Downloads · 30 days
0
EdwardoSunny/ttt-discover-v3-qwen3-8b-ac1-s2-step1
ttt-discover-v3-qwen3-8b-ac1-s2-step1 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 0 (0-indexed). Strict upstream eval parity: 1100s hard k…
Downloads · 30 days
0
Access
Public
Updated Jul 18, 2026
Repo size
703 MB
Likes
1
Public
Click a slice to open those files.
.safetensors349 MB · 99%
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 0 (0-indexed). Strict upstream eval parity:
1100s hard kill, verbatim prompts/entrypoints, group 64x8, T=1.0, kl 0.1.
{
"step": 0,
"progress/batch": 0,
"optim/lr": 4e-05,
"progress/done_frac": 0.02,
"puct/buffer_size": 8,
"puct/sampled_size": 8,
"puct/T": 0,
"puct/scale_last": 1e-06,
"puct/buffer_value/mean": -2.000000000000002,
"puct/buffer_value/std": 3.3820826609804605e-15,
"puct/buffer_value/min": -2.000000000000007,
"puct/buffer_value/max": -1.999999999999996,
"puct/buffer_timestep/mean": -1.0,
"puct/buffer_timestep/std": 0.0,
"puct/buffer_timestep/min": -1.0,
"puct/buffer_timestep/max": -1.0,
"puct/buffer_construction_len/mean": 5458.5,
"puct/buffer_construction_len/std": 1297.3183880605409,
"puct/buffer_construction_len/min": 3267.0,
"puct/buffer_construction_len/max": 7850.0,
"puct/sampled_value/mean": -2.000000000000002,
"puct/sampled_value/std": 3.3820826609804605e-15,
"puct/sampled_value/min": -2.000000000000007,
"puct/sampled_value/max": -1.999999999999996,
"puct/sampled_timestep/mean": -1.0,
"puct/sampled_timestep/std": 0.0,
"puct/sampled_timestep/min": -1.0,
"puct/sampled_timestep/max": -1.0,
"puct/sampled_construction_len/mean": 5458.5,
"puct/sampled_construction_len/std": 1297.3183880605409,
"puct/sampled_construction_len/min": 3267.0,
"puct/sampled_construction_len/max": 7850.0,
"time/sampling": 4343.812570810318,
"env/all/ac_tokens_per_turn": 8618.517578125,
"env/all/ob_tokens_per_turn": 2472.0,
"env/all/turns_per_episode": 1.0,
"env/all/total_episodes": 512,
"env/all/total_turns": 512,
"env/all/total_ac_tokens": 4412681,
"env/all/total_ob_tokens": 1265664,
"env/all/time/sampling_mean": 506.7895562471822,
"env/all/time/sampling_max": 716.1196627616882,
"env/all/time/env_step_mean": 1472.5427056960762,
"env/all/time/env_step_max": 3628.097250699997,
"env/all/reward/mean": 0.3453682776377632,
"env/all/reward/max": 0.6619519887167399,
"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": 0.3453682776377632,
"env/all/correctness": 0.65234375,
"env/all/correctness/min": 0.0,
"env/all/correctness/max": 1.0,
"env/all/raw_score": 7.3341255051568215,
"env/all/raw_score/min": 1.5106835698448162,
"env/all/raw_score/max": 999.9999999999999,
"env/all/initial_raw_score": -2.000000000000002,
"env/all/initial_raw_score/min": -2.000000000000007,
"env/all/initial_raw_score/max": -1.999999999999996,
"env/all/msg": "Program must define a top-level `run_code()` function.",
"env/all/parsed_code": "```python\ndef evaluate_sequence(sequence):\n \"\"\"\n C1 verifier helper, matching evaluator logic.\n Returns np.inf if the input is invalid.\n \"\"\"\n if not isinstance(sequence, list):\n return np.inf\n if 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\n sequence = [float(x) for x in sequence]\n sequence = [max(0.0, x) for x in sequence]\n sequence = [min(1000.0, x) for x in sequence]\n\n n = len(sequence)\n b_sequence = np.convolve(sequence, sequence)\n max_b = max(b_sequence)\n sum_a = np.sum(sequence)\n\n if sum_a < 0.01:\n return np.inf\n\n return float(2 * n * max_b / (sum_a**2))\n\n\ndef _time_budget_s(default=1000.0):\n \"\"\"Self-chosen search budget, capped by the harness's advisory budget.\"\"\"\n import os\n\n raw = os.environ.get(\"AB_TIME_BUDGET_S\")\n if not raw:\n return default\n try:\n return max(10.0, min(default, float(raw) - 30.0))\n except ValueError:\n return default\n\n\ndef run_code():\n \"\"\"Run the C1 constructor and return (solution, self-reported score).\"\"\"\n heights = propose_candidate(seed=42, budget_s=_time_budget_s())\n c1_value = evaluate_sequence(heights)\n return heights, c1_value\n```",
"env/all/time/policy": 506.7895562471822,
"env/all/time/policy/min": 173.39177536964417,
"env/all/time/policy/max": 716.1196627616882,
"env/all/time/env_step": 1472.5427056960762,
"env/all/time/env_step/min": 0.006741762161254883,
"env/all/time/env_step/max": 3628.097250699997,
"env/all/time/reward_compute": 1.1213123798370361e-06,
"env/all/time/reward_compute/min": 2.384185791015625e-07,
"env/all/time/reward_compute/max": 1.7434358596801758e-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": 0.025558048859238625,
"advantage/min": -0.9994414448738098,
"advantage/max": 4.336101055145264,
"time/assemble_training_data": 9.382949829101562,
"time/kl_vs_base": 136.39276003837585,
"kl_policy_base": 0.0004983046092092991,
"time/train": 1042.797997713089,
"time/save_checkpoint": 13.900818824768066,
"time/total": 5551.387796878815
}
[2026-07-09T06:24:18+00:00] job=1812630 node=node-14 ngpu=3 ntrain=1 replicas=2 flash_attn=no
[2026-07-09T09:14:17+00:00] job=1813129 node=node-28 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-09T13:46:36+00:00] job=1813130 node=node-22 ngpu=6 ntrain=2 replicas=4 flash_attn=yes