KoHRM-Text-1.4B-lora-comp-swe-zero-30m-v1 / lora_step_2000_info.json
gyung's picture
Upload KoHRM LoRA adapter epoch_1
b634de6 verified
Raw
History Blame Contribute Delete
8.14 kB
{
"arch": {
"name": "baselines.hrm_nocarry_bp_warmup@HierarchicalReasoningModel",
"head": "lm_head@LMHead",
"half_layers": true,
"H_cycles": 2,
"L_cycles": 3,
"H_override": {},
"bp_warmup_ratio": 0.0,
"bp_max_steps": 5,
"n_layers": 32,
"hidden_size": 1536,
"num_heads": 12,
"expansion": 4,
"norm_type": "pre",
"norm_eps": 1e-06,
"rope_theta": 10000.0,
"pos_emb_type": "rope",
"init_type": "lecun_normal"
},
"data": {
"path": "/home/work/.data/hrm_text_prepared/kohrm_sft_comp_swe_zero_30m_v1",
"target_only": true
},
"global_batch_size": 16384,
"epochs": 1,
"lr": 8e-05,
"lr_min_ratio": 0.1,
"lr_warmup_steps": 100,
"weight_decay": 0.0,
"beta1": 0.9,
"beta2": 0.95,
"ema": null,
"fwd_bwd_dtype": "bfloat16",
"project_name": "KoHRM-Text-LoRA",
"run_name": "KoHRM-Text-1.4B-lora-comp-swe-zero-30m-v1",
"checkpoint_path": "/home/work/.data/hrm_text_lora/component_parallel_20260604_gbs16k/KoHRM-Text-1.4B-lora-comp-swe-zero-30m-v1",
"resume_from": "/home/work/.data/hrm_text_checkpoints/KoHRM-Text-1.4B-stage4d-korean-tool-finance-repeat3-gbs180",
"resume_epoch": null,
"resume_step": null,
"weights_only_resume_from_ema": true,
"resume_step_offset": 0,
"total_steps_override": null,
"skip_batches": 0,
"seed": 0,
"checkpoint_interval": 1,
"checkpoint_step_interval": 1000,
"checkpoint_keep_last": 2,
"log_interval": 5,
"lora": {
"rank": 16,
"alpha": 32.0,
"dropout": 0.0,
"target_suffixes": [
"gqkv_proj",
"o_proj",
"gate_up_proj",
"down_proj",
"lm_head"
]
},
"tag": "step_2000",
"step": 2000,
"matched_modules": [
"model.H_level.core.layers.0.attn.gqkv_proj",
"model.H_level.core.layers.0.attn.o_proj",
"model.H_level.core.layers.0.mlp.gate_up_proj",
"model.H_level.core.layers.0.mlp.down_proj",
"model.H_level.core.layers.1.attn.gqkv_proj",
"model.H_level.core.layers.1.attn.o_proj",
"model.H_level.core.layers.1.mlp.gate_up_proj",
"model.H_level.core.layers.1.mlp.down_proj",
"model.H_level.core.layers.2.attn.gqkv_proj",
"model.H_level.core.layers.2.attn.o_proj",
"model.H_level.core.layers.2.mlp.gate_up_proj",
"model.H_level.core.layers.2.mlp.down_proj",
"model.H_level.core.layers.3.attn.gqkv_proj",
"model.H_level.core.layers.3.attn.o_proj",
"model.H_level.core.layers.3.mlp.gate_up_proj",
"model.H_level.core.layers.3.mlp.down_proj",
"model.H_level.core.layers.4.attn.gqkv_proj",
"model.H_level.core.layers.4.attn.o_proj",
"model.H_level.core.layers.4.mlp.gate_up_proj",
"model.H_level.core.layers.4.mlp.down_proj",
"model.H_level.core.layers.5.attn.gqkv_proj",
"model.H_level.core.layers.5.attn.o_proj",
"model.H_level.core.layers.5.mlp.gate_up_proj",
"model.H_level.core.layers.5.mlp.down_proj",
"model.H_level.core.layers.6.attn.gqkv_proj",
"model.H_level.core.layers.6.attn.o_proj",
"model.H_level.core.layers.6.mlp.gate_up_proj",
"model.H_level.core.layers.6.mlp.down_proj",
"model.H_level.core.layers.7.attn.gqkv_proj",
"model.H_level.core.layers.7.attn.o_proj",
"model.H_level.core.layers.7.mlp.gate_up_proj",
"model.H_level.core.layers.7.mlp.down_proj",
"model.H_level.core.layers.8.attn.gqkv_proj",
"model.H_level.core.layers.8.attn.o_proj",
"model.H_level.core.layers.8.mlp.gate_up_proj",
"model.H_level.core.layers.8.mlp.down_proj",
"model.H_level.core.layers.9.attn.gqkv_proj",
"model.H_level.core.layers.9.attn.o_proj",
"model.H_level.core.layers.9.mlp.gate_up_proj",
"model.H_level.core.layers.9.mlp.down_proj",
"model.H_level.core.layers.10.attn.gqkv_proj",
"model.H_level.core.layers.10.attn.o_proj",
"model.H_level.core.layers.10.mlp.gate_up_proj",
"model.H_level.core.layers.10.mlp.down_proj",
"model.H_level.core.layers.11.attn.gqkv_proj",
"model.H_level.core.layers.11.attn.o_proj",
"model.H_level.core.layers.11.mlp.gate_up_proj",
"model.H_level.core.layers.11.mlp.down_proj",
"model.H_level.core.layers.12.attn.gqkv_proj",
"model.H_level.core.layers.12.attn.o_proj",
"model.H_level.core.layers.12.mlp.gate_up_proj",
"model.H_level.core.layers.12.mlp.down_proj",
"model.H_level.core.layers.13.attn.gqkv_proj",
"model.H_level.core.layers.13.attn.o_proj",
"model.H_level.core.layers.13.mlp.gate_up_proj",
"model.H_level.core.layers.13.mlp.down_proj",
"model.H_level.core.layers.14.attn.gqkv_proj",
"model.H_level.core.layers.14.attn.o_proj",
"model.H_level.core.layers.14.mlp.gate_up_proj",
"model.H_level.core.layers.14.mlp.down_proj",
"model.H_level.core.layers.15.attn.gqkv_proj",
"model.H_level.core.layers.15.attn.o_proj",
"model.H_level.core.layers.15.mlp.gate_up_proj",
"model.H_level.core.layers.15.mlp.down_proj",
"model.L_level.core.layers.0.attn.gqkv_proj",
"model.L_level.core.layers.0.attn.o_proj",
"model.L_level.core.layers.0.mlp.gate_up_proj",
"model.L_level.core.layers.0.mlp.down_proj",
"model.L_level.core.layers.1.attn.gqkv_proj",
"model.L_level.core.layers.1.attn.o_proj",
"model.L_level.core.layers.1.mlp.gate_up_proj",
"model.L_level.core.layers.1.mlp.down_proj",
"model.L_level.core.layers.2.attn.gqkv_proj",
"model.L_level.core.layers.2.attn.o_proj",
"model.L_level.core.layers.2.mlp.gate_up_proj",
"model.L_level.core.layers.2.mlp.down_proj",
"model.L_level.core.layers.3.attn.gqkv_proj",
"model.L_level.core.layers.3.attn.o_proj",
"model.L_level.core.layers.3.mlp.gate_up_proj",
"model.L_level.core.layers.3.mlp.down_proj",
"model.L_level.core.layers.4.attn.gqkv_proj",
"model.L_level.core.layers.4.attn.o_proj",
"model.L_level.core.layers.4.mlp.gate_up_proj",
"model.L_level.core.layers.4.mlp.down_proj",
"model.L_level.core.layers.5.attn.gqkv_proj",
"model.L_level.core.layers.5.attn.o_proj",
"model.L_level.core.layers.5.mlp.gate_up_proj",
"model.L_level.core.layers.5.mlp.down_proj",
"model.L_level.core.layers.6.attn.gqkv_proj",
"model.L_level.core.layers.6.attn.o_proj",
"model.L_level.core.layers.6.mlp.gate_up_proj",
"model.L_level.core.layers.6.mlp.down_proj",
"model.L_level.core.layers.7.attn.gqkv_proj",
"model.L_level.core.layers.7.attn.o_proj",
"model.L_level.core.layers.7.mlp.gate_up_proj",
"model.L_level.core.layers.7.mlp.down_proj",
"model.L_level.core.layers.8.attn.gqkv_proj",
"model.L_level.core.layers.8.attn.o_proj",
"model.L_level.core.layers.8.mlp.gate_up_proj",
"model.L_level.core.layers.8.mlp.down_proj",
"model.L_level.core.layers.9.attn.gqkv_proj",
"model.L_level.core.layers.9.attn.o_proj",
"model.L_level.core.layers.9.mlp.gate_up_proj",
"model.L_level.core.layers.9.mlp.down_proj",
"model.L_level.core.layers.10.attn.gqkv_proj",
"model.L_level.core.layers.10.attn.o_proj",
"model.L_level.core.layers.10.mlp.gate_up_proj",
"model.L_level.core.layers.10.mlp.down_proj",
"model.L_level.core.layers.11.attn.gqkv_proj",
"model.L_level.core.layers.11.attn.o_proj",
"model.L_level.core.layers.11.mlp.gate_up_proj",
"model.L_level.core.layers.11.mlp.down_proj",
"model.L_level.core.layers.12.attn.gqkv_proj",
"model.L_level.core.layers.12.attn.o_proj",
"model.L_level.core.layers.12.mlp.gate_up_proj",
"model.L_level.core.layers.12.mlp.down_proj",
"model.L_level.core.layers.13.attn.gqkv_proj",
"model.L_level.core.layers.13.attn.o_proj",
"model.L_level.core.layers.13.mlp.gate_up_proj",
"model.L_level.core.layers.13.mlp.down_proj",
"model.L_level.core.layers.14.attn.gqkv_proj",
"model.L_level.core.layers.14.attn.o_proj",
"model.L_level.core.layers.14.mlp.gate_up_proj",
"model.L_level.core.layers.14.mlp.down_proj",
"model.L_level.core.layers.15.attn.gqkv_proj",
"model.L_level.core.layers.15.attn.o_proj",
"model.L_level.core.layers.15.mlp.gate_up_proj",
"model.L_level.core.layers.15.mlp.down_proj",
"lm_head"
],
"num_lora_tensors": 258
}