Skip to content

Refactor full evaluation of SIMBA with dspy.Evaluate #8022

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

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Refactor full evaluation of SIMBA with dspy.Evaluate #8022

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
16 changes: 7 additions & 9 deletions dspy/teleprompt/simba.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -252,6 +252,7 @@ def register_new_program(prog: dspy.Module, score_list: list[float]):
break

# STEP 5: Evaluate these new system_candidates on the same mini-batch
# TODO: refactor the step 5 and 6 with dspy.Evaluate after https://github.com/stanfordnlp/dspy/pull/8003 is merged
logger.info(f"Batch {batch_idx+1}: Evaluating {len(system_candidates)} programs on {self.bsize} examples.")

exec_pairs = [(wrap_program(sys, self.metric), ex) for sys in system_candidates for ex in batch]
Expand Down Expand Up @@ -299,16 +300,13 @@ def register_new_program(prog: dspy.Module, score_list: list[float]):
exec_pairs = [(wrap_program(sys, self.metric), ex) for sys in candidate_programs for ex in trainset]
outputs = run_parallel(exec_pairs)

scores = []
for idx_prog, prog in enumerate(candidate_programs):
start = idx_prog * len(trainset)
end = (idx_prog + 1) * len(trainset)
sys_scores = [outputs[i]["score"] for i in range(start, end)]
avg_score = sum(sys_scores) / len(sys_scores) if sys_scores else 0.0
scores.append(avg_score)
evaluate = dspy.Evaluate(devset=trainset, metric=self.metric, num_threads=self.num_threads, display_progress=False, display_table=False)
scores = [evaluate(program, callback_metadata={"metric_key": "eval_full"}) for program in candidate_programs]

for idx_prog, score in enumerate(scores):
if idx_prog != 0:
trial_logs[idx_prog-1]["train_score"] = avg_score
trial_logs[idx_prog-1]["train_score"] = score

best_idx = scores.index(max(scores)) if scores else 0
best_program = candidate_programs[best_idx]
logger.info(
Expand Down
Loading