arg1, arg2) {
+    imports.wbg.__wbg_subarray_6ca5cfa7fbb9abbe = function(arg0, arg1, arg2) {
         const ret = getObject(arg0).subarray(arg1 >>> 0, arg2 >>> 0);
         return addHeapObject(ret);
     };

build/m_bg.wasm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:31acdbb8b18f216db0b64e628bd398f63260b35b6dfb3d1e23d01d757b8b9163
-size 4544545
+oid sha256:a62b31d8872b8c4c1f0f98dad8f2375d05ad120060731eb02ca21f827d188ac5
+size 4566571

index.html

@@ -1,7 +1,7 @@
 <html>
   <head>
     <meta content="text/html;charset=utf-8" http-equiv="Content-Type" />
-    <title>Candle Phi 1.5 Rust/WASM</title>
+    <title>Candle Phi 1.5 / Phi 2.0 Rust/WASM</title>
   </head>
   <body></body>
 </html>
@@ -38,10 +38,10 @@
       import snarkdown from "https://cdn.skypack.dev/snarkdown";
       import hljs from "https://cdn.skypack.dev/highlight.js";
       // models base url
-      const MODELS = {
-        phi_1_5_quantized: {
-          base_url:
-            "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
+        const MODELS = {
+            phi_1_5_q4k: {
+            base_url:
+          "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
           model: "model-q4k.gguf",
           tokenizer: "tokenizer.json",
           config: "phi-1_5.json",
@@ -49,9 +49,9 @@
           seq_len: 2048,
           size: "800 MB",
         },
-        phi_1_5_quantized_2: {
-          base_url:
-            "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
+          phi_1_5_q80: {
+            base_url:
+          "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
           model: "model-q80.gguf",
           tokenizer: "tokenizer.json",
           config: "phi-1_5.json",
@@ -59,9 +59,19 @@
           seq_len: 2048,
           size: "1.51 GB",
         },
-        puffin_phi_v2_quantized: {
-          base_url:
-            "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
+          phi_2_0_q4k: {
+            base_url:
+          "https://huggingface.co/radames/phi-2-quantized/resolve/main/",
+          model: ["model-v2-q4k.gguf_aa.part", "model-v2-q4k.gguf_ab.part", "model-v2-q4k.gguf_ac.part"],
+          tokenizer: "tokenizer.json",
+          config: "config.json",
+          quantized: true,
+          seq_len: 2048,
+          size: "1.57GB",
+        },
+          puffin_phi_v2_q4k: {
+            base_url:
+          "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
           model: "model-puffin-phi-v2-q4k.gguf",
           tokenizer: "tokenizer-puffin-phi-v2.json",
           config: "puffin-phi-v2.json",
@@ -69,9 +79,9 @@
           seq_len: 2048,
           size: "798 MB",
         },
-        puffin_phi_v2_quantized_2: {
-          base_url:
-            "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
+          puffin_phi_v2_q80: {
+            base_url:
+          "https://huggingface.co/lmz/candle-quantized-phi/resolve/main/",
           model: "model-puffin-phi-v2-q80.gguf",
           tokenizer: "tokenizer-puffin-phi-v2.json",
           config: "puffin-phi-v2.json",
@@ -106,8 +116,8 @@ Let’s think step by step.`,
         },
         {
           title: "Question answering",
-          prompt: `What is the capital of France?  
-Answer:`,
+          prompt: `Instruct: What is the capital of France?  
+Output:`,
         },
         {
           title: "Chat mode",
@@ -148,7 +158,7 @@ Very polite review:`,
         const getValue = (id) => document.querySelector(`#${id}`).value;
         const modelID = getValue("model");
         const model = MODELS[modelID];
-        const weightsURL = model.base_url + model.model;
+        const weightsURL =  model.model instanceof Array ? model.model.map((m) => model.base_url + m) : model.base_url + model.model;
         const tokenizerURL = model.base_url + model.tokenizer;
         const configURL = model.base_url + model.config;
 
@@ -246,6 +256,13 @@ Very polite review:`,
           option.innerText = `${id} (${model.size})`;
           modelSelect.appendChild(option);
         }
+        const query = new URLSearchParams(window.location.search);
+        const modelID = query.get("model");
+        if (modelID) {
+            modelSelect.value = modelID;
+        } else {
+            modelSelect.value = "phi_1_5_q4k";
+        }
 
         for (const [i, { title, prompt }] of TEMPLATES.entries()) {
           const div = document.createElement("div");
@@ -257,7 +274,7 @@ Very polite review:`,
           input.value = prompt;
           const label = document.createElement("label");
           label.htmlFor = `templates-${i}`;
-          label.classList.add("cursor-pointer", "px-1");
+          label.classList.add("cursor-pointer");
           label.innerText = title;
           div.appendChild(input);
           div.appendChild(label);
@@ -272,6 +289,14 @@ Very polite review:`,
         prompt.style.height = prompt.scrollHeight + "px";
       });
       modelSelect.addEventListener("change", (e) => {
+        const query = new URLSearchParams(window.location.search);
+        query.set("model", e.target.value);
+        window.history.replaceState(
+        { },
+        "",
+        `${window.location.pathname}?${query}`
+        );
+        window.parent.postMessage({queryString: "?" + query }, "*")
         const model = MODELS[e.target.value];
         document.querySelector("#max-seq").max = model.seq_len;
         document.querySelector("#max-seq").nextElementSibling.value = 200;
@@ -320,42 +345,25 @@ Very polite review:`,
     <main class="grid grid-cols-1 gap-8 relative">
       <span class="absolute text-5xl -ml-[1em]"> 🕯️ </span>
       <div>
-        <h1 class="text-5xl font-bold">Candle Phi 1.5</h1>
+        <h1 class="text-5xl font-bold">Candle Phi 1.5 / Phi 2.0 </h1>
         <h2 class="text-2xl font-bold">Rust/WASM Demo</h2>
         <p class="max-w-lg">
           The
-          <a
-            href="https://huggingface.co/microsoft/phi-1_5"
-            class="link"
-            target="_blank"
-            >Phi-1.5</a
-          >
-          model achieves state-of-the-art performance with only 1.3 billion
-          parameters, compared to models with up to 10 billion. You can try the
-          quantized version of the model here. Additional prompt examples are
+          <a href="https://huggingface.co/microsoft/phi-1_5" class="link" target="_blank">Phi-1.5</a> and
+          <a href="https://huggingface.co/microsoft/phi-2" class="link" target="_blank">Phi-2</a> models achieves
+          state-of-the-art performance with only 1.3 billion and 2.7 billion parameters, compared to larger models with up
+          to 13
+          billion parameters. Here you can try the quantized versions. Additional prompt examples are
           available in the
-          <a
-            href="https://arxiv.org/pdf/2309.05463.pdf#page=8"
-            class="link"
-            target="_blank"
-          >
-            technical report </a
-          >.
+          <a href="https://arxiv.org/pdf/2309.05463.pdf#page=8" class="link" target="_blank">
+            technical report </a>.
         </p>
         <p class="max-w-lg">
           You can also try
-          <a
-            href="https://huggingface.co/teknium/Puffin-Phi-v2"
-            class="link"
-            target="_blank"
-            >Puffin-Phi V2
+          <a href="https://huggingface.co/teknium/Puffin-Phi-v2" class="link" target="_blank">Puffin-Phi V2
           </a>
-          quantized version model, a fine-tuned version of Phi-1.5 on the
-          <a
-            href="https://huggingface.co/datasets/LDJnr/Puffin"
-            class="link"
-            target="_blank"
-            >Puffin dataset
+          quantized version, a fine-tuned version of Phi-1.5 on the
+          <a href="https://huggingface.co/datasets/LDJnr/Puffin" class="link" target="_blank">Puffin dataset
           </a>
         </p>
       </div>
@@ -390,8 +398,8 @@ Very polite review:`,
           oninput="this.style.height = 0;this.style.height = this.scrollHeight + 'px'"
           placeholder="Add your prompt here..."
         >
-Write a detailed analogy between mathematics and a lighthouse.  
-Answer:</textarea
+Instruct: Write a detailed analogy between mathematics and a lighthouse.  
+Output:</textarea
         >
         <button id="clear-btn">
           <svg

phiWorker.js

@@ -12,6 +12,20 @@ async function fetchArrayBuffer(url) {
   cache.put(url, res.clone());
   return new Uint8Array(await res.arrayBuffer());
 }
+async function concatenateArrayBuffers(urls) {
+  const arrayBuffers = await Promise.all(urls.map(url => fetchArrayBuffer(url)));
+
+  let totalLength = arrayBuffers.reduce((acc, arrayBuffer) => acc + arrayBuffer.byteLength, 0);
+  let concatenatedBuffer = new Uint8Array(totalLength);
+
+  let offset = 0;
+  arrayBuffers.forEach(buffer => {
+    concatenatedBuffer.set(new Uint8Array(buffer), offset);
+    offset += buffer.byteLength;
+  });
+  return concatenatedBuffer;
+}
+
 class Phi {
   static instance = {};
 
@@ -27,10 +41,9 @@ class Phi {
       await init();
 
       self.postMessage({ status: "loading", message: "Loading Model" });
-
       const [weightsArrayU8, tokenizerArrayU8, configArrayU8] =
         await Promise.all([
-          fetchArrayBuffer(weightsURL),
+          weightsURL instanceof Array ? concatenateArrayBuffers(weightsURL) : fetchArrayBuffer(weightsURL),
           fetchArrayBuffer(tokenizerURL),
           fetchArrayBuffer(configURL),
         ]);