Merge pull request #147 from filopedraz/feat/fix-cht-llama

Fixed general errors about Mistral model

cht-llama-cpp/main.py

@@ -10,7 +10,7 @@
 
 load_dotenv()
 
-MODEL_PATH = f"./ml/models/{os.getenv('MODEL_ID', 'mistral-7b-instruct-v0.1.Q5_0')}.gguf"
+MODEL_PATH = f"./ml/models/{os.getenv('MODEL_ID', 'yarn-mistral-7b-128k.Q4_K_M')}.gguf"
 # Mistral gguf follows ChatML syntax
 # https://github.com/openai/openai-python/blob/main/chatml.md
 PROMPT_TEMPLATE_STRING = '{"system_prompt_template": "<|im_start|>system\\n{}\\n<|im_end|>\\n", "default_system_text": "You are an helpful AI assistant.", "user_prompt_template": "<|im_start|>user\\n{}\\n<|im_end|>\\n", "assistant_prompt_template": "<|im_start|>assistant\\n{}\\n<|im_end|>\\n", "request_assistant_response_token": "<|im_start|>assistant\\n", "template_format": "chatml"}'  # noqa
@@ -19,8 +19,10 @@
     parser = argparse.ArgumentParser()
     parser.add_argument("--model-path", help="Path to GGUF", default=MODEL_PATH)
     parser.add_argument("--port", help="Port to run model server on", type=int, default=8000)
+    parser.add_argument("--ctx", help="Context dimension", type=int, default=4096)
     args = parser.parse_args()
     MODEL_PATH = args.model_path
+    MODEL_CTX = args.ctx
 
 logging.basicConfig(
     format="%(asctime)s %(levelname)-8s %(message)s",
@@ -33,7 +35,7 @@ def create_start_app_handler(app: FastAPI):
     def start_app() -> None:
         from models import LLaMACPPBasedModel
 
-        LLaMACPPBasedModel.get_model(MODEL_PATH, PROMPT_TEMPLATE_STRING)
+        LLaMACPPBasedModel.get_model(MODEL_PATH, PROMPT_TEMPLATE_STRING, MODEL_CTX)
 
     return start_app
 

cht-llama-cpp/models.py

@@ -26,20 +26,6 @@ class LLaMACPPBasedModel(object):
     def tokenize(cls, prompt):
         return cls.model.tokenize(b" " + prompt.encode("utf-8"))
 
-    @classmethod
-    def reduce_number_of_messages(cls, messages, max_tokens):
-        buffer_tokens = 32
-        ctx_max_tokens = 4096
-        num_messages = len(messages)
-
-        tokens = [len(cls.tokenize(doc["content"])) for doc in messages]
-
-        token_count = sum(tokens[:num_messages])
-        while token_count + max_tokens + buffer_tokens > ctx_max_tokens:
-            num_messages -= 1
-            token_count -= tokens[num_messages]
-        return messages[:num_messages]
-
     @classmethod
     def generate(
         cls,
@@ -55,7 +41,6 @@ def generate(
     ):
         if stop is None:
             stop = []
-        messages = cls.reduce_number_of_messages(messages[::-1], max_tokens)[::-1]
         cls.model.n_threads = n_threads
         cht_resp = cls.model.create_chat_completion(
             messages,
@@ -75,13 +60,13 @@ def generate(
         return cht_resp
 
     @classmethod
-    def get_model(cls, model_path, prompt_template_jsonstr):
+    def get_model(cls, model_path, prompt_template_jsonstr, n_ctx):
         chat_format = "llama-2"
         if "mistral" in model_path:
             cls.PROMPT_TEMPLATE = json.loads(prompt_template_jsonstr)
             chat_format = cls.PROMPT_TEMPLATE.get("template_format", "chatml")
         if cls.model is None:
-            cls.model = Llama(model_path, chat_format=chat_format)
+            cls.model = Llama(model_path, chat_format=chat_format, n_ctx=n_ctx)
 
         return cls.model
 

cht-llama-cpp/requirements.txt

@@ -6,4 +6,4 @@ tqdm==4.65.0
 httpx==0.23.3
 python-dotenv==1.0.0
 tenacity==8.2.2
-llama-cpp-python==0.2.11
+llama-cpp-python==0.2.14