Downloads · 30 days
0
EdwardoSunny/ttt-discover-v3-qwen3-8b-ac2-step29
ttt-discover-v3-qwen3-8b-ac2-step29 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 ac2. Checkpoint saved after training step 28 (0-indexed). Strict upstream eval parity: 1100s hard…
Downloads · 30 days
0
Access
Public
Updated Jul 11, 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 ac2. Checkpoint saved
after training step 28 (0-indexed). Strict upstream eval parity:
1100s hard kill, verbatim prompts/entrypoints, group 64x8, T=1.0, kl 0.1.
{
"step": 28,
"progress/batch": 28,
"optim/lr": 4e-05,
"progress/done_frac": 0.58,
"puct/buffer_size": 454,
"puct/sampled_size": 8,
"puct/T": 14336,
"puct/scale_last": 0.43969346846112234,
"puct/buffer_value/mean": 0.929770515121919,
"puct/buffer_value/std": 0.045464698125460876,
"puct/buffer_value/min": 0.5041471954736918,
"puct/buffer_value/max": 0.9438406639348141,
"puct/buffer_timestep/mean": 13.222466960352422,
"puct/buffer_timestep/std": 8.239869074946299,
"puct/buffer_timestep/min": -1.0,
"puct/buffer_timestep/max": 27.0,
"puct/buffer_construction_len/mean": 2682.409691629956,
"puct/buffer_construction_len/std": 2565.9851255905373,
"puct/buffer_construction_len/min": 1024.0,
"puct/buffer_construction_len/max": 32768.0,
"puct/sampled_value/mean": 0.9438066130045593,
"puct/sampled_value/std": 2.026613330693627e-05,
"puct/sampled_value/min": 0.9437716626922124,
"puct/sampled_value/max": 0.9438406639348141,
"puct/sampled_timestep/mean": 27.0,
"puct/sampled_timestep/std": 0.0,
"puct/sampled_timestep/min": 27.0,
"puct/sampled_timestep/max": 27.0,
"puct/sampled_construction_len/mean": 4096.0,
"puct/sampled_construction_len/std": 0.0,
"puct/sampled_construction_len/min": 4096.0,
"puct/sampled_construction_len/max": 4096.0,
"time/sampling": 4418.597546815872,
"env/all/ac_tokens_per_turn": 9545.271484375,
"env/all/ob_tokens_per_turn": 4165.625,
"env/all/turns_per_episode": 1.0,
"env/all/total_episodes": 512,
"env/all/total_turns": 512,
"env/all/total_ac_tokens": 4887179,
"env/all/total_ob_tokens": 2132800,
"env/all/time/sampling_mean": 391.37988067697734,
"env/all/time/sampling_max": 502.60800886154175,
"env/all/time/env_step_mean": 1829.290811994113,
"env/all/time/env_step_max": 3926.9222033023834,
"env/all/reward/mean": 0.6986357145507958,
"env/all/reward/max": 0.9440545618752789,
"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": 0.6986357145507958,
"env/all/correctness": 0.771484375,
"env/all/correctness/min": 0.0,
"env/all/correctness/max": 1.0,
"env/all/raw_score": 0.905573381898753,
"env/all/raw_score/min": 0.028615092282257632,
"env/all/raw_score/max": 0.9440545618752789,
"env/all/initial_raw_score": 0.9438066130045591,
"env/all/initial_raw_score/min": 0.9437716626922124,
"env/all/initial_raw_score/max": 0.9438406639348141,
"env/all/msg": "Program must define a top-level `run_code()` function.",
"env/all/parsed_code": "```python\ndef _simpson_l2sq(conv: np.ndarray) -> Tuple[float, np.ndarray]:\n \"\"\"Compute ||f*f||_2^2 via Simpson's rule with endpoint zeros and gradient.\"\"\"\n m = conv.size\n if m == 0:\n return 0.0, np.zeros_like(conv)\n dx = 1.0 / (m + 1)\n y = np.zeros(m + 2, dtype=conv.dtype)\n y[0] = 0.0\n y[1:-1] = conv\n y[-1] = 0.0\n lhs = y[:-1]\n rhs = y[1:]\n l2_sq = (dx / 3.0) * np.sum(lhs * lhs + lhs * rhs + rhs * rhs)\n grad_y = (dx / 3.0) * (4.0 * y + np.roll(y, 1) + np.roll(y, -1))\n grad_conv = grad_y[1:-1]\n return float(l2_sq), grad_conv\n\ndef _l1(conv: np.ndarray) -> Tuple[float, np.ndarray]:\n \"\"\"Compute ||f*f||_1 and its gradient.\"\"\"\n m = conv.size\n dx = 1.0 / (m + 1) if m > 0 else 1.0\n val = dx * float(np.sum(conv)) if m > 0 else 0.0\n grad = np.full_like(conv, dx)\n return val, grad\n\ndef _linf(conv: np.ndarray) -> Tuple[float, np.ndarray]:\n \"\"\"Compute ||f*f||_inf and its subgradient.\"\"\"\n if conv.size == 0:\n return 0.0, np.zeros_like(conv)\n m = float(np.max(conv))\n mask = conv == m\n count = int(mask.sum())\n if count == 0 or m <= 0.0:\n return m, np.zeros_like(conv)\n grad = mask.astype(conv.dtype)\n return m, grad\n\ndef _objective_and_grad_conv(conv: np.ndarray) -> Tuple[float, np.ndarray]:\n \"\"\"Compute C = l2_sq / (l1 * linf) and its gradient.\"\"\"\n l2_sq, g_l2 = _simpson_l2sq(conv)\n l1, g_l1 = _l1(conv)\n linf, g_linf = _linf(conv)\n if l1 <= 0.0 or linf <= 0.0:\n return 0.0, np.zeros_like(conv)\n denom = l1 * linf\n c_value = l2_sq / denom\n num_grad = g_l2 * denom - l2_sq * (g_l1 * linf + l1 * g_linf)\n g_conv = num_grad / (denom * denom)\n return float(c_value), g_conv\n\ndef _grad_h_from_conv_grad(h: np.ndarray, g_conv: np.ndarray) -> np.ndarray:\n \"\"\"Compute gradient of C w.r.t h from gradient of C w.r.t conv.\"\"\"\n h_rev = h[::-1]\n g_h = np.convolve(g_conv, h_rev, mode=\"full\")\n N = len(h)\n # Extract the central N elements for the gradient\n gradient = g_h[N-1 : N-1 + N]\n return gradient\n```",
"env/all/time/policy": 391.37988067697734,
"env/all/time/policy/min": 100.39073753356934,
"env/all/time/policy/max": 502.60800886154175,
"env/all/time/env_step": 1829.290811994113,
"env/all/time/env_step/min": 0.006700992584228516,
"env/all/time/env_step/max": 3926.9222033023834,
"env/all/time/reward_compute": 2.230517566204071e-07,
"env/all/time/reward_compute/min": 1.6763806343078613e-07,
"env/all/time/reward_compute/max": 3.8743019104003906e-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": 0.01771220937371254,
"advantage/min": -1.0,
"advantage/max": 2.453498363494873,
"time/assemble_training_data": 10.460225105285645,
"time/kl_vs_base": 101.31567573547363,
"kl_policy_base": 0.000734988076146692,
"time/train": 675.9371919631958,
"time/save_checkpoint": 11.815765619277954,
"time/total": 5222.140110254288
}
[2026-07-09T06:38:34+00:00] job=1812632 node=node-30 ngpu=3 ntrain=1 replicas=2 flash_attn=no
[2026-07-09T06:45:52+00:00] job=1812704 node=node-30 ngpu=3 ntrain=1 replicas=2 flash_attn=no
[2026-07-09T07:00:42+00:00] job=1812735 node=node-1 ngpu=3 ntrain=1 replicas=2 flash_attn=no
[2026-07-09T07:26:33+00:00] job=1812827 node=node-14 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-09T09:21:09+00:00] job=1813131 node=node-1 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-09T14:53:48+00:00] job=1813132 node=node-2 ngpu=6 ntrain=2 replicas=4 flash_attn=yes
[2026-07-10T03:31:51+00:00] job=1816627 node=node-14 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-10T03:54:03+00:00] job=1816628 node=node-29 ngpu=6 ntrain=2 replicas=4 flash_attn=yes
[2026-07-10T07:56:44+00:00] job=1817463 node=node-27 ngpu=3 ntrain=1 replicas=2 flash_attn=yes
[2026-07-10T09:10:26+00:00] job=1817464 node=node-7 ngpu=6 ntrain=2 replicas=4 flash_attn=yes