Downloads · 30 days
0
EdwardoSunny/ttt-discover-v3-qwen3-8b-ac1-step12
ttt-discover-v3-qwen3-8b-ac1-step12 is a machine learning model from EdwardoSunny. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
LoRA adapter (rank 32, all-linear) from the upstream-faithful local reproduction of TTT-Discover on autoresearch-bench ac1. Checkpoint saved after training step 11 (0-indexed). Strict upstream eval parity: 1100s hard…
Downloads · 30 days
0
Access
Public
Updated Jul 10, 2026
Repo size
349 MB
Likes
0
Public
Click a slice to open those files.
.safetensors349 MB · 100%
From the Hugging Face model README
LoRA adapter (rank 32, all-linear) from the upstream-faithful local
reproduction of TTT-Discover on autoresearch-bench ac1. Checkpoint saved
after training step 11 (0-indexed). Strict upstream eval parity:
1100s hard kill, verbatim prompts/entrypoints, group 64x8, T=1.0, kl 0.1.
{
"step": 11,
"progress/batch": 11,
"optim/lr": 4e-05,
"progress/done_frac": 0.24,
"puct/buffer_size": 184,
"puct/sampled_size": 8,
"puct/T": 5632,
"puct/scale_last": 0.3990266118483723,
"puct/buffer_value/mean": -1.541618149620317,
"puct/buffer_value/std": 0.1124599614527512,
"puct/buffer_value/min": -2.000000000000008,
"puct/buffer_value/max": -1.5065964289456981,
"puct/buffer_timestep/mean": 4.739130434782608,
"puct/buffer_timestep/std": 3.326015912853092,
"puct/buffer_timestep/min": -1.0,
"puct/buffer_timestep/max": 10.0,
"puct/buffer_construction_len/mean": 1236.5815217391305,
"puct/buffer_construction_len/std": 969.9643672779889,
"puct/buffer_construction_len/min": 1000.0,
"puct/buffer_construction_len/max": 7397.0,
"puct/sampled_value/mean": -1.5070493288421054,
"puct/sampled_value/std": 0.0002388470136256805,
"puct/sampled_value/min": -1.5071867255611344,
"puct/sampled_value/max": -1.5065964289456981,
"puct/sampled_timestep/mean": 10.0,
"puct/sampled_timestep/std": 0.0,
"puct/sampled_timestep/min": 10.0,
"puct/sampled_timestep/max": 10.0,
"puct/sampled_construction_len/mean": 1000.0,
"puct/sampled_construction_len/std": 0.0,
"puct/sampled_construction_len/min": 1000.0,
"puct/sampled_construction_len/max": 1000.0,
"time/sampling": 5307.9264969825745,
"env/all/ac_tokens_per_turn": 8860.203125,
"env/all/ob_tokens_per_turn": 3142.375,
"env/all/turns_per_episode": 1.0,
"env/all/total_episodes": 512,
"env/all/total_turns": 512,
"env/all/total_ac_tokens": 4536424,
"env/all/total_ob_tokens": 1608896,
"env/all/time/sampling_mean": 324.81165285455063,
"env/all/time/sampling_max": 452.124662399292,
"env/all/time/env_step_mean": 2544.5612419000827,
"env/all/time/env_step_max": 4920.267135381699,
"env/all/reward/mean": 0.5611883985062822,
"env/all/reward/max": 0.6637885161058397,
"env/all/reward/min": 0.0,
"env/all/format": 1.0,
"env/all/format/min": 1.0,
"env/all/format/max": 1.0,
"env/all/reward": 0.5611883985062822,
"env/all/correctness": 0.88671875,
"env/all/correctness/min": 0.0,
"env/all/correctness/max": 1.0,
"env/all/raw_score": 1.743000714310765,
"env/all/raw_score/min": 1.5065039076431792,
"env/all/raw_score/max": 45.7151279174971,
"env/all/initial_raw_score": -1.5070493288421054,
"env/all/initial_raw_score/min": -1.5071867255611344,
"env/all/initial_raw_score/max": -1.5065964289456981,
"env/all/msg": "Program must define a top-level `run_code()` function.",
"env/all/parsed_code": "```python\nimport numpy as np\nfrom typing import List\n\ndef evaluate_sequence(sequence: List[float]) -> float:\n \"\"\"\n Evaluates a sequence of coefficients with enhanced security checks.\n Returns np.inf if the input is invalid.\n \"\"\"\n # --- Security Checks ---\n if not isinstance(sequence, list) or not sequence:\n return np.inf\n\n for x in sequence:\n if isinstance(x, bool) or not isinstance(x, (int, float)):\n return np.inf\n if np.isnan(x) or np.isinf(x):\n return np.inf\n sequence = [float(x) for x in sequence]\n sequence = [max(0, x) for x in sequence]\n sequence = [min(1000.0, x) for x in sequence]\n n = len(sequence)\n b_sequence = np.convolve(sequence, sequence)\n max_b = max(b_sequence)\n sum_a = np.sum(sequence)\n if sum_a < 0.01:\n return np.inf\n return float(2 * n * max_b / (sum_a ** 2))\n\n# The proposed_candidate function is defined above\npropose_candidate()\n```",
"env/all/time/policy": 324.81165285455063,
"env/all/time/policy/min": 146.2706732749939,
"env/all/time/policy/max": 452.124662399292,
"env/all/time/env_step": 2544.5612419000827,
"env/all/time/env_step/min": 0.008902788162231445,
"env/all/time/env_step/max": 4920.267135381699,
"env/all/time/reward_compute": 4.614703357219696e-07,
"env/all/time/reward_compute/min": 2.3096799850463867e-07,
"env/all/time/reward_compute/max": 1.5720725059509277e-06,
"env/all/by_group/frac_mixed": 1.0,
"env/all/by_group/frac_all_good": 0.0,
"env/all/by_group/frac_all_bad": 0.0,
"advantage/mean": 0.008608231320977211,
"advantage/min": -1.0,
"advantage/max": 1.4131360054016113,
"time/assemble_training_data": 10.262456893920898,
"time/kl_vs_base": 93.2112672328949,
"kl_policy_base": 0.00073909608181566,
"time/train": 576.3283305168152,
"time/save_checkpoint": 36.255486249923706,
"time/total": 6025.731567144394
}
[2026-07-09T06:24:18+00:00] job=1812628 node=node-12 ngpu=3 ntrain=1 replicas=2 flash_attn=no
[2026-07-09T08:30:32+00:00] job=1813127 node=node-22 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-09T13:07:08+00:00] job=1813128 node=node-6 ngpu=6 ntrain=2 replicas=4 flash_attn=yes