Fix: Phi-3 doesn't display (#2928)

* fix: params correction

* add phi

* version bump

extensions/inference-nitro-extension/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@janhq/inference-cortex-extension",
   "productName": "Cortex Inference Engine",
-  "version": "1.0.7",
+  "version": "1.0.8",
   "description": "This extension embeds cortex.cpp, a lightweight inference engine written in C++. See https://nitro.jan.ai.\nAdditional dependencies could be installed to run without Cuda Toolkit installation.",
   "main": "dist/index.js",
   "node": "dist/node/index.cjs.js",

extensions/inference-nitro-extension/resources/models/phi3-3.8b/model.json

@@ -8,17 +8,23 @@
     "id": "phi3-3.8b",
     "object": "model",
     "name": "Phi-3 Mini",
-    "version": "1.0",
+    "version": "1.1",
     "description": "Phi-3 Mini is Microsoft's newest, compact model designed for mobile use.",
     "format": "gguf",
     "settings": {
       "ctx_len": 4096,
       "prompt_template": "<|user|>\n{prompt}<|end|>\n<|assistant|>\n",
-      "llama_model_path": "Phi-3-mini-4k-instruct-q4.gguf"
+      "llama_model_path": "Phi-3-mini-4k-instruct-q4.gguf",
+      "ngl": 32
     },
     "parameters": {
       "max_tokens": 4096,
-      "stop": ["<|end|>"]
+      "stop": ["<|end|>"],
+      "temperature": 0.7,
+      "top_p": 0.95,
+      "stream": true,
+      "frequency_penalty": 0,
+      "presence_penalty": 0
     },
     "metadata": {
       "author": "Microsoft",

extensions/inference-nitro-extension/rollup.config.ts

@@ -23,6 +23,7 @@ const mistralIns7bq4Json = require('./resources/models/mistral-ins-7b-q4/model.j
 const mixtral8x7bInstructJson = require('./resources/models/mixtral-8x7b-instruct/model.json')
 const noromaid7bJson = require('./resources/models/noromaid-7b/model.json')
 const openchat357bJson = require('./resources/models/openchat-3.5-7b/model.json')
+const phi3bJson = require('./resources/models/phi3-3.8b/model.json')
 const phind34bJson = require('./resources/models/phind-34b/model.json')
 const qwen7bJson = require('./resources/models/qwen-7b/model.json')
 const stableZephyr3bJson = require('./resources/models/stable-zephyr-3b/model.json')
@@ -64,6 +65,7 @@ export default [
           mixtral8x7bInstructJson,
           noromaid7bJson,
           openchat357bJson,
+          phi3bJson,
           phind34bJson,
           qwen7bJson,
           stableZephyr3bJson,