Alexander Watson
commited on
Commit
•
06594f2
1
Parent(s):
fb73a92
update interfaces
Browse files
app.py
CHANGED
@@ -11,7 +11,11 @@ import requests
|
|
11 |
import streamlit as st
|
12 |
from datasets import load_dataset
|
13 |
from gretel_client import Gretel
|
14 |
-
from navigator_helpers import
|
|
|
|
|
|
|
|
|
15 |
|
16 |
# Create a StringIO buffer to capture the logging output
|
17 |
log_buffer = StringIO()
|
@@ -103,6 +107,9 @@ def main():
|
|
103 |
)
|
104 |
if "gretel" not in st.session_state:
|
105 |
st.session_state.gretel = None
|
|
|
|
|
|
|
106 |
if st.button("Validate API Key"):
|
107 |
if api_key:
|
108 |
try:
|
@@ -340,6 +347,22 @@ def main():
|
|
340 |
st.markdown("---")
|
341 |
st.markdown("### Format Prompts")
|
342 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
343 |
instruction_format_prompt = st.text_area(
|
344 |
"Instruction Format Prompt",
|
345 |
value=st.session_state.get(
|
@@ -365,58 +388,53 @@ def main():
|
|
365 |
st.write(
|
366 |
"Get started with your current configuration using the SDK code below:"
|
367 |
)
|
|
|
368 |
|
369 |
-
|
370 |
-
|
371 |
-
|
372 |
-
|
373 |
-
|
374 |
-
|
375 |
-
|
376 |
-
|
377 |
-
|
378 |
-
|
379 |
-
|
380 |
-
|
381 |
-
|
382 |
-
|
383 |
-
|
384 |
-
|
385 |
-
|
386 |
-
|
387 |
-
|
388 |
-
|
389 |
-
|
390 |
-
|
391 |
-
|
392 |
-
|
393 |
-
|
394 |
-
|
395 |
-
|
396 |
-
|
397 |
-
|
398 |
-
|
399 |
-
|
400 |
-
|
401 |
-
|
402 |
-
|
403 |
-
|
404 |
-
|
405 |
-
|
406 |
-
|
407 |
-
|
408 |
-
|
409 |
-
use_aaa={use_aaa},
|
410 |
-
output_file="results.csv",
|
411 |
-
verbose=True,
|
412 |
-
)
|
413 |
-
new_df = augmenter.augment()
|
414 |
-
"""
|
415 |
st.code(config_text, language="python")
|
416 |
st.download_button(
|
417 |
label="Download SDK Code",
|
418 |
data=config_text,
|
419 |
-
file_name="
|
420 |
mime="text/plain",
|
421 |
)
|
422 |
|
@@ -431,20 +449,20 @@ def main():
|
|
431 |
if "logs" not in st.session_state:
|
432 |
st.session_state.logs = []
|
433 |
|
434 |
-
if "
|
435 |
-
st.session_state.
|
436 |
|
437 |
if start_button:
|
438 |
-
# Clear the
|
439 |
-
st.session_state.
|
440 |
st.session_state.logs = []
|
441 |
|
442 |
with st.expander("Synthetic Data", expanded=True):
|
443 |
st.subheader("Synthetic Data Generation")
|
444 |
progress_bar = st.progress(0)
|
445 |
-
tab1, tab2 = st.tabs(["
|
446 |
with tab1:
|
447 |
-
|
448 |
st.info(
|
449 |
"Click on the 'Logs' tab to see and debug real-time logging for each record as it is generated by the agents."
|
450 |
)
|
@@ -467,7 +485,7 @@ def main():
|
|
467 |
handler = StreamlitLogHandler(custom_log_handler)
|
468 |
logger.addHandler(handler)
|
469 |
|
470 |
-
config =
|
471 |
input_fields=selected_fields,
|
472 |
output_instruction_field=output_instruction_field,
|
473 |
output_response_field=output_response_field,
|
@@ -480,26 +498,28 @@ def main():
|
|
480 |
navigator_tabular=navigator_tabular,
|
481 |
navigator_llm=navigator_llm,
|
482 |
co_teach_llms=co_teach_llms,
|
|
|
483 |
instruction_format_prompt=instruction_format_prompt,
|
484 |
response_format_prompt=response_format_prompt,
|
485 |
)
|
|
|
486 |
start_time = time.time()
|
487 |
with st.spinner("Generating synthetic data..."):
|
488 |
for index in range(num_records):
|
489 |
row = df.iloc[index]
|
490 |
-
|
491 |
pd.DataFrame([row]),
|
492 |
config,
|
493 |
use_aaa=use_aaa,
|
494 |
output_file="results.csv",
|
495 |
verbose=True,
|
496 |
)
|
497 |
-
new_df =
|
498 |
-
st.session_state.
|
499 |
-
|
500 |
-
|
501 |
pd.concat(
|
502 |
-
st.session_state.
|
503 |
)
|
504 |
)
|
505 |
progress = (index + 1) / num_records
|
@@ -520,11 +540,11 @@ def main():
|
|
520 |
|
521 |
time.sleep(0.1)
|
522 |
logger.removeHandler(handler)
|
523 |
-
st.success("Data
|
524 |
st.stop()
|
525 |
|
526 |
if stop_button:
|
527 |
-
st.warning("
|
528 |
|
529 |
# Get the complete logs from the session state
|
530 |
complete_logs = st.session_state.logs
|
@@ -532,22 +552,22 @@ def main():
|
|
532 |
# Convert complete logs to JSONL format
|
533 |
log_jsonl = "\n".join([json.dumps({"log": log}) for log in complete_logs])
|
534 |
|
535 |
-
# Convert
|
536 |
-
if st.session_state.
|
537 |
-
|
538 |
-
st.session_state.
|
539 |
)
|
540 |
-
if not
|
541 |
-
|
542 |
[
|
543 |
json.dumps(row.to_dict())
|
544 |
-
for _, row in
|
545 |
]
|
546 |
)
|
547 |
else:
|
548 |
-
|
549 |
else:
|
550 |
-
|
551 |
|
552 |
# Create a temporary directory to store the files
|
553 |
with tempfile.TemporaryDirectory() as temp_dir:
|
@@ -556,26 +576,28 @@ def main():
|
|
556 |
with open(log_file_path, "w") as log_file:
|
557 |
log_file.write(log_jsonl)
|
558 |
|
559 |
-
# Write the
|
560 |
-
if
|
561 |
-
|
562 |
temp_dir, "synthetic_data.jsonl"
|
563 |
)
|
564 |
-
with open(
|
565 |
-
|
566 |
|
567 |
# Write the SDK code to a file
|
568 |
-
sdk_file_path = os.path.join(temp_dir, "
|
569 |
with open(sdk_file_path, "w") as sdk_file:
|
570 |
sdk_file.write(config_text)
|
571 |
|
572 |
-
# Create a ZIP file containing the logs,
|
573 |
-
zip_file_path = os.path.join(temp_dir, "
|
574 |
with zipfile.ZipFile(zip_file_path, "w") as zip_file:
|
575 |
zip_file.write(log_file_path, "complete_logs.jsonl")
|
576 |
-
if
|
577 |
-
zip_file.write(
|
578 |
-
|
|
|
|
|
579 |
|
580 |
# Download the ZIP file
|
581 |
with open(zip_file_path, "rb") as zip_file:
|
|
|
11 |
import streamlit as st
|
12 |
from datasets import load_dataset
|
13 |
from gretel_client import Gretel
|
14 |
+
from navigator_helpers import (
|
15 |
+
DataSynthesisConfig,
|
16 |
+
TrainingDataSynthesizer,
|
17 |
+
StreamlitLogHandler,
|
18 |
+
)
|
19 |
|
20 |
# Create a StringIO buffer to capture the logging output
|
21 |
log_buffer = StringIO()
|
|
|
107 |
)
|
108 |
if "gretel" not in st.session_state:
|
109 |
st.session_state.gretel = None
|
110 |
+
if "synthesized_data" not in st.session_state:
|
111 |
+
st.session_state.synthesized_data = []
|
112 |
+
|
113 |
if st.button("Validate API Key"):
|
114 |
if api_key:
|
115 |
try:
|
|
|
347 |
st.markdown("---")
|
348 |
st.markdown("### Format Prompts")
|
349 |
|
350 |
+
st.markdown("---")
|
351 |
+
st.markdown("### Format Prompts")
|
352 |
+
|
353 |
+
system_prompt = st.text_area(
|
354 |
+
"System Prompt",
|
355 |
+
value=st.session_state.get(
|
356 |
+
"system_prompt",
|
357 |
+
"You are an AI assistant tasked with generating high-quality instruction-response pairs.\n"
|
358 |
+
"Your goal is to create diverse, engaging, and informative content that covers a wide range of topics.\n"
|
359 |
+
"When generating instructions, aim for clear, concise questions or commands that prompt thoughtful responses.\n"
|
360 |
+
"When generating responses, provide detailed, accurate, and helpful information that directly addresses the instruction.",
|
361 |
+
),
|
362 |
+
help="Specify the system prompt for the LLM",
|
363 |
+
)
|
364 |
+
st.session_state.system_prompt = system_prompt
|
365 |
+
|
366 |
instruction_format_prompt = st.text_area(
|
367 |
"Instruction Format Prompt",
|
368 |
value=st.session_state.get(
|
|
|
388 |
st.write(
|
389 |
"Get started with your current configuration using the SDK code below:"
|
390 |
)
|
391 |
+
config_text = f"""#!pip install -Uqq git+https://github.com/gretelai/navigator-helpers.git
|
392 |
|
393 |
+
import logging
|
394 |
+
import pandas as pd
|
395 |
+
from navigator_helpers import DataSynthesisConfig, TrainingDataSynthesizer
|
396 |
+
|
397 |
+
# Configure the logger
|
398 |
+
logging.basicConfig(level=logging.INFO, format="%(message)s")
|
399 |
+
|
400 |
+
DATASET = "YOUR_DATASET"
|
401 |
+
API_KEY = "YOUR_API_KEY"
|
402 |
+
|
403 |
+
df = pd.read_csv(DATASET)
|
404 |
+
|
405 |
+
# Create the data synthesis configuration
|
406 |
+
config = DataSynthesisConfig(
|
407 |
+
input_fields={st.session_state.selected_fields},
|
408 |
+
output_instruction_field="{output_instruction_field}",
|
409 |
+
output_response_field="{output_response_field}",
|
410 |
+
num_instructions={num_instructions},
|
411 |
+
num_responses={num_responses},
|
412 |
+
temperature={temperature},
|
413 |
+
max_tokens_instruction={max_tokens_instruction},
|
414 |
+
max_tokens_response={max_tokens_response},
|
415 |
+
api_key=API_KEY,
|
416 |
+
navigator_tabular="{navigator_tabular}",
|
417 |
+
navigator_llm="{navigator_llm}",
|
418 |
+
co_teach_llms={co_teach_llms},
|
419 |
+
system_prompt='''{system_prompt}''',
|
420 |
+
instruction_format_prompt='''{instruction_format_prompt}''',
|
421 |
+
response_format_prompt='''{response_format_prompt}'''
|
422 |
+
)
|
423 |
+
|
424 |
+
# Create the training data synthesizer and perform synthesis
|
425 |
+
synthesizer = TrainingDataSynthesizer(
|
426 |
+
df,
|
427 |
+
config,
|
428 |
+
use_aaa={use_aaa},
|
429 |
+
output_file="results.csv",
|
430 |
+
verbose=True,
|
431 |
+
)
|
432 |
+
new_df = synthesizer.generate()"""
|
|
|
|
|
|
|
|
|
|
|
|
|
433 |
st.code(config_text, language="python")
|
434 |
st.download_button(
|
435 |
label="Download SDK Code",
|
436 |
data=config_text,
|
437 |
+
file_name="data_synthesis_code.py",
|
438 |
mime="text/plain",
|
439 |
)
|
440 |
|
|
|
449 |
if "logs" not in st.session_state:
|
450 |
st.session_state.logs = []
|
451 |
|
452 |
+
if "synthetic_data" not in st.session_state:
|
453 |
+
st.session_state.synthetic_data = []
|
454 |
|
455 |
if start_button:
|
456 |
+
# Clear the synthetic data and logs before starting a new generation
|
457 |
+
st.session_state.synthetic_data = []
|
458 |
st.session_state.logs = []
|
459 |
|
460 |
with st.expander("Synthetic Data", expanded=True):
|
461 |
st.subheader("Synthetic Data Generation")
|
462 |
progress_bar = st.progress(0)
|
463 |
+
tab1, tab2 = st.tabs(["synthetic Data", "Logs"])
|
464 |
with tab1:
|
465 |
+
synthetic_data_placeholder = st.empty()
|
466 |
st.info(
|
467 |
"Click on the 'Logs' tab to see and debug real-time logging for each record as it is generated by the agents."
|
468 |
)
|
|
|
485 |
handler = StreamlitLogHandler(custom_log_handler)
|
486 |
logger.addHandler(handler)
|
487 |
|
488 |
+
config = DataSynthesisConfig(
|
489 |
input_fields=selected_fields,
|
490 |
output_instruction_field=output_instruction_field,
|
491 |
output_response_field=output_response_field,
|
|
|
498 |
navigator_tabular=navigator_tabular,
|
499 |
navigator_llm=navigator_llm,
|
500 |
co_teach_llms=co_teach_llms,
|
501 |
+
system_prompt=system_prompt,
|
502 |
instruction_format_prompt=instruction_format_prompt,
|
503 |
response_format_prompt=response_format_prompt,
|
504 |
)
|
505 |
+
|
506 |
start_time = time.time()
|
507 |
with st.spinner("Generating synthetic data..."):
|
508 |
for index in range(num_records):
|
509 |
row = df.iloc[index]
|
510 |
+
synthesizer = TrainingDataSynthesizer(
|
511 |
pd.DataFrame([row]),
|
512 |
config,
|
513 |
use_aaa=use_aaa,
|
514 |
output_file="results.csv",
|
515 |
verbose=True,
|
516 |
)
|
517 |
+
new_df = synthesizer.generate()
|
518 |
+
st.session_state.synthetic_data.append(new_df)
|
519 |
+
synthetic_data_placeholder.subheader("Synthetic Data")
|
520 |
+
synthetic_data_placeholder.dataframe(
|
521 |
pd.concat(
|
522 |
+
st.session_state.synthetic_data, ignore_index=True
|
523 |
)
|
524 |
)
|
525 |
progress = (index + 1) / num_records
|
|
|
540 |
|
541 |
time.sleep(0.1)
|
542 |
logger.removeHandler(handler)
|
543 |
+
st.success("Data synthetic completed!")
|
544 |
st.stop()
|
545 |
|
546 |
if stop_button:
|
547 |
+
st.warning("Synthesis stopped by the user.")
|
548 |
|
549 |
# Get the complete logs from the session state
|
550 |
complete_logs = st.session_state.logs
|
|
|
552 |
# Convert complete logs to JSONL format
|
553 |
log_jsonl = "\n".join([json.dumps({"log": log}) for log in complete_logs])
|
554 |
|
555 |
+
# Convert synthesized data to JSONL format if it exists
|
556 |
+
if st.session_state.synthesized_data:
|
557 |
+
synthesized_df = pd.concat(
|
558 |
+
st.session_state.synthesized_data, ignore_index=True
|
559 |
)
|
560 |
+
if not synthesized_df.empty:
|
561 |
+
synthesized_data_jsonl = "\n".join(
|
562 |
[
|
563 |
json.dumps(row.to_dict())
|
564 |
+
for _, row in synthesized_df.iterrows()
|
565 |
]
|
566 |
)
|
567 |
else:
|
568 |
+
synthesized_data_jsonl = None
|
569 |
else:
|
570 |
+
synthesized_data_jsonl = None
|
571 |
|
572 |
# Create a temporary directory to store the files
|
573 |
with tempfile.TemporaryDirectory() as temp_dir:
|
|
|
576 |
with open(log_file_path, "w") as log_file:
|
577 |
log_file.write(log_jsonl)
|
578 |
|
579 |
+
# Write the synthesized data to a file if it exists
|
580 |
+
if synthesized_data_jsonl:
|
581 |
+
synthesized_data_file_path = os.path.join(
|
582 |
temp_dir, "synthetic_data.jsonl"
|
583 |
)
|
584 |
+
with open(synthesized_data_file_path, "w") as synthesized_data_file:
|
585 |
+
synthesized_data_file.write(synthesized_data_jsonl)
|
586 |
|
587 |
# Write the SDK code to a file
|
588 |
+
sdk_file_path = os.path.join(temp_dir, "data_synthesis_code.py")
|
589 |
with open(sdk_file_path, "w") as sdk_file:
|
590 |
sdk_file.write(config_text)
|
591 |
|
592 |
+
# Create a ZIP file containing the logs, synthesized data, and SDK code
|
593 |
+
zip_file_path = os.path.join(temp_dir, "synthesis_results.zip")
|
594 |
with zipfile.ZipFile(zip_file_path, "w") as zip_file:
|
595 |
zip_file.write(log_file_path, "complete_logs.jsonl")
|
596 |
+
if synthesized_data_jsonl:
|
597 |
+
zip_file.write(
|
598 |
+
synthesized_data_file_path, "synthesized_data.jsonl"
|
599 |
+
)
|
600 |
+
zip_file.write(sdk_file_path, "data_synthesis_code.py")
|
601 |
|
602 |
# Download the ZIP file
|
603 |
with open(zip_file_path, "rb") as zip_file:
|