[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file vocab.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file merges.txt
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file tokenizer.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file added_tokens.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file special_tokens_map.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file tokenizer_config.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file chat_template.jinja
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2336 >> Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
[INFO|2026-01-03 17:49:53] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 17:49:53] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file vocab.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file merges.txt
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file tokenizer.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file added_tokens.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file special_tokens_map.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file tokenizer_config.json
[INFO|2026-01-03 17:49:53] tokenization_utils_base.py:2065 >> loading file chat_template.jinja
[INFO|2026-01-03 17:49:54] tokenization_utils_base.py:2336 >> Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.
[INFO|2026-01-03 17:49:54] logging.py:143 >> Loading dataset gpu_aiAgent_zh.jsonl...
[INFO|2026-01-03 17:49:55] logging.py:143 >> Loading dataset processed_dataset1.jsonl...
[INFO|2026-01-03 17:49:59] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 17:49:59] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 17:49:59] logging.py:143 >> KV cache is disabled during training.
[INFO|2026-01-03 17:49:59] modeling_utils.py:1305 >> loading weights file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/model.safetensors
[INFO|2026-01-03 17:49:59] modeling_utils.py:2411 >> Instantiating Qwen3ForCausalLM model under default dtype torch.bfloat16.
[INFO|2026-01-03 17:49:59] configuration_utils.py:1098 >> Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "use_cache": false
}

[INFO|2026-01-03 17:50:00] modeling_utils.py:5606 >> All model checkpoint weights were used when initializing Qwen3ForCausalLM.

[INFO|2026-01-03 17:50:00] modeling_utils.py:5614 >> All the weights of Qwen3ForCausalLM were initialized from the model checkpoint at /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base.
If your task is similar to the task the model of the checkpoint was trained on, you can already use Qwen3ForCausalLM for predictions without further training.
[INFO|2026-01-03 17:50:00] configuration_utils.py:1051 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/generation_config.json
[INFO|2026-01-03 17:50:00] configuration_utils.py:1098 >> Generate config GenerationConfig {
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "max_new_tokens": 2048
}

[INFO|2026-01-03 17:50:00] logging.py:143 >> Gradient checkpointing enabled.
[INFO|2026-01-03 17:50:00] logging.py:143 >> Using torch SDPA for faster training and inference.
[INFO|2026-01-03 17:50:00] logging.py:143 >> Upcasting trainable params to float32.
[INFO|2026-01-03 17:50:00] logging.py:143 >> Fine-tuning method: LoRA
[INFO|2026-01-03 17:50:00] logging.py:143 >> Found linear modules: k_proj,q_proj,v_proj,down_proj,o_proj,up_proj,gate_proj
[INFO|2026-01-03 17:50:03] logging.py:143 >> trainable params: 69,730,304 || all params: 1,790,305,280 || trainable%: 3.8949
[INFO|2026-01-03 17:50:03] trainer.py:757 >> Using auto half precision backend
[INFO|2026-01-03 17:50:04] trainer.py:2433 >> ***** Running training *****
[INFO|2026-01-03 17:50:04] trainer.py:2434 >>   Num examples = 8,762
[INFO|2026-01-03 17:50:04] trainer.py:2435 >>   Num Epochs = 3
[INFO|2026-01-03 17:50:04] trainer.py:2436 >>   Instantaneous batch size per device = 2
[INFO|2026-01-03 17:50:04] trainer.py:2439 >>   Total train batch size (w. parallel, distributed & accumulation) = 16
[INFO|2026-01-03 17:50:04] trainer.py:2440 >>   Gradient Accumulation steps = 8
[INFO|2026-01-03 17:50:04] trainer.py:2441 >>   Total optimization steps = 1,644
[INFO|2026-01-03 17:50:04] trainer.py:2442 >>   Number of trainable parameters = 69,730,304
[INFO|2026-01-03 17:50:41] logging.py:143 >> {'loss': 1.4690, 'learning_rate': 4.9999e-05, 'epoch': 0.01, 'throughput': 834.18}
[INFO|2026-01-03 17:51:14] logging.py:143 >> {'loss': 1.5127, 'learning_rate': 4.9996e-05, 'epoch': 0.02, 'throughput': 839.03}
[INFO|2026-01-03 17:51:47] logging.py:143 >> {'loss': 1.4400, 'learning_rate': 4.9991e-05, 'epoch': 0.03, 'throughput': 844.64}
[INFO|2026-01-03 17:52:20] logging.py:143 >> {'loss': 1.4987, 'learning_rate': 4.9984e-05, 'epoch': 0.04, 'throughput': 842.78}
[INFO|2026-01-03 17:52:53] logging.py:143 >> {'loss': 1.4058, 'learning_rate': 4.9974e-05, 'epoch': 0.05, 'throughput': 847.67}
[INFO|2026-01-03 17:53:26] logging.py:143 >> {'loss': 1.4049, 'learning_rate': 4.9962e-05, 'epoch': 0.05, 'throughput': 850.44}
[INFO|2026-01-03 17:53:59] logging.py:143 >> {'loss': 1.4239, 'learning_rate': 4.9947e-05, 'epoch': 0.06, 'throughput': 850.70}
[INFO|2026-01-03 17:54:32] logging.py:143 >> {'loss': 1.3890, 'learning_rate': 4.9931e-05, 'epoch': 0.07, 'throughput': 849.34}
[INFO|2026-01-03 17:55:05] logging.py:143 >> {'loss': 1.3187, 'learning_rate': 4.9912e-05, 'epoch': 0.08, 'throughput': 851.42}
[INFO|2026-01-03 17:55:39] logging.py:143 >> {'loss': 1.3100, 'learning_rate': 4.9890e-05, 'epoch': 0.09, 'throughput': 856.96}
[INFO|2026-01-03 17:56:11] logging.py:143 >> {'loss': 1.3295, 'learning_rate': 4.9867e-05, 'epoch': 0.10, 'throughput': 856.18}
[INFO|2026-01-03 17:56:44] logging.py:143 >> {'loss': 1.2718, 'learning_rate': 4.9841e-05, 'epoch': 0.11, 'throughput': 856.16}
[INFO|2026-01-03 17:57:17] logging.py:143 >> {'loss': 1.2547, 'learning_rate': 4.9813e-05, 'epoch': 0.12, 'throughput': 854.86}
[INFO|2026-01-03 17:57:50] logging.py:143 >> {'loss': 1.3020, 'learning_rate': 4.9783e-05, 'epoch': 0.13, 'throughput': 856.68}
[INFO|2026-01-03 17:58:24] logging.py:143 >> {'loss': 1.2328, 'learning_rate': 4.9750e-05, 'epoch': 0.14, 'throughput': 860.35}
[INFO|2026-01-03 17:58:58] logging.py:143 >> {'loss': 1.2960, 'learning_rate': 4.9716e-05, 'epoch': 0.15, 'throughput': 861.72}
[INFO|2026-01-03 17:59:31] logging.py:143 >> {'loss': 1.3248, 'learning_rate': 4.9679e-05, 'epoch': 0.16, 'throughput': 861.08}
[INFO|2026-01-03 18:00:04] logging.py:143 >> {'loss': 1.2070, 'learning_rate': 4.9639e-05, 'epoch': 0.16, 'throughput': 861.29}
[INFO|2026-01-03 18:00:38] logging.py:143 >> {'loss': 1.2342, 'learning_rate': 4.9598e-05, 'epoch': 0.17, 'throughput': 863.20}
[INFO|2026-01-03 18:01:11] logging.py:143 >> {'loss': 1.1931, 'learning_rate': 4.9554e-05, 'epoch': 0.18, 'throughput': 864.21}
[INFO|2026-01-03 18:01:11] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-100
[INFO|2026-01-03 18:01:11] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 18:01:11] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 18:01:12] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-100/chat_template.jinja
[INFO|2026-01-03 18:01:12] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-100/tokenizer_config.json
[INFO|2026-01-03 18:01:12] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-100/special_tokens_map.json
[INFO|2026-01-03 18:01:46] logging.py:143 >> {'loss': 1.2122, 'learning_rate': 4.9508e-05, 'epoch': 0.19, 'throughput': 863.90}
[INFO|2026-01-03 18:02:19] logging.py:143 >> {'loss': 1.2973, 'learning_rate': 4.9460e-05, 'epoch': 0.20, 'throughput': 862.94}
[INFO|2026-01-03 18:02:51] logging.py:143 >> {'loss': 1.2483, 'learning_rate': 4.9409e-05, 'epoch': 0.21, 'throughput': 860.58}
[INFO|2026-01-03 18:03:24] logging.py:143 >> {'loss': 1.1153, 'learning_rate': 4.9356e-05, 'epoch': 0.22, 'throughput': 859.99}
[INFO|2026-01-03 18:03:57] logging.py:143 >> {'loss': 1.2269, 'learning_rate': 4.9301e-05, 'epoch': 0.23, 'throughput': 860.43}
[INFO|2026-01-03 18:04:30] logging.py:143 >> {'loss': 1.1324, 'learning_rate': 4.9244e-05, 'epoch': 0.24, 'throughput': 860.09}
[INFO|2026-01-03 18:05:02] logging.py:143 >> {'loss': 1.1818, 'learning_rate': 4.9185e-05, 'epoch': 0.25, 'throughput': 859.22}
[INFO|2026-01-03 18:05:37] logging.py:143 >> {'loss': 1.1243, 'learning_rate': 4.9123e-05, 'epoch': 0.26, 'throughput': 861.88}
[INFO|2026-01-03 18:06:09] logging.py:143 >> {'loss': 1.2134, 'learning_rate': 4.9059e-05, 'epoch': 0.26, 'throughput': 860.64}
[INFO|2026-01-03 18:06:44] logging.py:143 >> {'loss': 1.0775, 'learning_rate': 4.8993e-05, 'epoch': 0.27, 'throughput': 862.50}
[INFO|2026-01-03 18:07:16] logging.py:143 >> {'loss': 1.2069, 'learning_rate': 4.8925e-05, 'epoch': 0.28, 'throughput': 861.61}
[INFO|2026-01-03 18:07:48] logging.py:143 >> {'loss': 1.1649, 'learning_rate': 4.8855e-05, 'epoch': 0.29, 'throughput': 861.33}
[INFO|2026-01-03 18:08:22] logging.py:143 >> {'loss': 1.1495, 'learning_rate': 4.8782e-05, 'epoch': 0.30, 'throughput': 862.32}
[INFO|2026-01-03 18:08:56] logging.py:143 >> {'loss': 1.1231, 'learning_rate': 4.8708e-05, 'epoch': 0.31, 'throughput': 864.00}
[INFO|2026-01-03 18:09:29] logging.py:143 >> {'loss': 1.0885, 'learning_rate': 4.8631e-05, 'epoch': 0.32, 'throughput': 863.35}
[INFO|2026-01-03 18:10:02] logging.py:143 >> {'loss': 1.1022, 'learning_rate': 4.8552e-05, 'epoch': 0.33, 'throughput': 863.43}
[INFO|2026-01-03 18:10:33] logging.py:143 >> {'loss': 1.1223, 'learning_rate': 4.8470e-05, 'epoch': 0.34, 'throughput': 861.53}
[INFO|2026-01-03 18:11:07] logging.py:143 >> {'loss': 1.0287, 'learning_rate': 4.8387e-05, 'epoch': 0.35, 'throughput': 862.14}
[INFO|2026-01-03 18:11:38] logging.py:143 >> {'loss': 1.1138, 'learning_rate': 4.8302e-05, 'epoch': 0.36, 'throughput': 860.49}
[INFO|2026-01-03 18:12:11] logging.py:143 >> {'loss': 1.1149, 'learning_rate': 4.8214e-05, 'epoch': 0.37, 'throughput': 860.36}
[INFO|2026-01-03 18:12:11] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-200
[INFO|2026-01-03 18:12:11] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 18:12:11] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 18:12:11] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-200/chat_template.jinja
[INFO|2026-01-03 18:12:11] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-200/tokenizer_config.json
[INFO|2026-01-03 18:12:11] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-200/special_tokens_map.json
[INFO|2026-01-03 18:12:46] logging.py:143 >> {'loss': 1.1954, 'learning_rate': 4.8124e-05, 'epoch': 0.37, 'throughput': 860.43}
[INFO|2026-01-03 18:13:19] logging.py:143 >> {'loss': 1.0426, 'learning_rate': 4.8032e-05, 'epoch': 0.38, 'throughput': 860.83}
[INFO|2026-01-03 18:13:51] logging.py:143 >> {'loss': 1.2254, 'learning_rate': 4.7939e-05, 'epoch': 0.39, 'throughput': 860.08}
[INFO|2026-01-03 18:14:25] logging.py:143 >> {'loss': 1.0323, 'learning_rate': 4.7843e-05, 'epoch': 0.40, 'throughput': 860.93}
[INFO|2026-01-03 18:14:59] logging.py:143 >> {'loss': 0.9971, 'learning_rate': 4.7744e-05, 'epoch': 0.41, 'throughput': 862.28}
[INFO|2026-01-03 18:15:31] logging.py:143 >> {'loss': 1.0328, 'learning_rate': 4.7644e-05, 'epoch': 0.42, 'throughput': 861.03}
[INFO|2026-01-03 18:16:04] logging.py:143 >> {'loss': 1.0546, 'learning_rate': 4.7542e-05, 'epoch': 0.43, 'throughput': 861.65}
[INFO|2026-01-03 18:16:37] logging.py:143 >> {'loss': 1.1701, 'learning_rate': 4.7438e-05, 'epoch': 0.44, 'throughput': 861.08}
[INFO|2026-01-03 18:17:09] logging.py:143 >> {'loss': 1.0949, 'learning_rate': 4.7331e-05, 'epoch': 0.45, 'throughput': 860.64}
[INFO|2026-01-03 18:17:41] logging.py:143 >> {'loss': 1.1142, 'learning_rate': 4.7223e-05, 'epoch': 0.46, 'throughput': 859.97}
[INFO|2026-01-03 18:18:14] logging.py:143 >> {'loss': 1.0364, 'learning_rate': 4.7112e-05, 'epoch': 0.47, 'throughput': 859.48}
[INFO|2026-01-03 18:18:47] logging.py:143 >> {'loss': 1.0669, 'learning_rate': 4.7000e-05, 'epoch': 0.47, 'throughput': 859.84}
[INFO|2026-01-03 18:19:20] logging.py:143 >> {'loss': 1.0070, 'learning_rate': 4.6886e-05, 'epoch': 0.48, 'throughput': 859.99}
[INFO|2026-01-03 18:19:53] logging.py:143 >> {'loss': 0.9444, 'learning_rate': 4.6769e-05, 'epoch': 0.49, 'throughput': 859.86}
[INFO|2026-01-03 18:20:26] logging.py:143 >> {'loss': 1.0157, 'learning_rate': 4.6651e-05, 'epoch': 0.50, 'throughput': 859.98}
[INFO|2026-01-03 18:20:58] logging.py:143 >> {'loss': 1.0188, 'learning_rate': 4.6530e-05, 'epoch': 0.51, 'throughput': 859.53}
[INFO|2026-01-03 18:21:31] logging.py:143 >> {'loss': 0.9333, 'learning_rate': 4.6408e-05, 'epoch': 0.52, 'throughput': 859.24}
[INFO|2026-01-03 18:22:04] logging.py:143 >> {'loss': 0.9435, 'learning_rate': 4.6283e-05, 'epoch': 0.53, 'throughput': 859.31}
[INFO|2026-01-03 18:22:36] logging.py:143 >> {'loss': 0.9418, 'learning_rate': 4.6157e-05, 'epoch': 0.54, 'throughput': 859.02}
[INFO|2026-01-03 18:23:10] logging.py:143 >> {'loss': 0.9934, 'learning_rate': 4.6029e-05, 'epoch': 0.55, 'throughput': 859.28}
[INFO|2026-01-03 18:23:10] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-300
[INFO|2026-01-03 18:23:10] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 18:23:10] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 18:23:10] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-300/chat_template.jinja
[INFO|2026-01-03 18:23:10] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-300/tokenizer_config.json
[INFO|2026-01-03 18:23:10] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-300/special_tokens_map.json
[INFO|2026-01-03 18:23:44] logging.py:143 >> {'loss': 1.0952, 'learning_rate': 4.5899e-05, 'epoch': 0.56, 'throughput': 858.69}
[INFO|2026-01-03 18:24:17] logging.py:143 >> {'loss': 0.9632, 'learning_rate': 4.5767e-05, 'epoch': 0.57, 'throughput': 858.75}
[INFO|2026-01-03 18:24:49] logging.py:143 >> {'loss': 0.9310, 'learning_rate': 4.5633e-05, 'epoch': 0.58, 'throughput': 858.17}
[INFO|2026-01-03 18:25:22] logging.py:143 >> {'loss': 1.0103, 'learning_rate': 4.5497e-05, 'epoch': 0.58, 'throughput': 858.16}
[INFO|2026-01-03 18:25:55] logging.py:143 >> {'loss': 0.9518, 'learning_rate': 4.5359e-05, 'epoch': 0.59, 'throughput': 858.08}
[INFO|2026-01-03 18:26:28] logging.py:143 >> {'loss': 0.8974, 'learning_rate': 4.5220e-05, 'epoch': 0.60, 'throughput': 858.47}
[INFO|2026-01-03 18:27:01] logging.py:143 >> {'loss': 0.9171, 'learning_rate': 4.5078e-05, 'epoch': 0.61, 'throughput': 858.14}
[INFO|2026-01-03 18:27:34] logging.py:143 >> {'loss': 0.8990, 'learning_rate': 4.4935e-05, 'epoch': 0.62, 'throughput': 858.26}
[INFO|2026-01-03 18:28:07] logging.py:143 >> {'loss': 0.8964, 'learning_rate': 4.4790e-05, 'epoch': 0.63, 'throughput': 858.48}
[INFO|2026-01-03 18:28:40] logging.py:143 >> {'loss': 1.0919, 'learning_rate': 4.4643e-05, 'epoch': 0.64, 'throughput': 858.03}
[INFO|2026-01-03 18:29:12] logging.py:143 >> {'loss': 1.0469, 'learning_rate': 4.4495e-05, 'epoch': 0.65, 'throughput': 857.86}
[INFO|2026-01-03 18:29:46] logging.py:143 >> {'loss': 0.8841, 'learning_rate': 4.4344e-05, 'epoch': 0.66, 'throughput': 858.02}
[INFO|2026-01-03 18:30:19] logging.py:143 >> {'loss': 1.0016, 'learning_rate': 4.4192e-05, 'epoch': 0.67, 'throughput': 858.21}
[INFO|2026-01-03 18:30:51] logging.py:143 >> {'loss': 0.9584, 'learning_rate': 4.4038e-05, 'epoch': 0.68, 'throughput': 857.44}
[INFO|2026-01-03 18:31:24] logging.py:143 >> {'loss': 0.9964, 'learning_rate': 4.3882e-05, 'epoch': 0.68, 'throughput': 857.55}
[INFO|2026-01-03 18:31:57] logging.py:143 >> {'loss': 0.9808, 'learning_rate': 4.3725e-05, 'epoch': 0.69, 'throughput': 857.40}
[INFO|2026-01-03 18:32:31] logging.py:143 >> {'loss': 0.9126, 'learning_rate': 4.3566e-05, 'epoch': 0.70, 'throughput': 857.82}
[INFO|2026-01-03 18:33:04] logging.py:143 >> {'loss': 0.8749, 'learning_rate': 4.3405e-05, 'epoch': 0.71, 'throughput': 858.10}
[INFO|2026-01-03 18:33:37] logging.py:143 >> {'loss': 0.9203, 'learning_rate': 4.3243e-05, 'epoch': 0.72, 'throughput': 858.29}
[INFO|2026-01-03 18:34:11] logging.py:143 >> {'loss': 0.8857, 'learning_rate': 4.3078e-05, 'epoch': 0.73, 'throughput': 858.70}
[INFO|2026-01-03 18:34:11] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-400
[INFO|2026-01-03 18:34:11] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 18:34:11] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 18:34:11] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-400/chat_template.jinja
[INFO|2026-01-03 18:34:11] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-400/tokenizer_config.json
[INFO|2026-01-03 18:34:11] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-400/special_tokens_map.json
[INFO|2026-01-03 18:34:46] logging.py:143 >> {'loss': 0.9175, 'learning_rate': 4.2913e-05, 'epoch': 0.74, 'throughput': 858.62}
[INFO|2026-01-03 18:35:18] logging.py:143 >> {'loss': 0.9572, 'learning_rate': 4.2745e-05, 'epoch': 0.75, 'throughput': 858.42}
[INFO|2026-01-03 18:35:51] logging.py:143 >> {'loss': 1.0534, 'learning_rate': 4.2576e-05, 'epoch': 0.76, 'throughput': 858.30}
[INFO|2026-01-03 18:36:24] logging.py:143 >> {'loss': 0.8583, 'learning_rate': 4.2405e-05, 'epoch': 0.77, 'throughput': 858.40}
[INFO|2026-01-03 18:36:58] logging.py:143 >> {'loss': 0.8049, 'learning_rate': 4.2233e-05, 'epoch': 0.78, 'throughput': 858.68}
[INFO|2026-01-03 18:37:31] logging.py:143 >> {'loss': 0.9599, 'learning_rate': 4.2059e-05, 'epoch': 0.79, 'throughput': 858.76}
[INFO|2026-01-03 18:38:04] logging.py:143 >> {'loss': 0.8874, 'learning_rate': 4.1884e-05, 'epoch': 0.79, 'throughput': 858.73}
[INFO|2026-01-03 18:38:36] logging.py:143 >> {'loss': 0.8944, 'learning_rate': 4.1707e-05, 'epoch': 0.80, 'throughput': 858.55}
[INFO|2026-01-03 18:39:10] logging.py:143 >> {'loss': 0.9774, 'learning_rate': 4.1528e-05, 'epoch': 0.81, 'throughput': 858.61}
[INFO|2026-01-03 18:39:43] logging.py:143 >> {'loss': 0.8543, 'learning_rate': 4.1349e-05, 'epoch': 0.82, 'throughput': 859.04}
[INFO|2026-01-03 18:40:16] logging.py:143 >> {'loss': 0.8293, 'learning_rate': 4.1167e-05, 'epoch': 0.83, 'throughput': 859.07}
[INFO|2026-01-03 18:40:50] logging.py:143 >> {'loss': 0.9264, 'learning_rate': 4.0984e-05, 'epoch': 0.84, 'throughput': 859.36}
[INFO|2026-01-03 18:41:23] logging.py:143 >> {'loss': 0.9641, 'learning_rate': 4.0800e-05, 'epoch': 0.85, 'throughput': 859.44}
[INFO|2026-01-03 18:41:56] logging.py:143 >> {'loss': 0.8050, 'learning_rate': 4.0614e-05, 'epoch': 0.86, 'throughput': 859.28}
[INFO|2026-01-03 18:42:28] logging.py:143 >> {'loss': 1.0411, 'learning_rate': 4.0427e-05, 'epoch': 0.87, 'throughput': 858.91}
[INFO|2026-01-03 18:43:02] logging.py:143 >> {'loss': 0.8638, 'learning_rate': 4.0238e-05, 'epoch': 0.88, 'throughput': 859.55}
[INFO|2026-01-03 18:43:34] logging.py:143 >> {'loss': 1.0152, 'learning_rate': 4.0048e-05, 'epoch': 0.89, 'throughput': 859.07}
[INFO|2026-01-03 18:44:07] logging.py:143 >> {'loss': 0.9046, 'learning_rate': 3.9856e-05, 'epoch': 0.89, 'throughput': 859.04}
[INFO|2026-01-03 18:44:39] logging.py:143 >> {'loss': 0.8428, 'learning_rate': 3.9664e-05, 'epoch': 0.90, 'throughput': 858.64}
[INFO|2026-01-03 18:45:12] logging.py:143 >> {'loss': 0.8138, 'learning_rate': 3.9470e-05, 'epoch': 0.91, 'throughput': 858.66}
[INFO|2026-01-03 18:45:12] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-500
[INFO|2026-01-03 18:45:12] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 18:45:12] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 18:45:13] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-500/chat_template.jinja
[INFO|2026-01-03 18:45:13] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-500/tokenizer_config.json
[INFO|2026-01-03 18:45:13] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-500/special_tokens_map.json
[INFO|2026-01-03 18:45:47] logging.py:143 >> {'loss': 0.8349, 'learning_rate': 3.9274e-05, 'epoch': 0.92, 'throughput': 858.45}
[INFO|2026-01-03 18:46:20] logging.py:143 >> {'loss': 0.8830, 'learning_rate': 3.9077e-05, 'epoch': 0.93, 'throughput': 858.66}
[INFO|2026-01-03 18:46:54] logging.py:143 >> {'loss': 0.8141, 'learning_rate': 3.8879e-05, 'epoch': 0.94, 'throughput': 858.89}
[INFO|2026-01-03 18:47:26] logging.py:143 >> {'loss': 0.8964, 'learning_rate': 3.8680e-05, 'epoch': 0.95, 'throughput': 858.74}
[INFO|2026-01-03 18:48:00] logging.py:143 >> {'loss': 0.7845, 'learning_rate': 3.8479e-05, 'epoch': 0.96, 'throughput': 858.79}
[INFO|2026-01-03 18:48:32] logging.py:143 >> {'loss': 0.9675, 'learning_rate': 3.8278e-05, 'epoch': 0.97, 'throughput': 858.76}
[INFO|2026-01-03 18:49:06] logging.py:143 >> {'loss': 0.9429, 'learning_rate': 3.8075e-05, 'epoch': 0.98, 'throughput': 859.21}
[INFO|2026-01-03 18:49:39] logging.py:143 >> {'loss': 0.9034, 'learning_rate': 3.7870e-05, 'epoch': 0.99, 'throughput': 858.95}
[INFO|2026-01-03 18:50:12] logging.py:143 >> {'loss': 0.8696, 'learning_rate': 3.7665e-05, 'epoch': 1.00, 'throughput': 859.09}
[INFO|2026-01-03 18:50:42] logging.py:143 >> {'loss': 0.9914, 'learning_rate': 3.7459e-05, 'epoch': 1.00, 'throughput': 858.83}
[INFO|2026-01-03 18:51:15] logging.py:143 >> {'loss': 0.7924, 'learning_rate': 3.7251e-05, 'epoch': 1.01, 'throughput': 858.78}
[INFO|2026-01-03 18:51:48] logging.py:143 >> {'loss': 0.7542, 'learning_rate': 3.7042e-05, 'epoch': 1.02, 'throughput': 858.67}
[INFO|2026-01-03 18:52:20] logging.py:143 >> {'loss': 0.7216, 'learning_rate': 3.6832e-05, 'epoch': 1.03, 'throughput': 858.51}
[INFO|2026-01-03 18:52:53] logging.py:143 >> {'loss': 0.8276, 'learning_rate': 3.6621e-05, 'epoch': 1.04, 'throughput': 858.31}
[INFO|2026-01-03 18:53:26] logging.py:143 >> {'loss': 0.8081, 'learning_rate': 3.6409e-05, 'epoch': 1.05, 'throughput': 858.44}
[INFO|2026-01-03 18:53:58] logging.py:143 >> {'loss': 0.7503, 'learning_rate': 3.6196e-05, 'epoch': 1.06, 'throughput': 858.31}
[INFO|2026-01-03 18:54:32] logging.py:143 >> {'loss': 0.7449, 'learning_rate': 3.5982e-05, 'epoch': 1.07, 'throughput': 858.46}
[INFO|2026-01-03 18:55:05] logging.py:143 >> {'loss': 0.7779, 'learning_rate': 3.5767e-05, 'epoch': 1.08, 'throughput': 858.72}
[INFO|2026-01-03 18:55:39] logging.py:143 >> {'loss': 0.7257, 'learning_rate': 3.5551e-05, 'epoch': 1.09, 'throughput': 859.04}
[INFO|2026-01-03 18:56:12] logging.py:143 >> {'loss': 0.7348, 'learning_rate': 3.5334e-05, 'epoch': 1.09, 'throughput': 859.01}
[INFO|2026-01-03 18:56:12] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-600
[INFO|2026-01-03 18:56:12] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 18:56:12] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 18:56:12] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-600/chat_template.jinja
[INFO|2026-01-03 18:56:12] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-600/tokenizer_config.json
[INFO|2026-01-03 18:56:12] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-600/special_tokens_map.json
[INFO|2026-01-03 18:56:46] logging.py:143 >> {'loss': 0.8140, 'learning_rate': 3.5116e-05, 'epoch': 1.10, 'throughput': 858.80}
[INFO|2026-01-03 18:57:19] logging.py:143 >> {'loss': 0.6645, 'learning_rate': 3.4897e-05, 'epoch': 1.11, 'throughput': 858.74}
[INFO|2026-01-03 18:57:52] logging.py:143 >> {'loss': 0.7804, 'learning_rate': 3.4677e-05, 'epoch': 1.12, 'throughput': 858.73}
[INFO|2026-01-03 18:58:24] logging.py:143 >> {'loss': 0.8645, 'learning_rate': 3.4457e-05, 'epoch': 1.13, 'throughput': 858.48}
[INFO|2026-01-03 18:58:58] logging.py:143 >> {'loss': 0.7509, 'learning_rate': 3.4235e-05, 'epoch': 1.14, 'throughput': 858.70}
[INFO|2026-01-03 18:59:32] logging.py:143 >> {'loss': 0.7380, 'learning_rate': 3.4013e-05, 'epoch': 1.15, 'throughput': 858.81}
[INFO|2026-01-03 19:00:05] logging.py:143 >> {'loss': 0.7310, 'learning_rate': 3.3789e-05, 'epoch': 1.16, 'throughput': 858.83}
[INFO|2026-01-03 19:00:38] logging.py:143 >> {'loss': 0.6878, 'learning_rate': 3.3565e-05, 'epoch': 1.17, 'throughput': 858.79}
[INFO|2026-01-03 19:01:11] logging.py:143 >> {'loss': 0.6777, 'learning_rate': 3.3341e-05, 'epoch': 1.18, 'throughput': 858.84}
[INFO|2026-01-03 19:01:44] logging.py:143 >> {'loss': 0.8185, 'learning_rate': 3.3115e-05, 'epoch': 1.19, 'throughput': 858.76}
[INFO|2026-01-03 19:02:17] logging.py:143 >> {'loss': 0.7367, 'learning_rate': 3.2889e-05, 'epoch': 1.20, 'throughput': 858.67}
[INFO|2026-01-03 19:02:49] logging.py:143 >> {'loss': 0.7812, 'learning_rate': 3.2662e-05, 'epoch': 1.20, 'throughput': 858.63}
[INFO|2026-01-03 19:03:23] logging.py:143 >> {'loss': 0.6602, 'learning_rate': 3.2434e-05, 'epoch': 1.21, 'throughput': 858.96}
[INFO|2026-01-03 19:03:56] logging.py:143 >> {'loss': 0.7918, 'learning_rate': 3.2206e-05, 'epoch': 1.22, 'throughput': 858.90}
[INFO|2026-01-03 19:04:29] logging.py:143 >> {'loss': 0.6596, 'learning_rate': 3.1977e-05, 'epoch': 1.23, 'throughput': 858.90}
[INFO|2026-01-03 19:05:03] logging.py:143 >> {'loss': 0.5950, 'learning_rate': 3.1747e-05, 'epoch': 1.24, 'throughput': 859.24}
[INFO|2026-01-03 19:05:36] logging.py:143 >> {'loss': 0.7421, 'learning_rate': 3.1517e-05, 'epoch': 1.25, 'throughput': 859.23}
[INFO|2026-01-03 19:06:07] logging.py:143 >> {'loss': 0.8241, 'learning_rate': 3.1286e-05, 'epoch': 1.26, 'throughput': 858.60}
[INFO|2026-01-03 19:06:39] logging.py:143 >> {'loss': 0.7889, 'learning_rate': 3.1054e-05, 'epoch': 1.27, 'throughput': 858.29}
[INFO|2026-01-03 19:07:11] logging.py:143 >> {'loss': 0.8462, 'learning_rate': 3.0822e-05, 'epoch': 1.28, 'throughput': 858.12}
[INFO|2026-01-03 19:07:11] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-700
[INFO|2026-01-03 19:07:11] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 19:07:11] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 19:07:12] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-700/chat_template.jinja
[INFO|2026-01-03 19:07:12] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-700/tokenizer_config.json
[INFO|2026-01-03 19:07:12] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-700/special_tokens_map.json
[INFO|2026-01-03 19:07:45] logging.py:143 >> {'loss': 0.8064, 'learning_rate': 3.0590e-05, 'epoch': 1.29, 'throughput': 857.86}
[INFO|2026-01-03 19:08:17] logging.py:143 >> {'loss': 0.7113, 'learning_rate': 3.0357e-05, 'epoch': 1.30, 'throughput': 857.75}
[INFO|2026-01-03 19:08:50] logging.py:143 >> {'loss': 0.7086, 'learning_rate': 3.0123e-05, 'epoch': 1.30, 'throughput': 857.58}
[INFO|2026-01-03 19:09:23] logging.py:143 >> {'loss': 0.6923, 'learning_rate': 2.9889e-05, 'epoch': 1.31, 'throughput': 857.53}
[INFO|2026-01-03 19:09:56] logging.py:143 >> {'loss': 0.7034, 'learning_rate': 2.9654e-05, 'epoch': 1.32, 'throughput': 857.46}
[INFO|2026-01-03 19:10:29] logging.py:143 >> {'loss': 0.7029, 'learning_rate': 2.9420e-05, 'epoch': 1.33, 'throughput': 857.55}
[INFO|2026-01-03 19:11:02] logging.py:143 >> {'loss': 0.7442, 'learning_rate': 2.9184e-05, 'epoch': 1.34, 'throughput': 857.69}
[INFO|2026-01-03 19:11:36] logging.py:143 >> {'loss': 0.6896, 'learning_rate': 2.8949e-05, 'epoch': 1.35, 'throughput': 857.92}
[INFO|2026-01-03 19:12:10] logging.py:143 >> {'loss': 0.7188, 'learning_rate': 2.8713e-05, 'epoch': 1.36, 'throughput': 858.05}
[INFO|2026-01-03 19:12:44] logging.py:143 >> {'loss': 0.7536, 'learning_rate': 2.8476e-05, 'epoch': 1.37, 'throughput': 858.33}
[INFO|2026-01-03 19:13:17] logging.py:143 >> {'loss': 0.6742, 'learning_rate': 2.8239e-05, 'epoch': 1.38, 'throughput': 858.27}
[INFO|2026-01-03 19:13:50] logging.py:143 >> {'loss': 0.6787, 'learning_rate': 2.8002e-05, 'epoch': 1.39, 'throughput': 858.28}
[INFO|2026-01-03 19:14:22] logging.py:143 >> {'loss': 0.8232, 'learning_rate': 2.7765e-05, 'epoch': 1.40, 'throughput': 858.22}
[INFO|2026-01-03 19:14:55] logging.py:143 >> {'loss': 0.7529, 'learning_rate': 2.7528e-05, 'epoch': 1.41, 'throughput': 858.10}
[INFO|2026-01-03 19:15:28] logging.py:143 >> {'loss': 0.7612, 'learning_rate': 2.7290e-05, 'epoch': 1.41, 'throughput': 858.08}
[INFO|2026-01-03 19:16:00] logging.py:143 >> {'loss': 0.7770, 'learning_rate': 2.7052e-05, 'epoch': 1.42, 'throughput': 857.80}
[INFO|2026-01-03 19:16:33] logging.py:143 >> {'loss': 0.7732, 'learning_rate': 2.6814e-05, 'epoch': 1.43, 'throughput': 857.81}
[INFO|2026-01-03 19:17:06] logging.py:143 >> {'loss': 0.6874, 'learning_rate': 2.6575e-05, 'epoch': 1.44, 'throughput': 857.88}
[INFO|2026-01-03 19:17:40] logging.py:143 >> {'loss': 0.6866, 'learning_rate': 2.6337e-05, 'epoch': 1.45, 'throughput': 858.01}
[INFO|2026-01-03 19:18:12] logging.py:143 >> {'loss': 0.6184, 'learning_rate': 2.6098e-05, 'epoch': 1.46, 'throughput': 857.91}
[INFO|2026-01-03 19:18:12] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-800
[INFO|2026-01-03 19:18:12] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 19:18:12] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 19:18:13] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-800/chat_template.jinja
[INFO|2026-01-03 19:18:13] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-800/tokenizer_config.json
[INFO|2026-01-03 19:18:13] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-800/special_tokens_map.json
[INFO|2026-01-03 19:18:46] logging.py:143 >> {'loss': 0.6364, 'learning_rate': 2.5860e-05, 'epoch': 1.47, 'throughput': 857.71}
[INFO|2026-01-03 19:19:19] logging.py:143 >> {'loss': 0.7312, 'learning_rate': 2.5621e-05, 'epoch': 1.48, 'throughput': 857.63}
[INFO|2026-01-03 19:19:52] logging.py:143 >> {'loss': 0.6735, 'learning_rate': 2.5382e-05, 'epoch': 1.49, 'throughput': 857.79}
[INFO|2026-01-03 19:20:25] logging.py:143 >> {'loss': 0.5681, 'learning_rate': 2.5143e-05, 'epoch': 1.50, 'throughput': 857.56}
[INFO|2026-01-03 19:20:57] logging.py:143 >> {'loss': 0.7154, 'learning_rate': 2.4904e-05, 'epoch': 1.51, 'throughput': 857.25}
[INFO|2026-01-03 19:21:28] logging.py:143 >> {'loss': 0.6590, 'learning_rate': 2.4666e-05, 'epoch': 1.51, 'throughput': 856.95}
[INFO|2026-01-03 19:22:02] logging.py:143 >> {'loss': 0.6600, 'learning_rate': 2.4427e-05, 'epoch': 1.52, 'throughput': 857.08}
[INFO|2026-01-03 19:22:35] logging.py:143 >> {'loss': 0.8048, 'learning_rate': 2.4188e-05, 'epoch': 1.53, 'throughput': 857.23}
[INFO|2026-01-03 19:23:08] logging.py:143 >> {'loss': 0.6516, 'learning_rate': 2.3949e-05, 'epoch': 1.54, 'throughput': 857.24}
[INFO|2026-01-03 19:23:41] logging.py:143 >> {'loss': 0.6813, 'learning_rate': 2.3711e-05, 'epoch': 1.55, 'throughput': 857.25}
[INFO|2026-01-03 19:24:15] logging.py:143 >> {'loss': 0.7308, 'learning_rate': 2.3472e-05, 'epoch': 1.56, 'throughput': 857.41}
[INFO|2026-01-03 19:24:48] logging.py:143 >> {'loss': 0.7728, 'learning_rate': 2.3234e-05, 'epoch': 1.57, 'throughput': 857.47}
[INFO|2026-01-03 19:25:20] logging.py:143 >> {'loss': 0.6673, 'learning_rate': 2.2996e-05, 'epoch': 1.58, 'throughput': 857.38}
[INFO|2026-01-03 19:25:53] logging.py:143 >> {'loss': 0.7155, 'learning_rate': 2.2758e-05, 'epoch': 1.59, 'throughput': 857.40}
[INFO|2026-01-03 19:26:27] logging.py:143 >> {'loss': 0.6372, 'learning_rate': 2.2520e-05, 'epoch': 1.60, 'throughput': 857.54}
[INFO|2026-01-03 19:27:00] logging.py:143 >> {'loss': 0.7021, 'learning_rate': 2.2282e-05, 'epoch': 1.61, 'throughput': 857.66}
[INFO|2026-01-03 19:27:34] logging.py:143 >> {'loss': 0.6923, 'learning_rate': 2.2045e-05, 'epoch': 1.62, 'throughput': 857.84}
[INFO|2026-01-03 19:28:07] logging.py:143 >> {'loss': 0.6426, 'learning_rate': 2.1808e-05, 'epoch': 1.62, 'throughput': 857.93}
[INFO|2026-01-03 19:28:41] logging.py:143 >> {'loss': 0.5618, 'learning_rate': 2.1571e-05, 'epoch': 1.63, 'throughput': 857.97}
[INFO|2026-01-03 19:29:12] logging.py:143 >> {'loss': 0.6160, 'learning_rate': 2.1335e-05, 'epoch': 1.64, 'throughput': 857.64}
[INFO|2026-01-03 19:29:12] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-900
[INFO|2026-01-03 19:29:12] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 19:29:12] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 19:29:12] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-900/chat_template.jinja
[INFO|2026-01-03 19:29:12] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-900/tokenizer_config.json
[INFO|2026-01-03 19:29:12] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-900/special_tokens_map.json
[INFO|2026-01-03 19:29:46] logging.py:143 >> {'loss': 0.6574, 'learning_rate': 2.1099e-05, 'epoch': 1.65, 'throughput': 857.40}
[INFO|2026-01-03 19:30:19] logging.py:143 >> {'loss': 0.6823, 'learning_rate': 2.0863e-05, 'epoch': 1.66, 'throughput': 857.31}
[INFO|2026-01-03 19:30:51] logging.py:143 >> {'loss': 0.5728, 'learning_rate': 2.0627e-05, 'epoch': 1.67, 'throughput': 857.31}
[INFO|2026-01-03 19:31:24] logging.py:143 >> {'loss': 0.7025, 'learning_rate': 2.0392e-05, 'epoch': 1.68, 'throughput': 857.28}
[INFO|2026-01-03 19:31:58] logging.py:143 >> {'loss': 0.5986, 'learning_rate': 2.0158e-05, 'epoch': 1.69, 'throughput': 857.53}
[INFO|2026-01-03 19:32:30] logging.py:143 >> {'loss': 0.6195, 'learning_rate': 1.9924e-05, 'epoch': 1.70, 'throughput': 857.32}
[INFO|2026-01-03 19:33:04] logging.py:143 >> {'loss': 0.6283, 'learning_rate': 1.9690e-05, 'epoch': 1.71, 'throughput': 857.41}
[INFO|2026-01-03 19:33:38] logging.py:143 >> {'loss': 0.5767, 'learning_rate': 1.9457e-05, 'epoch': 1.72, 'throughput': 857.51}
[INFO|2026-01-03 19:34:10] logging.py:143 >> {'loss': 0.5237, 'learning_rate': 1.9224e-05, 'epoch': 1.72, 'throughput': 857.34}
[INFO|2026-01-03 19:34:43] logging.py:143 >> {'loss': 0.7184, 'learning_rate': 1.8992e-05, 'epoch': 1.73, 'throughput': 857.40}
[INFO|2026-01-03 19:35:16] logging.py:143 >> {'loss': 0.6595, 'learning_rate': 1.8761e-05, 'epoch': 1.74, 'throughput': 857.47}
[INFO|2026-01-03 19:35:49] logging.py:143 >> {'loss': 0.6161, 'learning_rate': 1.8530e-05, 'epoch': 1.75, 'throughput': 857.50}
[INFO|2026-01-03 19:36:23] logging.py:143 >> {'loss': 0.6076, 'learning_rate': 1.8299e-05, 'epoch': 1.76, 'throughput': 857.63}
[INFO|2026-01-03 19:36:56] logging.py:143 >> {'loss': 0.5543, 'learning_rate': 1.8069e-05, 'epoch': 1.77, 'throughput': 857.76}
[INFO|2026-01-03 19:37:29] logging.py:143 >> {'loss': 0.6496, 'learning_rate': 1.7840e-05, 'epoch': 1.78, 'throughput': 857.70}
[INFO|2026-01-03 19:38:02] logging.py:143 >> {'loss': 0.6315, 'learning_rate': 1.7612e-05, 'epoch': 1.79, 'throughput': 857.79}
[INFO|2026-01-03 19:38:35] logging.py:143 >> {'loss': 0.5654, 'learning_rate': 1.7384e-05, 'epoch': 1.80, 'throughput': 857.71}
[INFO|2026-01-03 19:39:08] logging.py:143 >> {'loss': 0.5803, 'learning_rate': 1.7157e-05, 'epoch': 1.81, 'throughput': 857.74}
[INFO|2026-01-03 19:39:42] logging.py:143 >> {'loss': 0.5516, 'learning_rate': 1.6930e-05, 'epoch': 1.82, 'throughput': 857.88}
[INFO|2026-01-03 19:40:15] logging.py:143 >> {'loss': 0.6530, 'learning_rate': 1.6704e-05, 'epoch': 1.83, 'throughput': 857.90}
[INFO|2026-01-03 19:40:15] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1000
[INFO|2026-01-03 19:40:15] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 19:40:15] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 19:40:15] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1000/chat_template.jinja
[INFO|2026-01-03 19:40:15] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1000/tokenizer_config.json
[INFO|2026-01-03 19:40:15] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1000/special_tokens_map.json
[INFO|2026-01-03 19:40:50] logging.py:143 >> {'loss': 0.6456, 'learning_rate': 1.6479e-05, 'epoch': 1.83, 'throughput': 857.82}
[INFO|2026-01-03 19:41:22] logging.py:143 >> {'loss': 0.6778, 'learning_rate': 1.6255e-05, 'epoch': 1.84, 'throughput': 857.68}
[INFO|2026-01-03 19:41:55] logging.py:143 >> {'loss': 0.6779, 'learning_rate': 1.6032e-05, 'epoch': 1.85, 'throughput': 857.65}
[INFO|2026-01-03 19:42:29] logging.py:143 >> {'loss': 0.5039, 'learning_rate': 1.5809e-05, 'epoch': 1.86, 'throughput': 857.77}
[INFO|2026-01-03 19:43:02] logging.py:143 >> {'loss': 0.6694, 'learning_rate': 1.5588e-05, 'epoch': 1.87, 'throughput': 857.78}
[INFO|2026-01-03 19:43:35] logging.py:143 >> {'loss': 0.5749, 'learning_rate': 1.5367e-05, 'epoch': 1.88, 'throughput': 857.95}
[INFO|2026-01-03 19:44:08] logging.py:143 >> {'loss': 0.5602, 'learning_rate': 1.5147e-05, 'epoch': 1.89, 'throughput': 857.95}
[INFO|2026-01-03 19:44:41] logging.py:143 >> {'loss': 0.6450, 'learning_rate': 1.4928e-05, 'epoch': 1.90, 'throughput': 857.78}
[INFO|2026-01-03 19:45:15] logging.py:143 >> {'loss': 0.6209, 'learning_rate': 1.4710e-05, 'epoch': 1.91, 'throughput': 858.00}
[INFO|2026-01-03 19:45:47] logging.py:143 >> {'loss': 0.5908, 'learning_rate': 1.4492e-05, 'epoch': 1.92, 'throughput': 857.90}
[INFO|2026-01-03 19:46:20] logging.py:143 >> {'loss': 0.5957, 'learning_rate': 1.4276e-05, 'epoch': 1.93, 'throughput': 857.91}
[INFO|2026-01-03 19:46:52] logging.py:143 >> {'loss': 0.6543, 'learning_rate': 1.4061e-05, 'epoch': 1.93, 'throughput': 857.70}
[INFO|2026-01-03 19:47:26] logging.py:143 >> {'loss': 0.5055, 'learning_rate': 1.3846e-05, 'epoch': 1.94, 'throughput': 857.84}
[INFO|2026-01-03 19:47:59] logging.py:143 >> {'loss': 0.5246, 'learning_rate': 1.3633e-05, 'epoch': 1.95, 'throughput': 857.97}
[INFO|2026-01-03 19:48:33] logging.py:143 >> {'loss': 0.5412, 'learning_rate': 1.3421e-05, 'epoch': 1.96, 'throughput': 858.06}
[INFO|2026-01-03 19:49:05] logging.py:143 >> {'loss': 0.6240, 'learning_rate': 1.3210e-05, 'epoch': 1.97, 'throughput': 857.94}
[INFO|2026-01-03 19:49:38] logging.py:143 >> {'loss': 0.5854, 'learning_rate': 1.3000e-05, 'epoch': 1.98, 'throughput': 857.89}
[INFO|2026-01-03 19:50:10] logging.py:143 >> {'loss': 0.5972, 'learning_rate': 1.2791e-05, 'epoch': 1.99, 'throughput': 857.79}
[INFO|2026-01-03 19:50:44] logging.py:143 >> {'loss': 0.5627, 'learning_rate': 1.2583e-05, 'epoch': 2.00, 'throughput': 857.96}
[INFO|2026-01-03 19:51:15] logging.py:143 >> {'loss': 0.5440, 'learning_rate': 1.2376e-05, 'epoch': 2.01, 'throughput': 857.97}
[INFO|2026-01-03 19:51:15] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1100
[INFO|2026-01-03 19:51:15] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 19:51:15] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 19:51:15] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1100/chat_template.jinja
[INFO|2026-01-03 19:51:15] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1100/tokenizer_config.json
[INFO|2026-01-03 19:51:15] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1100/special_tokens_map.json
[INFO|2026-01-03 19:51:49] logging.py:143 >> {'loss': 0.4701, 'learning_rate': 1.2170e-05, 'epoch': 2.02, 'throughput': 857.99}
[INFO|2026-01-03 19:52:23] logging.py:143 >> {'loss': 0.4359, 'learning_rate': 1.1966e-05, 'epoch': 2.03, 'throughput': 858.21}
[INFO|2026-01-03 19:52:57] logging.py:143 >> {'loss': 0.4436, 'learning_rate': 1.1763e-05, 'epoch': 2.03, 'throughput': 858.35}
[INFO|2026-01-03 19:53:30] logging.py:143 >> {'loss': 0.5898, 'learning_rate': 1.1561e-05, 'epoch': 2.04, 'throughput': 858.30}
[INFO|2026-01-03 19:54:02] logging.py:143 >> {'loss': 0.5882, 'learning_rate': 1.1360e-05, 'epoch': 2.05, 'throughput': 858.05}
[INFO|2026-01-03 19:54:35] logging.py:143 >> {'loss': 0.4388, 'learning_rate': 1.1160e-05, 'epoch': 2.06, 'throughput': 858.00}
[INFO|2026-01-03 19:55:08] logging.py:143 >> {'loss': 0.5827, 'learning_rate': 1.0962e-05, 'epoch': 2.07, 'throughput': 858.03}
[INFO|2026-01-03 19:55:40] logging.py:143 >> {'loss': 0.5261, 'learning_rate': 1.0765e-05, 'epoch': 2.08, 'throughput': 857.97}
[INFO|2026-01-03 19:56:13] logging.py:143 >> {'loss': 0.5661, 'learning_rate': 1.0569e-05, 'epoch': 2.09, 'throughput': 857.88}
[INFO|2026-01-03 19:56:46] logging.py:143 >> {'loss': 0.5380, 'learning_rate': 1.0375e-05, 'epoch': 2.10, 'throughput': 857.82}
[INFO|2026-01-03 19:57:18] logging.py:143 >> {'loss': 0.7164, 'learning_rate': 1.0182e-05, 'epoch': 2.11, 'throughput': 857.75}
[INFO|2026-01-03 19:57:52] logging.py:143 >> {'loss': 0.5611, 'learning_rate': 9.9903e-06, 'epoch': 2.12, 'throughput': 857.80}
[INFO|2026-01-03 19:58:24] logging.py:143 >> {'loss': 0.5341, 'learning_rate': 9.7999e-06, 'epoch': 2.13, 'throughput': 857.70}
[INFO|2026-01-03 19:58:56] logging.py:143 >> {'loss': 0.4899, 'learning_rate': 9.6110e-06, 'epoch': 2.14, 'throughput': 857.60}
[INFO|2026-01-03 19:59:29] logging.py:143 >> {'loss': 0.6115, 'learning_rate': 9.4234e-06, 'epoch': 2.14, 'throughput': 857.59}
[INFO|2026-01-03 20:00:03] logging.py:143 >> {'loss': 0.5713, 'learning_rate': 9.2373e-06, 'epoch': 2.15, 'throughput': 857.66}
[INFO|2026-01-03 20:00:36] logging.py:143 >> {'loss': 0.5144, 'learning_rate': 9.0526e-06, 'epoch': 2.16, 'throughput': 857.70}
[INFO|2026-01-03 20:01:09] logging.py:143 >> {'loss': 0.5139, 'learning_rate': 8.8694e-06, 'epoch': 2.17, 'throughput': 857.78}
[INFO|2026-01-03 20:01:42] logging.py:143 >> {'loss': 0.5763, 'learning_rate': 8.6876e-06, 'epoch': 2.18, 'throughput': 857.78}
[INFO|2026-01-03 20:02:17] logging.py:143 >> {'loss': 0.5669, 'learning_rate': 8.5074e-06, 'epoch': 2.19, 'throughput': 858.01}
[INFO|2026-01-03 20:02:17] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1200
[INFO|2026-01-03 20:02:17] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 20:02:17] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 20:02:17] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1200/chat_template.jinja
[INFO|2026-01-03 20:02:17] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1200/tokenizer_config.json
[INFO|2026-01-03 20:02:17] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1200/special_tokens_map.json
[INFO|2026-01-03 20:02:51] logging.py:143 >> {'loss': 0.5508, 'learning_rate': 8.3286e-06, 'epoch': 2.20, 'throughput': 857.95}
[INFO|2026-01-03 20:03:24] logging.py:143 >> {'loss': 0.5538, 'learning_rate': 8.1514e-06, 'epoch': 2.21, 'throughput': 858.01}
[INFO|2026-01-03 20:03:57] logging.py:143 >> {'loss': 0.4806, 'learning_rate': 7.9757e-06, 'epoch': 2.22, 'throughput': 858.04}
[INFO|2026-01-03 20:04:31] logging.py:143 >> {'loss': 0.4594, 'learning_rate': 7.8015e-06, 'epoch': 2.23, 'throughput': 858.10}
[INFO|2026-01-03 20:05:04] logging.py:143 >> {'loss': 0.4625, 'learning_rate': 7.6290e-06, 'epoch': 2.24, 'throughput': 858.25}
[INFO|2026-01-03 20:05:37] logging.py:143 >> {'loss': 0.5248, 'learning_rate': 7.4580e-06, 'epoch': 2.24, 'throughput': 858.06}
[INFO|2026-01-03 20:06:09] logging.py:143 >> {'loss': 0.5796, 'learning_rate': 7.2886e-06, 'epoch': 2.25, 'throughput': 858.05}
[INFO|2026-01-03 20:06:42] logging.py:143 >> {'loss': 0.4509, 'learning_rate': 7.1208e-06, 'epoch': 2.26, 'throughput': 858.04}
[INFO|2026-01-03 20:07:16] logging.py:143 >> {'loss': 0.4935, 'learning_rate': 6.9547e-06, 'epoch': 2.27, 'throughput': 858.16}
[INFO|2026-01-03 20:07:49] logging.py:143 >> {'loss': 0.5073, 'learning_rate': 6.7902e-06, 'epoch': 2.28, 'throughput': 858.20}
[INFO|2026-01-03 20:08:23] logging.py:143 >> {'loss': 0.4892, 'learning_rate': 6.6274e-06, 'epoch': 2.29, 'throughput': 858.22}
[INFO|2026-01-03 20:08:55] logging.py:143 >> {'loss': 0.5268, 'learning_rate': 6.4662e-06, 'epoch': 2.30, 'throughput': 858.15}
[INFO|2026-01-03 20:09:27] logging.py:143 >> {'loss': 0.5226, 'learning_rate': 6.3067e-06, 'epoch': 2.31, 'throughput': 857.94}
[INFO|2026-01-03 20:10:01] logging.py:143 >> {'loss': 0.4098, 'learning_rate': 6.1490e-06, 'epoch': 2.32, 'throughput': 858.07}
[INFO|2026-01-03 20:10:33] logging.py:143 >> {'loss': 0.5064, 'learning_rate': 5.9930e-06, 'epoch': 2.33, 'throughput': 857.86}
[INFO|2026-01-03 20:11:05] logging.py:143 >> {'loss': 0.4860, 'learning_rate': 5.8387e-06, 'epoch': 2.34, 'throughput': 857.82}
[INFO|2026-01-03 20:11:39] logging.py:143 >> {'loss': 0.5659, 'learning_rate': 5.6861e-06, 'epoch': 2.35, 'throughput': 857.87}
[INFO|2026-01-03 20:12:11] logging.py:143 >> {'loss': 0.5572, 'learning_rate': 5.5353e-06, 'epoch': 2.35, 'throughput': 857.82}
[INFO|2026-01-03 20:12:43] logging.py:143 >> {'loss': 0.5891, 'learning_rate': 5.3863e-06, 'epoch': 2.36, 'throughput': 857.72}
[INFO|2026-01-03 20:13:16] logging.py:143 >> {'loss': 0.5121, 'learning_rate': 5.2391e-06, 'epoch': 2.37, 'throughput': 857.66}
[INFO|2026-01-03 20:13:16] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1300
[INFO|2026-01-03 20:13:16] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 20:13:16] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 20:13:16] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1300/chat_template.jinja
[INFO|2026-01-03 20:13:16] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1300/tokenizer_config.json
[INFO|2026-01-03 20:13:16] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1300/special_tokens_map.json
[INFO|2026-01-03 20:13:51] logging.py:143 >> {'loss': 0.4585, 'learning_rate': 5.0937e-06, 'epoch': 2.38, 'throughput': 857.71}
[INFO|2026-01-03 20:14:23] logging.py:143 >> {'loss': 0.5115, 'learning_rate': 4.9501e-06, 'epoch': 2.39, 'throughput': 857.56}
[INFO|2026-01-03 20:15:05] logging.py:143 >> {'loss': 0.3663, 'learning_rate': 4.8083e-06, 'epoch': 2.40, 'throughput': 856.67}
[INFO|2026-01-03 20:15:39] logging.py:143 >> {'loss': 0.4918, 'learning_rate': 4.6684e-06, 'epoch': 2.41, 'throughput': 856.82}
[INFO|2026-01-03 20:16:13] logging.py:143 >> {'loss': 0.4168, 'learning_rate': 4.5303e-06, 'epoch': 2.42, 'throughput': 856.92}
[INFO|2026-01-03 20:16:47] logging.py:143 >> {'loss': 0.4658, 'learning_rate': 4.3941e-06, 'epoch': 2.43, 'throughput': 857.11}
[INFO|2026-01-03 20:17:21] logging.py:143 >> {'loss': 0.4081, 'learning_rate': 4.2598e-06, 'epoch': 2.44, 'throughput': 857.20}
[INFO|2026-01-03 20:17:54] logging.py:143 >> {'loss': 0.4739, 'learning_rate': 4.1274e-06, 'epoch': 2.45, 'throughput': 857.23}
[INFO|2026-01-03 20:18:29] logging.py:143 >> {'loss': 0.3149, 'learning_rate': 3.9969e-06, 'epoch': 2.45, 'throughput': 857.50}
[INFO|2026-01-03 20:19:02] logging.py:143 >> {'loss': 0.5650, 'learning_rate': 3.8683e-06, 'epoch': 2.46, 'throughput': 857.55}
[INFO|2026-01-03 20:19:34] logging.py:143 >> {'loss': 0.4656, 'learning_rate': 3.7416e-06, 'epoch': 2.47, 'throughput': 857.47}
[INFO|2026-01-03 20:20:07] logging.py:143 >> {'loss': 0.4552, 'learning_rate': 3.6169e-06, 'epoch': 2.48, 'throughput': 857.44}
[INFO|2026-01-03 20:20:41] logging.py:143 >> {'loss': 0.4571, 'learning_rate': 3.4941e-06, 'epoch': 2.49, 'throughput': 857.63}
[INFO|2026-01-03 20:21:13] logging.py:143 >> {'loss': 0.5344, 'learning_rate': 3.3733e-06, 'epoch': 2.50, 'throughput': 857.46}
[INFO|2026-01-03 20:21:46] logging.py:143 >> {'loss': 0.5010, 'learning_rate': 3.2545e-06, 'epoch': 2.51, 'throughput': 857.44}
[INFO|2026-01-03 20:22:20] logging.py:143 >> {'loss': 0.5723, 'learning_rate': 3.1376e-06, 'epoch': 2.52, 'throughput': 857.51}
[INFO|2026-01-03 20:22:52] logging.py:143 >> {'loss': 0.5347, 'learning_rate': 3.0227e-06, 'epoch': 2.53, 'throughput': 857.49}
[INFO|2026-01-03 20:23:25] logging.py:143 >> {'loss': 0.4727, 'learning_rate': 2.9099e-06, 'epoch': 2.54, 'throughput': 857.44}
[INFO|2026-01-03 20:23:58] logging.py:143 >> {'loss': 0.6187, 'learning_rate': 2.7990e-06, 'epoch': 2.55, 'throughput': 857.44}
[INFO|2026-01-03 20:24:31] logging.py:143 >> {'loss': 0.5945, 'learning_rate': 2.6902e-06, 'epoch': 2.56, 'throughput': 857.44}
[INFO|2026-01-03 20:24:31] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1400
[INFO|2026-01-03 20:24:31] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 20:24:31] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 20:24:31] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1400/chat_template.jinja
[INFO|2026-01-03 20:24:31] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1400/tokenizer_config.json
[INFO|2026-01-03 20:24:31] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1400/special_tokens_map.json
[INFO|2026-01-03 20:25:06] logging.py:143 >> {'loss': 0.5123, 'learning_rate': 2.5835e-06, 'epoch': 2.56, 'throughput': 857.49}
[INFO|2026-01-03 20:25:39] logging.py:143 >> {'loss': 0.4564, 'learning_rate': 2.4787e-06, 'epoch': 2.57, 'throughput': 857.52}
[INFO|2026-01-03 20:26:11] logging.py:143 >> {'loss': 0.5329, 'learning_rate': 2.3761e-06, 'epoch': 2.58, 'throughput': 857.44}
[INFO|2026-01-03 20:26:44] logging.py:143 >> {'loss': 0.5803, 'learning_rate': 2.2755e-06, 'epoch': 2.59, 'throughput': 857.36}
[INFO|2026-01-03 20:27:18] logging.py:143 >> {'loss': 0.4759, 'learning_rate': 2.1769e-06, 'epoch': 2.60, 'throughput': 857.53}
[INFO|2026-01-03 20:27:51] logging.py:143 >> {'loss': 0.4964, 'learning_rate': 2.0805e-06, 'epoch': 2.61, 'throughput': 857.44}
[INFO|2026-01-03 20:28:23] logging.py:143 >> {'loss': 0.4652, 'learning_rate': 1.9861e-06, 'epoch': 2.62, 'throughput': 857.43}
[INFO|2026-01-03 20:28:55] logging.py:143 >> {'loss': 0.5859, 'learning_rate': 1.8939e-06, 'epoch': 2.63, 'throughput': 857.30}
[INFO|2026-01-03 20:29:28] logging.py:143 >> {'loss': 0.6222, 'learning_rate': 1.8037e-06, 'epoch': 2.64, 'throughput': 857.27}
[INFO|2026-01-03 20:30:01] logging.py:143 >> {'loss': 0.3722, 'learning_rate': 1.7157e-06, 'epoch': 2.65, 'throughput': 857.22}
[INFO|2026-01-03 20:30:34] logging.py:143 >> {'loss': 0.5299, 'learning_rate': 1.6298e-06, 'epoch': 2.66, 'throughput': 857.23}
[INFO|2026-01-03 20:31:06] logging.py:143 >> {'loss': 0.4470, 'learning_rate': 1.5460e-06, 'epoch': 2.66, 'throughput': 857.12}
[INFO|2026-01-03 20:31:39] logging.py:143 >> {'loss': 0.3965, 'learning_rate': 1.4644e-06, 'epoch': 2.67, 'throughput': 857.14}
[INFO|2026-01-03 20:32:13] logging.py:143 >> {'loss': 0.4579, 'learning_rate': 1.3849e-06, 'epoch': 2.68, 'throughput': 857.25}
[INFO|2026-01-03 20:32:46] logging.py:143 >> {'loss': 0.5361, 'learning_rate': 1.3076e-06, 'epoch': 2.69, 'throughput': 857.23}
[INFO|2026-01-03 20:33:20] logging.py:143 >> {'loss': 0.4439, 'learning_rate': 1.2325e-06, 'epoch': 2.70, 'throughput': 857.38}
[INFO|2026-01-03 20:33:52] logging.py:143 >> {'loss': 0.3843, 'learning_rate': 1.1595e-06, 'epoch': 2.71, 'throughput': 857.30}
[INFO|2026-01-03 20:34:25] logging.py:143 >> {'loss': 0.4191, 'learning_rate': 1.0887e-06, 'epoch': 2.72, 'throughput': 857.21}
[INFO|2026-01-03 20:34:56] logging.py:143 >> {'loss': 0.4898, 'learning_rate': 1.0200e-06, 'epoch': 2.73, 'throughput': 857.05}
[INFO|2026-01-03 20:35:29] logging.py:143 >> {'loss': 0.4288, 'learning_rate': 9.5359e-07, 'epoch': 2.74, 'throughput': 857.00}
[INFO|2026-01-03 20:35:29] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1500
[INFO|2026-01-03 20:35:29] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 20:35:29] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 20:35:30] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1500/chat_template.jinja
[INFO|2026-01-03 20:35:30] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1500/tokenizer_config.json
[INFO|2026-01-03 20:35:30] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1500/special_tokens_map.json
[INFO|2026-01-03 20:36:03] logging.py:143 >> {'loss': 0.4114, 'learning_rate': 8.8935e-07, 'epoch': 2.75, 'throughput': 856.85}
[INFO|2026-01-03 20:36:35] logging.py:143 >> {'loss': 0.4987, 'learning_rate': 8.2730e-07, 'epoch': 2.76, 'throughput': 856.68}
[INFO|2026-01-03 20:49:25] logging.py:143 >> {'loss': 0.4648, 'learning_rate': 7.6746e-07, 'epoch': 2.77, 'throughput': 798.02}
[INFO|2026-01-03 20:49:57] logging.py:143 >> {'loss': 0.5213, 'learning_rate': 7.0984e-07, 'epoch': 2.77, 'throughput': 798.11}
[INFO|2026-01-03 20:50:30] logging.py:143 >> {'loss': 0.4990, 'learning_rate': 6.5443e-07, 'epoch': 2.78, 'throughput': 798.29}
[INFO|2026-01-03 20:51:04] logging.py:143 >> {'loss': 0.4285, 'learning_rate': 6.0125e-07, 'epoch': 2.79, 'throughput': 798.47}
[INFO|2026-01-03 20:51:36] logging.py:143 >> {'loss': 0.4615, 'learning_rate': 5.5029e-07, 'epoch': 2.80, 'throughput': 798.61}
[INFO|2026-01-03 20:52:09] logging.py:143 >> {'loss': 0.4648, 'learning_rate': 5.0156e-07, 'epoch': 2.81, 'throughput': 798.77}
[INFO|2026-01-03 20:52:41] logging.py:143 >> {'loss': 0.4858, 'learning_rate': 4.5508e-07, 'epoch': 2.82, 'throughput': 798.76}
[INFO|2026-01-03 20:53:14] logging.py:143 >> {'loss': 0.4944, 'learning_rate': 4.1083e-07, 'epoch': 2.83, 'throughput': 798.92}
[INFO|2026-01-03 20:53:45] logging.py:143 >> {'loss': 0.4598, 'learning_rate': 3.6883e-07, 'epoch': 2.84, 'throughput': 798.90}
[INFO|2026-01-03 20:54:19] logging.py:143 >> {'loss': 0.4981, 'learning_rate': 3.2907e-07, 'epoch': 2.85, 'throughput': 799.16}
[INFO|2026-01-03 20:54:51] logging.py:143 >> {'loss': 0.5183, 'learning_rate': 2.9157e-07, 'epoch': 2.86, 'throughput': 799.31}
[INFO|2026-01-03 20:55:25] logging.py:143 >> {'loss': 0.4789, 'learning_rate': 2.5632e-07, 'epoch': 2.87, 'throughput': 799.60}
[INFO|2026-01-03 20:55:57] logging.py:143 >> {'loss': 0.4666, 'learning_rate': 2.2333e-07, 'epoch': 2.87, 'throughput': 799.65}
[INFO|2026-01-03 20:56:32] logging.py:143 >> {'loss': 0.4199, 'learning_rate': 1.9261e-07, 'epoch': 2.88, 'throughput': 800.12}
[INFO|2026-01-03 20:57:06] logging.py:143 >> {'loss': 0.4318, 'learning_rate': 1.6415e-07, 'epoch': 2.89, 'throughput': 800.36}
[INFO|2026-01-03 20:57:40] logging.py:143 >> {'loss': 0.3600, 'learning_rate': 1.3795e-07, 'epoch': 2.90, 'throughput': 800.67}
[INFO|2026-01-03 20:58:13] logging.py:143 >> {'loss': 0.4392, 'learning_rate': 1.1403e-07, 'epoch': 2.91, 'throughput': 800.81}
[INFO|2026-01-03 20:58:47] logging.py:143 >> {'loss': 0.4489, 'learning_rate': 9.2377e-08, 'epoch': 2.92, 'throughput': 801.10}
[INFO|2026-01-03 20:58:47] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1600
[INFO|2026-01-03 20:58:47] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 20:58:47] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 20:58:47] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1600/chat_template.jinja
[INFO|2026-01-03 20:58:47] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1600/tokenizer_config.json
[INFO|2026-01-03 20:58:47] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1600/special_tokens_map.json
[INFO|2026-01-03 20:59:22] logging.py:143 >> {'loss': 0.5492, 'learning_rate': 7.2999e-08, 'epoch': 2.93, 'throughput': 801.33}
[INFO|2026-01-03 20:59:56] logging.py:143 >> {'loss': 0.5030, 'learning_rate': 5.5896e-08, 'epoch': 2.94, 'throughput': 801.57}
[INFO|2026-01-03 21:00:30] logging.py:143 >> {'loss': 0.4689, 'learning_rate': 4.1070e-08, 'epoch': 2.95, 'throughput': 801.85}
[INFO|2026-01-03 21:01:03] logging.py:143 >> {'loss': 0.5420, 'learning_rate': 2.8524e-08, 'epoch': 2.96, 'throughput': 801.98}
[INFO|2026-01-03 21:01:36] logging.py:143 >> {'loss': 0.3969, 'learning_rate': 1.8256e-08, 'epoch': 2.97, 'throughput': 802.24}
[INFO|2026-01-03 21:02:09] logging.py:143 >> {'loss': 0.4960, 'learning_rate': 1.0270e-08, 'epoch': 2.98, 'throughput': 802.42}
[INFO|2026-01-03 21:02:42] logging.py:143 >> {'loss': 0.5585, 'learning_rate': 4.5645e-09, 'epoch': 2.98, 'throughput': 802.55}
[INFO|2026-01-03 21:03:16] logging.py:143 >> {'loss': 0.4553, 'learning_rate': 1.1412e-09, 'epoch': 2.99, 'throughput': 802.80}
[INFO|2026-01-03 21:03:40] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1644
[INFO|2026-01-03 21:03:40] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 21:03:40] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 21:03:41] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1644/chat_template.jinja
[INFO|2026-01-03 21:03:41] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1644/tokenizer_config.json
[INFO|2026-01-03 21:03:41] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/checkpoint-1644/special_tokens_map.json
[INFO|2026-01-03 21:03:42] trainer.py:2718 >> 

Training completed. Do not forget to share your model on huggingface.co/models =)


[INFO|2026-01-03 21:03:42] trainer.py:4074 >> Saving model checkpoint to saves/Qwen3-1.7B-Base/lora/train_2026-01-03
[INFO|2026-01-03 21:03:42] configuration_utils.py:750 >> loading configuration file /root/autodl-tmp/models/modelscope_cache/models/Qwen/Qwen3-1___7B-Base/config.json
[INFO|2026-01-03 21:03:42] configuration_utils.py:817 >> Model config Qwen3Config {
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 151643,
  "eos_token_id": 151643,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "max_window_layers": 28,
  "model_type": "qwen3",
  "num_attention_heads": 16,
  "num_hidden_layers": 28,
  "num_key_value_heads": 8,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000,
  "sliding_window": null,
  "tie_word_embeddings": true,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.55.0",
  "use_cache": true,
  "use_sliding_window": false,
  "vocab_size": 151936
}

[INFO|2026-01-03 21:03:42] tokenization_utils_base.py:2393 >> chat template saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/chat_template.jinja
[INFO|2026-01-03 21:03:42] tokenization_utils_base.py:2562 >> tokenizer config file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/tokenizer_config.json
[INFO|2026-01-03 21:03:42] tokenization_utils_base.py:2571 >> Special tokens file saved in saves/Qwen3-1.7B-Base/lora/train_2026-01-03/special_tokens_map.json
[WARNING|2026-01-03 21:03:42] logging.py:148 >> No metric eval_loss to plot.
[WARNING|2026-01-03 21:03:42] logging.py:148 >> No metric eval_accuracy to plot.
[INFO|2026-01-03 21:03:42] modelcard.py:456 >> Dropping the following result as it does not have all the necessary fields:
{'task': {'name': 'Causal Language Modeling', 'type': 'text-generation'}}
