Update tokenizer, configuration and training reports
Browse files- reports/run/resolved.json +83 -0
- reports/status.json +1 -1
reports/run/resolved.json
ADDED
|
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": {
|
| 3 |
+
"architectures": [
|
| 4 |
+
"BalancedLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 0,
|
| 9 |
+
"dtype": "float32",
|
| 10 |
+
"eos_token_id": 0,
|
| 11 |
+
"head_dim": 64,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 576,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 2496,
|
| 16 |
+
"layer_types": [
|
| 17 |
+
"full_attention",
|
| 18 |
+
"full_attention",
|
| 19 |
+
"full_attention",
|
| 20 |
+
"full_attention",
|
| 21 |
+
"full_attention",
|
| 22 |
+
"full_attention",
|
| 23 |
+
"full_attention",
|
| 24 |
+
"full_attention",
|
| 25 |
+
"full_attention",
|
| 26 |
+
"full_attention",
|
| 27 |
+
"full_attention",
|
| 28 |
+
"full_attention",
|
| 29 |
+
"full_attention",
|
| 30 |
+
"full_attention",
|
| 31 |
+
"full_attention",
|
| 32 |
+
"full_attention",
|
| 33 |
+
"full_attention",
|
| 34 |
+
"full_attention",
|
| 35 |
+
"full_attention",
|
| 36 |
+
"full_attention",
|
| 37 |
+
"full_attention",
|
| 38 |
+
"full_attention",
|
| 39 |
+
"full_attention",
|
| 40 |
+
"full_attention",
|
| 41 |
+
"full_attention",
|
| 42 |
+
"full_attention"
|
| 43 |
+
],
|
| 44 |
+
"max_position_embeddings": 2048,
|
| 45 |
+
"max_window_layers": 28,
|
| 46 |
+
"model_type": "slayer_balanced",
|
| 47 |
+
"num_attention_heads": 9,
|
| 48 |
+
"num_hidden_layers": 26,
|
| 49 |
+
"num_key_value_heads": 3,
|
| 50 |
+
"pad_token_id": null,
|
| 51 |
+
"rms_norm_eps": 1e-06,
|
| 52 |
+
"rope_parameters": {
|
| 53 |
+
"rope_theta": 10000.0,
|
| 54 |
+
"rope_type": "default"
|
| 55 |
+
},
|
| 56 |
+
"sliding_window": null,
|
| 57 |
+
"tie_word_embeddings": true,
|
| 58 |
+
"transformers_version": "5.16.1",
|
| 59 |
+
"use_cache": true,
|
| 60 |
+
"use_sliding_window": false,
|
| 61 |
+
"vocab_size": 24576
|
| 62 |
+
},
|
| 63 |
+
"train": {
|
| 64 |
+
"seed": 1337,
|
| 65 |
+
"block": 2048,
|
| 66 |
+
"micro_batch": 16,
|
| 67 |
+
"accumulation": 2,
|
| 68 |
+
"adam_lr": 0.0015,
|
| 69 |
+
"muon_lr": 0.02,
|
| 70 |
+
"weight_decay": 0.1,
|
| 71 |
+
"clip": 1.0,
|
| 72 |
+
"checkpoint_every": 500,
|
| 73 |
+
"log_every": 20,
|
| 74 |
+
"compile": true
|
| 75 |
+
},
|
| 76 |
+
"parameters": 149333081,
|
| 77 |
+
"world_size": 8,
|
| 78 |
+
"tokens_per_update": 524288,
|
| 79 |
+
"total_steps": 59368,
|
| 80 |
+
"total_tokens": 31125929984,
|
| 81 |
+
"warmup_steps": 2000,
|
| 82 |
+
"initialization": "from_scratch"
|
| 83 |
+
}
|
reports/status.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"stage":"
|
|
|
|
| 1 |
+
{"stage": "training", "initialization": "from_scratch", "planned_tokens": 31125929984, "parameters": 149333081, "ready_by_utc": "2026-10-01T13:00:00+00:00"}
|