Adding a simple monkey search for Leetcode - Darn LeetMonkey

app.py

@@ -5,7 +5,7 @@ from pinecone_text.sparse import SpladeEncoder
 from sentence_transformers import SentenceTransformer
 import transformers
 transformers.logging.set_verbosity_error()
-from transformers import AutoTokenizer, AutoModelForCausalLM
+from transformers import AutoTokenizer, AutoModelForCausalLM, GPTQConfig
 
 
 import os
@@ -16,6 +16,11 @@ pc = Pinecone(api_key=PINECONE_API_KEY)
 index_name = "leetmonkey-sparse-dense"
 index = pc.Index(index_name)
 
+quantization_config = GPTQConfig(
+    disable_exllama=True
+)
+
+
 
 # Initialize models
 device = 'cpu'
@@ -25,7 +30,7 @@ dense_model = SentenceTransformer('sentence-transformers/all-Mpnet-base-v2', dev
 # Load the quantized Llama 2 model and tokenizer
 model_name = "TheBloke/Llama-2-7B-Chat-GPTQ"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", quantization_config=quantization_config)
 
 def search_problems(query, top_k=5):
     dense_query = dense_model.encode([query])[0].tolist()