final (runs/alephlm0-s0e-attn/config)
Browse files
runs/alephlm0-s0e-attn/config/config.json
ADDED
|
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"profile": "_applied",
|
| 3 |
+
"run_name": "alephlm0-s0e-attn",
|
| 4 |
+
"hf_repo": "AbstractPhil/alephlm-0",
|
| 5 |
+
"sources": [
|
| 6 |
+
{
|
| 7 |
+
"repo": "AbstractPhil/conceptual-captions-12m-webdataset-berts",
|
| 8 |
+
"n_chunks": 1,
|
| 9 |
+
"chunk_rows": 500000,
|
| 10 |
+
"missing": {}
|
| 11 |
+
}
|
| 12 |
+
],
|
| 13 |
+
"experts": [
|
| 14 |
+
"bert",
|
| 15 |
+
"modern",
|
| 16 |
+
"roberta",
|
| 17 |
+
"albert",
|
| 18 |
+
"distil"
|
| 19 |
+
],
|
| 20 |
+
"ref_expert": "bert",
|
| 21 |
+
"ref_hf_name": "google-bert/bert-base-uncased",
|
| 22 |
+
"require_all_experts": true,
|
| 23 |
+
"caption_field": "caption_llava",
|
| 24 |
+
"caption_field_candidates": [
|
| 25 |
+
"caption_llava",
|
| 26 |
+
"caption",
|
| 27 |
+
"caption_llava_short"
|
| 28 |
+
],
|
| 29 |
+
"work_dir": "/workspace/alephlm0",
|
| 30 |
+
"keep_expert_shards": false,
|
| 31 |
+
"disk_floor_gb": 25.0,
|
| 32 |
+
"ram_resident": true,
|
| 33 |
+
"preflight": true,
|
| 34 |
+
"artifact_repo": "AbstractPhil/captionbert-8192-v2",
|
| 35 |
+
"targets_repo": "AbstractPhil/captionbert-8192-v2-consensus",
|
| 36 |
+
"push_targets": true,
|
| 37 |
+
"hf_push": true,
|
| 38 |
+
"push_every_min": 30.0,
|
| 39 |
+
"keep_local_ckpts": 3,
|
| 40 |
+
"parity_chunk": 0,
|
| 41 |
+
"parity_n": 64,
|
| 42 |
+
"parity_min_cos": 0.999,
|
| 43 |
+
"fit_chunks": [
|
| 44 |
+
0,
|
| 45 |
+
11,
|
| 46 |
+
22,
|
| 47 |
+
33,
|
| 48 |
+
44,
|
| 49 |
+
55
|
| 50 |
+
],
|
| 51 |
+
"fit_rows_per_chunk": 4000,
|
| 52 |
+
"holdout_chunks": [
|
| 53 |
+
60,
|
| 54 |
+
61
|
| 55 |
+
],
|
| 56 |
+
"fit_seed": 0,
|
| 57 |
+
"d_model": 512,
|
| 58 |
+
"n_heads": 8,
|
| 59 |
+
"n_layers": 12,
|
| 60 |
+
"d_ff": 2048,
|
| 61 |
+
"max_len": 8192,
|
| 62 |
+
"output_dim": 768,
|
| 63 |
+
"dropout": 0.1,
|
| 64 |
+
"pooling": "attn",
|
| 65 |
+
"max_tokens": 256,
|
| 66 |
+
"length_bucketing": true,
|
| 67 |
+
"bucket_window": 64,
|
| 68 |
+
"grad_checkpointing": true,
|
| 69 |
+
"vram_probe": true,
|
| 70 |
+
"epochs": 2,
|
| 71 |
+
"batch_size": 2048,
|
| 72 |
+
"lr": 0.0006,
|
| 73 |
+
"min_lr": 1e-06,
|
| 74 |
+
"warmup_steps": 100,
|
| 75 |
+
"grad_clip": 1.0,
|
| 76 |
+
"seed": 1772344645,
|
| 77 |
+
"amp": true,
|
| 78 |
+
"num_workers": 0,
|
| 79 |
+
"log_every": 50,
|
| 80 |
+
"eval_every": 100,
|
| 81 |
+
"ckpt_every_min": 20.0,
|
| 82 |
+
"fresh_start": true,
|
| 83 |
+
"allow_mismatched_resume": false,
|
| 84 |
+
"nce_weight": 1.0,
|
| 85 |
+
"mse_weight": 1.0,
|
| 86 |
+
"nce_temperature": 0.07,
|
| 87 |
+
"cv_weight": 0.0,
|
| 88 |
+
"cv_target": 0.084,
|
| 89 |
+
"run_stage0": false,
|
| 90 |
+
"run_stage1": false,
|
| 91 |
+
"run_stage2": false,
|
| 92 |
+
"run_stage3": true,
|
| 93 |
+
"resume": true
|
| 94 |
+
}
|