Upload mini-beatrix-3-la2-6-12-20-c249k/manifest.json with huggingface_hub
Browse files
mini-beatrix-3-la2-6-12-20-c249k/manifest.json
ADDED
|
@@ -0,0 +1,142 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"preset": "mini-beatrix-3-la2-6-12-20-c249k",
|
| 3 |
+
"model_config": {
|
| 4 |
+
"name": "mini-beatrix-3-la2-6-12-20-c249k",
|
| 5 |
+
"d_model": 1024,
|
| 6 |
+
"n_layers": 32,
|
| 7 |
+
"n_heads": 16,
|
| 8 |
+
"context": 4096,
|
| 9 |
+
"vocab_size": 256,
|
| 10 |
+
"tokenizer": "byte-trigram",
|
| 11 |
+
"hub_layers": [
|
| 12 |
+
0,
|
| 13 |
+
1,
|
| 14 |
+
2,
|
| 15 |
+
3,
|
| 16 |
+
4,
|
| 17 |
+
5,
|
| 18 |
+
6,
|
| 19 |
+
7,
|
| 20 |
+
8,
|
| 21 |
+
9,
|
| 22 |
+
10,
|
| 23 |
+
11,
|
| 24 |
+
12,
|
| 25 |
+
13,
|
| 26 |
+
14,
|
| 27 |
+
15,
|
| 28 |
+
16,
|
| 29 |
+
17,
|
| 30 |
+
18,
|
| 31 |
+
19,
|
| 32 |
+
20,
|
| 33 |
+
21,
|
| 34 |
+
22,
|
| 35 |
+
23,
|
| 36 |
+
24,
|
| 37 |
+
25,
|
| 38 |
+
26,
|
| 39 |
+
27,
|
| 40 |
+
28,
|
| 41 |
+
29,
|
| 42 |
+
30,
|
| 43 |
+
31
|
| 44 |
+
],
|
| 45 |
+
"hub_K": 64,
|
| 46 |
+
"hub_D": 128,
|
| 47 |
+
"tau": 0.1,
|
| 48 |
+
"bank_experts": 3,
|
| 49 |
+
"bank_ff": 1024,
|
| 50 |
+
"head_K": 256,
|
| 51 |
+
"head_D": 256,
|
| 52 |
+
"gate_init": -3.0,
|
| 53 |
+
"tie_embeddings": false,
|
| 54 |
+
"hub_chunk": 256,
|
| 55 |
+
"hub_const": 4,
|
| 56 |
+
"hub_ckpt": 0,
|
| 57 |
+
"fusion": null,
|
| 58 |
+
"qk_norm": "rms",
|
| 59 |
+
"attn_fp32": true,
|
| 60 |
+
"attn_kernel": "fp16",
|
| 61 |
+
"lookup_arm_sites": [
|
| 62 |
+
2,
|
| 63 |
+
6,
|
| 64 |
+
12,
|
| 65 |
+
20
|
| 66 |
+
]
|
| 67 |
+
},
|
| 68 |
+
"created_utc": "2026-10-11T01:04:31Z",
|
| 69 |
+
"updated_utc": "2026-10-11T01:35:52Z",
|
| 70 |
+
"steps": 250000,
|
| 71 |
+
"tokens_seen": 65536000000,
|
| 72 |
+
"wall_hours": 0.0,
|
| 73 |
+
"phases": [
|
| 74 |
+
{
|
| 75 |
+
"name": "v3_trunk",
|
| 76 |
+
"dataset": "fineweb-edu",
|
| 77 |
+
"planned_tokens": 65450541056,
|
| 78 |
+
"status": "done",
|
| 79 |
+
"tokens_done": 65450541056
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"name": "lookup_arm",
|
| 83 |
+
"dataset": "fineweb-recall-far-5",
|
| 84 |
+
"planned_tokens": 1048576000,
|
| 85 |
+
"status": "active",
|
| 86 |
+
"tokens_done": 85458944
|
| 87 |
+
}
|
| 88 |
+
],
|
| 89 |
+
"checkpoints": [
|
| 90 |
+
{
|
| 91 |
+
"step": 250000,
|
| 92 |
+
"tokens": 65536000000,
|
| 93 |
+
"kind": "safetensors",
|
| 94 |
+
"path": "checkpoints/step_00250000.safetensors",
|
| 95 |
+
"val_bpb": 0.9816863979239622,
|
| 96 |
+
"utc": "2026-10-11T01:35:47Z"
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"step": 250000,
|
| 100 |
+
"tokens": 65536000000,
|
| 101 |
+
"kind": "resume",
|
| 102 |
+
"path": "resume/latest.pt",
|
| 103 |
+
"val_bpb": null,
|
| 104 |
+
"utc": "2026-10-11T01:35:47Z"
|
| 105 |
+
}
|
| 106 |
+
],
|
| 107 |
+
"data_state": {
|
| 108 |
+
"dataset": "fineweb-recall-far-5",
|
| 109 |
+
"rows_consumed": 18026,
|
| 110 |
+
"epoch": 0
|
| 111 |
+
},
|
| 112 |
+
"notes": [
|
| 113 |
+
{
|
| 114 |
+
"utc": "2026-10-11T01:04:31Z",
|
| 115 |
+
"msg": "manifest created"
|
| 116 |
+
},
|
| 117 |
+
{
|
| 118 |
+
"utc": "2026-10-11T01:04:31Z",
|
| 119 |
+
"msg": "WEIGHTS-ONLY START at step 249,674 from AbstractPhil/alephllm-mini-beatrix-training/mini-beatrix-3-la2-6-12-20/checkpoints/step_00249674.safetensors: fresh optimizer states; the curriculum cursor set from the step (active phase 'lookup_arm', 65.451B tokens counted); its stream opens at seed offset 7919 (a data-order discontinuity by design: the source run consumed the shuffle head)"
|
| 120 |
+
}
|
| 121 |
+
],
|
| 122 |
+
"halt": null,
|
| 123 |
+
"data_plane": {
|
| 124 |
+
"data_scale": 4.0,
|
| 125 |
+
"epoch_cap": 2.0,
|
| 126 |
+
"rebalance_to": "generators",
|
| 127 |
+
"recipe_hash": "6d64747bdbd1",
|
| 128 |
+
"minted_lexicon": null
|
| 129 |
+
},
|
| 130 |
+
"init_from": {
|
| 131 |
+
"source": "AbstractPhil/alephllm-mini-beatrix-training/mini-beatrix-3-la2-6-12-20/checkpoints/step_00249674.safetensors",
|
| 132 |
+
"step": 249674,
|
| 133 |
+
"phase": "lookup_arm",
|
| 134 |
+
"seed_offset": 7919,
|
| 135 |
+
"tokens_per_step": 262144,
|
| 136 |
+
"live_tokens_per_step": 262144,
|
| 137 |
+
"weights": "the shipped bf16 checkpoint, upcast to the fp32 masters",
|
| 138 |
+
"optimizers": "fresh (Muon momentum and Adam moments start at zero)",
|
| 139 |
+
"guards_added": [],
|
| 140 |
+
"utc": "2026-10-11T01:04:31Z"
|
| 141 |
+
}
|
| 142 |
+
}
|