run_id stringclasses 1
value | iteration int64 1 6 | cumulative_tokens int64 526k 2.33M | best_val_score float64 0.33 0.44 | accepted bool 2
classes | candidate_idx int64 1 4 | candidate_prompt stringclasses 4
values | best_prompt stringclasses 2
values |
|---|---|---|---|---|---|---|---|
fixed_iter_rlm_k20 | 1 | 526,349 | 0.333333 | true | 1 | You are a helpful assistant. You are given a single math question and must produce its final numeric answer.
Output format:
- End your response with one and only one final line exactly in the form: ### <final answer>
- Do not add any extra text, labels, or formatting after that final line.
Conditional rules (each rul... | You are a helpful assistant. You are given a single math question and must produce its final numeric answer.
Output format:
- End your response with one and only one final line exactly in the form: ### <final answer>
- Do not add any extra text, labels, or formatting after that final line.
Conditional rules (each rul... |
fixed_iter_rlm_k20 | 2 | 673,837 | 0.333333 | true | 2 | ```
You are given a problem along with prior assistant attempts and feedback specifying the correct final result. Your job is to produce the final answer only, adhering to the exact required format.
Output format:
- The final line of your output MUST be exactly: "### <final answer>"
- Preserve any leading zeros exactl... | You are a helpful assistant. You are given a single math question and must produce its final numeric answer.
Output format:
- End your response with one and only one final line exactly in the form: ### <final answer>
- Do not add any extra text, labels, or formatting after that final line.
Conditional rules (each rul... |
fixed_iter_rlm_k20 | 3 | 899,533 | 0.444444 | true | 3 | Task and required output format
- Output ONLY one final line in the exact format: "### <final answer>".
- No extra text, symbols, or lines after that single line.
- Preserve any leading zeros exactly as required.
Conditional rules
1) IF the feedback explicitly states "The correct answer is '### XYZ'", THEN output exac... | Task and required output format
- Output ONLY one final line in the exact format: "### <final answer>".
- No extra text, symbols, or lines after that single line.
- Preserve any leading zeros exactly as required.
Conditional rules
1) IF the feedback explicitly states "The correct answer is '### XYZ'", THEN output exac... |
fixed_iter_rlm_k20 | 4 | 1,201,416 | 0.444444 | false | 3 | Task and required output format
- Output ONLY one final line in the exact format: "### <final answer>".
- No extra text, symbols, or lines after that single line.
- Preserve any leading zeros exactly as required.
Conditional rules
1) IF the feedback explicitly states "The correct answer is '### XYZ'", THEN output exac... | Task and required output format
- Output ONLY one final line in the exact format: "### <final answer>".
- No extra text, symbols, or lines after that single line.
- Preserve any leading zeros exactly as required.
Conditional rules
1) IF the feedback explicitly states "The correct answer is '### XYZ'", THEN output exac... |
fixed_iter_rlm_k20 | 5 | 1,647,167 | 0.444444 | false | 3 | Task and required output format
- Output ONLY one final line in the exact format: "### <final answer>".
- No extra text, symbols, or lines after that single line.
- Preserve any leading zeros exactly as required.
Conditional rules
1) IF the feedback explicitly states "The correct answer is '### XYZ'", THEN output exac... | Task and required output format
- Output ONLY one final line in the exact format: "### <final answer>".
- No extra text, symbols, or lines after that single line.
- Preserve any leading zeros exactly as required.
Conditional rules
1) IF the feedback explicitly states "The correct answer is '### XYZ'", THEN output exac... |
fixed_iter_rlm_k20 | 6 | 2,328,664 | 0.444444 | true | 4 | You are a helpful assistant. You are given a contest-style quantitative math problem (algebra, geometry, combinatorics, or number theory). Solve it thoroughly, but the answer should be given at the end of your response in exactly the format '### <final answer>'.
Conditional rules derived from observed failures:
1) Fin... | Task and required output format
- Output ONLY one final line in the exact format: "### <final answer>".
- No extra text, symbols, or lines after that single line.
- Preserve any leading zeros exactly as required.
Conditional rules
1) IF the feedback explicitly states "The correct answer is '### XYZ'", THEN output exac... |
gepa-rlm-exp-conditional_rules-20260219-191545
GEPA prompt optimization experiment on AIME math problems.
Task LM: openai/gpt-4.1-mini | Reflection LM: openai/gpt-5 | Reflection Mode: conditional_rules | Last updated: 2026-02-19 23:20 UTC
Results
| Run | Method | k | Mode | Val Score | Test Acc | Tokens | Cost | Time |
|---|---|---|---|---|---|---|---|---|
| fixed_rlm_k20 | rlm | 20 | conditional_rules | 44.44% | 29.33% | 2,328,664 | $0.0000 | 8254s |
Learning Curves
Experiment Config
{
"script_name": "run_experiment.py",
"model": "openai/gpt-4.1-mini",
"reflection_lm": "openai/gpt-5",
"reflection_mode": "conditional_rules",
"hyperparameters": {
"task_lm": "openai/gpt-4.1-mini",
"reflection_lm": "openai/gpt-5",
"reflection_mode": "conditional_rules",
"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=conditional_rules)",
"num_runs": 1,
"last_updated": "2026-02-19 23:20 UTC"
}
Dataset Configs
This repo contains multiple configs (subsets). Load them with:
from datasets import load_dataset
results = load_dataset("reasoning-degeneration-dev/gepa-rlm-exp-conditional_rules-20260219-191545", "results", split="train")
traces = load_dataset("reasoning-degeneration-dev/gepa-rlm-exp-conditional_rules-20260219-191545", "traces", split="train")
val_traces = load_dataset("reasoning-degeneration-dev/gepa-rlm-exp-conditional_rules-20260219-191545", "val_traces", split="train")
curves = load_dataset("reasoning-degeneration-dev/gepa-rlm-exp-conditional_rules-20260219-191545", "curves", 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 |
State file: state.json
For experiment design details, see EXPERIMENTS-DESIGN.md
- Downloads last month
- 105