run_id string | iteration int64 | cumulative_tokens int64 | best_val_score float64 | accepted bool | candidate_idx int64 | candidate_prompt string | best_prompt string |
|---|---|---|---|---|---|---|---|
fixed_iter_rlm_k3 | 1 | 64,986 | 0.311111 | true | 1 | You are a math problem–solving assistant. You will be given a single contest-style math question drawn from domains such as:
- Base and digit representation puzzles (e.g., relating a decimal number to a permutation of its digits in another base).
- Palindromes across bases (e.g., numbers that are palindromes in base 10... | You are a math problem–solving assistant. You will be given a single contest-style math question drawn from domains such as:
- Base and digit representation puzzles (e.g., relating a decimal number to a permutation of its digits in another base).
- Palindromes across bases (e.g., numbers that are palindromes in base 10... |
fixed_iter_rlm_k3 | 2 | 228,657 | 0.355556 | true | 2 | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... |
fixed_iter_rlm_k3 | 3 | 257,400 | 0.355556 | false | 2 | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... |
fixed_iter_rlm_k3 | 4 | 331,578 | 0.355556 | false | 2 | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... |
fixed_iter_rlm_k3 | 5 | 362,633 | 0.355556 | false | 2 | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... |
fixed_iter_rlm_k3 | 6 | 645,403 | 0.355556 | false | 2 | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... | You are a math problem–solving assistant. You will be given a single contest-style math question (geometry, algebra/number theory, or combinatorics/geometry), possibly with a brief diagram description (e.g., Asymptote code). Your job is to solve the problem correctly and present a clear, logically sound solution. At th... |
gepa-exp-inputs_only-rlm-20260220-083526
GEPA prompt optimization experiment on AIME math problems.
Task LM: openai/gpt-4.1-mini | Reflection LM: openai/gpt-5 | Reflection Mode: inputs_only | Last updated: 2026-02-20 19:47 UTC
Results
| Run | Method | k | Mode | Val Score | Test Acc | Tokens | Cost | Time |
|---|---|---|---|---|---|---|---|---|
| fixed_rlm_k3 | rlm | 3 | inputs_only | 35.56% | 48.67% | 645,403 | $0.0000 | 34052s |
Learning Curves
Experiment Config
{
"script_name": "run_experiment.py",
"model": "openai/gpt-4.1-mini",
"reflection_lm": "openai/gpt-5",
"reflection_mode": "inputs_only",
"hyperparameters": {
"task_lm": "openai/gpt-4.1-mini",
"reflection_lm": "openai/gpt-5",
"reflection_mode": "inputs_only",
"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=inputs_only)",
"num_runs": 1,
"last_updated": "2026-02-20 19:47 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-inputs_only-rlm-20260220-083526", "results", split="train")
traces = load_dataset("reasoning-degeneration-dev/gepa-exp-inputs_only-rlm-20260220-083526", "traces", split="train")
val_traces = load_dataset("reasoning-degeneration-dev/gepa-exp-inputs_only-rlm-20260220-083526", "val_traces", split="train")
curves = load_dataset("reasoning-degeneration-dev/gepa-exp-inputs_only-rlm-20260220-083526", "curves", split="train")
rlm_call_traces = load_dataset("reasoning-degeneration-dev/gepa-exp-inputs_only-rlm-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
- 61