checkpoint step=19600 val_loss=2.6220
Browse files
checkpoints/step_00019600/training_meta.json
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"step": 19600,
|
| 3 |
+
"epoch": 0,
|
| 4 |
+
"total_tokens": 1284505600,
|
| 5 |
+
"val_loss": 2.6220367693901063,
|
| 6 |
+
"config": {
|
| 7 |
+
"vocab_size": 50304,
|
| 8 |
+
"d_model": 4096,
|
| 9 |
+
"n_layers": 56,
|
| 10 |
+
"n_heads": 32,
|
| 11 |
+
"n_kv_heads": 8,
|
| 12 |
+
"max_len": 2048,
|
| 13 |
+
"rope_theta": 5000000.0,
|
| 14 |
+
"ffn_hidden": 11008,
|
| 15 |
+
"use_moe": false,
|
| 16 |
+
"moe_experts": 8,
|
| 17 |
+
"moe_top_k": 2,
|
| 18 |
+
"moe_shared": 1,
|
| 19 |
+
"moe_hidden": 1408,
|
| 20 |
+
"moe_every": 4,
|
| 21 |
+
"moe_bias_speed": 0.001,
|
| 22 |
+
"use_qk_norm": true,
|
| 23 |
+
"use_structural_bias": true,
|
| 24 |
+
"n_struct_rel": 4,
|
| 25 |
+
"use_fim": true,
|
| 26 |
+
"fim_rate": 0.5,
|
| 27 |
+
"fim_pre_id": 50269,
|
| 28 |
+
"fim_suf_id": 50270,
|
| 29 |
+
"fim_mid_id": 50271,
|
| 30 |
+
"grad_checkpointing": true,
|
| 31 |
+
"lr": 0.0001,
|
| 32 |
+
"min_lr": 1e-05,
|
| 33 |
+
"weight_decay": 0.1,
|
| 34 |
+
"beta1": 0.9,
|
| 35 |
+
"beta2": 0.95,
|
| 36 |
+
"grad_clip": 1.0,
|
| 37 |
+
"warmup_tokens": 500000,
|
| 38 |
+
"target_total_tokens": 10000000000,
|
| 39 |
+
"micro_batch": 4,
|
| 40 |
+
"grad_accum": 8,
|
| 41 |
+
"log_every": 10,
|
| 42 |
+
"ckpt_every_steps": 200,
|
| 43 |
+
"eval_every_steps": 200,
|
| 44 |
+
"seed": 42,
|
| 45 |
+
"max_session_hours": 11.5
|
| 46 |
+
},
|
| 47 |
+
"timestamp": 1785469022.05987
|
| 48 |
+
}
|