Update app.py
Browse files
app.py
CHANGED
|
@@ -1,20 +1,54 @@
|
|
| 1 |
import gradio as gr
|
|
|
|
| 2 |
import json
|
| 3 |
import re
|
| 4 |
from datetime import datetime
|
| 5 |
from typing import Literal
|
| 6 |
import os
|
| 7 |
import importlib
|
| 8 |
-
from llm_handler import send_to_llm
|
| 9 |
from main import generate_data, PROMPT_1
|
| 10 |
from topics import TOPICS
|
| 11 |
from system_messages import SYSTEM_MESSAGES_VODALUS
|
| 12 |
import random
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
|
| 15 |
ANNOTATION_CONFIG_FILE = "annotation_config.json"
|
| 16 |
OUTPUT_FILE_PATH = "dataset.jsonl"
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
def load_annotation_config():
|
| 19 |
try:
|
| 20 |
with open(ANNOTATION_CONFIG_FILE, 'r') as f:
|
|
@@ -57,6 +91,19 @@ def load_annotation_config():
|
|
| 57 |
]
|
| 58 |
}
|
| 59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
def save_annotation_config(config):
|
| 61 |
with open(ANNOTATION_CONFIG_FILE, 'w') as f:
|
| 62 |
json.dump(config, f, indent=2)
|
|
@@ -66,8 +113,44 @@ def load_jsonl_dataset(file_path):
|
|
| 66 |
return []
|
| 67 |
with open(file_path, 'r') as f:
|
| 68 |
return [json.loads(line.strip()) for line in f if line.strip()]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
|
| 70 |
def save_row(file_path, index, row_data):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
with open(file_path, 'r') as f:
|
| 72 |
lines = f.readlines()
|
| 73 |
|
|
@@ -75,8 +158,23 @@ def save_row(file_path, index, row_data):
|
|
| 75 |
|
| 76 |
with open(file_path, 'w') as f:
|
| 77 |
f.writelines(lines)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
|
| 81 |
def get_row(file_path, index):
|
| 82 |
data = load_jsonl_dataset(file_path)
|
|
@@ -106,19 +204,19 @@ def markdown_to_json(markdown_str):
|
|
| 106 |
}
|
| 107 |
return json.dumps(json_data, indent=2)
|
| 108 |
|
| 109 |
-
def navigate_rows(file_path: str, current_index: int, direction: Literal[
|
| 110 |
-
new_index = max(0, current_index + direction)
|
| 111 |
return load_and_show_row(file_path, new_index, metadata_config)
|
| 112 |
|
| 113 |
def load_and_show_row(file_path, index, metadata_config):
|
| 114 |
row_data, total = get_row(file_path, index)
|
| 115 |
if not row_data:
|
| 116 |
-
return ("", index, total, "3", [], [], [], "")
|
| 117 |
|
| 118 |
try:
|
| 119 |
data = json.loads(row_data)
|
| 120 |
except json.JSONDecodeError:
|
| 121 |
-
return (row_data, index, total, "3", [], [], [], "Error: Invalid JSON")
|
| 122 |
|
| 123 |
metadata = data.get("metadata", {}).get("annotation", {})
|
| 124 |
|
|
@@ -128,7 +226,7 @@ def load_and_show_row(file_path, index, metadata_config):
|
|
| 128 |
toxic_tags = metadata.get("tags", {}).get("toxic", [])
|
| 129 |
other = metadata.get("free_text", {}).get("Additional Notes", "")
|
| 130 |
|
| 131 |
-
return (row_data, index, total, quality,
|
| 132 |
high_quality_tags, low_quality_tags, toxic_tags, other)
|
| 133 |
|
| 134 |
def save_row_with_metadata(file_path, index, row_data, config, quality, high_quality_tags, low_quality_tags, toxic_tags, other):
|
|
@@ -182,7 +280,12 @@ def load_config_to_ui(config):
|
|
| 182 |
[[field["name"], field["description"]] for field in config["free_text_fields"]]
|
| 183 |
)
|
| 184 |
|
| 185 |
-
def save_config_from_ui(name, description, scale, categories, fields):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 186 |
new_config = {
|
| 187 |
"quality_scale": {
|
| 188 |
"name": name,
|
|
@@ -190,7 +293,8 @@ def save_config_from_ui(name, description, scale, categories, fields):
|
|
| 190 |
"scale": [{"value": row[0], "label": row[1]} for row in scale]
|
| 191 |
},
|
| 192 |
"tag_categories": [{"name": row[0], "type": row[1], "tags": row[2].split(", ")} for row in categories],
|
| 193 |
-
"free_text_fields": [{"name": row[0], "description": row[1]} for row in fields]
|
|
|
|
| 194 |
}
|
| 195 |
save_annotation_config(new_config)
|
| 196 |
return "Configuration saved successfully", new_config
|
|
@@ -218,7 +322,7 @@ def generate_preview(row_data, quality, high_quality_tags, low_quality_tags, tox
|
|
| 218 |
return "Error: Invalid JSON in the current row data"
|
| 219 |
|
| 220 |
def load_dataset_config():
|
| 221 |
-
|
| 222 |
with open("system_messages.py", "r") as f:
|
| 223 |
system_messages_content = f.read()
|
| 224 |
vodalus_system_message = re.search(r'SYSTEM_MESSAGES_VODALUS = \[(.*?)\]', system_messages_content, re.DOTALL).group(1).strip()[3:-3] # Extract the content between triple quotes
|
|
@@ -232,9 +336,37 @@ def load_dataset_config():
|
|
| 232 |
topics_module = importlib.import_module("topics")
|
| 233 |
topics_list = topics_module.TOPICS
|
| 234 |
|
| 235 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 236 |
|
| 237 |
-
def save_dataset_config(system_messages, prompt_1, topics):
|
| 238 |
# Save VODALUS_SYSTEM_MESSAGE to system_messages.py
|
| 239 |
with open("system_messages.py", "w") as f:
|
| 240 |
f.write(f'SYSTEM_MESSAGES_VODALUS = [\n"""\n{system_messages}\n""",\n]\n')
|
|
@@ -261,8 +393,17 @@ def save_dataset_config(system_messages, prompt_1, topics):
|
|
| 261 |
|
| 262 |
with open("topics.py", "w") as f:
|
| 263 |
f.write(topics_content)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 264 |
|
| 265 |
return "Dataset configuration saved successfully"
|
|
|
|
| 266 |
|
| 267 |
|
| 268 |
def chat_with_llm(message, history):
|
|
@@ -273,7 +414,12 @@ def chat_with_llm(message, history):
|
|
| 273 |
msg_list.append({"role": "assistant", "content": h[1]})
|
| 274 |
msg_list.append({"role": "user", "content": message})
|
| 275 |
|
| 276 |
-
response, _ = send_to_llm(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 277 |
|
| 278 |
return history + [[message, response]]
|
| 279 |
except Exception as e:
|
|
@@ -283,14 +429,15 @@ def chat_with_llm(message, history):
|
|
| 283 |
def update_chat_context(row_data, index, total, quality, high_quality_tags, low_quality_tags, toxic_tags, other):
|
| 284 |
context = f"""Current app state:
|
| 285 |
Row: {index + 1}/{total}
|
| 286 |
-
Data: {row_data}
|
| 287 |
Quality: {quality}
|
| 288 |
High Quality Tags: {', '.join(high_quality_tags)}
|
| 289 |
Low Quality Tags: {', '.join(low_quality_tags)}
|
| 290 |
Toxic Tags: {', '.join(toxic_tags)}
|
| 291 |
Additional Notes: {other}
|
|
|
|
|
|
|
| 292 |
"""
|
| 293 |
-
return [[None, context]]
|
| 294 |
|
| 295 |
|
| 296 |
async def run_generate_dataset(num_workers, num_generations, output_file_path):
|
|
@@ -309,34 +456,191 @@ async def run_generate_dataset(num_workers, num_generations, output_file_path):
|
|
| 309 |
|
| 310 |
return f"Generated {num_generations} entries and saved to {output_file_path}", "\n".join(generated_data[:5]) + "\n..."
|
| 311 |
|
| 312 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 313 |
|
| 314 |
with demo:
|
| 315 |
-
gr.Markdown("#
|
| 316 |
|
| 317 |
config = gr.State(load_annotation_config())
|
| 318 |
|
| 319 |
with gr.Row():
|
| 320 |
-
with gr.Column(
|
| 321 |
with gr.Tab("Dataset Editor"):
|
| 322 |
-
with gr.Row():
|
| 323 |
-
|
| 324 |
-
|
|
|
|
|
|
|
| 325 |
|
| 326 |
with gr.Row():
|
| 327 |
prev_button = gr.Button("← Previous")
|
| 328 |
-
row_index = gr.
|
| 329 |
-
total_rows = gr.
|
|
|
|
| 330 |
next_button = gr.Button("Next →")
|
| 331 |
|
| 332 |
with gr.Row():
|
| 333 |
with gr.Column(scale=3):
|
| 334 |
-
row_editor = gr.TextArea(label="Edit Row", lines=
|
| 335 |
|
| 336 |
with gr.Column(scale=2):
|
| 337 |
quality_label = gr.Radio(label="Relevance for Training", choices=[])
|
| 338 |
tag_components = [gr.CheckboxGroup(label=f"Tag Group {i+1}", choices=[]) for i in range(3)]
|
| 339 |
other_description = gr.Textbox(label="Additional annotations", lines=3)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 340 |
|
| 341 |
with gr.Row():
|
| 342 |
to_markdown_button = gr.Button("Convert to Markdown")
|
|
@@ -349,50 +653,94 @@ with demo:
|
|
| 349 |
|
| 350 |
with gr.Tab("Annotation Configuration"):
|
| 351 |
with gr.Row():
|
| 352 |
-
with gr.Column():
|
| 353 |
-
|
| 354 |
-
|
|
|
|
|
|
|
|
|
|
| 355 |
quality_scale = gr.Dataframe(
|
| 356 |
headers=["Value", "Label"],
|
| 357 |
datatype=["str", "str"],
|
| 358 |
-
label="Quality Scale",
|
| 359 |
-
interactive=True
|
|
|
|
|
|
|
|
|
|
|
|
|
| 360 |
)
|
| 361 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 362 |
with gr.Row():
|
| 363 |
-
|
| 364 |
-
|
| 365 |
-
|
| 366 |
-
|
| 367 |
-
|
| 368 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 369 |
|
| 370 |
with gr.Row():
|
| 371 |
-
|
| 372 |
-
|
| 373 |
-
datatype=["str", "str"],
|
| 374 |
-
label="Free Text Fields",
|
| 375 |
-
interactive=True
|
| 376 |
-
)
|
| 377 |
|
| 378 |
-
|
| 379 |
-
|
|
|
|
|
|
|
| 380 |
|
| 381 |
with gr.Tab("Dataset Configuration"):
|
| 382 |
with gr.Row():
|
| 383 |
-
vodalus_system_message = gr.TextArea(label="
|
| 384 |
-
prompt_1 = gr.TextArea(label="
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 385 |
|
| 386 |
with gr.Row():
|
| 387 |
-
|
| 388 |
-
|
| 389 |
-
datatype=["str"],
|
| 390 |
-
label="TOPICS",
|
| 391 |
-
interactive=True
|
| 392 |
-
)
|
| 393 |
|
| 394 |
-
save_dataset_config_btn = gr.Button("Save Dataset Configuration")
|
| 395 |
-
dataset_config_status = gr.Textbox(label="Status")
|
| 396 |
|
| 397 |
with gr.Tab("Dataset Generation"):
|
| 398 |
with gr.Row():
|
|
@@ -406,16 +754,54 @@ with demo:
|
|
| 406 |
generation_status = gr.Textbox(label="Generation Status")
|
| 407 |
generation_output = gr.TextArea(label="Generation Output", lines=10)
|
| 408 |
|
| 409 |
-
|
| 410 |
-
|
| 411 |
-
|
| 412 |
-
|
| 413 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 414 |
|
| 415 |
load_button.click(
|
| 416 |
-
|
| 417 |
-
inputs=[
|
| 418 |
-
outputs=[row_editor, row_index, total_rows, quality_label, *tag_components, other_description]
|
| 419 |
).then(
|
| 420 |
update_annotation_ui,
|
| 421 |
inputs=[config],
|
|
@@ -424,8 +810,8 @@ with demo:
|
|
| 424 |
|
| 425 |
prev_button.click(
|
| 426 |
navigate_rows,
|
| 427 |
-
inputs=[
|
| 428 |
-
outputs=[row_editor, row_index, total_rows, quality_label, *tag_components, other_description]
|
| 429 |
).then(
|
| 430 |
update_annotation_ui,
|
| 431 |
inputs=[config],
|
|
@@ -434,8 +820,8 @@ with demo:
|
|
| 434 |
|
| 435 |
next_button.click(
|
| 436 |
navigate_rows,
|
| 437 |
-
inputs=[
|
| 438 |
-
outputs=[row_editor, row_index, total_rows, quality_label, *tag_components, other_description]
|
| 439 |
).then(
|
| 440 |
update_annotation_ui,
|
| 441 |
inputs=[config],
|
|
@@ -444,7 +830,7 @@ with demo:
|
|
| 444 |
|
| 445 |
save_row_button.click(
|
| 446 |
save_row_with_metadata,
|
| 447 |
-
inputs=[
|
| 448 |
tag_components[0], tag_components[1], tag_components[2], other_description],
|
| 449 |
outputs=[editor_status]
|
| 450 |
).then(
|
|
@@ -476,7 +862,7 @@ with demo:
|
|
| 476 |
|
| 477 |
save_config_btn.click(
|
| 478 |
save_config_from_ui,
|
| 479 |
-
inputs=[quality_scale_name, quality_scale_description, quality_scale, tag_categories, free_text_fields],
|
| 480 |
outputs=[config_status, config]
|
| 481 |
).then(
|
| 482 |
update_annotation_ui,
|
|
@@ -492,12 +878,12 @@ with demo:
|
|
| 492 |
|
| 493 |
demo.load(
|
| 494 |
load_dataset_config,
|
| 495 |
-
outputs=[vodalus_system_message, prompt_1, topics]
|
| 496 |
)
|
| 497 |
|
| 498 |
save_dataset_config_btn.click(
|
| 499 |
save_dataset_config,
|
| 500 |
-
inputs=[vodalus_system_message, prompt_1, topics],
|
| 501 |
outputs=[dataset_config_status]
|
| 502 |
)
|
| 503 |
|
|
@@ -507,10 +893,21 @@ with demo:
|
|
| 507 |
outputs=[generation_status, generation_output]
|
| 508 |
)
|
| 509 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 510 |
msg.submit(chat_with_llm, [msg, chatbot], [chatbot])
|
| 511 |
clear.click(lambda: None, None, chatbot, queue=False)
|
| 512 |
|
| 513 |
-
|
| 514 |
for button in [load_button, prev_button, next_button]:
|
| 515 |
button.click(
|
| 516 |
update_chat_context,
|
|
@@ -518,6 +915,30 @@ with demo:
|
|
| 518 |
outputs=[chatbot]
|
| 519 |
)
|
| 520 |
|
| 521 |
-
if __name__ == "__main__":
|
| 522 |
-
demo.launch(share=True)
|
| 523 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
from gradio import update
|
| 3 |
import json
|
| 4 |
import re
|
| 5 |
from datetime import datetime
|
| 6 |
from typing import Literal
|
| 7 |
import os
|
| 8 |
import importlib
|
| 9 |
+
from llm_handler import send_to_llm
|
| 10 |
from main import generate_data, PROMPT_1
|
| 11 |
from topics import TOPICS
|
| 12 |
from system_messages import SYSTEM_MESSAGES_VODALUS
|
| 13 |
import random
|
| 14 |
+
from params import load_params, save_params
|
| 15 |
+
import pandas as pd
|
| 16 |
+
import csv
|
| 17 |
+
|
| 18 |
|
| 19 |
|
| 20 |
ANNOTATION_CONFIG_FILE = "annotation_config.json"
|
| 21 |
OUTPUT_FILE_PATH = "dataset.jsonl"
|
| 22 |
|
| 23 |
+
def load_llm_config():
|
| 24 |
+
params = load_params()
|
| 25 |
+
return (
|
| 26 |
+
params.get('PROVIDER', ''),
|
| 27 |
+
params.get('BASE_URL', ''),
|
| 28 |
+
params.get('WORKSPACE', ''),
|
| 29 |
+
params.get('API_KEY', ''),
|
| 30 |
+
params.get('max_tokens', 2048),
|
| 31 |
+
params.get('temperature', 0.7),
|
| 32 |
+
params.get('top_p', 0.9),
|
| 33 |
+
params.get('frequency_penalty', 0.0),
|
| 34 |
+
params.get('presence_penalty', 0.0)
|
| 35 |
+
)
|
| 36 |
+
|
| 37 |
+
def save_llm_config(provider, base_url, workspace, api_key, max_tokens, temperature, top_p, frequency_penalty, presence_penalty):
|
| 38 |
+
save_params({
|
| 39 |
+
'PROVIDER': provider,
|
| 40 |
+
'BASE_URL': base_url,
|
| 41 |
+
'WORKSPACE': workspace,
|
| 42 |
+
'API_KEY': api_key,
|
| 43 |
+
'max_tokens': max_tokens,
|
| 44 |
+
'temperature': temperature,
|
| 45 |
+
'top_p': top_p,
|
| 46 |
+
'frequency_penalty': frequency_penalty,
|
| 47 |
+
'presence_penalty': presence_penalty
|
| 48 |
+
})
|
| 49 |
+
return "LLM configuration saved successfully"
|
| 50 |
+
|
| 51 |
+
|
| 52 |
def load_annotation_config():
|
| 53 |
try:
|
| 54 |
with open(ANNOTATION_CONFIG_FILE, 'r') as f:
|
|
|
|
| 91 |
]
|
| 92 |
}
|
| 93 |
|
| 94 |
+
|
| 95 |
+
def load_csv_dataset(file_path):
|
| 96 |
+
data = []
|
| 97 |
+
with open(file_path, 'r') as f:
|
| 98 |
+
reader = csv.DictReader(f)
|
| 99 |
+
for row in reader:
|
| 100 |
+
data.append(row)
|
| 101 |
+
return data
|
| 102 |
+
|
| 103 |
+
def load_txt_dataset(file_path):
|
| 104 |
+
with open(file_path, 'r') as f:
|
| 105 |
+
return [{"content": line.strip()} for line in f if line.strip()]
|
| 106 |
+
|
| 107 |
def save_annotation_config(config):
|
| 108 |
with open(ANNOTATION_CONFIG_FILE, 'w') as f:
|
| 109 |
json.dump(config, f, indent=2)
|
|
|
|
| 113 |
return []
|
| 114 |
with open(file_path, 'r') as f:
|
| 115 |
return [json.loads(line.strip()) for line in f if line.strip()]
|
| 116 |
+
|
| 117 |
+
def load_dataset(file):
|
| 118 |
+
if file is None:
|
| 119 |
+
return "", 0, 0, "No file uploaded", "3", [], [], [], ""
|
| 120 |
+
|
| 121 |
+
file_path = file.name
|
| 122 |
+
file_extension = os.path.splitext(file_path)[1].lower()
|
| 123 |
+
|
| 124 |
+
if file_extension == '.csv':
|
| 125 |
+
data = load_csv_dataset(file_path)
|
| 126 |
+
elif file_extension == '.txt':
|
| 127 |
+
data = load_txt_dataset(file_path)
|
| 128 |
+
elif file_extension == '.jsonl':
|
| 129 |
+
data = load_jsonl_dataset(file_path)
|
| 130 |
+
else:
|
| 131 |
+
return "", 0, 0, f"Unsupported file type: {file_extension}", "3", [], [], [], ""
|
| 132 |
+
|
| 133 |
+
if not data:
|
| 134 |
+
return "", 0, 0, "No data found in the file", "3", [], [], [], ""
|
| 135 |
+
|
| 136 |
+
first_row = json.dumps(data[0], indent=2)
|
| 137 |
+
return first_row, 0, len(data), f"Row: 1/{len(data)}", "3", [], [], [], ""
|
| 138 |
|
| 139 |
def save_row(file_path, index, row_data):
|
| 140 |
+
file_extension = file_path.split('.')[-1].lower()
|
| 141 |
+
|
| 142 |
+
if file_extension == 'jsonl':
|
| 143 |
+
save_jsonl_row(file_path, index, row_data)
|
| 144 |
+
elif file_extension == 'csv':
|
| 145 |
+
save_csv_row(file_path, index, row_data)
|
| 146 |
+
elif file_extension == 'txt':
|
| 147 |
+
save_txt_row(file_path, index, row_data)
|
| 148 |
+
else:
|
| 149 |
+
raise ValueError(f"Unsupported file format: {file_extension}")
|
| 150 |
+
|
| 151 |
+
return f"Row {index} saved successfully"
|
| 152 |
+
|
| 153 |
+
def save_jsonl_row(file_path, index, row_data):
|
| 154 |
with open(file_path, 'r') as f:
|
| 155 |
lines = f.readlines()
|
| 156 |
|
|
|
|
| 158 |
|
| 159 |
with open(file_path, 'w') as f:
|
| 160 |
f.writelines(lines)
|
| 161 |
+
|
| 162 |
+
def save_csv_row(file_path, index, row_data):
|
| 163 |
+
df = pd.read_csv(file_path)
|
| 164 |
+
row_dict = json.loads(row_data)
|
| 165 |
+
for col, value in row_dict.items():
|
| 166 |
+
df.at[index, col] = value
|
| 167 |
+
df.to_csv(file_path, index=False)
|
| 168 |
+
|
| 169 |
+
def save_txt_row(file_path, index, row_data):
|
| 170 |
+
with open(file_path, 'r') as f:
|
| 171 |
+
lines = f.readlines()
|
| 172 |
|
| 173 |
+
row_dict = json.loads(row_data)
|
| 174 |
+
lines[index] = row_dict.get('content', '') + '\n'
|
| 175 |
+
|
| 176 |
+
with open(file_path, 'w') as f:
|
| 177 |
+
f.writelines(lines)
|
| 178 |
|
| 179 |
def get_row(file_path, index):
|
| 180 |
data = load_jsonl_dataset(file_path)
|
|
|
|
| 204 |
}
|
| 205 |
return json.dumps(json_data, indent=2)
|
| 206 |
|
| 207 |
+
def navigate_rows(file_path: str, current_index: int, direction: Literal["prev", "next"], metadata_config):
|
| 208 |
+
new_index = max(0, current_index + (-1 if direction == "prev" else 1))
|
| 209 |
return load_and_show_row(file_path, new_index, metadata_config)
|
| 210 |
|
| 211 |
def load_and_show_row(file_path, index, metadata_config):
|
| 212 |
row_data, total = get_row(file_path, index)
|
| 213 |
if not row_data:
|
| 214 |
+
return ("", index, total, f"Row: {index + 1}/{total}", "3", [], [], [], "")
|
| 215 |
|
| 216 |
try:
|
| 217 |
data = json.loads(row_data)
|
| 218 |
except json.JSONDecodeError:
|
| 219 |
+
return (row_data, index, total, f"Row: {index + 1}/{total}", "3", [], [], [], "Error: Invalid JSON")
|
| 220 |
|
| 221 |
metadata = data.get("metadata", {}).get("annotation", {})
|
| 222 |
|
|
|
|
| 226 |
toxic_tags = metadata.get("tags", {}).get("toxic", [])
|
| 227 |
other = metadata.get("free_text", {}).get("Additional Notes", "")
|
| 228 |
|
| 229 |
+
return (row_data, index, total, f"Row: {index + 1}/{total}", quality,
|
| 230 |
high_quality_tags, low_quality_tags, toxic_tags, other)
|
| 231 |
|
| 232 |
def save_row_with_metadata(file_path, index, row_data, config, quality, high_quality_tags, low_quality_tags, toxic_tags, other):
|
|
|
|
| 280 |
[[field["name"], field["description"]] for field in config["free_text_fields"]]
|
| 281 |
)
|
| 282 |
|
| 283 |
+
def save_config_from_ui(name, description, scale, categories, fields, topics, all_topics_text):
|
| 284 |
+
if all_topics_text.visible:
|
| 285 |
+
topics_list = [topic.strip() for topic in all_topics_text.split("\n") if topic.strip()]
|
| 286 |
+
else:
|
| 287 |
+
topics_list = [topic[0] for topic in topics]
|
| 288 |
+
|
| 289 |
new_config = {
|
| 290 |
"quality_scale": {
|
| 291 |
"name": name,
|
|
|
|
| 293 |
"scale": [{"value": row[0], "label": row[1]} for row in scale]
|
| 294 |
},
|
| 295 |
"tag_categories": [{"name": row[0], "type": row[1], "tags": row[2].split(", ")} for row in categories],
|
| 296 |
+
"free_text_fields": [{"name": row[0], "description": row[1]} for row in fields],
|
| 297 |
+
"topics": topics_list
|
| 298 |
}
|
| 299 |
save_annotation_config(new_config)
|
| 300 |
return "Configuration saved successfully", new_config
|
|
|
|
| 322 |
return "Error: Invalid JSON in the current row data"
|
| 323 |
|
| 324 |
def load_dataset_config():
|
| 325 |
+
params = load_params()
|
| 326 |
with open("system_messages.py", "r") as f:
|
| 327 |
system_messages_content = f.read()
|
| 328 |
vodalus_system_message = re.search(r'SYSTEM_MESSAGES_VODALUS = \[(.*?)\]', system_messages_content, re.DOTALL).group(1).strip()[3:-3] # Extract the content between triple quotes
|
|
|
|
| 336 |
topics_module = importlib.import_module("topics")
|
| 337 |
topics_list = topics_module.TOPICS
|
| 338 |
|
| 339 |
+
return (
|
| 340 |
+
vodalus_system_message,
|
| 341 |
+
prompt_1,
|
| 342 |
+
[[topic] for topic in topics_list],
|
| 343 |
+
params.get('max_tokens', 2048),
|
| 344 |
+
params.get('temperature', 0.7),
|
| 345 |
+
params.get('top_p', 0.9),
|
| 346 |
+
params.get('frequency_penalty', 0.0),
|
| 347 |
+
params.get('presence_penalty', 0.0)
|
| 348 |
+
)
|
| 349 |
+
|
| 350 |
+
def edit_all_topics_func(topics):
|
| 351 |
+
topics_list = [topic[0] for topic in topics]
|
| 352 |
+
jsonl_rows = "\n".join([json.dumps({"topic": topic}) for topic in topics_list])
|
| 353 |
+
return (
|
| 354 |
+
gr.update(visible=False),
|
| 355 |
+
gr.update(value=jsonl_rows, visible=True),
|
| 356 |
+
gr.update(visible=True)
|
| 357 |
+
)
|
| 358 |
+
|
| 359 |
+
def update_topics_from_text(text):
|
| 360 |
+
try:
|
| 361 |
+
# Try parsing as JSONL
|
| 362 |
+
topics_list = [json.loads(line)["topic"] for line in text.split("\n") if line.strip()]
|
| 363 |
+
except json.JSONDecodeError:
|
| 364 |
+
# If parsing fails, treat as plain text
|
| 365 |
+
topics_list = [topic.strip() for topic in text.split("\n") if topic.strip()]
|
| 366 |
+
|
| 367 |
+
return gr.Dataframe.update(value=[[topic] for topic in topics_list], visible=True), gr.TextArea.update(visible=False)
|
| 368 |
|
| 369 |
+
def save_dataset_config(system_messages, prompt_1, topics, max_tokens, temperature, top_p, frequency_penalty, presence_penalty):
|
| 370 |
# Save VODALUS_SYSTEM_MESSAGE to system_messages.py
|
| 371 |
with open("system_messages.py", "w") as f:
|
| 372 |
f.write(f'SYSTEM_MESSAGES_VODALUS = [\n"""\n{system_messages}\n""",\n]\n')
|
|
|
|
| 393 |
|
| 394 |
with open("topics.py", "w") as f:
|
| 395 |
f.write(topics_content)
|
| 396 |
+
|
| 397 |
+
save_params({
|
| 398 |
+
'max_tokens': max_tokens,
|
| 399 |
+
'temperature': temperature,
|
| 400 |
+
'top_p': top_p,
|
| 401 |
+
'frequency_penalty': frequency_penalty,
|
| 402 |
+
'presence_penalty': presence_penalty
|
| 403 |
+
})
|
| 404 |
|
| 405 |
return "Dataset configuration saved successfully"
|
| 406 |
+
|
| 407 |
|
| 408 |
|
| 409 |
def chat_with_llm(message, history):
|
|
|
|
| 414 |
msg_list.append({"role": "assistant", "content": h[1]})
|
| 415 |
msg_list.append({"role": "user", "content": message})
|
| 416 |
|
| 417 |
+
response, _ = send_to_llm(msg_list)
|
| 418 |
+
|
| 419 |
+
return history + [[message, response]]
|
| 420 |
+
except Exception as e:
|
| 421 |
+
print(f"Error in chat_with_llm: {str(e)}")
|
| 422 |
+
return history + [[message, f"Error: {str(e)}"]]
|
| 423 |
|
| 424 |
return history + [[message, response]]
|
| 425 |
except Exception as e:
|
|
|
|
| 429 |
def update_chat_context(row_data, index, total, quality, high_quality_tags, low_quality_tags, toxic_tags, other):
|
| 430 |
context = f"""Current app state:
|
| 431 |
Row: {index + 1}/{total}
|
|
|
|
| 432 |
Quality: {quality}
|
| 433 |
High Quality Tags: {', '.join(high_quality_tags)}
|
| 434 |
Low Quality Tags: {', '.join(low_quality_tags)}
|
| 435 |
Toxic Tags: {', '.join(toxic_tags)}
|
| 436 |
Additional Notes: {other}
|
| 437 |
+
|
| 438 |
+
Data: {row_data}
|
| 439 |
"""
|
| 440 |
+
return [[None, context]]
|
| 441 |
|
| 442 |
|
| 443 |
async def run_generate_dataset(num_workers, num_generations, output_file_path):
|
|
|
|
| 456 |
|
| 457 |
return f"Generated {num_generations} entries and saved to {output_file_path}", "\n".join(generated_data[:5]) + "\n..."
|
| 458 |
|
| 459 |
+
def add_topic_row(data):
|
| 460 |
+
if isinstance(data, pd.DataFrame):
|
| 461 |
+
return pd.concat([data, pd.DataFrame({"Topic": ["New Topic"]})], ignore_index=True)
|
| 462 |
+
else:
|
| 463 |
+
return data + [["New Topic"]]
|
| 464 |
+
|
| 465 |
+
def remove_last_topic_row(data):
|
| 466 |
+
return data[:-1] if len(data) > 1 else data
|
| 467 |
+
|
| 468 |
+
def edit_all_topics_func(topics):
|
| 469 |
+
topics_list = [topic[0] for topic in topics]
|
| 470 |
+
jsonl_rows = "\n".join([json.dumps({"topic": topic}) for topic in topics_list])
|
| 471 |
+
return (
|
| 472 |
+
gr.update(visible=False),
|
| 473 |
+
gr.update(value=jsonl_rows, visible=True),
|
| 474 |
+
gr.update(visible=True)
|
| 475 |
+
)
|
| 476 |
+
|
| 477 |
+
def update_topics_from_text(text):
|
| 478 |
+
try:
|
| 479 |
+
# Try parsing as JSONL
|
| 480 |
+
topics_list = [json.loads(line)["topic"] for line in text.split("\n") if line.strip()]
|
| 481 |
+
except json.JSONDecodeError:
|
| 482 |
+
# If parsing fails, treat as plain text
|
| 483 |
+
topics_list = [topic.strip() for topic in text.split("\n") if topic.strip()]
|
| 484 |
+
|
| 485 |
+
return gr.Dataframe.update(value=[[topic] for topic in topics_list], visible=True), gr.TextArea.update(visible=False)
|
| 486 |
+
|
| 487 |
+
def update_topics_from_text(text):
|
| 488 |
+
try:
|
| 489 |
+
# Try parsing as JSONL
|
| 490 |
+
topics_list = [json.loads(line)["topic"] for line in text.split("\n") if line.strip()]
|
| 491 |
+
except json.JSONDecodeError:
|
| 492 |
+
# If parsing fails, treat as plain text
|
| 493 |
+
topics_list = [topic.strip() for topic in text.split("\n") if topic.strip()]
|
| 494 |
+
|
| 495 |
+
return gr.Dataframe.update(value=[[topic] for topic in topics_list], visible=True), gr.TextArea.update(visible=False)
|
| 496 |
+
|
| 497 |
+
css = """
|
| 498 |
+
body, #root {
|
| 499 |
+
margin: 0;
|
| 500 |
+
padding: 0;
|
| 501 |
+
width: 100%;
|
| 502 |
+
height: 100%;
|
| 503 |
+
overflow-x: hidden;
|
| 504 |
+
}
|
| 505 |
+
.gradio-container {
|
| 506 |
+
max-width: 100% !important;
|
| 507 |
+
width: 100% !important;
|
| 508 |
+
margin: 0 auto !important;
|
| 509 |
+
padding: 0 !important;
|
| 510 |
+
}
|
| 511 |
+
.message-row {
|
| 512 |
+
justify-content: space-evenly !important;
|
| 513 |
+
}
|
| 514 |
+
.message-bubble-border {
|
| 515 |
+
border-radius: 6px !important;
|
| 516 |
+
}
|
| 517 |
+
.message-buttons-bot, .message-buttons-user {
|
| 518 |
+
right: 10px !important;
|
| 519 |
+
left: auto !important;
|
| 520 |
+
bottom: 2px !important;
|
| 521 |
+
}
|
| 522 |
+
.dark.message-bubble-border {
|
| 523 |
+
border-color: #343140 !important;
|
| 524 |
+
}
|
| 525 |
+
.dark.user {
|
| 526 |
+
background: #1e1c26 !important;
|
| 527 |
+
}
|
| 528 |
+
.dark.assistant.dark, .dark.pending.dark {
|
| 529 |
+
background: #16141c !important;
|
| 530 |
+
}
|
| 531 |
+
.tab-nav {
|
| 532 |
+
border-bottom: 2px solid #e0e0e0 !important;
|
| 533 |
+
}
|
| 534 |
+
.tab-nav button {
|
| 535 |
+
font-size: 16px !important;
|
| 536 |
+
padding: 10px 20px !important;
|
| 537 |
+
}
|
| 538 |
+
.input-row {
|
| 539 |
+
margin-bottom: 20px !important;
|
| 540 |
+
}
|
| 541 |
+
.button-row {
|
| 542 |
+
display: flex !important;
|
| 543 |
+
justify-content: space-between !important;
|
| 544 |
+
margin-top: 20px !important;
|
| 545 |
+
}
|
| 546 |
+
#row-editor {
|
| 547 |
+
height: 80vh !important;
|
| 548 |
+
font-size: 16px !important;
|
| 549 |
+
}
|
| 550 |
+
|
| 551 |
+
.file-upload-row {
|
| 552 |
+
height: 50px !important;
|
| 553 |
+
margin-bottom: 1rem !important;
|
| 554 |
+
}
|
| 555 |
+
|
| 556 |
+
.file-upload-row > .gr-column {
|
| 557 |
+
min-width: 0 !important;
|
| 558 |
+
}
|
| 559 |
+
|
| 560 |
+
.compact-file-upload {
|
| 561 |
+
height: 50px !important;
|
| 562 |
+
overflow: hidden !important;
|
| 563 |
+
}
|
| 564 |
+
|
| 565 |
+
.compact-file-upload > .file-preview {
|
| 566 |
+
min-height: 0 !important;
|
| 567 |
+
max-height: 50px !important;
|
| 568 |
+
padding: 0 !important;
|
| 569 |
+
}
|
| 570 |
+
|
| 571 |
+
.compact-file-upload > .file-preview > .file-preview-handler {
|
| 572 |
+
height: 50px !important;
|
| 573 |
+
padding: 0 8px !important;
|
| 574 |
+
display: flex !important;
|
| 575 |
+
align-items: center !important;
|
| 576 |
+
}
|
| 577 |
+
|
| 578 |
+
.compact-file-upload > .file-preview > .file-preview-handler > .file-preview-title {
|
| 579 |
+
white-space: nowrap !important;
|
| 580 |
+
overflow: hidden !important;
|
| 581 |
+
text-overflow: ellipsis !important;
|
| 582 |
+
flex: 1 !important;
|
| 583 |
+
}
|
| 584 |
+
|
| 585 |
+
.compact-file-upload > .file-preview > .file-preview-handler > .file-preview-remove {
|
| 586 |
+
padding: 0 !important;
|
| 587 |
+
min-width: 24px !important;
|
| 588 |
+
width: 24px !important;
|
| 589 |
+
height: 24px !important;
|
| 590 |
+
}
|
| 591 |
+
|
| 592 |
+
.compact-button {
|
| 593 |
+
height: 50px !important;
|
| 594 |
+
min-height: 40px !important;
|
| 595 |
+
width: 100% !important;
|
| 596 |
+
}
|
| 597 |
+
|
| 598 |
+
.compact-file-upload > label {
|
| 599 |
+
height: 50px !important;
|
| 600 |
+
padding: 0 8px !important;
|
| 601 |
+
display: flex !important;
|
| 602 |
+
align-items: center !important;
|
| 603 |
+
justify-content: left !important;
|
| 604 |
+
}
|
| 605 |
+
"""
|
| 606 |
+
|
| 607 |
+
demo = gr.Blocks(theme='Ama434/neutral-barlow', css=css)
|
| 608 |
|
| 609 |
with demo:
|
| 610 |
+
gr.Markdown("# Dataset Editor and Annotation Tool")
|
| 611 |
|
| 612 |
config = gr.State(load_annotation_config())
|
| 613 |
|
| 614 |
with gr.Row():
|
| 615 |
+
with gr.Column(min_width=1000):
|
| 616 |
with gr.Tab("Dataset Editor"):
|
| 617 |
+
with gr.Row(elem_classes="file-upload-row"):
|
| 618 |
+
with gr.Column(scale=3, min_width=400):
|
| 619 |
+
file_upload = gr.File(label="Upload Dataset File (.txt, .jsonl, or .csv)", elem_classes="compact-file-upload")
|
| 620 |
+
with gr.Column(scale=1, min_width=100):
|
| 621 |
+
load_button = gr.Button("Load Dataset", elem_classes="compact-button")
|
| 622 |
|
| 623 |
with gr.Row():
|
| 624 |
prev_button = gr.Button("← Previous")
|
| 625 |
+
row_index = gr.State(value=0)
|
| 626 |
+
total_rows = gr.State(value=0)
|
| 627 |
+
current_row_display = gr.Textbox(label="Current Row", interactive=False)
|
| 628 |
next_button = gr.Button("Next →")
|
| 629 |
|
| 630 |
with gr.Row():
|
| 631 |
with gr.Column(scale=3):
|
| 632 |
+
row_editor = gr.TextArea(label="Edit Row", lines=40)
|
| 633 |
|
| 634 |
with gr.Column(scale=2):
|
| 635 |
quality_label = gr.Radio(label="Relevance for Training", choices=[])
|
| 636 |
tag_components = [gr.CheckboxGroup(label=f"Tag Group {i+1}", choices=[]) for i in range(3)]
|
| 637 |
other_description = gr.Textbox(label="Additional annotations", lines=3)
|
| 638 |
+
|
| 639 |
+
# Add the AI Assistant as a dropdown
|
| 640 |
+
with gr.Accordion("AI Assistant", open=False):
|
| 641 |
+
chatbot = gr.Chatbot(height=300)
|
| 642 |
+
msg = gr.Textbox(label="Chat with AI Assistant")
|
| 643 |
+
clear = gr.Button("Clear")
|
| 644 |
|
| 645 |
with gr.Row():
|
| 646 |
to_markdown_button = gr.Button("Convert to Markdown")
|
|
|
|
| 653 |
|
| 654 |
with gr.Tab("Annotation Configuration"):
|
| 655 |
with gr.Row():
|
| 656 |
+
with gr.Column(scale=1):
|
| 657 |
+
gr.Markdown("### Quality Scale")
|
| 658 |
+
quality_scale_name = gr.Textbox(label="Scale Name")
|
| 659 |
+
quality_scale_description = gr.Textbox(label="Scale Description", lines=2)
|
| 660 |
+
|
| 661 |
+
with gr.Column(scale=2):
|
| 662 |
quality_scale = gr.Dataframe(
|
| 663 |
headers=["Value", "Label"],
|
| 664 |
datatype=["str", "str"],
|
| 665 |
+
label="Quality Scale Options",
|
| 666 |
+
interactive=True,
|
| 667 |
+
col_count=(2, "fixed"),
|
| 668 |
+
row_count=(5, "dynamic"),
|
| 669 |
+
height=400,
|
| 670 |
+
wrap=True
|
| 671 |
)
|
| 672 |
|
| 673 |
+
gr.Markdown("### Tag Categories")
|
| 674 |
+
tag_categories = gr.Dataframe(
|
| 675 |
+
headers=["Name", "Type", "Tags"],
|
| 676 |
+
datatype=["str", "str", "str"],
|
| 677 |
+
label="Tag Categories",
|
| 678 |
+
interactive=True,
|
| 679 |
+
col_count=(3, "fixed"),
|
| 680 |
+
row_count=(3, "dynamic"),
|
| 681 |
+
height=250,
|
| 682 |
+
wrap=True
|
| 683 |
+
)
|
| 684 |
+
|
| 685 |
with gr.Row():
|
| 686 |
+
add_tag_category = gr.Button("Add Category")
|
| 687 |
+
remove_tag_category = gr.Button("Remove Last Category")
|
| 688 |
+
|
| 689 |
+
gr.Markdown("### Free Text Fields")
|
| 690 |
+
free_text_fields = gr.Dataframe(
|
| 691 |
+
headers=["Name", "Description"],
|
| 692 |
+
datatype=["str", "str"],
|
| 693 |
+
label="Free Text Fields",
|
| 694 |
+
interactive=True,
|
| 695 |
+
col_count=(2, "fixed"),
|
| 696 |
+
row_count=(2, "dynamic"),
|
| 697 |
+
height=300,
|
| 698 |
+
wrap=True
|
| 699 |
+
)
|
| 700 |
|
| 701 |
with gr.Row():
|
| 702 |
+
add_free_text_field = gr.Button("Add Field")
|
| 703 |
+
remove_free_text_field = gr.Button("Remove Last Field")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 704 |
|
| 705 |
+
|
| 706 |
+
with gr.Row():
|
| 707 |
+
save_config_btn = gr.Button("Save Configuration", variant="primary")
|
| 708 |
+
config_status = gr.Textbox(label="Status", interactive=False)
|
| 709 |
|
| 710 |
with gr.Tab("Dataset Configuration"):
|
| 711 |
with gr.Row():
|
| 712 |
+
vodalus_system_message = gr.TextArea(label="System Message for JSONL Dataset", lines=10)
|
| 713 |
+
prompt_1 = gr.TextArea(label="Dataset Gerenation Prompt", lines=10)
|
| 714 |
+
|
| 715 |
+
gr.Markdown("### Topics")
|
| 716 |
+
with gr.Row():
|
| 717 |
+
with gr.Column(scale=2):
|
| 718 |
+
topics = gr.Dataframe(
|
| 719 |
+
headers=["Topic"],
|
| 720 |
+
datatype=["str"],
|
| 721 |
+
label="Topics",
|
| 722 |
+
interactive=True,
|
| 723 |
+
col_count=(1, "fixed"),
|
| 724 |
+
row_count=(5, "dynamic"),
|
| 725 |
+
height=200,
|
| 726 |
+
wrap=True
|
| 727 |
+
)
|
| 728 |
+
|
| 729 |
+
with gr.Column(scale=1):
|
| 730 |
+
with gr.Row():
|
| 731 |
+
add_topic = gr.Button("Add Topic")
|
| 732 |
+
remove_topic = gr.Button("Remove Last Topic")
|
| 733 |
+
edit_all_topics = gr.Button("Edit All Topics")
|
| 734 |
+
all_topics_edit = gr.TextArea(label="Edit All Topics (JSONL or Plain Text)", visible=False, lines=10)
|
| 735 |
+
format_info = gr.Markdown("""
|
| 736 |
+
Enter topics as JSONL (e.g., {"topic": "Example Topic"}) or plain text (one topic per line).
|
| 737 |
+
JSONL format allows for additional metadata if needed.
|
| 738 |
+
""", visible=False)
|
| 739 |
|
| 740 |
with gr.Row():
|
| 741 |
+
save_dataset_config_btn = gr.Button("Save Dataset Configuration", variant="primary")
|
| 742 |
+
dataset_config_status = gr.Textbox(label="Status")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 743 |
|
|
|
|
|
|
|
| 744 |
|
| 745 |
with gr.Tab("Dataset Generation"):
|
| 746 |
with gr.Row():
|
|
|
|
| 754 |
generation_status = gr.Textbox(label="Generation Status")
|
| 755 |
generation_output = gr.TextArea(label="Generation Output", lines=10)
|
| 756 |
|
| 757 |
+
with gr.Tab("LLM Configuration"):
|
| 758 |
+
with gr.Row():
|
| 759 |
+
provider = gr.Dropdown(choices=["local-model", "anything-llm"], label="LLM Provider")
|
| 760 |
+
base_url = gr.Textbox(label="Base URL (for local model)")
|
| 761 |
+
with gr.Row():
|
| 762 |
+
workspace = gr.Textbox(label="Workspace (for AnythingLLM)")
|
| 763 |
+
api_key = gr.Textbox(label="API Key (for AnythingLLM)")
|
| 764 |
+
|
| 765 |
+
with gr.Accordion("Advanced Options", open=False):
|
| 766 |
+
with gr.Row():
|
| 767 |
+
max_tokens = gr.Slider(minimum=100, maximum=4096, value=2048, step=1, label="Max Tokens")
|
| 768 |
+
temperature = gr.Slider(minimum=0, maximum=1, value=0.7, step=0.01, label="Temperature")
|
| 769 |
+
with gr.Row():
|
| 770 |
+
top_p = gr.Slider(minimum=0, maximum=1, value=0.9, step=0.01, label="Top P")
|
| 771 |
+
frequency_penalty = gr.Slider(minimum=0, maximum=2, value=0.0, step=0.01, label="Frequency Penalty")
|
| 772 |
+
presence_penalty = gr.Slider(minimum=0, maximum=2, value=0.0, step=0.01, label="Presence Penalty")
|
| 773 |
+
|
| 774 |
+
save_llm_config_btn = gr.Button("Save LLM Configuration")
|
| 775 |
+
llm_config_status = gr.Textbox(label="Status")
|
| 776 |
+
|
| 777 |
+
add_topic.click(
|
| 778 |
+
lambda x: x + [["New Topic"]],
|
| 779 |
+
inputs=[topics],
|
| 780 |
+
outputs=[topics]
|
| 781 |
+
)
|
| 782 |
+
|
| 783 |
+
remove_topic.click(
|
| 784 |
+
lambda x: x[:-1] if len(x) > 0 else x,
|
| 785 |
+
inputs=[topics],
|
| 786 |
+
outputs=[topics]
|
| 787 |
+
)
|
| 788 |
+
|
| 789 |
+
edit_all_topics.click(
|
| 790 |
+
edit_all_topics_func,
|
| 791 |
+
inputs=[topics],
|
| 792 |
+
outputs=[topics, all_topics_edit, format_info]
|
| 793 |
+
)
|
| 794 |
+
|
| 795 |
+
all_topics_edit.submit(
|
| 796 |
+
update_topics_from_text,
|
| 797 |
+
inputs=[all_topics_edit],
|
| 798 |
+
outputs=[topics, all_topics_edit, format_info]
|
| 799 |
+
)
|
| 800 |
|
| 801 |
load_button.click(
|
| 802 |
+
load_dataset,
|
| 803 |
+
inputs=[file_upload],
|
| 804 |
+
outputs=[row_editor, row_index, total_rows, current_row_display, quality_label, *tag_components, other_description]
|
| 805 |
).then(
|
| 806 |
update_annotation_ui,
|
| 807 |
inputs=[config],
|
|
|
|
| 810 |
|
| 811 |
prev_button.click(
|
| 812 |
navigate_rows,
|
| 813 |
+
inputs=[file_upload, row_index, gr.State("prev"), config],
|
| 814 |
+
outputs=[row_editor, row_index, total_rows, current_row_display, quality_label, *tag_components, other_description]
|
| 815 |
).then(
|
| 816 |
update_annotation_ui,
|
| 817 |
inputs=[config],
|
|
|
|
| 820 |
|
| 821 |
next_button.click(
|
| 822 |
navigate_rows,
|
| 823 |
+
inputs=[file_upload, row_index, gr.State("next"), config],
|
| 824 |
+
outputs=[row_editor, row_index, total_rows, current_row_display, quality_label, *tag_components, other_description]
|
| 825 |
).then(
|
| 826 |
update_annotation_ui,
|
| 827 |
inputs=[config],
|
|
|
|
| 830 |
|
| 831 |
save_row_button.click(
|
| 832 |
save_row_with_metadata,
|
| 833 |
+
inputs=[file_upload, row_index, row_editor, config, quality_label,
|
| 834 |
tag_components[0], tag_components[1], tag_components[2], other_description],
|
| 835 |
outputs=[editor_status]
|
| 836 |
).then(
|
|
|
|
| 862 |
|
| 863 |
save_config_btn.click(
|
| 864 |
save_config_from_ui,
|
| 865 |
+
inputs=[quality_scale_name, quality_scale_description, quality_scale, tag_categories, free_text_fields, topics, all_topics_edit],
|
| 866 |
outputs=[config_status, config]
|
| 867 |
).then(
|
| 868 |
update_annotation_ui,
|
|
|
|
| 878 |
|
| 879 |
demo.load(
|
| 880 |
load_dataset_config,
|
| 881 |
+
outputs=[vodalus_system_message, prompt_1, topics, max_tokens, temperature, top_p, frequency_penalty, presence_penalty]
|
| 882 |
)
|
| 883 |
|
| 884 |
save_dataset_config_btn.click(
|
| 885 |
save_dataset_config,
|
| 886 |
+
inputs=[vodalus_system_message, prompt_1, topics, max_tokens, temperature, top_p, frequency_penalty, presence_penalty],
|
| 887 |
outputs=[dataset_config_status]
|
| 888 |
)
|
| 889 |
|
|
|
|
| 893 |
outputs=[generation_status, generation_output]
|
| 894 |
)
|
| 895 |
|
| 896 |
+
demo.load(
|
| 897 |
+
load_llm_config,
|
| 898 |
+
outputs=[provider, base_url, workspace, api_key, max_tokens, temperature, top_p, frequency_penalty, presence_penalty]
|
| 899 |
+
)
|
| 900 |
+
|
| 901 |
+
save_llm_config_btn.click(
|
| 902 |
+
save_llm_config,
|
| 903 |
+
inputs=[provider, base_url, workspace, api_key, max_tokens, temperature, top_p, frequency_penalty, presence_penalty],
|
| 904 |
+
outputs=[llm_config_status]
|
| 905 |
+
)
|
| 906 |
+
|
| 907 |
msg.submit(chat_with_llm, [msg, chatbot], [chatbot])
|
| 908 |
clear.click(lambda: None, None, chatbot, queue=False)
|
| 909 |
|
| 910 |
+
|
| 911 |
for button in [load_button, prev_button, next_button]:
|
| 912 |
button.click(
|
| 913 |
update_chat_context,
|
|
|
|
| 915 |
outputs=[chatbot]
|
| 916 |
)
|
| 917 |
|
|
|
|
|
|
|
| 918 |
|
| 919 |
+
demo.load(
|
| 920 |
+
lambda: (
|
| 921 |
+
initial_values := load_dataset_config(),
|
| 922 |
+
gr.update(value=initial_values[0]), # vodalus_system_message
|
| 923 |
+
gr.update(value=initial_values[1]), # prompt_1
|
| 924 |
+
gr.update(value=initial_values[2]), # topics_data
|
| 925 |
+
gr.update(value=initial_values[3]), # max_tokens_val
|
| 926 |
+
gr.update(value=initial_values[4]), # temperature_val
|
| 927 |
+
gr.update(value=initial_values[5]), # top_p_val
|
| 928 |
+
gr.update(value=initial_values[6]), # frequency_penalty_val
|
| 929 |
+
gr.update(value=initial_values[7]) # presence_penalty_val
|
| 930 |
+
)[1:], # We return a tuple slice to exclude the initial_values assignment
|
| 931 |
+
outputs=[
|
| 932 |
+
vodalus_system_message,
|
| 933 |
+
prompt_1,
|
| 934 |
+
topics,
|
| 935 |
+
max_tokens,
|
| 936 |
+
temperature,
|
| 937 |
+
top_p,
|
| 938 |
+
frequency_penalty,
|
| 939 |
+
presence_penalty
|
| 940 |
+
]
|
| 941 |
+
)
|
| 942 |
+
|
| 943 |
+
if __name__ == "__main__":
|
| 944 |
+
demo.launch(share=True)
|