{ "step": 17542, "tokens_seen": 10346692608, "raw_tokens_seen": 10346692608, "tst": { "bag": 1, "steps": 0 }, "train_seconds": 547795.9212001314, "ngram_canon_sha256": "0e9f25836e7da68ad664bb0908c7acba962458643ca92a9e2f2d75c54e8ebb2c", "skipped_updates": 0, "world_size": 1, "preset": "700m", "ngram_params": 1000000000, "args": { "preset": "700m", "ngram_params": 1000000000, "model_override": [ "d_model=768", "n_layers=56", "n_heads=6", "n_kv_heads=2", "head_dim=128", "gdn_heads=6", "gdn_head_dim=128", "ffn_dim=2304", "ngram.heads_per_order=4", "ngram.dim=32", "ngram.inject_layers=1,27", "ngram.canonical_tokens=True", "ngram.stochastic_round=True", "ngram.sign_pool=8192", "tie_embeddings=False" ], "tokenizer": "/exp_train/tokenizer/tokenizer.json", "data": "/exp_train/data/packed1-train.bin", "eval_data": "/exp_train/data/packed1-val.bin", "micro_tokens": 18432, "seq_len": 900, "grad_accum": 32, "steps": 17542, "stop_at": 0, "warmup_steps": 88, "decay_frac": 0.295, "min_lr_frac": 0.0, "tst_bag": 0, "tst_frac": 0.2, "optimizer": "dion3", "dion_lr": 0.02, "dion_fraction": 0.5, "dion_mu": 0.95, "dion_weight_decay": 0.01, "dion_triton": true, "dion_replicated": false, "resync_every": 50, "dion_compile": true, "scalar_lr": 0.003, "adamw_lr": 0.0003, "grad_clip": 1.0, "compile": "tail", "compile_mode": "default", "loss_chunks": 4, "residual_dtype": "float32", "fast_ce": true, "update_threads": 80, "ngram_sync": "default", "ddp_comm_hook": "none", "ddp_bucket_mb": 64, "fla_nosync": true, "layer_drop": 0.25, "layer_drop_dist": "ild", "layer_drop_schedule": "decreasing", "layer_drop_final": 0.05, "layer_drop_end": 0.705, "depth_attn": "off", "delta_block": true, "delta_blocks": 4, "delta_variant": "repo", "eval_every": 3508, "eval_tokens": 20000000, "eval_exit_at": 42, "ckpt_dir": "/exp_train/runs/packed1-500m-deep/ckpt", "ckpt_every": 877, "ckpt_minutes": 0.0, "keep": 2, "resume": "auto", "max_minutes": 0.0, "max_skips": 3, "metrics": "/exp_train/runs/packed1-500m-deep/metrics.jsonl", "output": "/exp_train/runs/packed1-500m-deep/results.jsonl", "run_name": "packed1-500m-deep-a0", "timing_skip": 2, "peak_tflops": 0.0, "gpu_mem_cap_gib": 59.0, "seed": 1234, "dist_backend": "nccl", "profile_micro": 0, "sync_debug": false, "wandb_project": "exp_train", "wandb_entity": "", "wandb_name": "packed1-500m-deep", "wandb_mode": "online", "hub_repo": "Ba2han/packed1-500m-deep", "hub_keep": 2, "hub_private": true, "hub_squash": true, "table_probe_rows": 65536, "eval_max_minutes": 30.0, "eval_ngram_ablation": true, "crash_at_step": -1, "crash_point": "mid-update" }, "model_config": { "vocab_size": 60800, "d_model": 768, "n_layers": 56, "full_attn_every": 4, "n_heads": 6, "n_kv_heads": 2, "head_dim": 128, "rope_theta": 1000000.0, "gdn_heads": 6, "gdn_head_dim": 128, "gdn_expand_v": 1.0, "gdn_conv_size": 4, "gdn_variant": "gdn", "gdn2_gate_rank": 0, "gdn2_lower_bound": 0.0, "ffn_dim": 2304, "norm_eps": 1e-06, "gated_residual": true, "tie_embeddings": false, "ngram": { "enabled": true, "orders": [ 2, 3 ], "heads_per_order": 4, "dim": 32, "total_params": 1000000000, "inject_layers": [ 1, 27 ], "table_dtype": "bfloat16", "lr": 0.02, "beta2": 0.99, "eps": 1e-08, "canonical_tokens": true, "stochastic_round": true, "sign_pool": 8192 } }, "data_blocks": 561351, "eval_batches": 1085, "saved_at": "2026-11-02 23:41:11", "checksums": { "optim_rank0": 14332326.093579413, "state_rank0": 318850.0, "model": 45744.409796107095, "ngram": -7987.134480161983 }, "format": "safetensors", "files": { "optim_rank0": [ "optim_rank0.safetensors", "optim_rank0.json" ], "state_rank0": [ "state_rank0.safetensors", "state_rank0.json" ], "model": [ "model.safetensors" ], "ngram": [ "ngram.safetensors" ], "assets": [ "tokenizer.json", "tokenizer_config.json", "special_tokens_map.json", "config.json" ] }, "write_seconds": 11.22276174009312 }