Downloads · 30 days
0
EdwardoSunny/ttt-discover-v3-qwen3-8b-erdos-step19
ttt-discover-v3-qwen3-8b-erdos-step19 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 erdos. Checkpoint saved after training step 18 (0-indexed). Strict upstream eval parity: 1100s har…
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 erdos. Checkpoint saved
after training step 18 (0-indexed). Strict upstream eval parity:
1100s hard kill, verbatim prompts/entrypoints, group 64x8, T=1.0, kl 0.1.
{
"step": 18,
"progress/batch": 18,
"optim/lr": 4e-05,
"progress/done_frac": 0.38,
"puct/buffer_size": 296,
"puct/sampled_size": 8,
"puct/T": 9216,
"puct/scale_last": 0.11905844328365012,
"puct/buffer_value/mean": -0.38563976920309456,
"puct/buffer_value/std": 0.021880386441203607,
"puct/buffer_value/min": -0.5236859987110454,
"puct/buffer_value/max": -0.38094155671635577,
"puct/buffer_timestep/mean": 8.243243243243244,
"puct/buffer_timestep/std": 5.344385310420472,
"puct/buffer_timestep/min": -1.0,
"puct/buffer_timestep/max": 17.0,
"puct/buffer_construction_len/mean": 78.4695945945946,
"puct/buffer_construction_len/std": 12.854398546715776,
"puct/buffer_construction_len/min": 42.0,
"puct/buffer_construction_len/max": 143.0,
"puct/sampled_value/mean": -0.38094157550797003,
"puct/sampled_value/std": 1.878667065836842e-08,
"puct/sampled_value/min": -0.3809415957664156,
"puct/sampled_value/max": -0.38094155671635577,
"puct/sampled_timestep/mean": 17.0,
"puct/sampled_timestep/std": 0.0,
"puct/sampled_timestep/min": 17.0,
"puct/sampled_timestep/max": 17.0,
"puct/sampled_construction_len/mean": 80.0,
"puct/sampled_construction_len/std": 0.0,
"puct/sampled_construction_len/min": 80.0,
"puct/sampled_construction_len/max": 80.0,
"time/sampling": 1613.2692370414734,
"env/all/ac_tokens_per_turn": 9428.919921875,
"env/all/ob_tokens_per_turn": 1564.25,
"env/all/turns_per_episode": 1.0,
"env/all/total_episodes": 512,
"env/all/total_turns": 512,
"env/all/total_ac_tokens": 4827607,
"env/all/total_ob_tokens": 800896,
"env/all/time/sampling_mean": 524.4651684728451,
"env/all/time/sampling_max": 742.6165125370026,
"env/all/time/env_step_mean": 78.85070284502581,
"env/all/time/env_step_max": 1100.121943473816,
"env/all/reward/mean": 1.0647895156204368,
"env/all/reward/max": 2.625074519821873,
"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": 1.0647895156204368,
"env/all/correctness": 0.4140625,
"env/all/correctness/min": 0.0,
"env/all/correctness/max": 1.0,
"env/all/raw_score": 0.39063227238859494,
"env/all/raw_score/min": 0.3809415565913576,
"env/all/raw_score/max": 0.525,
"env/all/initial_raw_score": -0.38094157550797003,
"env/all/initial_raw_score/min": -0.3809415957664156,
"env/all/initial_raw_score/max": -0.38094155671635577,
"env/all/msg": "RuntimeError: Program execution failed: ValueError: Some entries in x0 lay outside the specified bounds",
"env/all/parsed_code": "```python\nimport numpy as np\nfrom scipy.optimize import differential_evolution\nfrom scipy.signal import correlate\n\ndef evaluate_C5(h, dx):\n h1 = 1.0 - h\n corr = correlate(h, h1, mode='full')\n max_corr = np.max(corr)\n return max_corr * dx\n\ndef run(seed=42, budget_s=1000, **kwargs):\n np.random.seed(seed)\n import random\n random.seed(seed)\n\n n_points = len(initial_h_values)\n dx = 2.0 / n_points\n required_sum = n_points / 2 # required sum for the function to satisfy the integral constraint\n\n # Generate a randomized initial guess with required total sum\n initial_guess = np.random.rand(n_points)\n initial_guess = initial_guess * (required_sum / np.sum(initial_guess))\n\n def objective(x):\n sum_x = np.sum(x)\n if sum_x == 0:\n return float('inf')\n scaled_x = x * (required_sum / sum_x)\n return evaluate_C5(scaled_x, dx)\n\n bounds = [(0.0, 1.0) for _ in range(n_points)]\n\n result = differential_evolution(\n objective,\n bounds,\n strategy='rand1bin', # Increased diversity in mutation\n popsize=40, # Larger population for more exploration\n maxiter=500, # More iterations for convergence\n tol=1e-6,\n mutation=(0.5, 1.0), # Wider mutation range\n recombination=0.9, # Higher recombination for diversity\n disp=False,\n polish=True,\n x0=initial_guess\n )\n\n best_h_scaled = result.x * (required_sum / np.sum(result.x))\n best_c5 = result.fun\n return best_h_scaled, best_c5, n_points\n```",
"env/all/time/policy": 524.4651684728451,
"env/all/time/policy/min": 266.1613006591797,
"env/all/time/policy/max": 742.6165125370026,
"env/all/time/env_step": 78.85070284502581,
"env/all/time/env_step/min": 0.004856586456298828,
"env/all/time/env_step/max": 1100.121943473816,
"env/all/time/reward_compute": 3.4319236874580383e-07,
"env/all/time/reward_compute/min": 2.3469328880310059e-07,
"env/all/time/reward_compute/max": 5.103647708892822e-07,
"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": 0.02072959393262863,
"advantage/min": -1.0,
"advantage/max": 2.591762065887451,
"time/assemble_training_data": 7.424537658691406,
"time/kl_vs_base": 131.11628103256226,
"kl_policy_base": 0.0008124432060867548,
"time/train": 1030.1548042297363,
"time/save_checkpoint": 11.807781457901001,
"time/total": 2795.10022687912
}
[2026-07-09T06:24:18+00:00] job=1812624 node=node-6 ngpu=3 ntrain=1 replicas=2 flash_attn=no
[2026-07-09T07:31:44+00:00] job=1812955 node=node-1 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-09T07:59:00+00:00] job=1813123 node=node-14 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-09T09:28:32+00:00] job=1813124 node=node-3 ngpu=6 ntrain=2 replicas=4 flash_attn=yes
[2026-07-09T09:35:59+00:00] job=1813609 node=node-6 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-09T09:45:57+00:00] job=1813622 node=node-12 ngpu=3 ntrain=1 replicas=2 flash_attn=yes