{ "preset": "mini-beatrix-2s-control-attn-fp16", "model_config": { "name": "mini-beatrix-2s-control-attn-fp16", "d_model": 1024, "n_layers": 20, "n_heads": 16, "context": 4096, "vocab_size": 256, "tokenizer": "byte-trigram", "hub_layers": [], "hub_K": 64, "hub_D": 128, "tau": 0.1, "bank_experts": 3, "bank_ff": 1024, "head_K": 256, "head_D": 256, "gate_init": -3.0, "tie_embeddings": false, "hub_chunk": 256, "hub_const": 4, "hub_ckpt": 0, "fusion": null, "qk_norm": "", "attn_fp32": true, "attn_kernel": "fp16" }, "created_utc": "2026-10-10T00:19:49Z", "updated_utc": "2026-10-10T02:17:15Z", "steps": 18251, "tokens_seen": 4784390144, "wall_hours": 1.94674377726184, "phases": [ { "name": "warmup_wikitext", "dataset": "wikitext-103", "planned_tokens": 300000000, "status": "done", "tokens_done": 300154880 }, { "name": "fineweb_main", "dataset": "fineweb-edu", "planned_tokens": 5000000000, "status": "active", "tokens_done": 4484235264 }, { "name": "curriculum_s0", "dataset": "curriculum-s0", "planned_tokens": 700000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s1", "dataset": "curriculum-s1", "planned_tokens": 700000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s2", "dataset": "curriculum-s2", "planned_tokens": 1000000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s3", "dataset": "curriculum-s3", "planned_tokens": 1200000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s4", "dataset": "curriculum-s4", "planned_tokens": 1200000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s5", "dataset": "curriculum-s5", "planned_tokens": 1000000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s6", "dataset": "curriculum-s6", "planned_tokens": 800000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s7", "dataset": "curriculum-s7", "planned_tokens": 1400000000, "status": "planned", "tokens_done": 0 }, { "name": "curriculum_s8", "dataset": "curriculum-s8", "planned_tokens": 800000000, "status": "planned", "tokens_done": 0 }, { "name": "anneal_nochat", "dataset": "anneal-nochat", "planned_tokens": 1000000000, "status": "planned", "tokens_done": 0 }, { "name": "anneal_mix", "dataset": "anneal-mix", "planned_tokens": 1000000000, "status": "planned", "tokens_done": 0 } ], "checkpoints": [ { "step": 18000, "tokens": 4718592000, "kind": "safetensors", "path": "checkpoints/step_00018000.safetensors", "val_bpb": 1.0603969668180775, "utc": "2026-10-10T02:03:12Z" }, { "step": 18000, "tokens": 4718592000, "kind": "resume", "path": "resume/latest.pt", "val_bpb": null, "utc": "2026-10-10T02:03:12Z" }, { "step": 18251, "tokens": 4784390144, "kind": "safetensors", "path": "checkpoints/step_00018251.safetensors", "val_bpb": 1.0603969668180775, "utc": "2026-10-10T02:16:44Z" }, { "step": 18251, "tokens": 4784390144, "kind": "fp8", "path": "checkpoints/fp8/step_00018251.safetensors", "val_bpb": null, "utc": "2026-10-10T02:16:50Z" }, { "step": 18251, "tokens": 4784390144, "kind": "resume", "path": "resume/latest.pt", "val_bpb": null, "utc": "2026-10-10T02:16:50Z" } ], "data_state": { "dataset": "fineweb-edu", "rows_consumed": 123047, "epoch": 0 }, "notes": [ { "utc": "2026-10-10T00:19:49Z", "msg": "manifest created" }, { "utc": "2026-10-10T00:19:49Z", "msg": "WEIGHTS-ONLY START at step 16,000 from AbstractPhil/alephllm-mini-beatrix-training/mini-beatrix-2s-control/checkpoints/step_00016000.safetensors: fresh optimizer states; the curriculum cursor set from the step (active phase 'fineweb_main', 4.194B tokens counted); its stream opens at seed offset 7919 (a data-order discontinuity by design: the source run consumed the shuffle head)" } ], "halt": null, "data_plane": { "data_scale": 1.0, "epoch_cap": 4.0, "rebalance_to": null, "recipe_hash": "1b0cebcb3490", "minted_lexicon": null }, "init_from": { "source": "AbstractPhil/alephllm-mini-beatrix-training/mini-beatrix-2s-control/checkpoints/step_00016000.safetensors", "step": 16000, "phase": "fineweb_main", "seed_offset": 7919, "tokens_per_step": 262144, "live_tokens_per_step": 262144, "weights": "the shipped bf16 checkpoint, upcast to the fp32 masters", "optimizers": "fresh (Muon momentum and Adam moments start at zero)", "guards_added": [], "utc": "2026-10-10T00:19:49Z" } }