Downloads · 30 days
0
EdwardoSunny/ttt-discover-v3-qwen3-8b-erdos-step24
ttt-discover-v3-qwen3-8b-erdos-step24 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 23 (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 23 (0-indexed). Strict upstream eval parity:
1100s hard kill, verbatim prompts/entrypoints, group 64x8, T=1.0, kl 0.1.
{
"step": 23,
"progress/batch": 23,
"optim/lr": 4e-05,
"progress/done_frac": 0.48,
"puct/buffer_size": 376,
"puct/sampled_size": 8,
"puct/T": 11776,
"puct/scale_last": 0.11905845290956629,
"puct/buffer_value/mean": -0.3847110900468638,
"puct/buffer_value/std": 0.019532018678533106,
"puct/buffer_value/min": -0.5236859987110454,
"puct/buffer_value/max": -0.3809415470904396,
"puct/buffer_timestep/mean": 10.74468085106383,
"puct/buffer_timestep/std": 6.786933895851247,
"puct/buffer_timestep/min": -1.0,
"puct/buffer_timestep/max": 22.0,
"puct/buffer_construction_len/mean": 78.79521276595744,
"puct/buffer_construction_len/std": 11.422406053941245,
"puct/buffer_construction_len/min": 42.0,
"puct/buffer_construction_len/max": 143.0,
"puct/sampled_value/mean": -0.38094155046750566,
"puct/sampled_value/std": 2.276995178380485e-09,
"puct/sampled_value/min": -0.38094155241092964,
"puct/sampled_value/max": -0.3809415470904396,
"puct/sampled_timestep/mean": 22.0,
"puct/sampled_timestep/std": 0.0,
"puct/sampled_timestep/min": 22.0,
"puct/sampled_timestep/max": 22.0,
"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": 4378.611856937408,
"env/all/ac_tokens_per_turn": 9112.66796875,
"env/all/ob_tokens_per_turn": 1256.0,
"env/all/turns_per_episode": 1.0,
"env/all/total_episodes": 512,
"env/all/total_turns": 512,
"env/all/total_ac_tokens": 4665686,
"env/all/total_ob_tokens": 643072,
"env/all/time/sampling_mean": 474.40575103089213,
"env/all/time/sampling_max": 697.0114929676056,
"env/all/time/env_step_mean": 1699.136680111289,
"env/all/time/env_step_max": 3699.9477863311768,
"env/all/reward/mean": 0.5780589035179919,
"env/all/reward/max": 2.62507466975591,
"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": 0.5780589035179919,
"env/all/correctness": 0.228515625,
"env/all/correctness/min": 0.0,
"env/all/correctness/max": 1.0,
"env/all/raw_score": 0.3975674014906814,
"env/all/raw_score/min": 0.38094153483345955,
"env/all/raw_score/max": 0.5,
"env/all/initial_raw_score": -0.3809415504675056,
"env/all/initial_raw_score/min": -0.38094155241092964,
"env/all/initial_raw_score/max": -0.3809415470904396,
"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\n# Pre-defined initial guess (from the problem)\ninitial_h_values = [0.8, 0.8, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.6, 0.6, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.4, 0.4, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.9, 0.7,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.2, 0.0, 0.0, 0.0, 0.0, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.6, 0.0, 0.0, 0.0, 0.0, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,\n 0.0, 0.0, 0.0, 0.0, 0.0]\n\ndef generate_random_binary(n_points):\n indices = np.random.choice(n_points, size=n_points // 2, replace=False)\n h = np.zeros(n_points)\n h[indices] = 1.0\n return h\n\ndef run(seed=42, budget_s=1000, **kwargs):\n np.random.seed(seed)\n \n n_points = len(initial_h_values)\n dx = 2.0 / n_points\n required_sum = n_points / 2.0\n\n def generate_alternating_guess(n_points):\n return np.array([1.0 if i % 2 == 0 else 0.0 for i in range(n_points)])\n\n # Generate three diverse initial guesses\n initial_guesses = [\n initial_h_values.copy(), # Provided initial guess\n generate_alternating_guess(n_points), # Alternating pattern\n generate_random_binary(n_points) # Random binary pattern\n ]\n\n best_c5 = float('inf')\n best_h = None\n\n for guess in initial_guesses:\n # Ensure the initial guess is properly scaled\n guess = (guess / np.sum(guess)) * required_sum if np.sum(guess) > 0 else guess\n\n def objective(x):\n h = x\n h1 = 1.0 - h\n corr = correlate(h, h1, mode='full')\n max_corr = np.max(corr)\n penalty = 100.0 * (np.sum(h) - required_sum) ** 2 # Stronger penalty\n return max_corr * dx + penalty\n\n bounds = [(0.0, 1.0) for _ in range(n_points)]\n\n result = differential_evolution(\n objective,\n bounds,\n strategy='best1bin',\n popsize=100, # Larger population for better exploration\n maxiter=500, # Sufficient iterations\n tol=1e-8,\n mutation=(0.5, 0.8), # Mutation range\n recombination=0.8,\n disp=False,\n polish=True,\n x0=guess # Use initial guess for warm start\n )\n\n h_current = result.x\n h1_current = 1.0 - h_current\n corr_current = correlate(h_current, h1_current, mode='full')\n max_corr_current = np.max(corr_current)\n true_c5 = max_corr_current * dx\n\n if true_c5 < best_c5:\n best_c5 = true_c5\n best_h = h_current\n\n # Return the best found solution\n return best_h, best_c5, n_points\n```",
"env/all/time/policy": 474.40575103089213,
"env/all/time/policy/min": 226.13372039794922,
"env/all/time/policy/max": 697.0114929676056,
"env/all/time/env_step": 1699.136680111289,
"env/all/time/env_step/min": 0.006023406982421875,
"env/all/time/env_step/max": 3699.9477863311768,
"env/all/time/reward_compute": 4.1117891669273376e-07,
"env/all/time/reward_compute/min": 2.7194619178771973e-07,
"env/all/time/reward_compute/max": 7.003545761108398e-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": 0.03390217199921608,
"advantage/min": -0.9973665475845337,
"advantage/max": 6.477557182312012,
"time/assemble_training_data": 7.135315656661987,
"time/kl_vs_base": 130.33649349212646,
"kl_policy_base": 0.0008708504028618336,
"time/train": 963.7112393379211,
"time/save_checkpoint": 11.815401554107666,
"time/total": 5493.163318872452
}
[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