feat: use langchain

app.py

@@ -83,7 +83,8 @@ def display_streamlit_sidebar():
     st.sidebar.write('**Parameters**')
     form = st.sidebar.form("config_form", clear_on_submit=True)
 
-    model_repo_id = form.text_input(label="Repo", value=st.session_state["model_repo_id"])
+    model_option = form.selectbox("Quickly select a model", ("llama", "meditron"))
+    model_repo_id = form.text_input(label="Repo", value=model_option)#value=st.session_state["model_repo_id"])
     model_filename = form.text_input(label="File name", value=st.session_state["model_filename"])
     model_type = form.text_input(label="Model type", value=st.session_state["model_type"])
     gpu_layers = form.slider('GPU Layers', min_value=0,

utils/epfl_meditron_utils.py

@@ -1,6 +1,8 @@
 from ctransformers import AutoModelForCausalLM, AutoTokenizer
 from transformers import pipeline
 import streamlit as st
+from langchain.chains import LLMChain
+from langchain.prompts import PromptTemplate
 
 # Simple inference example
 # output = llm(
@@ -12,14 +14,25 @@ import streamlit as st
 
 prompt_format = "<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant"
 
+
+template = """Question: {question}
+
+Answer:"""
+
+
+
+
 def get_llm_response(repo, filename, model_type, gpu_layers, prompt):
     print("Loading model")
-    model = AutoModelForCausalLM.from_pretrained(repo, model_file=filename, model_type=model_type, gpu_layers=gpu_layers)
+    llm = AutoModelForCausalLM.from_pretrained(repo, model_file=filename, model_type=model_type, gpu_layers=gpu_layers)
     print("Model loaded")
 
     #llm_prompt = prompt_format.format(system_message=system_prompt, prompt=prompt)
     print(f"LLM prompt: {prompt}")
-    response = model(prompt, stop=["</s>"])
-    print(f"Response: {response}")
+    
+    prompt = PromptTemplate(template=template, input_variables=["question"])
+
+    llm_chain = LLMChain(prompt=prompt, llm=llm)
+    response = llm_chain.run(prompt)
     
     return response