Sync Seed-72 Mixed checkpoint-44000
Browse files- FINAL5_20261014/seed72_mixed/checkpoint-44000/config.json +46 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_epoch_start_state_rank0.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_epoch_start_state_rank1.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_epoch_start_state_rank2.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_gen_state_rank0.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_gen_state_rank1.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_gen_state_rank2.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/model.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/rng_state_rank0.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/rng_state_rank1.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/rng_state_rank2.pt +3 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/training_config.json +74 -0
- FINAL5_20261014/seed72_mixed/checkpoint-44000/training_state.pt +3 -0
FINAL5_20261014/seed72_mixed/checkpoint-44000/config.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"vocab_size": 151936,
|
| 3 |
+
"hidden_size": 1536,
|
| 4 |
+
"intermediate_size": 6144,
|
| 5 |
+
"num_hidden_layers": 28,
|
| 6 |
+
"num_attention_heads": 12,
|
| 7 |
+
"num_key_value_heads": 4,
|
| 8 |
+
"head_dim": 128,
|
| 9 |
+
"hidden_act": "silu",
|
| 10 |
+
"max_position_embeddings": 32768,
|
| 11 |
+
"initializer_range": 0.02,
|
| 12 |
+
"rms_norm_eps": 1e-06,
|
| 13 |
+
"use_cache": true,
|
| 14 |
+
"tie_word_embeddings": true,
|
| 15 |
+
"rope_theta": 1000000,
|
| 16 |
+
"rope_scaling": null,
|
| 17 |
+
"positional": "rope",
|
| 18 |
+
"qk_gain_fp32": true,
|
| 19 |
+
"fp32_master": "none",
|
| 20 |
+
"rope_train_position_scale_set": null,
|
| 21 |
+
"attention_bias": false,
|
| 22 |
+
"attention_dropout": 0.0,
|
| 23 |
+
"bos_token_id": 151643,
|
| 24 |
+
"eos_token_id": 151643,
|
| 25 |
+
"checkpoint_activations": false,
|
| 26 |
+
"attention_type": "sparse_stochastic",
|
| 27 |
+
"sparse_block_size": 128,
|
| 28 |
+
"sparse_num_random_blocks": 35,
|
| 29 |
+
"sparse_mask_pool_size": 10000,
|
| 30 |
+
"sparse_max_seq_length": 8192,
|
| 31 |
+
"sparse_layer_indices": null,
|
| 32 |
+
"sparse_mask_cache_dir": "../../sparse/mask_cache_8k_mixed_seed72",
|
| 33 |
+
"sparse_mask_seed": 72,
|
| 34 |
+
"sparse_mixed_nrb_set": [
|
| 35 |
+
63,
|
| 36 |
+
47,
|
| 37 |
+
31,
|
| 38 |
+
21,
|
| 39 |
+
11
|
| 40 |
+
],
|
| 41 |
+
"moba_block_size": 128,
|
| 42 |
+
"moba_top_k": 32,
|
| 43 |
+
"asentmax_alpha": 1.5,
|
| 44 |
+
"asentmax_learn_temp": true,
|
| 45 |
+
"asentmax_init_temp": 1.0
|
| 46 |
+
}
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_epoch_start_state_rank0.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0656ddbff730b3ef739cdb4d980f54ff169c931544ce7a093c21a7dc6fa6f49e
|
| 3 |
+
size 6858
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_epoch_start_state_rank1.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c8f9144eba7ef521ca55f126341e4d11d4584a2c550212432b63364585392f0d
|
| 3 |
+
size 6858
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_epoch_start_state_rank2.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:587ab30e72f1d5694ef66c539c75743ea5b18c1d61abb5f03cfa4b66d9ef6119
|
| 3 |
+
size 6858
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_gen_state_rank0.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b21b4160791b11c535b1414390199117f5857a94ac58d7b970468583a72bb0bc
|
| 3 |
+
size 6802
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_gen_state_rank1.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:32ec129fe94d28aa0ddb1827422f7723c72b31ebc7a9306e98894b6131bb7fe9
|
| 3 |
+
size 6802
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/loader_gen_state_rank2.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9c5588d56d5c8519bb2c45d4b62d3684da8142cd70467b7c6b989cd944f4e533
|
| 3 |
+
size 6802
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:93e49691cc61baecf002591de2456b35b24c4088126ae0a62bd199bd2e4f4bb9
|
| 3 |
+
size 2404833883
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/rng_state_rank0.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5511be6d04d8f7e269fb0e5e4891be12cb296d1910425f419edec1f656a98bf7
|
| 3 |
+
size 14821
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/rng_state_rank1.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:61070ecc5635d95a84794e5e85c825aa18c4e007f303d7522ef3d55f429e7c2b
|
| 3 |
+
size 14821
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/rng_state_rank2.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:848353d56afb2726c65fdbd03b7aeb4fd39fe162f825849c0d5d81939677cc0d
|
| 3 |
+
size 14821
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/training_config.json
ADDED
|
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_path": "../../qwen3_06b_base",
|
| 3 |
+
"model_dtype": "bfloat16",
|
| 4 |
+
"init_from_scratch": true,
|
| 5 |
+
"train_path": "../../data/fineweb_large/tokenized",
|
| 6 |
+
"val_path": null,
|
| 7 |
+
"max_seq_length": 8192,
|
| 8 |
+
"num_workers": 0,
|
| 9 |
+
"data_seed": 72,
|
| 10 |
+
"output_dir": "models/FINAL5_20261014/1p2b_mixed_qkgainfp32_s72/run_20261015_193033",
|
| 11 |
+
"resume_from_checkpoint": "models/FINAL5_20261014/1p2b_mixed_qkgainfp32_s72/run_20261015_193033/checkpoint-41000",
|
| 12 |
+
"force_new_run": false,
|
| 13 |
+
"num_epochs": 1,
|
| 14 |
+
"max_steps": 50462,
|
| 15 |
+
"per_device_batch_size": 1,
|
| 16 |
+
"gradient_accumulation_steps": 20,
|
| 17 |
+
"global_batch_size": 60,
|
| 18 |
+
"learning_rate": 0.0005,
|
| 19 |
+
"weight_decay": 0.01,
|
| 20 |
+
"warmup_steps": 750,
|
| 21 |
+
"lr_scheduler": "cosine",
|
| 22 |
+
"max_grad_norm": 1.0,
|
| 23 |
+
"save_steps": 1000,
|
| 24 |
+
"eval_steps": 250,
|
| 25 |
+
"eval_batches": 50,
|
| 26 |
+
"logging_steps": 10,
|
| 27 |
+
"seed": 72,
|
| 28 |
+
"verbose": true,
|
| 29 |
+
"checkpoint_activations": false,
|
| 30 |
+
"resume_micro_step_override": 820000,
|
| 31 |
+
"strategy": "ddp",
|
| 32 |
+
"mixed_precision": "bf16",
|
| 33 |
+
"wandb_enabled": false,
|
| 34 |
+
"wandb_project": "stochastic_sparsity",
|
| 35 |
+
"wandb_name": "FINAL5_20261014 1p2b_mixed_qkgainfp32_s72",
|
| 36 |
+
"wandb_tags": [
|
| 37 |
+
"qwen3",
|
| 38 |
+
"1.2b",
|
| 39 |
+
"8k",
|
| 40 |
+
"p2",
|
| 41 |
+
"mixed",
|
| 42 |
+
"seed72"
|
| 43 |
+
],
|
| 44 |
+
"hidden_size": 1536,
|
| 45 |
+
"intermediate_size": 6144,
|
| 46 |
+
"num_attention_heads": 12,
|
| 47 |
+
"num_key_value_heads": 4,
|
| 48 |
+
"num_hidden_layers": null,
|
| 49 |
+
"head_dim": null,
|
| 50 |
+
"attention_dropout": null,
|
| 51 |
+
"qk_gain_fp32": true,
|
| 52 |
+
"positional": "rope",
|
| 53 |
+
"fp32_master": "none",
|
| 54 |
+
"keep_last_checkpoints": 3,
|
| 55 |
+
"keep_every_n_steps": 10000,
|
| 56 |
+
"rope_train_position_scale_set": null,
|
| 57 |
+
"attention_type": "sparse_stochastic",
|
| 58 |
+
"sparse_block_size": 128,
|
| 59 |
+
"sparse_num_random_blocks": 35,
|
| 60 |
+
"sparse_mask_pool_size": 10000,
|
| 61 |
+
"sparse_val_dense_compare": false,
|
| 62 |
+
"sparse_layer_indices": null,
|
| 63 |
+
"sparse_mask_cache_dir": "../../sparse/mask_cache_8k_mixed_seed72",
|
| 64 |
+
"sparse_mask_seed": 72,
|
| 65 |
+
"sparse_mixed_nrb_set": [
|
| 66 |
+
63,
|
| 67 |
+
47,
|
| 68 |
+
31,
|
| 69 |
+
21,
|
| 70 |
+
11
|
| 71 |
+
],
|
| 72 |
+
"dense_mode": false,
|
| 73 |
+
"dense_step_intervals": null
|
| 74 |
+
}
|
FINAL5_20261014/seed72_mixed/checkpoint-44000/training_state.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e4986e8cae219ef44b69b95dd406fc5fd830e63108ac7759a14712e49f62b9c5
|
| 3 |
+
size 4809712543
|