Zwea Htet commited on
Commit
e058c61
1 Parent(s): c6e1770

added prompts

Browse files
assets/prompts/__init__.py ADDED
File without changes
assets/prompts/custom_prompts.py CHANGED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ text_qa_template_str = (
2
+ "You are a helpful assistant that can give users information about california water regulations."
3
+ "Your task is to answer the question based on the context below and using your own knowledge.\n"
4
+ "Context: {context_str}\n"
5
+ "Question: {query_str}\n"
6
+ "Answer: \n"
7
+ "If the question is relevant, provide the name of the chapter, the article and the section."
8
+ "If possible, also provide the page number of the document from which your answer was found.\n"
9
+ )
10
+
11
+ refine_template_str = (
12
+ "The original question is as follows: {query_str}\n"
13
+ "We have provided an existing answer: {existing_answer}\n"
14
+ "We have the opportunity to refine the existing answer "
15
+ "(only if needed) with some more context below.\n"
16
+ "------------\n"
17
+ "{context_msg}\n"
18
+ "------------\n"
19
+ "Using both the new context and your own knowledege, update or repeat the existing answer.\n"
20
+ )
models/llamaCustom.py CHANGED
@@ -21,8 +21,11 @@ from llama_index import (
21
  load_index_from_storage,
22
  )
23
  from llama_index.llms import CompletionResponse, CustomLLM, LLMMetadata
 
24
  from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
25
 
 
 
26
  load_dotenv()
27
  # openai.api_key = os.getenv("OPENAI_API_KEY")
28
  fs = HfFileSystem()
@@ -35,6 +38,10 @@ NUM_OUTPUT = 525
35
  # set maximum chunk overlap
36
  CHUNK_OVERLAP_RATION = 0.2
37
 
 
 
 
 
38
 
39
  @st.cache_resource
40
  def load_model(model_name: str):
@@ -136,7 +143,10 @@ class LlamaCustom:
136
 
137
  def get_response(self, query_str):
138
  print("query_str: ", query_str)
139
- query_engine = self.vector_index.as_query_engine()
 
 
 
140
  response = query_engine.query(query_str)
141
  print("metadata: ", response.metadata)
142
  return str(response)
 
21
  load_index_from_storage,
22
  )
23
  from llama_index.llms import CompletionResponse, CustomLLM, LLMMetadata
24
+ from llama_index.prompts import Prompt
25
  from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
26
 
27
+ from assets.prompts import custom_prompts
28
+
29
  load_dotenv()
30
  # openai.api_key = os.getenv("OPENAI_API_KEY")
31
  fs = HfFileSystem()
 
38
  # set maximum chunk overlap
39
  CHUNK_OVERLAP_RATION = 0.2
40
 
41
+ text_qa_template = Prompt(custom_prompts.text_qa_template_str)
42
+
43
+ refine_template = Prompt(custom_prompts.refine_template_str)
44
+
45
 
46
  @st.cache_resource
47
  def load_model(model_name: str):
 
143
 
144
  def get_response(self, query_str):
145
  print("query_str: ", query_str)
146
+ # query_engine = self.vector_index.as_query_engine()
147
+ query_engine = self.vector_index.as_query_engine(
148
+ text_qa_template=text_qa_template, refine_template=refine_template
149
+ )
150
  response = query_engine.query(query_str)
151
  print("metadata: ", response.metadata)
152
  return str(response)