1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
|
from langchain import PromptTemplate, LLMChain
from kg_rag.utility import *
QUESTION_PATH = config_data["MCQ_PATH"]
SYSTEM_PROMPT = system_prompts["MCQ_QUESTION_PROMPT_BASED"]
SAVE_PATH = config_data["SAVE_RESULTS_PATH"]
MODEL_NAME = config_data["LLAMA_MODEL_NAME"]
BRANCH_NAME = config_data["LLAMA_MODEL_BRANCH"]
CACHE_DIR = config_data["LLM_CACHE_DIR"]
save_name = "_".join(MODEL_NAME.split("/")[-1].split("-"))+"_prompt_based_two_hop_mcq_from_monarch_and_robokop_response.csv"
INSTRUCTION = "Question: {question}"
def main():
start_time = time.time()
llm = llama_model(MODEL_NAME, BRANCH_NAME, CACHE_DIR)
template = get_prompt(INSTRUCTION, SYSTEM_PROMPT)
prompt = PromptTemplate(template=template, input_variables=["question"])
llm_chain = LLMChain(prompt=prompt, llm=llm)
question_df = pd.read_csv(QUESTION_PATH)
answer_list = []
for index, row in question_df.iterrows():
question = row["text"]
output = llm_chain.run(question)
answer_list.append((row["text"], row["correct_node"], output))
answer_df = pd.DataFrame(answer_list, columns=["question", "correct_answer", "llm_answer"])
answer_df.to_csv(os.path.join(SAVE_PATH, save_name), index=False, header=True)
print("Completed in {} min".format((time.time()-start_time)/60))
if __name__ == "__main__":
main()
|
