iteration int64 1 50 | island_id int64 0 2 | score float64 -3.22 0 | best_score float64 -0.99 0 | delta float64 -2.21 2.2 | adaptation_type large_stringclasses 2
values | exploration_intensity float64 0.42 0.43 | is_valid bool 2
classes | task_id large_stringclasses 1
value | meta_guidance_tactic large_stringclasses 1
value | tactic_approach_type large_stringclasses 3
values | prompt_text large_stringlengths 64.4k 109k | reasoning_trace large_stringlengths 710 12.6k | program_code large_stringlengths 30.3k 34k |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | 0 | 0 | 0 | 0 | L1_explore | 0.415911 | false | autoresearch | exploration | ## system
You are optimizing a GPT language model training script. Your goal is to minimize validation bits-per-byte (val_bpb). Lower is better.
## Constraints
- Maximum model size: 50 million parameters.
- Fixed training time budget (approximately 5 minutes wall-clock).
- Only `train.py` is editable. The data prepar... | <<<<<<< SEARCH
# ---------------------------------------------------------------------------
# Hyperparameters (edit these directly, no CLI flags needed)
# ---------------------------------------------------------------------------
# Model architecture
ASPECT_RATIO = 64 # model_dim = depth * ASPECT_RATIO
HEAD_DI... | """
Autoresearch pretraining script. Single-GPU, single-file.
Cherry-picked and simplified from nanochat.
Supports CUDA (with FA3) and MPS (with SDPA fallback).
Usage: AUTORESEARCH_DEVICE=mps python train.py # Mac
AUTORESEARCH_DEVICE=cuda python train.py # GPU cluster
uv run train.py ... | |
2 | 2 | -0.991392 | -0.991392 | -0.991392 | L1_exploit | 0.415911 | true | autoresearch | exploitation | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)\n===(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
3 | 0 | -1.139375 | -0.991392 | -0.147983 | L1_explore | 0.419231 | true | autoresearch | balanced | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n def setup_optimizer(self, unembedding_lr=0.004, embedding_lr=0.2, matrix_lr=0.0(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
4 | 1 | -1.203 | -0.991392 | -0.063625 | L1_explore | 0.415911 | true | autoresearch | balanced | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n for shape in sorted({p.shape for p in matrix_params}):\n group_p(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
5 | 1 | -2.544342 | -0.991392 | -1.341342 | L1_exploit | 0.419231 | true | autoresearch | exploitation | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n# ---------------------------------------------------------------------------\n# Hy(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
6 | 2 | -1.314421 | -0.991392 | 1.229921 | L1_explore | 0.419231 | true | autoresearch | exploration | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n def setup_optimizer(self, unembedding_lr=0.004, embedding_lr=0.2, matrix_lr=0.0(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
7 | 2 | -1.138823 | -0.991392 | 0.175598 | L1_explore | 0.422457 | true | autoresearch | exploration | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n def setup_optimizer(self, unembedding_lr=0.004, embedding_lr=0.2, matrix_lr=0.0(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
8 | 1 | -2.619382 | -0.991392 | -1.480559 | L1_explore | 0.422457 | true | autoresearch | balanced | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)\n (...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
9 | 1 | -1.032124 | -0.991392 | 1.587258 | L1_explore | 0.422457 | true | autoresearch | exploration | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n# ---------------------------------------------------------------------------\n# Hy(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) | |
10 | 2 | -0.999858 | -0.991392 | 0.032266 | L1_explore | 0.422457 | true | autoresearch | exploration | "## system\nYou are optimizing a GPT language model training script. Your goal is to minimize valida(...TRUNCATED) | "<<<<<<< SEARCH\n# ---------------------------------------------------------------------------\n# Hy(...TRUNCATED) | "\"\"\"\nAutoresearch pretraining script. Single-GPU, single-file.\nCherry-picked and simplified fro(...TRUNCATED) |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 8