add completion to prompt function for phi3 (#2299)

Author: openvino-dev-samples

notebooks/llm-rag-llamaindex/llm-rag-llamaindex.ipynb

@@ -1245,6 +1245,9 @@
     "\n",
     "ov_config = {\"PERFORMANCE_HINT\": \"LATENCY\", \"NUM_STREAMS\": \"1\", \"CACHE_DIR\": \"\"}\n",
     "\n",
+    "stop_tokens = llm_model_configuration.get(\"stop_tokens\")\n",
+    "completion_to_prompt = llm_model_configuration.get(\"completion_to_prompt\")\n",
+    "\n",
     "if \"GPU\" in llm_device.value and \"qwen2-7b-instruct\" in llm_model_id.value:\n",
     "    ov_config[\"GPU_ENABLE_SDPA_OPTIMIZATION\"] = \"NO\"\n",
     "\n",
@@ -1259,6 +1262,7 @@
     "    max_new_tokens=2,\n",
     "    model_kwargs={\"ov_config\": ov_config, \"trust_remote_code\": True},\n",
     "    generate_kwargs={\"temperature\": 0.7, \"top_k\": 50, \"top_p\": 0.95},\n",
+    "    completion_to_prompt=completion_to_prompt,\n",
     "    device_map=llm_device.value,\n",
     ")\n",
     "\n",
@@ -1314,8 +1318,6 @@
     "else:\n",
     "    text_example_path = \"text_example_cn.pdf\"\n",
     "\n",
-    "stop_tokens = llm_model_configuration.get(\"stop_tokens\")\n",
-    "\n",
     "\n",
     "class StopOnTokens(StoppingCriteria):\n",
     "    def __init__(self, token_ids):\n",

utils/llm_config.py

@@ -54,6 +54,10 @@ def internlm_partial_text_processor(partial_text, new_text):
     return partial_text.split("<|im_end|>")[0]
 
 
+def phi_completion_to_prompt(completion):
+    return f"<|system|><|end|><|user|>{completion}<|end|><|assistant|>\n"
+
+
 SUPPORTED_LLM_MODELS = {
     "English": {
         "qwen2-0.5b-instruct": {
@@ -252,6 +256,7 @@ def internlm_partial_text_processor(partial_text, new_text):
             Context: {context} 
             Answer: <|end|>
             <|assistant|>""",
+            "completion_to_prompt": phi_completion_to_prompt,
         },
     },
     "Chinese": {