Spaces:
Runtime error
Runtime error
Update processing.py
Browse files- processing.py +35 -16
processing.py
CHANGED
@@ -1,8 +1,10 @@
|
|
1 |
import os
|
2 |
import time
|
3 |
import re
|
|
|
4 |
from huggingface_hub import login
|
5 |
import torch
|
|
|
6 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
7 |
from langdetect import detect
|
8 |
from langchain.chains import RetrievalQA
|
@@ -28,6 +30,7 @@ def load_instructions(file_path):
|
|
28 |
with open(file_path, 'r') as file:
|
29 |
return file.read().strip()
|
30 |
|
|
|
31 |
attachments_task = load_instructions("tasks/Attachments_task.txt")
|
32 |
bigfive_task = load_instructions("tasks/BigFive_task.txt")
|
33 |
personalities_task = load_instructions("tasks/Personalities_task.txt")
|
@@ -74,7 +77,6 @@ class SequentialAnalyzer:
|
|
74 |
use_cache=False,
|
75 |
load_in_4bit=False
|
76 |
)
|
77 |
-
model.gradient_checkpointing_enable()
|
78 |
return model
|
79 |
|
80 |
def create_pipeline(self, model):
|
@@ -109,18 +111,30 @@ class SequentialAnalyzer:
|
|
109 |
print(f"Warning: Input was truncated from {input_tokens} to {max_input_length} tokens.")
|
110 |
|
111 |
llm = HuggingFacePipeline(pipeline=self.pipe)
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
cleaned_output = self.post_process_output(output)
|
125 |
|
126 |
return cleaned_output, input_tokens
|
@@ -161,7 +175,7 @@ def process_input(input_file, progress=None):
|
|
161 |
transcription = content # Store the transcription
|
162 |
os.remove(srt_path)
|
163 |
else:
|
164 |
-
return "Unsupported file format. Please upload a TXT, SRT, PDF, or video file.", None, None, None, None, None, None, None, None, None, None
|
165 |
|
166 |
detected_language = detect_language(content)
|
167 |
|
@@ -172,6 +186,11 @@ def process_input(input_file, progress=None):
|
|
172 |
|
173 |
analyzer = SequentialAnalyzer(hf_token)
|
174 |
|
|
|
|
|
|
|
|
|
|
|
175 |
safe_progress(0.5, desc="Analyzing attachments...")
|
176 |
attachments_answer, attachments_tokens = analyzer.analyze_task(content, attachments_task, attachments_db)
|
177 |
print("Attachments output:\n", attachments_answer)
|
@@ -195,5 +214,5 @@ def process_input(input_file, progress=None):
|
|
195 |
safe_progress(1.0, desc="Analysis complete!")
|
196 |
|
197 |
return ("Analysis complete!", execution_info, detected_language,
|
198 |
-
attachments_answer, bigfive_answer, personalities_answer,
|
199 |
-
original_tokens, attachments_tokens, bigfive_tokens, personalities_tokens, transcription)
|
|
|
1 |
import os
|
2 |
import time
|
3 |
import re
|
4 |
+
import numpy as np
|
5 |
from huggingface_hub import login
|
6 |
import torch
|
7 |
+
import random
|
8 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
9 |
from langdetect import detect
|
10 |
from langchain.chains import RetrievalQA
|
|
|
30 |
with open(file_path, 'r') as file:
|
31 |
return file.read().strip()
|
32 |
|
33 |
+
general_task = load_instructions("tasks/general_task.txt")
|
34 |
attachments_task = load_instructions("tasks/Attachments_task.txt")
|
35 |
bigfive_task = load_instructions("tasks/BigFive_task.txt")
|
36 |
personalities_task = load_instructions("tasks/Personalities_task.txt")
|
|
|
77 |
use_cache=False,
|
78 |
load_in_4bit=False
|
79 |
)
|
|
|
80 |
return model
|
81 |
|
82 |
def create_pipeline(self, model):
|
|
|
111 |
print(f"Warning: Input was truncated from {input_tokens} to {max_input_length} tokens.")
|
112 |
|
113 |
llm = HuggingFacePipeline(pipeline=self.pipe)
|
114 |
+
|
115 |
+
if knowledge_db is None:
|
116 |
+
# For general task without specific knowledge DB
|
117 |
+
prompt = PromptTemplate(
|
118 |
+
template=task + "\n\n{question}\n\n-----------\n\nAnswer: ",
|
119 |
+
input_variables=["question"]
|
120 |
+
)
|
121 |
+
chain = prompt | llm
|
122 |
+
result = chain.invoke({"question": truncated_content})
|
123 |
+
output = result.split("-----------\n\nAnswer:")[-1].strip()
|
124 |
+
else:
|
125 |
+
# For tasks with specific knowledge DB
|
126 |
+
chain = RetrievalQA.from_chain_type(
|
127 |
+
llm=llm,
|
128 |
+
chain_type="stuff",
|
129 |
+
retriever=knowledge_db.as_retriever(),
|
130 |
+
chain_type_kwargs={"prompt": PromptTemplate(
|
131 |
+
template=task + "\n\n{context}\n\n{question}\n\n-----------\n\nAnswer: ",
|
132 |
+
input_variables=["context", "question"]
|
133 |
+
)}
|
134 |
+
)
|
135 |
+
result = chain({"query": truncated_content})
|
136 |
+
output = result['result'].split("-----------\n\nAnswer:")[-1].strip()
|
137 |
+
|
138 |
cleaned_output = self.post_process_output(output)
|
139 |
|
140 |
return cleaned_output, input_tokens
|
|
|
175 |
transcription = content # Store the transcription
|
176 |
os.remove(srt_path)
|
177 |
else:
|
178 |
+
return "Unsupported file format. Please upload a TXT, SRT, PDF, or video file.", None, None, None, None, None, None, None, None, None, None, None, None
|
179 |
|
180 |
detected_language = detect_language(content)
|
181 |
|
|
|
186 |
|
187 |
analyzer = SequentialAnalyzer(hf_token)
|
188 |
|
189 |
+
safe_progress(0.3, desc="Performing general analysis...")
|
190 |
+
general_answer, general_tokens = analyzer.analyze_task(content, general_task, None)
|
191 |
+
print("General output:\n", general_answer)
|
192 |
+
print(f"General input tokens (before truncation): {general_tokens}")
|
193 |
+
|
194 |
safe_progress(0.5, desc="Analyzing attachments...")
|
195 |
attachments_answer, attachments_tokens = analyzer.analyze_task(content, attachments_task, attachments_db)
|
196 |
print("Attachments output:\n", attachments_answer)
|
|
|
214 |
safe_progress(1.0, desc="Analysis complete!")
|
215 |
|
216 |
return ("Analysis complete!", execution_info, detected_language,
|
217 |
+
general_answer, attachments_answer, bigfive_answer, personalities_answer,
|
218 |
+
original_tokens, general_tokens, attachments_tokens, bigfive_tokens, personalities_tokens, transcription)
|