LanceY2004
commited on
RAG
Browse files- localrag_no_rewrite.py +193 -0
localrag_no_rewrite.py
ADDED
@@ -0,0 +1,193 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import ollama
|
3 |
+
import os
|
4 |
+
from openai import OpenAI
|
5 |
+
import argparse
|
6 |
+
|
7 |
+
# ANSI escape codes for colors
|
8 |
+
PINK = '\033[95m'
|
9 |
+
CYAN = '\033[96m'
|
10 |
+
YELLOW = '\033[93m'
|
11 |
+
NEON_GREEN = '\033[92m'
|
12 |
+
RESET_COLOR = '\033[0m'
|
13 |
+
|
14 |
+
# Function to open a file and return its contents as a string
|
15 |
+
def open_file(filepath):
|
16 |
+
with open(filepath, 'r', encoding='utf-8') as infile:
|
17 |
+
return infile.read()
|
18 |
+
|
19 |
+
# Function to get relevant context from the vault based on user input
|
20 |
+
def get_relevant_context(rewritten_input, vault_embeddings, vault_content, top_k=3):
|
21 |
+
if vault_embeddings.nelement() == 0: # Check if the tensor has any elements
|
22 |
+
return []
|
23 |
+
# Encode the rewritten input
|
24 |
+
input_embedding = ollama.embeddings(model='mxbai-embed-large', prompt=rewritten_input)["embedding"]
|
25 |
+
# Compute cosine similarity between the input and vault embeddings
|
26 |
+
cos_scores = torch.cosine_similarity(torch.tensor(input_embedding).unsqueeze(0), vault_embeddings)
|
27 |
+
# Adjust top_k if it's greater than the number of available scores
|
28 |
+
top_k = min(top_k, len(cos_scores))
|
29 |
+
# Sort the scores and get the top-k indices
|
30 |
+
top_indices = torch.topk(cos_scores, k=top_k)[1].tolist()
|
31 |
+
# Get the corresponding context from the vault
|
32 |
+
relevant_context = [vault_content[idx].strip() for idx in top_indices]
|
33 |
+
return relevant_context
|
34 |
+
|
35 |
+
# Function to interact with the Ollama model
|
36 |
+
def ollama_chat(user_input, system_message, vault_embeddings, vault_content, ollama_model, conversation_history):
|
37 |
+
# Get relevant context from the vault
|
38 |
+
relevant_context = get_relevant_context(user_input, vault_embeddings, vault_content, top_k=3)
|
39 |
+
if relevant_context:
|
40 |
+
# Convert list to a single string with newlines between items
|
41 |
+
context_str = "\n".join(relevant_context)
|
42 |
+
print("Context Pulled from Documents: \n\n" + CYAN + context_str + RESET_COLOR)
|
43 |
+
else:
|
44 |
+
print(CYAN + "No relevant context found." + RESET_COLOR)
|
45 |
+
|
46 |
+
# Prepare the user's input by concatenating it with the relevant context
|
47 |
+
user_input_with_context = user_input
|
48 |
+
if relevant_context:
|
49 |
+
user_input_with_context = context_str + "\n\n" + user_input
|
50 |
+
|
51 |
+
# Append the user's input to the conversation history
|
52 |
+
conversation_history.append({"role": "user", "content": user_input_with_context})
|
53 |
+
|
54 |
+
# Create a message history including the system message and the conversation history
|
55 |
+
messages = [
|
56 |
+
{"role": "system", "content": system_message},
|
57 |
+
*conversation_history
|
58 |
+
]
|
59 |
+
|
60 |
+
# Send the completion request to the Ollama model
|
61 |
+
response = client.chat.completions.create(
|
62 |
+
model=ollama_model,
|
63 |
+
messages=messages
|
64 |
+
)
|
65 |
+
|
66 |
+
# Append the model's response to the conversation history
|
67 |
+
conversation_history.append({"role": "assistant", "content": response.choices[0].message.content})
|
68 |
+
|
69 |
+
# Return the content of the response from the model
|
70 |
+
return response.choices[0].message.content
|
71 |
+
|
72 |
+
def process_text_files(user_input):
|
73 |
+
text_parse_directory = os.path.join("local-rag", "text_parse")
|
74 |
+
temp_file_path = os.path.join("local-rag", "temp.txt")
|
75 |
+
|
76 |
+
# Check if text_parse directory exists
|
77 |
+
if not os.path.exists(text_parse_directory):
|
78 |
+
print(f"Directory '{text_parse_directory}' does not exist.")
|
79 |
+
return False
|
80 |
+
|
81 |
+
# Check if temp.txt exists
|
82 |
+
if not os.path.exists(temp_file_path):
|
83 |
+
print("temp.txt does not exist.")
|
84 |
+
return False
|
85 |
+
|
86 |
+
# Read the first line of temp.txt
|
87 |
+
with open(temp_file_path, 'r', encoding='utf-8') as temp_file:
|
88 |
+
first_line = temp_file.readline().strip()
|
89 |
+
|
90 |
+
# Get all text files in the text_parse directory
|
91 |
+
text_files = [f for f in os.listdir(text_parse_directory) if f.endswith('.txt')]
|
92 |
+
|
93 |
+
# Check if the first line matches any of the text files
|
94 |
+
if f"{first_line}" not in text_files:
|
95 |
+
print(f"No matching file found for '{first_line}.txt' in text_parse directory.")
|
96 |
+
return False
|
97 |
+
|
98 |
+
# Proceed to check for the NOT FINISHED flag
|
99 |
+
file_path = os.path.join(text_parse_directory, f"{first_line}")
|
100 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
101 |
+
lines = f.readlines()
|
102 |
+
|
103 |
+
|
104 |
+
# Check if there are any lines after NOT FINISHED
|
105 |
+
if lines[-2].strip() == "====================NOT FINISHED====================":
|
106 |
+
print(f"'{first_line}' contains the 'NOT FINISHED' flag. Computing embeddings.")
|
107 |
+
|
108 |
+
vault_content = []
|
109 |
+
if os.path.exists(temp_file_path):
|
110 |
+
with open(temp_file_path, "r", encoding='utf-8') as vault_file:
|
111 |
+
vault_content = vault_file.readlines()
|
112 |
+
|
113 |
+
|
114 |
+
# Generate embeddings for the vault content using Ollama
|
115 |
+
vault_embeddings = []
|
116 |
+
for content in vault_content:
|
117 |
+
response = ollama.embeddings(model='mxbai-embed-large', prompt=content)
|
118 |
+
vault_embeddings.append(response["embedding"])
|
119 |
+
|
120 |
+
# Convert to tensor and print embeddings
|
121 |
+
vault_embeddings_tensor = torch.tensor(vault_embeddings)
|
122 |
+
print("Embeddings for each line in the vault:")
|
123 |
+
print(vault_embeddings_tensor)
|
124 |
+
|
125 |
+
# Save the tensor result to a file or variable as needed
|
126 |
+
with open(os.path.join(text_parse_directory, f"{first_line}_embedding.pt"), "wb") as tensor_file:
|
127 |
+
torch.save(vault_embeddings_tensor, tensor_file)
|
128 |
+
|
129 |
+
# Remove the NOT FINISHED line from the original file
|
130 |
+
with open(file_path, 'w', encoding='utf-8') as f:
|
131 |
+
f.writelines(lines[:-1]) # Write back all lines except the NOT FINISHED line
|
132 |
+
|
133 |
+
else:
|
134 |
+
print(f"'{first_line}' does not contain the 'NOT FINISHED' flag or is already complete. Loading tensor if it exists.")
|
135 |
+
|
136 |
+
# Try to load the tensor from the corresponding file
|
137 |
+
tensor_file_path = os.path.join(text_parse_directory, f"{first_line}_embedding.pt")
|
138 |
+
if os.path.exists(tensor_file_path):
|
139 |
+
vault_embeddings_tensor = torch.load(tensor_file_path)
|
140 |
+
print("Loaded Vault Embedding Tensor:")
|
141 |
+
print(vault_embeddings_tensor)
|
142 |
+
|
143 |
+
vault_content = []
|
144 |
+
|
145 |
+
if os.path.exists(temp_file_path):
|
146 |
+
with open(temp_file_path, "r", encoding='utf-8') as vault_file:
|
147 |
+
vault_content = vault_file.readlines()
|
148 |
+
|
149 |
+
else:
|
150 |
+
print(f"No tensor file found for '{text_files}'.")
|
151 |
+
|
152 |
+
|
153 |
+
|
154 |
+
# Conversation loop
|
155 |
+
conversation_history = []
|
156 |
+
system_message = "You are a helpful assistant that is an expert at extracting the most useful information from a given text"
|
157 |
+
|
158 |
+
response = ollama_chat(user_input, system_message, vault_embeddings_tensor, vault_content, args.model, conversation_history)
|
159 |
+
|
160 |
+
return response
|
161 |
+
|
162 |
+
|
163 |
+
# # Read each file in the text_parse directory and check for the NOT FINISHED flag
|
164 |
+
# for txt_file in text_files:
|
165 |
+
# file_path = os.path.join(text_parse_directory, txt_file)
|
166 |
+
# with open(file_path, 'r', encoding='utf-8') as f:
|
167 |
+
# lines = f.readlines()
|
168 |
+
# # Check if the last line contains the "NOT FINISHED" flag
|
169 |
+
# if lines and lines[-1].strip() == "==========NOT FINISHED==========":
|
170 |
+
# print(f"'{txt_file}' contains the 'NOT FINISHED' flag. Proceeding to next step.")
|
171 |
+
# # Append the content of this file to the vault
|
172 |
+
# with open(temp_file_path, 'a', encoding='utf-8') as vault_file:
|
173 |
+
# vault_file.write('\n'.join(lines[:-1]) + '\n') # Append content without the last flag line
|
174 |
+
# else:
|
175 |
+
# print(f"'{txt_file}' does not contain the 'NOT FINISHED' flag. Skipping.")
|
176 |
+
|
177 |
+
# Parse command-line arguments
|
178 |
+
parser = argparse.ArgumentParser(description="Ollama Chat")
|
179 |
+
parser.add_argument("--model", default="llama3", help="Ollama model to use (default: llama3)")
|
180 |
+
args = parser.parse_args()
|
181 |
+
|
182 |
+
# Configuration for the Ollama API client
|
183 |
+
client = OpenAI(
|
184 |
+
base_url='http://localhost:11434/v1',
|
185 |
+
api_key='llama3'
|
186 |
+
)
|
187 |
+
|
188 |
+
if __name__ == "__main__":
|
189 |
+
|
190 |
+
|
191 |
+
print(process_text_files("tell me about iterators"))
|
192 |
+
|
193 |
+
|