| [ckpt] runs/lta_owt_gpt2cached_len1024_fixedwrong70_c1024_ddit768x12_muon_ema_gbs512_8gpu_1m_20260513_171557/eval_snapshot_step_0091000_ema.pt step=91000 | |
| [decode-base] n=16 max_len=1024 steps=128 model_t=flow | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 2/16 | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 4/16 | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 6/16 | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 8/16 | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 10/16 | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 12/16 | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 14/16 | |
| [decode] temp=1.45 final=blend rule=dirichlet_resample support=1 semantic=1.5 anchor=state cfg=0/1@0:uniform decode_freq_penalty=0/0/0-1^1 final_sample=argmax/1/k64/p0.95 freq_penalty=0/0/0 start_t=0 start_init=noise generated 16/16 | |
| [summary] {"type": "summary", "checkpoint": "runs/lta_owt_gpt2cached_len1024_fixedwrong70_c1024_ddit768x12_muon_ema_gbs512_8gpu_1m_20260513_171557/eval_snapshot_step_0091000_ema.pt", "step": 91000, "decode": {"steps": 128, "model_t_mode": "flow", "decode_rule": "dirichlet_resample", "support_power": 1.0, "semantic_power": 1.5, "anchor_mode": "state", "cfg_scale": 0.0, "cfg_power": 1.0, "cfg_start": 0.0, "cfg_prior": "uniform", "decode_freq_penalty_alpha": 0.0, "decode_freq_penalty_beta": 0.0, "decode_freq_penalty_floor": 0.0, "decode_freq_penalty_start": 0.0, "decode_freq_penalty_end": 1.0, "decode_freq_penalty_power": 1.0, "start_t": 0.0, "start_init": "noise", "noise_init": "dirichlet", "noise_sigma": -1.0, "dirichlet_concentration": 1.0, "concentration_min": 1.0, "concentration_max": 1024.0, "target_prob": 1.0, "endpoint_temp": 1.45, "final_from": "blend", "final_sample_mode": "argmax", "final_sample_temp": 1.0, "final_top_k": 64, "final_top_p": 0.95, "final_freq_penalty_alpha": 0.0, "final_freq_penalty_beta": 0.0, "final_freq_penalty_floor": 0.0, "lock_bos": false, "n_samples": 16, "seed": 20260514}, "raw_genppl": {"ppl": 9.15153130996079, "nll_per_token": 2.2139212215647976, "tokens": 4080, "kept_samples": 16, "total_samples": 16, "empty_rate": 0.0, "skipped_samples": 0}, "stripped_genppl": {"ppl": 11.897193857793793, "nll_per_token": 2.4763025620404413, "tokens": 4080, "kept_samples": 16, "total_samples": 16, "empty_rate": 0.0, "skipped_samples": 0}, "diversity": {"sample_entropy": 2.4467596776773615, "unique_tokens": 209, "token_count": 16384, "distinct_1": 0.01275634765625, "distinct_2": 0.09567448680351906, "top_token_mass": 0.2486572265625}} | |
| [done] docs/lta_samples/metrics_20260514/fixedwrong70_step91000_decode_sweep_fast/step91000_ema_base_argmax.jsonl | |