run_id stringclasses 1
value | iteration int64 1 6 | cumulative_tokens int64 9.41k 60.8k | best_val_score float64 0.29 0.38 | accepted bool 1
class | candidate_idx int64 1 6 | candidate_prompt stringclasses 6
values | best_prompt stringclasses 3
values |
|---|---|---|---|---|---|---|---|
fixed_iter_vanilla_k3 | 1 | 9,408 | 0.288889 | true | 1 | You are a helpful math problem–solving assistant. Solve the given problem correctly and concisely, then put the final result on the very last line exactly in the form:
### <final answer>
Do not add any text after that final-answer line.
General approach and requirements:
- Read the problem carefully, identify the stru... | You are a helpful math problem–solving assistant. Solve the given problem correctly and concisely, then put the final result on the very last line exactly in the form:
### <final answer>
Do not add any text after that final-answer line.
General approach and requirements:
- Read the problem carefully, identify the stru... |
fixed_iter_vanilla_k3 | 2 | 16,550 | 0.355556 | true | 2 | You are a helpful math problem-solving assistant. You will be given a single math problem (typically contest-style). Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other... | You are a helpful math problem-solving assistant. You will be given a single math problem (typically contest-style). Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other... |
fixed_iter_vanilla_k3 | 3 | 29,769 | 0.355556 | true | 3 | You are a helpful math problem-solving assistant. You will be given one contest-style math problem. Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other trailing punctua... | You are a helpful math problem-solving assistant. You will be given a single math problem (typically contest-style). Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other... |
fixed_iter_vanilla_k3 | 4 | 41,126 | 0.377778 | true | 4 | You are a helpful math problem-solving assistant. You will be given a single math problem (typically contest-style). Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other... | You are a helpful math problem-solving assistant. You will be given a single math problem (typically contest-style). Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other... |
fixed_iter_vanilla_k3 | 5 | 52,889 | 0.377778 | true | 5 | You are a helpful contest-math problem–solving assistant. Solve the given problem correctly and concisely, then put the final result on the very last line exactly in the form:
### <final answer>
Do not add any text after that final-answer line.
General approach and requirements:
- Read the problem carefully, classify ... | You are a helpful math problem-solving assistant. You will be given a single math problem (typically contest-style). Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other... |
fixed_iter_vanilla_k3 | 6 | 60,806 | 0.377778 | true | 6 | You are a helpful contest-math problem–solving assistant. Solve the given problem correctly and concisely, then put the final result on the very last line exactly in the form:
### <final answer>
Do not add any text after that final-answer line.
General approach and requirements:
- Read the problem carefully, classify ... | You are a helpful math problem-solving assistant. You will be given a single math problem (typically contest-style). Solve it clearly and correctly, and put the final numeric or algebraic answer at the very end of your response in exactly this format:
### <final answer>
with a single space after the hashes and no other... |
gepa-exp-vanilla-20260220-083526
GEPA prompt optimization experiment on AIME math problems.
Task LM: openai/gpt-4.1-mini | Reflection LM: openai/gpt-5 | Reflection Mode: default | Last updated: 2026-02-20 20:28 UTC
Results
| Run | Method | k | Mode | Val Score | Test Acc | Tokens | Cost | Time |
|---|---|---|---|---|---|---|---|---|
| fixed_vanilla_k3 | vanilla | 3 | default | 37.78% | 45.33% | 60,806 | $0.1671 | 7021s |
Learning Curves
Experiment Config
{
"script_name": "run_experiment.py",
"model": "openai/gpt-4.1-mini",
"reflection_lm": "openai/gpt-5",
"reflection_mode": "default",
"hyperparameters": {
"task_lm": "openai/gpt-4.1-mini",
"reflection_lm": "openai/gpt-5",
"reflection_mode": "default",
"seed_prompt": "You are a helpful assistant. You are given a question and you need to answer it. The answer should b..."
},
"input_datasets": [
"AI-MO/aimo-validation-aime",
"MathArena/aime_2025"
],
"description": "GEPA prompt optimization experiment on AIME (reflection_mode=default)",
"num_runs": 1,
"last_updated": "2026-02-20 20:28 UTC"
}
Dataset Configs
This repo contains multiple configs (subsets). Load them with:
from datasets import load_dataset
results = load_dataset("reasoning-degeneration-dev/gepa-exp-vanilla-20260220-083526", "results", split="train")
traces = load_dataset("reasoning-degeneration-dev/gepa-exp-vanilla-20260220-083526", "traces", split="train")
val_traces = load_dataset("reasoning-degeneration-dev/gepa-exp-vanilla-20260220-083526", "val_traces", split="train")
curves = load_dataset("reasoning-degeneration-dev/gepa-exp-vanilla-20260220-083526", "curves", split="train")
rlm_call_traces = load_dataset("reasoning-degeneration-dev/gepa-exp-vanilla-20260220-083526", "rlm_call_traces", split="train")
| Config | Description |
|---|---|
results |
Per-run summary: method, k, val/test scores, tokens, cost, optimized prompt |
traces |
Per-example test set traces with model responses and scores |
val_traces |
Per-iteration validation traces with the prompt used at each step |
curves |
Per-iteration learning curves: tokens, scores, candidate prompts |
rlm_call_traces |
Hierarchical LLM call traces: every RLM iteration, sub-call, prompt, and response |
State file: state.json
For experiment design details, see EXPERIMENTS-DESIGN.md
- Downloads last month
- 69