added streaming

app/chains.py

@@ -20,7 +20,8 @@ llm = HuggingFaceEndpoint(
     repo_id="meta-llama/Meta-Llama-3-8B-Instruct",
     huggingfacehub_api_token=os.environ['HF_TOKEN'],
     max_new_tokens=512,
-    stop_sequences=[tokenizer.eos_token]
+    stop_sequences=[tokenizer.eos_token],
+    streaming=True,
 )
 
 simple_chain = (raw_prompt | llm).with_types(input_type=schemas.UserQuestion)

app/prompts.py

@@ -36,20 +36,6 @@ Answer the question based only on the following context:
 Question: {standalone_question}
 """
 
-map_prompt = """
-Given the following list of file paths, return a comma separated list of the most likely files to have content that could potentially help answer the question. Return nothing if none of those would help. 
-Make sure to return the complete full paths as it is writen in the original list
-
-File list:
-{file_list}
-
-Question: {question}
-
-Return a comma separated list of files and nothing else!
-Comma separated list:
-"""
-
-
 
 def format_prompt(prompt):
     chat = [
@@ -81,5 +67,4 @@ raw_prompt = PromptTemplate.from_template(raw_prompt)
 history_prompt_formatted = format_prompt(history_prompt)
 question_prompt_formatted = format_prompt(question_prompt)
 context_prompt_formatted = format_prompt(context_prompt)
-map_prompt_formatted = format_prompt(map_prompt)