Skip to content

Bug/136 preference dataset interruption #137

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
patrickfleith merged 2 commits into from
Nov 20, 2025
Merged

Bug/136 preference dataset interruption #137

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
897c127
Fixing num rows logged
patrickfleith Nov 20, 2025
c3e389d
Adding try/except to catch question, responses, or judge score genera…
patrickfleith Nov 20, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
105 changes: 62 additions & 43 deletions datafast/datasets.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1121,7 +1121,7 @@ def generate(self,
raise ValueError("input_documents must be provided in the configuration")

start_time = time.time()
expected_rows = self.get_num_expected_rows([question_gen_llm, chosen_response_gen_llm, rejected_response_gen_llm])
expected_rows = self.get_num_expected_rows([question_gen_llm])
logger.info(
f"Starting PreferenceDataset.generate() | "
f"Expected rows: {expected_rows}"
Expand All @@ -1135,28 +1135,40 @@ def generate(self,
# Process each input document
for doc in self.config.input_documents:
# Generate questions for each document
questions = self._generate_questions(doc, question_gen_llm, language_name)
try:
questions = self._generate_questions(doc, question_gen_llm, language_name)
except Exception as e:
logger.warning(
f"Failed to generate questions for document '{doc.get('title', 'unknown')}' | Error: {e}"
)
continue

# For each question, generate chosen and rejected responses
for question in questions:
# Track preference pair generation start time
pair_start_time = time.time()

# Generate chosen response
chosen_response = self._generate_chosen_response(
doc,
question,
chosen_response_gen_llm,
language_name
)
try:
# Track preference pair generation start time
pair_start_time = time.time()
# Generate chosen response
chosen_response = self._generate_chosen_response(
doc,
question,
chosen_response_gen_llm,
language_name
)

# Generate rejected response
rejected_response = self._generate_rejected_response(
doc,
question,
rejected_response_gen_llm,
language_name
)
# Generate rejected response
rejected_response = self._generate_rejected_response(
doc,
question,
rejected_response_gen_llm,
language_name
)

except Exception as e:
logger.warning(
f"Failed to generate responses for question '{question}' | Error: {e}"
)
continue

# If evolutionary instruction is enabled, refine the instruction and response
if self.config.evol_instruct and evolution_llm:
Expand All @@ -1181,31 +1193,38 @@ def generate(self,

# If LLM as judge is enabled, use the judge LLM to evaluate the preference pair
if self.config.llm_as_judge and judge_llm:
# Get judge scores for chosen response
chosen_response_result = self._judge_scoring(
doc, question, chosen_response, judge_llm
)
chosen_response_score = chosen_response_result.score
chosen_response_assessment = chosen_response_result.assessment
try:
# Get judge scores for chosen response
chosen_response_result = self._judge_scoring(
doc, question, chosen_response, judge_llm
)
chosen_response_score = chosen_response_result.score
chosen_response_assessment = chosen_response_result.assessment

# Get judge scores for rejected response
rejected_response_result = self._judge_scoring(
doc, question, rejected_response, judge_llm
)
rejected_response_score = rejected_response_result.score
rejected_response_assessment = rejected_response_result.assessment

# Swap chosen and rejected responses based on scores if needed
# This ensures the higher-scored response is always the chosen one
if rejected_response_score > chosen_response_score:
# Swap responses
chosen_response, rejected_response = rejected_response, chosen_response
# Swap scores
chosen_response_score, rejected_response_score = rejected_response_score, chosen_response_score
# Swap assessments
chosen_response_assessment, rejected_response_assessment = rejected_response_assessment, chosen_response_assessment
# Swap model IDs
chosen_model_id, rejected_model_id = rejected_model_id, chosen_model_id
# Get judge scores for rejected response
rejected_response_result = self._judge_scoring(
doc, question, rejected_response, judge_llm
)
rejected_response_score = rejected_response_result.score
rejected_response_assessment = rejected_response_result.assessment

# Swap chosen and rejected responses based on scores if needed
# This ensures the higher-scored response is always the chosen one
if rejected_response_score > chosen_response_score:
# Swap responses
chosen_response, rejected_response = rejected_response, chosen_response
# Swap scores
chosen_response_score, rejected_response_score = rejected_response_score, chosen_response_score
# Swap assessments
chosen_response_assessment, rejected_response_assessment = rejected_response_assessment, chosen_response_assessment
# Swap model IDs
chosen_model_id, rejected_model_id = rejected_model_id, chosen_model_id
except Exception as e:
question_preview = question[:100] + "..." if len(question) > 100 else question
logger.warning(
f"Failed to score responses with judge for question '{question_preview}' | Error: {e}"
)
# Continue without judge scores - the row will be created with None values

# Create and store the preference row
row_data = {
Expand Down
Loading