Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -3,7 +3,7 @@ import json
|
|
3 |
import tempfile
|
4 |
import os
|
5 |
import re # For parsing conversation
|
6 |
-
from typing import Union, Optional #
|
7 |
# Import the actual functions from synthgen
|
8 |
from synthgen import (
|
9 |
generate_synthetic_text,
|
@@ -154,25 +154,39 @@ def generate_prompts_ui(
|
|
154 |
|
155 |
# --- Modified Generation Wrappers ---
|
156 |
|
157 |
-
# Wrapper for text generation + JSON preparation
|
158 |
def run_generation_and_prepare_json(
|
159 |
prompt: str,
|
160 |
model: str,
|
161 |
num_samples: int,
|
162 |
-
temperature: float,
|
163 |
top_p: float,
|
164 |
max_tokens: int
|
165 |
-
):
|
166 |
"""Generates text samples and prepares a JSON file for download."""
|
167 |
-
# Handle optional settings
|
168 |
-
temp_val = temperature if temperature > 0 else None
|
169 |
-
top_p_val = top_p if 0 < top_p <= 1 else None
|
170 |
-
max_tokens_val = max_tokens if max_tokens > 0 else None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
171 |
|
172 |
if not prompt:
|
173 |
-
|
|
|
|
|
|
|
|
|
174 |
if num_samples <= 0:
|
175 |
-
|
|
|
|
|
|
|
176 |
|
177 |
output_str = f"Generating {num_samples} samples using model '{model}'...\n"
|
178 |
output_str += f"(Settings: Temp={temp_val}, Top-P={top_p_val}, MaxTokens={max_tokens_val})\n"
|
@@ -180,48 +194,58 @@ def run_generation_and_prepare_json(
|
|
180 |
results_list = []
|
181 |
|
182 |
for i in range(num_samples):
|
183 |
-
# Pass settings to the backend function
|
184 |
generated_text = generate_synthetic_text(
|
185 |
-
prompt,
|
186 |
-
model,
|
187 |
-
temperature=temp_val,
|
188 |
-
top_p=top_p_val,
|
189 |
-
max_tokens=max_tokens_val
|
190 |
)
|
191 |
output_str += f"--- Sample {i+1} ---\n"
|
192 |
output_str += generated_text + "\n\n"
|
193 |
if not generated_text.startswith("Error:"):
|
194 |
results_list.append(generated_text)
|
195 |
-
else:
|
196 |
-
pass
|
197 |
|
198 |
output_str += "="*20 + "\nGeneration complete (check results above for errors)."
|
199 |
json_filepath = create_json_file(results_list, "text_samples.json")
|
200 |
-
|
|
|
|
|
|
|
|
|
|
|
201 |
|
202 |
|
203 |
-
# Wrapper for conversation generation + JSON preparation
|
204 |
def run_conversation_generation_and_prepare_json(
|
205 |
system_prompts_text: str,
|
206 |
model: str,
|
207 |
num_turns: int,
|
208 |
-
temperature: float,
|
209 |
top_p: float,
|
210 |
max_tokens: int
|
211 |
-
):
|
212 |
"""Generates conversations and prepares a JSON file for download."""
|
213 |
temp_val = temperature if temperature > 0 else None
|
214 |
top_p_val = top_p if 0 < top_p <= 1 else None
|
215 |
max_tokens_val = max_tokens if max_tokens > 0 else None
|
216 |
|
|
|
|
|
|
|
217 |
if not system_prompts_text:
|
218 |
-
return
|
|
|
|
|
|
|
219 |
if num_turns <= 0:
|
220 |
-
|
|
|
|
|
|
|
221 |
|
222 |
prompts = [p.strip() for p in system_prompts_text.strip().split('\n') if p.strip()]
|
223 |
if not prompts:
|
224 |
-
return
|
|
|
|
|
|
|
225 |
|
226 |
output_str = f"Generating {len(prompts)} conversations ({num_turns} turns each) using model '{model}'...\n"
|
227 |
output_str += f"(Settings: Temp={temp_val}, Top-P={top_p_val}, MaxTokens={max_tokens_val})\n"
|
@@ -229,70 +253,33 @@ def run_conversation_generation_and_prepare_json(
|
|
229 |
results_list_structured = []
|
230 |
|
231 |
for i, prompt in enumerate(prompts):
|
232 |
-
# Pass settings to the backend function
|
233 |
conversation_text = generate_synthetic_conversation(
|
234 |
-
prompt,
|
235 |
-
model,
|
236 |
-
num_turns,
|
237 |
-
temperature=temp_val,
|
238 |
-
top_p=top_p_val,
|
239 |
-
max_tokens=max_tokens_val
|
240 |
)
|
241 |
-
|
242 |
output_str += f"--- Conversation {i+1}/{len(prompts)} ---\n"
|
243 |
output_str += conversation_text + "\n\n"
|
244 |
-
|
245 |
-
# Parse the generated text block for JSON structure
|
246 |
-
# Note: generate_synthetic_conversation includes a title like "Generated conversation for..."
|
247 |
-
# We might want to remove that before parsing or adjust the parser.
|
248 |
-
# Let's assume the core conversation starts after the first line break if a title exists.
|
249 |
core_conversation_text = conversation_text
|
250 |
-
if "
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
if len(parts) > 1:
|
255 |
-
core_conversation_text = parts[1]
|
256 |
-
else: # Handle case where title might not have double newline
|
257 |
-
core_conversation_text = conversation_text # Fallback to full text
|
258 |
-
else:
|
259 |
-
core_conversation_text = None # Don't try to parse errors
|
260 |
-
elif conversation_text.startswith("Error:"):
|
261 |
-
core_conversation_text = None # Don't try to parse errors
|
262 |
-
# Else: No double newline, assume the whole text is the conversation (or error)
|
263 |
-
|
264 |
if core_conversation_text:
|
265 |
messages = parse_conversation_string(core_conversation_text)
|
266 |
-
if messages:
|
267 |
-
|
268 |
-
|
269 |
-
|
270 |
-
|
271 |
-
else: # Parsing failed, optionally add raw text or error placeholder
|
272 |
-
results_list_structured.append({
|
273 |
-
"prompt": prompt,
|
274 |
-
"error": "Failed to parse conversation structure.",
|
275 |
-
"raw_text": core_conversation_text # Include raw text if parsing failed
|
276 |
-
})
|
277 |
-
elif conversation_text.startswith("Error:"):
|
278 |
-
results_list_structured.append({
|
279 |
-
"prompt": prompt,
|
280 |
-
"error": conversation_text # Include the error message from generation
|
281 |
-
})
|
282 |
-
else: # Handle case where core_conversation_text became None unexpectedly or original text was just a title
|
283 |
-
results_list_structured.append({
|
284 |
-
"prompt": prompt,
|
285 |
-
"error": "Could not extract conversation content for parsing.",
|
286 |
-
"raw_text": conversation_text
|
287 |
-
})
|
288 |
-
|
289 |
|
290 |
output_str += "="*40 + "\nGeneration complete (check results above for errors)."
|
291 |
-
|
292 |
-
# Create JSON file from the structured list
|
293 |
json_filepath = create_json_file(results_list_structured, "conversations.json")
|
294 |
|
295 |
-
|
|
|
|
|
|
|
|
|
296 |
|
297 |
|
298 |
# --- Gradio Interface Definition ---
|
@@ -419,4 +406,4 @@ if __name__ == "__main__":
|
|
419 |
print("Launching Gradio App...")
|
420 |
print("Make sure the OPENROUTER_API_KEY environment variable is set.")
|
421 |
# Use share=True for temporary public link if running locally and need to test
|
422 |
-
demo.launch(
|
|
|
3 |
import tempfile
|
4 |
import os
|
5 |
import re # For parsing conversation
|
6 |
+
from typing import Union, Optional, Dict # Import Dict
|
7 |
# Import the actual functions from synthgen
|
8 |
from synthgen import (
|
9 |
generate_synthetic_text,
|
|
|
154 |
|
155 |
# --- Modified Generation Wrappers ---
|
156 |
|
157 |
+
# Wrapper for text generation + JSON preparation - RETURNS DICT
|
158 |
def run_generation_and_prepare_json(
|
159 |
prompt: str,
|
160 |
model: str,
|
161 |
num_samples: int,
|
162 |
+
temperature: float,
|
163 |
top_p: float,
|
164 |
max_tokens: int
|
165 |
+
) -> Dict[gr.Textbox, str]: # Return type hint (optional but good practice)
|
166 |
"""Generates text samples and prepares a JSON file for download."""
|
167 |
+
# Handle optional settings
|
168 |
+
temp_val = temperature if temperature > 0 else None
|
169 |
+
top_p_val = top_p if 0 < top_p <= 1 else None
|
170 |
+
max_tokens_val = max_tokens if max_tokens > 0 else None
|
171 |
+
|
172 |
+
# Define component objects used in return dict keys - MUST MATCH OUTPUTS
|
173 |
+
# This requires the components to be defined *before* this function,
|
174 |
+
# which isn't the case. So we cannot use component objects as keys here.
|
175 |
+
# Gradio handles mapping if the keys are strings matching component labels
|
176 |
+
# OR if we return gr.update targeting components.
|
177 |
+
# Let's return explicit gr.update for clarity and robustness.
|
178 |
|
179 |
if not prompt:
|
180 |
+
# Return updates for both outputs
|
181 |
+
return {
|
182 |
+
output_text: gr.update(value="Error: Please enter a prompt."),
|
183 |
+
download_file_text: gr.update(value=None) # Clear file output
|
184 |
+
}
|
185 |
if num_samples <= 0:
|
186 |
+
return {
|
187 |
+
output_text: gr.update(value="Error: Number of samples must be positive."),
|
188 |
+
download_file_text: gr.update(value=None)
|
189 |
+
}
|
190 |
|
191 |
output_str = f"Generating {num_samples} samples using model '{model}'...\n"
|
192 |
output_str += f"(Settings: Temp={temp_val}, Top-P={top_p_val}, MaxTokens={max_tokens_val})\n"
|
|
|
194 |
results_list = []
|
195 |
|
196 |
for i in range(num_samples):
|
|
|
197 |
generated_text = generate_synthetic_text(
|
198 |
+
prompt, model, temperature=temp_val, top_p=top_p_val, max_tokens=max_tokens_val
|
|
|
|
|
|
|
|
|
199 |
)
|
200 |
output_str += f"--- Sample {i+1} ---\n"
|
201 |
output_str += generated_text + "\n\n"
|
202 |
if not generated_text.startswith("Error:"):
|
203 |
results_list.append(generated_text)
|
|
|
|
|
204 |
|
205 |
output_str += "="*20 + "\nGeneration complete (check results above for errors)."
|
206 |
json_filepath = create_json_file(results_list, "text_samples.json")
|
207 |
+
|
208 |
+
# Return dictionary mapping components to updates
|
209 |
+
return {
|
210 |
+
output_text: gr.update(value=output_str),
|
211 |
+
download_file_text: gr.update(value=json_filepath) # Update file path
|
212 |
+
}
|
213 |
|
214 |
|
215 |
+
# Wrapper for conversation generation + JSON preparation - RETURNS DICT
|
216 |
def run_conversation_generation_and_prepare_json(
|
217 |
system_prompts_text: str,
|
218 |
model: str,
|
219 |
num_turns: int,
|
220 |
+
temperature: float,
|
221 |
top_p: float,
|
222 |
max_tokens: int
|
223 |
+
) -> Dict[gr.Textbox, str]: # Return type hint (optional)
|
224 |
"""Generates conversations and prepares a JSON file for download."""
|
225 |
temp_val = temperature if temperature > 0 else None
|
226 |
top_p_val = top_p if 0 < top_p <= 1 else None
|
227 |
max_tokens_val = max_tokens if max_tokens > 0 else None
|
228 |
|
229 |
+
# Define component objects used in return dict keys - requires components defined first.
|
230 |
+
# Using explicit gr.update instead.
|
231 |
+
|
232 |
if not system_prompts_text:
|
233 |
+
return {
|
234 |
+
output_conv: gr.update(value="Error: Please enter or generate at least one system prompt/topic."),
|
235 |
+
download_file_conv: gr.update(value=None)
|
236 |
+
}
|
237 |
if num_turns <= 0:
|
238 |
+
return {
|
239 |
+
output_conv: gr.update(value="Error: Number of turns must be positive."),
|
240 |
+
download_file_conv: gr.update(value=None)
|
241 |
+
}
|
242 |
|
243 |
prompts = [p.strip() for p in system_prompts_text.strip().split('\n') if p.strip()]
|
244 |
if not prompts:
|
245 |
+
return {
|
246 |
+
output_conv: gr.update(value="Error: No valid prompts found in the input."),
|
247 |
+
download_file_conv: gr.update(value=None)
|
248 |
+
}
|
249 |
|
250 |
output_str = f"Generating {len(prompts)} conversations ({num_turns} turns each) using model '{model}'...\n"
|
251 |
output_str += f"(Settings: Temp={temp_val}, Top-P={top_p_val}, MaxTokens={max_tokens_val})\n"
|
|
|
253 |
results_list_structured = []
|
254 |
|
255 |
for i, prompt in enumerate(prompts):
|
|
|
256 |
conversation_text = generate_synthetic_conversation(
|
257 |
+
prompt, model, num_turns, temperature=temp_val, top_p=top_p_val, max_tokens=max_tokens_val
|
|
|
|
|
|
|
|
|
|
|
258 |
)
|
|
|
259 |
output_str += f"--- Conversation {i+1}/{len(prompts)} ---\n"
|
260 |
output_str += conversation_text + "\n\n"
|
261 |
+
# --- Parsing Logic ---
|
|
|
|
|
|
|
|
|
262 |
core_conversation_text = conversation_text
|
263 |
+
if conversation_text.startswith("Error:"): core_conversation_text = None
|
264 |
+
elif "\n\n" in conversation_text:
|
265 |
+
parts = conversation_text.split("\n\n", 1)
|
266 |
+
core_conversation_text = parts[1] if len(parts) > 1 else conversation_text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
267 |
if core_conversation_text:
|
268 |
messages = parse_conversation_string(core_conversation_text)
|
269 |
+
if messages: results_list_structured.append({"prompt": prompt, "messages": messages})
|
270 |
+
else: results_list_structured.append({"prompt": prompt, "error": "Failed to parse structure.", "raw_text": core_conversation_text})
|
271 |
+
elif conversation_text.startswith("Error:"): results_list_structured.append({"prompt": prompt, "error": conversation_text})
|
272 |
+
else: results_list_structured.append({"prompt": prompt, "error": "Could not extract content.", "raw_text": conversation_text})
|
273 |
+
# --- End Parsing Logic ---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
274 |
|
275 |
output_str += "="*40 + "\nGeneration complete (check results above for errors)."
|
|
|
|
|
276 |
json_filepath = create_json_file(results_list_structured, "conversations.json")
|
277 |
|
278 |
+
# Return dictionary mapping components to updates
|
279 |
+
return {
|
280 |
+
output_conv: gr.update(value=output_str),
|
281 |
+
download_file_conv: gr.update(value=json_filepath)
|
282 |
+
}
|
283 |
|
284 |
|
285 |
# --- Gradio Interface Definition ---
|
|
|
406 |
print("Launching Gradio App...")
|
407 |
print("Make sure the OPENROUTER_API_KEY environment variable is set.")
|
408 |
# Use share=True for temporary public link if running locally and need to test
|
409 |
+
demo.launch() # share=True
|