Add 1.2B Dense seed-70 step-25000 resume checkpoint for RCP transfer
Browse files- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/config.json +42 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank0.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank1.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank2.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank3.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/model.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank0.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank1.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank2.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank3.pt +3 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/train_config.json +69 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/trainer_state.json +9 -0
- checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/training_state.pt +3 -0
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/config.json
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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_norm": "on",
|
| 19 |
+
"qk_gain_fp32": true,
|
| 20 |
+
"norm_gain_fp32": false,
|
| 21 |
+
"qk_gain_mode": "vector",
|
| 22 |
+
"rope_train_position_scale_set": null,
|
| 23 |
+
"attention_bias": false,
|
| 24 |
+
"attention_dropout": 0.0,
|
| 25 |
+
"bos_token_id": 151643,
|
| 26 |
+
"eos_token_id": 151643,
|
| 27 |
+
"checkpoint_activations": true,
|
| 28 |
+
"attention_type": "sparse_stochastic",
|
| 29 |
+
"sparse_block_size": 128,
|
| 30 |
+
"sparse_num_random_blocks": 31,
|
| 31 |
+
"sparse_mask_pool_size": 1,
|
| 32 |
+
"sparse_max_seq_length": 8192,
|
| 33 |
+
"sparse_layer_indices": null,
|
| 34 |
+
"sparse_mask_cache_dir": null,
|
| 35 |
+
"sparse_mask_seed": 0,
|
| 36 |
+
"sparse_mixed_nrb_set": null,
|
| 37 |
+
"moba_block_size": 128,
|
| 38 |
+
"moba_top_k": 32,
|
| 39 |
+
"asentmax_alpha": 1.5,
|
| 40 |
+
"asentmax_learn_temp": true,
|
| 41 |
+
"asentmax_init_temp": 1.0
|
| 42 |
+
}
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank0.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d3d3d87a844d4fd6548dfc4c1235a1ee3e262506bc6260da75e7cbf9d9da819c
|
| 3 |
+
size 6802
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank1.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c2c029e9127ff4c346ab99e6af8377a6903003909fec0b073b70bfafc01d4517
|
| 3 |
+
size 6802
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank2.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:917e752ef93a10f1da3dc49f9c18adc43ae1bb568ee13d528c9824ee4513b511
|
| 3 |
+
size 6802
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/loader_gen_state_rank3.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:389dee6373df6f406c9ae897b97bc5acc1b426bc8c9810e2754f7d2f892a256f
|
| 3 |
+
size 6802
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4df93f77562ecb5b28a50c27b3e10c84e0b10609736333f76621d3c4a811cd3d
|
| 3 |
+
size 2404833883
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank0.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8f92dc12642f0e396c9aea648c347fadfcab8234a07ae545439e4a1e6521ed11
|
| 3 |
+
size 14821
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank1.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:59241ebd08eebd96b51f7a9bb0ab7333fc4665a17b548a25d73e390cb957c223
|
| 3 |
+
size 14821
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank2.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5f65c72fae8bdf73a4b4cd50d429dea879b957113952938b6982c2be4d55abd9
|
| 3 |
+
size 14821
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/rng_state_rank3.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8ed370cea6b4dd78fdab7056daa262fcf7c158dc34805993101942cfeec1072d
|
| 3 |
+
size 14821
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/train_config.json
ADDED
|
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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": 1,
|
| 9 |
+
"data_seed": 70,
|
| 10 |
+
"concatenate_seeded_8k": false,
|
| 11 |
+
"output_dir": "models/FINAL3_20260927/1p2b_dense_qkgainfp32_s70_trillium_resume_4gpu_ckpt6000/run_20260930_ckpt6000",
|
| 12 |
+
"resume_from_checkpoint": "models/FINAL3_20260927/1p2b_dense_qkgainfp32_s70_trillium_resume_4gpu_ckpt6000/run_20260930_ckpt6000/checkpoint-6000",
|
| 13 |
+
"force_new_run": false,
|
| 14 |
+
"num_epochs": 1,
|
| 15 |
+
"max_steps": 50462,
|
| 16 |
+
"per_device_batch_size": 1,
|
| 17 |
+
"gradient_accumulation_steps": 15,
|
| 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": 70,
|
| 28 |
+
"verbose": true,
|
| 29 |
+
"checkpoint_activations": true,
|
| 30 |
+
"resume_micro_step_override": 90000,
|
| 31 |
+
"strategy": "ddp",
|
| 32 |
+
"mixed_precision": "bf16",
|
| 33 |
+
"wandb_enabled": false,
|
| 34 |
+
"wandb_project": "stochastic_sparsity",
|
| 35 |
+
"wandb_name": "1p2b_dense_qkgainfp32_s70",
|
| 36 |
+
"wandb_tags": [
|
| 37 |
+
"qwen3",
|
| 38 |
+
"1.2b",
|
| 39 |
+
"8k",
|
| 40 |
+
"p2",
|
| 41 |
+
"dense",
|
| 42 |
+
"seed70"
|
| 43 |
+
],
|
| 44 |
+
"hidden_size": 1536,
|
| 45 |
+
"intermediate_size": 6144,
|
| 46 |
+
"num_attention_heads": 12,
|
| 47 |
+
"num_key_value_heads": 4,
|
| 48 |
+
"num_hidden_layers": 28,
|
| 49 |
+
"head_dim": null,
|
| 50 |
+
"attention_dropout": null,
|
| 51 |
+
"rope_theta": null,
|
| 52 |
+
"rope_train_position_scale_set": null,
|
| 53 |
+
"positional": "rope",
|
| 54 |
+
"qk_norm": "on",
|
| 55 |
+
"qk_gain_fp32": true,
|
| 56 |
+
"norm_gain_fp32": false,
|
| 57 |
+
"qk_gain_mode": "vector",
|
| 58 |
+
"attention_type": "sparse_stochastic",
|
| 59 |
+
"sparse_block_size": 128,
|
| 60 |
+
"sparse_num_random_blocks": 31,
|
| 61 |
+
"sparse_mask_pool_size": 1,
|
| 62 |
+
"sparse_val_dense_compare": false,
|
| 63 |
+
"sparse_layer_indices": null,
|
| 64 |
+
"sparse_mask_cache_dir": null,
|
| 65 |
+
"sparse_mask_seed": 0,
|
| 66 |
+
"sparse_mixed_nrb_set": null,
|
| 67 |
+
"dense_mode": true,
|
| 68 |
+
"dense_step_intervals": null
|
| 69 |
+
}
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/trainer_state.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"global_step": 25000,
|
| 3 |
+
"epoch": 0,
|
| 4 |
+
"micro_step": 375000,
|
| 5 |
+
"sampler_seed": 70,
|
| 6 |
+
"seed": 70,
|
| 7 |
+
"world_size": 4,
|
| 8 |
+
"optimizer_state_file": "training_state.pt"
|
| 9 |
+
}
|
checkpoints/FINAL3_20260927/part_t/1p2b_dense_qkgainfp32_s70/checkpoint-25000-trillium/training_state.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9c3000852374392d07739338bad444dd3d79b524e2542dbb0d917591826c80c4
|
| 3 |
+
size 4809712479
|