Downloads · 30 days
0
EdwardoSunny/ttt-discover-v3-qwen3-8b-erdos-step1
ttt-discover-v3-qwen3-8b-erdos-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 erdos. Checkpoint saved after training step 0 (0-indexed). Strict upstream eval parity: 1100s hard…
Downloads · 30 days
0
Access
Public
Updated Jul 18, 2026
Repo size
356 MB
Likes
0
Public
Click a slice to open those files.
.safetensors349 MB · 98%
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 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": 0.03507053895324075,
"puct/buffer_value/mean": -0.5065904991478537,
"puct/buffer_value/std": 0.010869062114838382,
"puct/buffer_value/min": -0.5236859987110454,
"puct/buffer_value/max": -0.4886154597578046,
"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": 71.25,
"puct/buffer_construction_len/std": 18.07449860992,
"puct/buffer_construction_len/min": 42.0,
"puct/buffer_construction_len/max": 89.0,
"puct/sampled_value/mean": -0.5065904991478538,
"puct/sampled_value/std": 0.010869062114838382,
"puct/sampled_value/min": -0.5236859987110454,
"puct/sampled_value/max": -0.4886154597578046,
"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": 71.25,
"puct/sampled_construction_len/std": 18.07449860992,
"puct/sampled_construction_len/min": 42.0,
"puct/sampled_construction_len/max": 89.0,
"time/sampling": 2225.0568001270294,
"env/all/ac_tokens_per_turn": 12429.11328125,
"env/all/ob_tokens_per_turn": 613.0,
"env/all/turns_per_episode": 1.0,
"env/all/total_episodes": 512,
"env/all/total_turns": 512,
"env/all/total_ac_tokens": 6363706,
"env/all/total_ob_tokens": 313856,
"env/all/time/sampling_mean": 763.4914457998239,
"env/all/time/sampling_max": 1155.2199728488922,
"env/all/time/env_step_mean": 17.696104682981968,
"env/all/time/env_step_max": 1100.0278153419495,
"env/all/reward/mean": 0.6804997348165548,
"env/all/reward/max": 2.6125916312517923,
"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": 0.6804997348165548,
"env/all/correctness": 0.310546875,
"env/all/correctness/min": 0.0,
"env/all/correctness/max": 1.0,
"env/all/raw_score": 0.4644364537939284,
"env/all/raw_score/min": 0.38276168457101944,
"env/all/raw_score/max": 0.88436076067983,
"env/all/initial_raw_score": -0.5065904991478536,
"env/all/initial_raw_score/min": -0.5236859987110454,
"env/all/initial_raw_score/max": -0.4886154597578046,
"env/all/msg": "Success; raw_score=0.3880920153109342",
"env/all/parsed_code": "```python\nimport numpy as np\nfrom scipy.optimize import differential_evolution\n\n# Assume initial_h_values is a pre-defined global array of length 86\n\ndef run(seed=42, budget_s=1000, **kwargs):\n # Choose number of points (smaller sequences are preferred)\n n_points = 86\n dx = 2.0 / n_points\n bounds = [(0.0, 1.0) for _ in range(n_points)]\n\n def objective(h_candidate):\n current_sum = np.sum(h_candidate)\n target_sum = n_points / 2\n if np.isclose(current_sum, 0):\n return np.inf # Avoid division by zero\n scale_factor = target_sum / current_sum\n h_scaled = h_candidate * scale_factor\n h_clipped = np.clip(h_scaled, 0.0, 1.0)\n # Compute cross-correlation\n crosscorr = np.correlate(h_clipped, 1 - h_clipped, mode='full')\n max_corr = np.max(crosscorr)\n c5 = max_corr * dx\n return c5\n\n # Execute the optimization\n result = differential_evolution(\n objective,\n bounds,\n strategy='best1bin',\n maxiter=1000,\n popsize=50,\n tol=1e-6,\n mutation=(0.5, 1),\n recombination=0.7,\n disp=False\n )\n\n # Extract best solution\n h_values = result.x\n c5_bound = result.fun\n return h_values, c5_bound, n_points\n```",
"env/all/time/policy": 763.4914457998239,
"env/all/time/policy/min": 349.5114736557007,
"env/all/time/policy/max": 1155.2199728488922,
"env/all/time/env_step": 17.696104682981968,
"env/all/time/env_step/min": 0.006815910339355469,
"env/all/time/env_step/max": 1100.0278153419495,
"env/all/time/reward_compute": 3.8463622331619263e-07,
"env/all/time/reward_compute/min": 1.8998980522155762e-07,
"env/all/time/reward_compute/max": 8.009374141693115e-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": 0.027355127036571503,
"advantage/min": -0.8776028156280518,
"advantage/max": 4.280961990356445,
"time/assemble_training_data": 11.05882453918457,
"time/kl_vs_base": 160.24207377433777,
"kl_policy_base": 0.0005018803058192134,
"time/train": 1277.938372850418,
"time/save_checkpoint": 14.13135814666748,
"time/total": 3694.365350008011
}
[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