[INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file vocab.json [WARNING|2025-10-19 08:28:09] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [WARNING|2025-10-19 08:28:08] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file merges.txt [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file tokenizer.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file added_tokens.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file special_tokens_map.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file tokenizer_config.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file chat_template.jinja [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2323 >> Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained. [INFO|2025-10-19 08:28:08] image_processing_base.py:379 >> loading configuration file /gemini-3/space/thu/zhaozhiyuan/wfy-mptsnet/Qwen2.5-VL-7B-Instruct/preprocessor_config.json [INFO|2025-10-19 08:28:08] image_processing_base.py:379 >> loading configuration file /gemini-3/space/thu/zhaozhiyuan/wfy-mptsnet/Qwen2.5-VL-7B-Instruct/preprocessor_config.json [WARNING|2025-10-19 08:28:08] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [INFO|2025-10-19 08:28:08] image_processing_base.py:434 >> Image processor Qwen2VLImageProcessor { "do_convert_rgb": true, "do_normalize": true, "do_rescale": true, "do_resize": true, "image_mean": [ 0.48145466, 0.4578275, 0.40821073 ], "image_processor_type": "Qwen2VLImageProcessor", "image_std": [ 0.26862954, 0.26130258, 0.27577711 ], "max_pixels": 12845056, "merge_size": 2, "min_pixels": 3136, "patch_size": 14, "processor_class": "Qwen2_5_VLProcessor", "resample": 3, "rescale_factor": 0.00392156862745098, "size": { "longest_edge": 12845056, "shortest_edge": 3136 }, "temporal_patch_size": 2 } [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file vocab.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file merges.txt [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file tokenizer.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file added_tokens.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file special_tokens_map.json [WARNING|2025-10-19 08:28:09] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [WARNING|2025-10-19 08:28:09] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [WARNING|2025-10-19 08:28:09] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [WARNING|2025-10-19 08:28:09] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [WARNING|2025-10-19 08:28:09] logging.py:329 >> Using a slow image processor as `use_fast` is unset and a slow processor was saved with this model. `use_fast=True` will be the default behavior in v4.50, even if the model was saved with a slow processor. This will result in minor differences in outputs. You'll still be able to use a slow processor with `use_fast=False`. [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file tokenizer_config.json [INFO|2025-10-19 08:28:08] tokenization_utils_base.py:2058 >> loading file chat_template.jinja [INFO|2025-10-19 08:28:09] tokenization_utils_base.py:2323 >> Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained. [INFO|2025-10-19 08:28:09] processing_utils.py:876 >> Processor Qwen2_5_VLProcessor: - image_processor: Qwen2VLImageProcessor { "do_convert_rgb": true, "do_normalize": true, "do_rescale": true, "do_resize": true, "image_mean": [ 0.48145466, 0.4578275, 0.40821073 ], "image_processor_type": "Qwen2VLImageProcessor", "image_std": [ 0.26862954, 0.26130258, 0.27577711 ], "max_pixels": 12845056, "merge_size": 2, "min_pixels": 3136, "patch_size": 14, "processor_class": "Qwen2_5_VLProcessor", "resample": 3, "rescale_factor": 0.00392156862745098, "size": { "longest_edge": 12845056, "shortest_edge": 3136 }, "temporal_patch_size": 2 } - tokenizer: Qwen2TokenizerFast(name_or_path='/gemini-3/space/thu/zhaozhiyuan/wfy-mptsnet/Qwen2.5-VL-7B-Instruct', vocab_size=151643, model_max_length=131072, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={ 151643: AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151644: AddedToken("<|im_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151645: AddedToken("<|im_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151646: AddedToken("<|object_ref_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151647: AddedToken("<|object_ref_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151648: AddedToken("<|box_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151649: AddedToken("<|box_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151650: AddedToken("<|quad_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151651: AddedToken("<|quad_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151652: AddedToken("<|vision_start|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151653: AddedToken("<|vision_end|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151654: AddedToken("<|vision_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151655: AddedToken("<|image_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151656: AddedToken("<|video_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True), 151657: AddedToken("", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), 151658: AddedToken("", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), 151659: AddedToken("<|fim_prefix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), 151660: AddedToken("<|fim_middle|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), 151661: AddedToken("<|fim_suffix|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), 151662: AddedToken("<|fim_pad|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), 151663: AddedToken("<|repo_name|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), 151664: AddedToken("<|file_sep|>", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False), } ) { "processor_class": "Qwen2_5_VLProcessor" } [INFO|2025-10-19 08:28:09] logging.py:157 >> Add <|im_end|> to stop words. [INFO|2025-10-19 08:28:09] logging.py:157 >> Loading dataset critic_training_data.json... [INFO|2025-10-19 08:28:33] logging.py:157 >> Loading dataset refined.json... [INFO|2025-10-19 08:28:57] logging.py:157 >> Loading dataset omnisvg.json... [INFO|2025-10-19 08:29:22] logging.py:157 >> Loading dataset svgen.json... [INFO|2025-10-19 08:29:51] logging.py:157 >> Loading dataset llm4svg.json... [INFO|2025-10-19 08:32:22] configuration_utils.py:697 >> loading configuration file /gemini-3/space/thu/zhaozhiyuan/wfy-mptsnet/Qwen2.5-VL-7B-Instruct/config.json [INFO|2025-10-19 08:32:22] configuration_utils.py:771 >> Model config Qwen2_5_VLConfig { "architectures": [ "Qwen2_5_VLForConditionalGeneration" ], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151645, "hidden_act": "silu", "hidden_size": 3584, "image_token_id": 151655, "initializer_range": 0.02, "intermediate_size": 18944, "max_position_embeddings": 128000, "max_window_layers": 28, "model_type": "qwen2_5_vl", "num_attention_heads": 28, "num_hidden_layers": 28, "num_key_value_heads": 4, "rms_norm_eps": 1e-06, "rope_scaling": { "mrope_section": [ 16, 24, 24 ], "rope_type": "default", "type": "default" }, "rope_theta": 1000000.0, "sliding_window": 32768, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "transformers_version": "4.50.0", "use_cache": true, "use_sliding_window": false, "video_token_id": 151656, "vision_config": { "depth": 32, "fullatt_block_indexes": [ 7, 15, 23, 31 ], "hidden_act": "silu", "hidden_size": 1280, "in_channels": 3, "in_chans": 3, "intermediate_size": 3420, "model_type": "qwen2_5_vl", "num_heads": 16, "out_hidden_size": 3584, "patch_size": 14, "spatial_merge_size": 2, "spatial_patch_size": 14, "temporal_patch_size": 2, "tokens_per_second": 2, "window_size": 112 }, "vision_end_token_id": 151653, "vision_start_token_id": 151652, "vision_token_id": 151654, "vocab_size": 152064 } [INFO|2025-10-19 08:32:22] modeling_utils.py:1151 >> loading weights file /gemini-3/space/thu/zhaozhiyuan/wfy-mptsnet/Qwen2.5-VL-7B-Instruct/model.safetensors.index.json [INFO|2025-10-19 08:32:22] modeling_utils.py:3747 >> Detected DeepSpeed ZeRO-3: activating zero.init() for this model [INFO|2025-10-19 08:32:22] configuration_utils.py:1139 >> Generate config GenerationConfig { "bos_token_id": 151643, "eos_token_id": 151645 } [INFO|2025-10-19 08:32:22] modeling_utils.py:2170 >> Instantiating Qwen2_5_VisionTransformerPretrainedModel model under default dtype torch.float32. [INFO|2025-10-19 08:32:43] modeling_utils.py:4987 >> All model checkpoint weights were used when initializing Qwen2_5_VLForConditionalGeneration. [INFO|2025-10-19 08:32:43] modeling_utils.py:4995 >> All the weights of Qwen2_5_VLForConditionalGeneration were initialized from the model checkpoint at /gemini-3/space/thu/zhaozhiyuan/wfy-mptsnet/Qwen2.5-VL-7B-Instruct. If your task is similar to the task the model of the checkpoint was trained on, you can already use Qwen2_5_VLForConditionalGeneration for predictions without further training. [INFO|2025-10-19 08:32:43] configuration_utils.py:1092 >> loading configuration file /gemini-3/space/thu/zhaozhiyuan/wfy-mptsnet/Qwen2.5-VL-7B-Instruct/generation_config.json [INFO|2025-10-19 08:32:43] configuration_utils.py:1139 >> Generate config GenerationConfig { "bos_token_id": 151643, "do_sample": true, "eos_token_id": [ 151645, 151643 ], "pad_token_id": 151643, "repetition_penalty": 1.05, "temperature": 1e-06 } [INFO|2025-10-19 08:32:43] logging.py:157 >> Gradient checkpointing enabled. [INFO|2025-10-19 08:32:43] logging.py:157 >> Using torch SDPA for faster training and inference. [INFO|2025-10-19 08:32:43] logging.py:157 >> ZeRO3 / FSDP detected, remaining trainable params in float32. [INFO|2025-10-19 08:32:43] logging.py:157 >> Fine-tuning method: Full [INFO|2025-10-19 08:32:43] logging.py:157 >> Set vision model not trainable: ['visual.patch_embed', 'visual.blocks']. [INFO|2025-10-19 08:32:43] logging.py:157 >> Set multi model projector not trainable: visual.merger. [INFO|2025-10-19 08:32:43] logging.py:157 >> trainable params: 7,615,616,512 || all params: 8,292,166,656 || trainable%: 91.8411 [INFO|2025-10-19 08:32:43] trainer.py:748 >> Using auto half precision backend [INFO|2025-10-19 08:32:46] trainer.py:2409 >> ***** Running training ***** [INFO|2025-10-19 08:32:46] trainer.py:2410 >> Num examples = 296,675 [INFO|2025-10-19 08:32:46] trainer.py:2411 >> Num Epochs = 3 [INFO|2025-10-19 08:32:46] trainer.py:2412 >> Instantaneous batch size per device = 2 [INFO|2025-10-19 08:32:46] trainer.py:2415 >> Total train batch size (w. parallel, distributed & accumulation) = 128 [INFO|2025-10-19 08:32:46] trainer.py:2416 >> Gradient Accumulation steps = 8 [INFO|2025-10-19 08:32:46] trainer.py:2417 >> Total optimization steps = 6,951 [INFO|2025-10-19 08:32:47] trainer.py:2418 >> Number of trainable parameters = 7,615,616,512 [INFO|2025-10-19 08:41:51] logging.py:157 >> {'loss': 1.4310, 'learning_rate': 9.9998e-05, 'epoch': 0.01, 'throughput': 9395.55} [INFO|2025-10-19 08:50:37] logging.py:157 >> {'loss': 0.5348, 'learning_rate': 9.9991e-05, 'epoch': 0.02, 'throughput': 9478.63} [INFO|2025-10-19 08:59:17] logging.py:157 >> {'loss': 0.4599, 'learning_rate': 9.9980e-05, 'epoch': 0.03, 'throughput': 9537.12} [INFO|2025-10-19 09:08:06] logging.py:157 >> {'loss': 0.4108, 'learning_rate': 9.9964e-05, 'epoch': 0.04, 'throughput': 9507.78} [INFO|2025-10-19 09:17:01] logging.py:157 >> {'loss': 0.3947, 'learning_rate': 9.9944e-05, 'epoch': 0.05, 'throughput': 9470.14} [INFO|2025-10-19 09:25:56] logging.py:157 >> {'loss': 0.3764, 'learning_rate': 9.9919e-05, 'epoch': 0.05, 'throughput': 9515.96} [INFO|2025-10-19 09:34:29] logging.py:157 >> {'loss': 0.3751, 'learning_rate': 9.9890e-05, 'epoch': 0.06, 'throughput': 9525.58} [INFO|2025-10-19 09:43:07] logging.py:157 >> {'loss': 0.3704, 'learning_rate': 9.9856e-05, 'epoch': 0.07, 'throughput': 9513.01} [INFO|2025-10-19 09:52:01] logging.py:157 >> {'loss': 0.3552, 'learning_rate': 9.9818e-05, 'epoch': 0.08, 'throughput': 9524.90} [INFO|2025-10-19 10:00:47] logging.py:157 >> {'loss': 0.3576, 'learning_rate': 9.9775e-05, 'epoch': 0.09, 'throughput': 9525.35} [INFO|2025-10-19 10:09:47] logging.py:157 >> {'loss': 0.3451, 'learning_rate': 9.9728e-05, 'epoch': 0.10, 'throughput': 9525.43} [INFO|2025-10-19 10:18:39] logging.py:157 >> {'loss': 0.3506, 'learning_rate': 9.9676e-05, 'epoch': 0.11, 'throughput': 9524.66} [INFO|2025-10-19 10:27:37] logging.py:157 >> {'loss': 0.3443, 'learning_rate': 9.9620e-05, 'epoch': 0.12, 'throughput': 9531.29} [INFO|2025-10-19 10:36:00] logging.py:157 >> {'loss': 0.3446, 'learning_rate': 9.9559e-05, 'epoch': 0.13, 'throughput': 9532.39} [INFO|2025-10-19 10:44:24] logging.py:157 >> {'loss': 0.3362, 'learning_rate': 9.9494e-05, 'epoch': 0.14, 'throughput': 9543.53} [INFO|2025-10-19 10:53:32] logging.py:157 >> {'loss': 0.3384, 'learning_rate': 9.9425e-05, 'epoch': 0.14, 'throughput': 9528.67} [INFO|2025-10-19 11:02:31] logging.py:157 >> {'loss': 0.3315, 'learning_rate': 9.9351e-05, 'epoch': 0.15, 'throughput': 9534.93} [INFO|2025-10-19 11:11:12] logging.py:157 >> {'loss': 0.3300, 'learning_rate': 9.9272e-05, 'epoch': 0.16, 'throughput': 9542.47} [INFO|2025-10-19 11:20:21] logging.py:157 >> {'loss': 0.3296, 'learning_rate': 9.9189e-05, 'epoch': 0.17, 'throughput': 9530.71} [INFO|2025-10-19 11:29:10] logging.py:157 >> {'loss': 0.3315, 'learning_rate': 9.9102e-05, 'epoch': 0.18, 'throughput': 9530.11} [INFO|2025-10-19 11:38:03] logging.py:157 >> {'loss': 0.3530, 'learning_rate': 9.9010e-05, 'epoch': 0.19, 'throughput': 9524.36} [INFO|2025-10-19 11:47:11] logging.py:157 >> {'loss': 0.3414, 'learning_rate': 9.8914e-05, 'epoch': 0.20, 'throughput': 9514.36} [INFO|2025-10-19 11:56:09] logging.py:157 >> {'loss': 0.3224, 'learning_rate': 9.8813e-05, 'epoch': 0.21, 'throughput': 9515.30} [INFO|2025-10-19 12:05:15] logging.py:157 >> {'loss': 0.3212, 'learning_rate': 9.8708e-05, 'epoch': 0.22, 'throughput': 9511.06} [INFO|2025-10-19 12:14:07] logging.py:157 >> {'loss': 0.3213, 'learning_rate': 9.8599e-05, 'epoch': 0.23, 'throughput': 9509.05} [INFO|2025-10-19 12:22:40] logging.py:157 >> {'loss': 0.3198, 'learning_rate': 9.8485e-05, 'epoch': 0.24, 'throughput': 9514.22} [INFO|2025-10-19 12:31:22] logging.py:157 >> {'loss': 0.3149, 'learning_rate': 9.8367e-05, 'epoch': 0.24, 'throughput': 9518.41} [INFO|2025-10-19 12:40:35] logging.py:157 >> {'loss': 0.3169, 'learning_rate': 9.8245e-05, 'epoch': 0.25, 'throughput': 9510.51} [INFO|2025-10-19 12:49:33] logging.py:157 >> {'loss': 0.3142, 'learning_rate': 9.8118e-05, 'epoch': 0.26, 'throughput': 9501.43} [INFO|2025-10-19 12:58:19] logging.py:157 >> {'loss': 0.3097, 'learning_rate': 9.7987e-05, 'epoch': 0.27, 'throughput': 9499.35} [INFO|2025-10-19 13:07:12] logging.py:157 >> {'loss': 0.3116, 'learning_rate': 9.7851e-05, 'epoch': 0.28, 'throughput': 9500.56} [INFO|2025-10-19 13:16:12] logging.py:157 >> {'loss': 0.3086, 'learning_rate': 9.7712e-05, 'epoch': 0.29, 'throughput': 9499.14} [INFO|2025-10-19 13:25:17] logging.py:157 >> {'loss': 0.3057, 'learning_rate': 9.7567e-05, 'epoch': 0.30, 'throughput': 9491.21} [INFO|2025-10-19 13:34:03] logging.py:157 >> {'loss': 0.3017, 'learning_rate': 9.7419e-05, 'epoch': 0.31, 'throughput': 9494.90} [INFO|2025-10-19 13:43:13] logging.py:157 >> {'loss': 0.3066, 'learning_rate': 9.7266e-05, 'epoch': 0.32, 'throughput': 9489.84} [INFO|2025-10-19 13:52:17] logging.py:157 >> {'loss': 0.2996, 'learning_rate': 9.7110e-05, 'epoch': 0.33, 'throughput': 9490.92} [INFO|2025-10-19 14:01:12] logging.py:157 >> {'loss': 0.2972, 'learning_rate': 9.6948e-05, 'epoch': 0.34, 'throughput': 9489.46} [INFO|2025-10-19 14:10:12] logging.py:157 >> {'loss': 0.2984, 'learning_rate': 9.6783e-05, 'epoch': 0.34, 'throughput': 9487.66} [INFO|2025-10-19 14:18:52] logging.py:157 >> {'loss': 0.2957, 'learning_rate': 9.6614e-05, 'epoch': 0.35, 'throughput': 9493.32} [INFO|2025-10-19 14:28:01] logging.py:157 >> {'loss': 0.2926, 'learning_rate': 9.6440e-05, 'epoch': 0.36, 'throughput': 9491.78} [INFO|2025-10-19 14:36:47] logging.py:157 >> {'loss': 0.2909, 'learning_rate': 9.6262e-05, 'epoch': 0.37, 'throughput': 9494.44} [INFO|2025-10-19 14:45:41] logging.py:157 >> {'loss': 0.2954, 'learning_rate': 9.6080e-05, 'epoch': 0.38, 'throughput': 9492.66} [INFO|2025-10-19 14:54:34] logging.py:157 >> {'loss': 0.2912, 'learning_rate': 9.5893e-05, 'epoch': 0.39, 'throughput': 9497.80} [INFO|2025-10-19 15:03:25] logging.py:157 >> {'loss': 0.2928, 'learning_rate': 9.5703e-05, 'epoch': 0.40, 'throughput': 9498.61} [INFO|2025-10-19 15:11:44] logging.py:157 >> {'loss': 0.2920, 'learning_rate': 9.5508e-05, 'epoch': 0.41, 'throughput': 9506.77} [INFO|2025-10-19 15:20:51] logging.py:157 >> {'loss': 0.2920, 'learning_rate': 9.5310e-05, 'epoch': 0.42, 'throughput': 9502.21} [INFO|2025-10-19 15:29:38] logging.py:157 >> {'loss': 0.2959, 'learning_rate': 9.5107e-05, 'epoch': 0.43, 'throughput': 9500.11} [INFO|2025-10-19 15:38:41] logging.py:157 >> {'loss': 0.2884, 'learning_rate': 9.4900e-05, 'epoch': 0.43, 'throughput': 9497.19} [INFO|2025-10-19 15:47:33] logging.py:157 >> {'loss': 0.2878, 'learning_rate': 9.4690e-05, 'epoch': 0.44, 'throughput': 9494.35} [INFO|2025-10-19 15:56:42] logging.py:157 >> {'loss': 0.2908, 'learning_rate': 9.4475e-05, 'epoch': 0.45, 'throughput': 9490.15} [INFO|2025-10-19 16:06:06] logging.py:157 >> {'loss': 0.2796, 'learning_rate': 9.4256e-05, 'epoch': 0.46, 'throughput': 9485.80} [INFO|2025-10-19 16:15:10] logging.py:157 >> {'loss': 0.2858, 'learning_rate': 9.4033e-05, 'epoch': 0.47, 'throughput': 9484.86} [INFO|2025-10-19 16:23:56] logging.py:157 >> {'loss': 0.2876, 'learning_rate': 9.3806e-05, 'epoch': 0.48, 'throughput': 9486.71} [INFO|2025-10-19 16:33:11] logging.py:157 >> {'loss': 0.2814, 'learning_rate': 9.3575e-05, 'epoch': 0.49, 'throughput': 9486.33} [INFO|2025-10-19 16:42:03] logging.py:157 >> {'loss': 0.2804, 'learning_rate': 9.3341e-05, 'epoch': 0.50, 'throughput': 9491.48} [INFO|2025-10-19 16:51:06] logging.py:157 >> {'loss': 0.2863, 'learning_rate': 9.3102e-05, 'epoch': 0.51, 'throughput': 9490.65} [INFO|2025-10-19 16:59:50] logging.py:157 >> {'loss': 0.2851, 'learning_rate': 9.2860e-05, 'epoch': 0.52, 'throughput': 9494.69} [INFO|2025-10-19 17:08:35] logging.py:157 >> {'loss': 0.2798, 'learning_rate': 9.2613e-05, 'epoch': 0.53, 'throughput': 9497.76} [INFO|2025-10-19 17:17:38] logging.py:157 >> {'loss': 0.2728, 'learning_rate': 9.2363e-05, 'epoch': 0.53, 'throughput': 9497.60} [INFO|2025-10-19 17:26:07] logging.py:157 >> {'loss': 0.2756, 'learning_rate': 9.2109e-05, 'epoch': 0.54, 'throughput': 9500.30} [INFO|2025-10-19 17:34:57] logging.py:157 >> {'loss': 0.2747, 'learning_rate': 9.1851e-05, 'epoch': 0.55, 'throughput': 9501.41} [INFO|2025-10-19 17:43:50] logging.py:157 >> {'loss': 0.2772, 'learning_rate': 9.1590e-05, 'epoch': 0.56, 'throughput': 9503.01} [INFO|2025-10-19 17:52:25] logging.py:157 >> {'loss': 0.2785, 'learning_rate': 9.1325e-05, 'epoch': 0.57, 'throughput': 9508.07} [INFO|2025-10-19 18:01:25] logging.py:157 >> {'loss': 0.2755, 'learning_rate': 9.1056e-05, 'epoch': 0.58, 'throughput': 9509.96} [INFO|2025-10-19 18:10:33] logging.py:157 >> {'loss': 0.2708, 'learning_rate': 9.0783e-05, 'epoch': 0.59, 'throughput': 9506.81} [INFO|2025-10-19 18:19:36] logging.py:157 >> {'loss': 0.2729, 'learning_rate': 9.0507e-05, 'epoch': 0.60, 'throughput': 9504.96} [INFO|2025-10-19 18:28:32] logging.py:157 >> {'loss': 0.2754, 'learning_rate': 9.0227e-05, 'epoch': 0.61, 'throughput': 9505.70} [INFO|2025-10-19 18:37:36] logging.py:157 >> {'loss': 0.2742, 'learning_rate': 8.9943e-05, 'epoch': 0.62, 'throughput': 9500.81} [INFO|2025-10-19 18:46:09] logging.py:157 >> {'loss': 0.2703, 'learning_rate': 8.9656e-05, 'epoch': 0.63, 'throughput': 9503.01} [INFO|2025-10-19 18:54:51] logging.py:157 >> {'loss': 0.2756, 'learning_rate': 8.9365e-05, 'epoch': 0.63, 'throughput': 9502.95} [INFO|2025-10-19 19:04:00] logging.py:157 >> {'loss': 0.2698, 'learning_rate': 8.9070e-05, 'epoch': 0.64, 'throughput': 9501.97} [INFO|2025-10-19 19:13:03] logging.py:157 >> {'loss': 0.2716, 'learning_rate': 8.8773e-05, 'epoch': 0.65, 'throughput': 9499.40} [INFO|2025-10-19 19:22:00] logging.py:157 >> {'loss': 0.2682, 'learning_rate': 8.8471e-05, 'epoch': 0.66, 'throughput': 9498.48} [INFO|2025-10-19 19:30:38] logging.py:157 >> {'loss': 0.2695, 'learning_rate': 8.8166e-05, 'epoch': 0.67, 'throughput': 9500.35} [INFO|2025-10-19 19:39:20] logging.py:157 >> {'loss': 0.2706, 'learning_rate': 8.7858e-05, 'epoch': 0.68, 'throughput': 9502.46} [INFO|2025-10-19 19:48:19] logging.py:157 >> {'loss': 0.2704, 'learning_rate': 8.7546e-05, 'epoch': 0.69, 'throughput': 9503.27} [INFO|2025-10-19 19:57:01] logging.py:157 >> {'loss': 0.2707, 'learning_rate': 8.7231e-05, 'epoch': 0.70, 'throughput': 9504.39} [INFO|2025-10-19 20:06:09] logging.py:157 >> {'loss': 0.2653, 'learning_rate': 8.6913e-05, 'epoch': 0.71, 'throughput': 9503.23} [INFO|2025-10-19 20:14:45] logging.py:157 >> {'loss': 0.2649, 'learning_rate': 8.6591e-05, 'epoch': 0.72, 'throughput': 9507.46} [INFO|2025-10-19 20:23:37] logging.py:157 >> {'loss': 0.2652, 'learning_rate': 8.6266e-05, 'epoch': 0.72, 'throughput': 9507.43} [INFO|2025-10-19 20:32:21] logging.py:157 >> {'loss': 0.2638, 'learning_rate': 8.5938e-05, 'epoch': 0.73, 'throughput': 9507.28} [INFO|2025-10-19 20:41:00] logging.py:157 >> {'loss': 0.2685, 'learning_rate': 8.5606e-05, 'epoch': 0.74, 'throughput': 9510.67} [INFO|2025-10-19 20:49:36] logging.py:157 >> {'loss': 0.2584, 'learning_rate': 8.5271e-05, 'epoch': 0.75, 'throughput': 9511.61} [INFO|2025-10-19 20:58:33] logging.py:157 >> {'loss': 0.2607, 'learning_rate': 8.4933e-05, 'epoch': 0.76, 'throughput': 9511.41} [INFO|2025-10-19 21:07:10] logging.py:157 >> {'loss': 0.2652, 'learning_rate': 8.4592e-05, 'epoch': 0.77, 'throughput': 9511.99} [INFO|2025-10-19 21:15:56] logging.py:157 >> {'loss': 0.2604, 'learning_rate': 8.4248e-05, 'epoch': 0.78, 'throughput': 9513.64} [INFO|2025-10-19 21:24:38] logging.py:157 >> {'loss': 0.2603, 'learning_rate': 8.3901e-05, 'epoch': 0.79, 'throughput': 9515.83} [INFO|2025-10-19 21:34:06] logging.py:157 >> {'loss': 0.2576, 'learning_rate': 8.3550e-05, 'epoch': 0.80, 'throughput': 9511.18} [INFO|2025-10-19 21:43:04] logging.py:157 >> {'loss': 0.2622, 'learning_rate': 8.3197e-05, 'epoch': 0.81, 'throughput': 9511.76} [INFO|2025-10-19 21:51:57] logging.py:157 >> {'loss': 0.2587, 'learning_rate': 8.2841e-05, 'epoch': 0.82, 'throughput': 9511.99} [INFO|2025-10-19 22:01:15] logging.py:157 >> {'loss': 0.2575, 'learning_rate': 8.2481e-05, 'epoch': 0.82, 'throughput': 9511.08} [INFO|2025-10-19 22:10:20] logging.py:157 >> {'loss': 0.2573, 'learning_rate': 8.2119e-05, 'epoch': 0.83, 'throughput': 9509.35} [INFO|2025-10-19 22:19:20] logging.py:157 >> {'loss': 0.2600, 'learning_rate': 8.1754e-05, 'epoch': 0.84, 'throughput': 9508.08} [INFO|2025-10-19 22:28:30] logging.py:157 >> {'loss': 0.2571, 'learning_rate': 8.1386e-05, 'epoch': 0.85, 'throughput': 9505.71} [INFO|2025-10-19 22:37:09] logging.py:157 >> {'loss': 0.2510, 'learning_rate': 8.1015e-05, 'epoch': 0.86, 'throughput': 9507.90} [INFO|2025-10-19 22:46:00] logging.py:157 >> {'loss': 0.2614, 'learning_rate': 8.0642e-05, 'epoch': 0.87, 'throughput': 9507.12} [INFO|2025-10-19 22:55:24] logging.py:157 >> {'loss': 0.2528, 'learning_rate': 8.0265e-05, 'epoch': 0.88, 'throughput': 9504.87} [INFO|2025-10-19 23:03:49] logging.py:157 >> {'loss': 0.2571, 'learning_rate': 7.9886e-05, 'epoch': 0.89, 'throughput': 9508.31} [INFO|2025-10-19 23:12:48] logging.py:157 >> {'loss': 0.2547, 'learning_rate': 7.9504e-05, 'epoch': 0.90, 'throughput': 9508.57} [INFO|2025-10-19 23:21:42] logging.py:157 >> {'loss': 0.2538, 'learning_rate': 7.9120e-05, 'epoch': 0.91, 'throughput': 9507.51} [INFO|2025-10-19 23:30:30] logging.py:157 >> {'loss': 0.2551, 'learning_rate': 7.8733e-05, 'epoch': 0.92, 'throughput': 9506.69} [INFO|2025-10-19 23:39:42] logging.py:157 >> {'loss': 0.2515, 'learning_rate': 7.8343e-05, 'epoch': 0.92, 'throughput': 9503.49} [INFO|2025-10-19 23:48:36] logging.py:157 >> {'loss': 0.2526, 'learning_rate': 7.7951e-05, 'epoch': 0.93, 'throughput': 9503.75} [INFO|2025-10-19 23:57:34] logging.py:157 >> {'loss': 0.2482, 'learning_rate': 7.7556e-05, 'epoch': 0.94, 'throughput': 9504.35} [INFO|2025-10-20 00:06:42] logging.py:157 >> {'loss': 0.2516, 'learning_rate': 7.7159e-05, 'epoch': 0.95, 'throughput': 9502.88} [INFO|2025-10-20 00:15:05] logging.py:157 >> {'loss': 0.2493, 'learning_rate': 7.6759e-05, 'epoch': 0.96, 'throughput': 9505.71} [INFO|2025-10-20 00:23:54] logging.py:157 >> {'loss': 0.2465, 'learning_rate': 7.6357e-05, 'epoch': 0.97, 'throughput': 9508.14} [INFO|2025-10-20 00:32:41] logging.py:157 >> {'loss': 0.2486, 'learning_rate': 7.5953e-05, 'epoch': 0.98, 'throughput': 9508.27} [INFO|2025-10-20 00:41:47] logging.py:157 >> {'loss': 0.2472, 'learning_rate': 7.5546e-05, 'epoch': 0.99, 'throughput': 9507.12} [INFO|2025-10-20 00:50:40] logging.py:157 >> {'loss': 0.2471, 'learning_rate': 7.5137e-05, 'epoch': 1.00, 'throughput': 9506.69} [INFO|2025-10-20 01:00:01] logging.py:157 >> {'loss': 0.2361, 'learning_rate': 7.4726e-05, 'epoch': 1.01, 'throughput': 9504.55} [INFO|2025-10-20 01:08:47] logging.py:157 >> {'loss': 0.2170, 'learning_rate': 7.4312e-05, 'epoch': 1.02, 'throughput': 9504.99} [INFO|2025-10-20 01:17:18] logging.py:157 >> {'loss': 0.2145, 'learning_rate': 7.3896e-05, 'epoch': 1.02, 'throughput': 9506.33} [INFO|2025-10-20 01:26:16] logging.py:157 >> {'loss': 0.2179, 'learning_rate': 7.3478e-05, 'epoch': 1.03, 'throughput': 9504.64} [INFO|2025-10-20 01:34:49] logging.py:157 >> {'loss': 0.2098, 'learning_rate': 7.3058e-05, 'epoch': 1.04, 'throughput': 9506.20} [INFO|2025-10-20 01:43:41] logging.py:157 >> {'loss': 0.2174, 'learning_rate': 7.2636e-05, 'epoch': 1.05, 'throughput': 9505.78} [INFO|2025-10-20 01:52:29] logging.py:157 >> {'loss': 0.2187, 'learning_rate': 7.2212e-05, 'epoch': 1.06, 'throughput': 9506.65} [INFO|2025-10-20 02:01:32] logging.py:157 >> {'loss': 0.2162, 'learning_rate': 7.1786e-05, 'epoch': 1.07, 'throughput': 9506.10} [INFO|2025-10-20 02:10:19] logging.py:157 >> {'loss': 0.2202, 'learning_rate': 7.1358e-05, 'epoch': 1.08, 'throughput': 9505.76} [INFO|2025-10-20 02:19:23] logging.py:157 >> {'loss': 0.2237, 'learning_rate': 7.0928e-05, 'epoch': 1.09, 'throughput': 9504.97} [INFO|2025-10-20 02:27:59] logging.py:157 >> {'loss': 0.2199, 'learning_rate': 7.0496e-05, 'epoch': 1.10, 'throughput': 9505.35} [INFO|2025-10-20 02:37:16] logging.py:157 >> {'loss': 0.2189, 'learning_rate': 7.0062e-05, 'epoch': 1.11, 'throughput': 9501.61} [INFO|2025-10-20 02:46:26] logging.py:157 >> {'loss': 0.2163, 'learning_rate': 6.9626e-05, 'epoch': 1.11, 'throughput': 9501.64} [INFO|2025-10-20 02:55:16] logging.py:157 >> {'loss': 0.2157, 'learning_rate': 6.9189e-05, 'epoch': 1.12, 'throughput': 9502.18} [INFO|2025-10-20 03:04:29] logging.py:157 >> {'loss': 0.2141, 'learning_rate': 6.8750e-05, 'epoch': 1.13, 'throughput': 9499.47} [INFO|2025-10-20 03:12:55] logging.py:157 >> {'loss': 0.2185, 'learning_rate': 6.8309e-05, 'epoch': 1.14, 'throughput': 9502.37} [INFO|2025-10-20 03:21:51] logging.py:157 >> {'loss': 0.2125, 'learning_rate': 6.7867e-05, 'epoch': 1.15, 'throughput': 9502.47} [INFO|2025-10-20 03:30:50] logging.py:157 >> {'loss': 0.2138, 'learning_rate': 6.7423e-05, 'epoch': 1.16, 'throughput': 9501.96} [INFO|2025-10-20 03:39:45] logging.py:157 >> {'loss': 0.2152, 'learning_rate': 6.6977e-05, 'epoch': 1.17, 'throughput': 9502.59} [INFO|2025-10-20 03:48:29] logging.py:157 >> {'loss': 0.2125, 'learning_rate': 6.6530e-05, 'epoch': 1.18, 'throughput': 9503.45} [INFO|2025-10-20 03:57:37] logging.py:157 >> {'loss': 0.2123, 'learning_rate': 6.6081e-05, 'epoch': 1.19, 'throughput': 9502.01} [INFO|2025-10-20 04:06:37] logging.py:157 >> {'loss': 0.2101, 'learning_rate': 6.5631e-05, 'epoch': 1.20, 'throughput': 9500.47} [INFO|2025-10-20 04:15:30] logging.py:157 >> {'loss': 0.2090, 'learning_rate': 6.5180e-05, 'epoch': 1.21, 'throughput': 9501.47} [INFO|2025-10-20 04:24:12] logging.py:157 >> {'loss': 0.2122, 'learning_rate': 6.4727e-05, 'epoch': 1.21, 'throughput': 9501.91} [INFO|2025-10-20 04:33:14] logging.py:157 >> {'loss': 0.2093, 'learning_rate': 6.4273e-05, 'epoch': 1.22, 'throughput': 9500.84} [INFO|2025-10-20 04:41:55] logging.py:157 >> {'loss': 0.2114, 'learning_rate': 6.3817e-05, 'epoch': 1.23, 'throughput': 9501.80} [INFO|2025-10-20 04:50:50] logging.py:157 >> {'loss': 0.2111, 'learning_rate': 6.3361e-05, 'epoch': 1.24, 'throughput': 9501.17} [INFO|2025-10-20 04:59:36] logging.py:157 >> {'loss': 0.2086, 'learning_rate': 6.2903e-05, 'epoch': 1.25, 'throughput': 9501.16} [INFO|2025-10-20 05:08:27] logging.py:157 >> {'loss': 0.2095, 'learning_rate': 6.2444e-05, 'epoch': 1.26, 'throughput': 9501.00} [INFO|2025-10-20 05:17:03] logging.py:157 >> {'loss': 0.2083, 'learning_rate': 6.1984e-05, 'epoch': 1.27, 'throughput': 9502.81} [INFO|2025-10-20 05:25:43] logging.py:157 >> {'loss': 0.2088, 'learning_rate': 6.1522e-05, 'epoch': 1.28, 'throughput': 9504.07} [INFO|2025-10-20 05:34:38] logging.py:157 >> {'loss': 0.2118, 'learning_rate': 6.1060e-05, 'epoch': 1.29, 'throughput': 9503.94} [INFO|2025-10-20 05:43:25] logging.py:157 >> {'loss': 0.2058, 'learning_rate': 6.0597e-05, 'epoch': 1.30, 'throughput': 9504.75} [INFO|2025-10-20 05:52:28] logging.py:157 >> {'loss': 0.2065, 'learning_rate': 6.0132e-05, 'epoch': 1.31, 'throughput': 9504.80} [INFO|2025-10-20 06:01:14] logging.py:157 >> {'loss': 0.2070, 'learning_rate': 5.9667e-05, 'epoch': 1.31, 'throughput': 9504.25} [INFO|2025-10-20 06:10:19] logging.py:157 >> {'loss': 0.2049, 'learning_rate': 5.9201e-05, 'epoch': 1.32, 'throughput': 9502.94} [INFO|2025-10-20 06:19:34] logging.py:157 >> {'loss': 0.2080, 'learning_rate': 5.8734e-05, 'epoch': 1.33, 'throughput': 9499.75} [INFO|2025-10-20 06:28:22] logging.py:157 >> {'loss': 0.2047, 'learning_rate': 5.8267e-05, 'epoch': 1.34, 'throughput': 9499.82} [INFO|2025-10-20 06:37:17] logging.py:157 >> {'loss': 0.2022, 'learning_rate': 5.7798e-05, 'epoch': 1.35, 'throughput': 9498.47} [INFO|2025-10-20 06:46:48] logging.py:157 >> {'loss': 0.2073, 'learning_rate': 5.7329e-05, 'epoch': 1.36, 'throughput': 9494.01} [INFO|2025-10-20 06:55:58] logging.py:157 >> {'loss': 0.2056, 'learning_rate': 5.6859e-05, 'epoch': 1.37, 'throughput': 9492.45} [INFO|2025-10-20 07:04:26] logging.py:157 >> {'loss': 0.2056, 'learning_rate': 5.6389e-05, 'epoch': 1.38, 'throughput': 9494.77} [INFO|2025-10-20 07:13:31] logging.py:157 >> {'loss': 0.2037, 'learning_rate': 5.5918e-05, 'epoch': 1.39, 'throughput': 9494.81} [INFO|2025-10-20 07:22:03] logging.py:157 >> {'loss': 0.2042, 'learning_rate': 5.5447e-05, 'epoch': 1.40, 'throughput': 9496.67} [INFO|2025-10-20 07:31:07] logging.py:157 >> {'loss': 0.2011, 'learning_rate': 5.4975e-05, 'epoch': 1.40, 'throughput': 9496.92} [INFO|2025-10-20 07:40:10] logging.py:157 >> {'loss': 0.2015, 'learning_rate': 5.4502e-05, 'epoch': 1.41, 'throughput': 9495.35} [INFO|2025-10-20 07:49:28] logging.py:157 >> {'loss': 0.2005, 'learning_rate': 5.4029e-05, 'epoch': 1.42, 'throughput': 9493.13} [INFO|2025-10-20 07:58:08] logging.py:157 >> {'loss': 0.2052, 'learning_rate': 5.3556e-05, 'epoch': 1.43, 'throughput': 9493.86} [INFO|2025-10-20 08:06:44] logging.py:157 >> {'loss': 0.2042, 'learning_rate': 5.3083e-05, 'epoch': 1.44, 'throughput': 9495.07} [INFO|2025-10-20 08:15:42] logging.py:157 >> {'loss': 0.1984, 'learning_rate': 5.2609e-05, 'epoch': 1.45, 'throughput': 9493.99} [INFO|2025-10-20 08:24:50] logging.py:157 >> {'loss': 0.2038, 'learning_rate': 5.2135e-05, 'epoch': 1.46, 'throughput': 9492.20} [INFO|2025-10-20 08:33:47] logging.py:157 >> {'loss': 0.2012, 'learning_rate': 5.1661e-05, 'epoch': 1.47, 'throughput': 9492.22} [INFO|2025-10-20 08:42:27] logging.py:157 >> {'loss': 0.1987, 'learning_rate': 5.1186e-05, 'epoch': 1.48, 'throughput': 9492.76} [INFO|2025-10-20 08:51:30] logging.py:157 >> {'loss': 0.1986, 'learning_rate': 5.0712e-05, 'epoch': 1.49, 'throughput': 9491.54} [INFO|2025-10-20 09:00:19] logging.py:157 >> {'loss': 0.1987, 'learning_rate': 5.0237e-05, 'epoch': 1.50, 'throughput': 9491.31} [INFO|2025-10-20 09:08:54] logging.py:157 >> {'loss': 0.1952, 'learning_rate': 4.9763e-05, 'epoch': 1.50, 'throughput': 9493.38} [INFO|2025-10-20 09:17:40] logging.py:157 >> {'loss': 0.1957, 'learning_rate': 4.9288e-05, 'epoch': 1.51, 'throughput': 9494.32} [INFO|2025-10-20 09:26:33] logging.py:157 >> {'loss': 0.1957, 'learning_rate': 4.8814e-05, 'epoch': 1.52, 'throughput': 9495.01} [INFO|2025-10-20 09:35:27] logging.py:157 >> {'loss': 0.1970, 'learning_rate': 4.8339e-05, 'epoch': 1.53, 'throughput': 9494.56} [INFO|2025-10-20 09:43:58] logging.py:157 >> {'loss': 0.1957, 'learning_rate': 4.7865e-05, 'epoch': 1.54, 'throughput': 9495.47} [INFO|2025-10-20 09:52:47] logging.py:157 >> {'loss': 0.1926, 'learning_rate': 4.7391e-05, 'epoch': 1.55, 'throughput': 9496.47} [INFO|2025-10-20 10:01:54] logging.py:157 >> {'loss': 0.1934, 'learning_rate': 4.6917e-05, 'epoch': 1.56, 'throughput': 9496.66} [INFO|2025-10-20 10:10:48] logging.py:157 >> {'loss': 0.1961, 'learning_rate': 4.6444e-05, 'epoch': 1.57, 'throughput': 9496.52} [INFO|2025-10-20 10:19:52] logging.py:157 >> {'loss': 0.1902, 'learning_rate': 4.5971e-05, 'epoch': 1.58, 'throughput': 9496.01} [INFO|2025-10-20 10:28:51] logging.py:157 >> {'loss': 0.1971, 'learning_rate': 4.5498e-05, 'epoch': 1.59, 'throughput': 9495.37} [INFO|2025-10-20 10:37:40] logging.py:157 >> {'loss': 0.1952, 'learning_rate': 4.5025e-05, 'epoch': 1.59, 'throughput': 9495.50} [INFO|2025-10-20 10:46:19] logging.py:157 >> {'loss': 0.1908, 'learning_rate': 4.4553e-05, 'epoch': 1.60, 'throughput': 9496.69} [INFO|2025-10-20 10:55:21] logging.py:157 >> {'loss': 0.1945, 'learning_rate': 4.4082e-05, 'epoch': 1.61, 'throughput': 9496.20} [INFO|2025-10-20 11:04:17] logging.py:157 >> {'loss': 0.1911, 'learning_rate': 4.3611e-05, 'epoch': 1.62, 'throughput': 9496.70} [INFO|2025-10-20 11:13:15] logging.py:157 >> {'loss': 0.1930, 'learning_rate': 4.3141e-05, 'epoch': 1.63, 'throughput': 9496.86} [INFO|2025-10-20 11:22:03] logging.py:157 >> {'loss': 0.1898, 'learning_rate': 4.2671e-05, 'epoch': 1.64, 'throughput': 9498.11} [INFO|2025-10-20 11:30:38] logging.py:157 >> {'loss': 0.1852, 'learning_rate': 4.2202e-05, 'epoch': 1.65, 'throughput': 9500.27} [INFO|2025-10-20 11:39:38] logging.py:157 >> {'loss': 0.1863, 'learning_rate': 4.1733e-05, 'epoch': 1.66, 'throughput': 9499.96} [INFO|2025-10-20 11:48:37] logging.py:157 >> {'loss': 0.1882, 'learning_rate': 4.1266e-05, 'epoch': 1.67, 'throughput': 9498.96} [INFO|2025-10-20 11:57:24] logging.py:157 >> {'loss': 0.1893, 'learning_rate': 4.0799e-05, 'epoch': 1.68, 'throughput': 9500.06} [INFO|2025-10-20 12:06:33] logging.py:157 >> {'loss': 0.1885, 'learning_rate': 4.0333e-05, 'epoch': 1.69, 'throughput': 9499.68} [INFO|2025-10-20 12:15:46] logging.py:157 >> {'loss': 0.1883, 'learning_rate': 3.9868e-05, 'epoch': 1.69, 'throughput': 9499.21} [INFO|2025-10-20 12:24:39] logging.py:157 >> {'loss': 0.1871, 'learning_rate': 3.9403e-05, 'epoch': 1.70, 'throughput': 9499.01} [INFO|2025-10-20 12:33:32] logging.py:157 >> {'loss': 0.1853, 'learning_rate': 3.8940e-05, 'epoch': 1.71, 'throughput': 9498.79} [INFO|2025-10-20 12:42:17] logging.py:157 >> {'loss': 0.1862, 'learning_rate': 3.8478e-05, 'epoch': 1.72, 'throughput': 9499.53} [INFO|2025-10-20 12:51:21] logging.py:157 >> {'loss': 0.1913, 'learning_rate': 3.8016e-05, 'epoch': 1.73, 'throughput': 9498.33} [INFO|2025-10-20 13:00:21] logging.py:157 >> {'loss': 0.1908, 'learning_rate': 3.7556e-05, 'epoch': 1.74, 'throughput': 9498.07} [INFO|2025-10-20 13:09:43] logging.py:157 >> {'loss': 0.1839, 'learning_rate': 3.7097e-05, 'epoch': 1.75, 'throughput': 9496.29} [INFO|2025-10-20 13:18:59] logging.py:157 >> {'loss': 0.1869, 'learning_rate': 3.6639e-05, 'epoch': 1.76, 'throughput': 9495.72} [INFO|2025-10-20 13:27:58] logging.py:157 >> {'loss': 0.1884, 'learning_rate': 3.6183e-05, 'epoch': 1.77, 'throughput': 9494.70} [INFO|2025-10-20 13:37:11] logging.py:157 >> {'loss': 0.1863, 'learning_rate': 3.5727e-05, 'epoch': 1.78, 'throughput': 9492.68} [INFO|2025-10-20 13:45:49] logging.py:157 >> {'loss': 0.1884, 'learning_rate': 3.5273e-05, 'epoch': 1.79, 'throughput': 9493.64} [INFO|2025-10-20 13:54:26] logging.py:157 >> {'loss': 0.1815, 'learning_rate': 3.4820e-05, 'epoch': 1.79, 'throughput': 9494.28} [INFO|2025-10-20 14:03:24] logging.py:157 >> {'loss': 0.1798, 'learning_rate': 3.4369e-05, 'epoch': 1.80, 'throughput': 9493.88} [INFO|2025-10-20 14:11:55] logging.py:157 >> {'loss': 0.1797, 'learning_rate': 3.3919e-05, 'epoch': 1.81, 'throughput': 9495.04} [INFO|2025-10-20 14:20:35] logging.py:157 >> {'loss': 0.1773, 'learning_rate': 3.3470e-05, 'epoch': 1.82, 'throughput': 9495.33} [INFO|2025-10-20 14:29:27] logging.py:157 >> {'loss': 0.1828, 'learning_rate': 3.3023e-05, 'epoch': 1.83, 'throughput': 9494.34} [INFO|2025-10-20 14:38:22] logging.py:157 >> {'loss': 0.1800, 'learning_rate': 3.2577e-05, 'epoch': 1.84, 'throughput': 9494.06} [INFO|2025-10-20 14:47:18] logging.py:157 >> {'loss': 0.1825, 'learning_rate': 3.2133e-05, 'epoch': 1.85, 'throughput': 9493.51} [INFO|2025-10-20 14:56:14] logging.py:157 >> {'loss': 0.1776, 'learning_rate': 3.1691e-05, 'epoch': 1.86, 'throughput': 9493.01} [INFO|2025-10-20 15:05:18] logging.py:157 >> {'loss': 0.1768, 'learning_rate': 3.1250e-05, 'epoch': 1.87, 'throughput': 9493.04} [INFO|2025-10-20 15:14:21] logging.py:157 >> {'loss': 0.1778, 'learning_rate': 3.0811e-05, 'epoch': 1.88, 'throughput': 9492.18} [INFO|2025-10-20 15:23:32] logging.py:157 >> {'loss': 0.1788, 'learning_rate': 3.0374e-05, 'epoch': 1.88, 'throughput': 9491.15} [INFO|2025-10-20 15:32:43] logging.py:157 >> {'loss': 0.1760, 'learning_rate': 2.9938e-05, 'epoch': 1.89, 'throughput': 9490.66} [INFO|2025-10-20 15:41:43] logging.py:157 >> {'loss': 0.1783, 'learning_rate': 2.9504e-05, 'epoch': 1.90, 'throughput': 9490.01} [INFO|2025-10-20 15:50:49] logging.py:157 >> {'loss': 0.1780, 'learning_rate': 2.9072e-05, 'epoch': 1.91, 'throughput': 9488.58} [INFO|2025-10-20 15:59:52] logging.py:157 >> {'loss': 0.1759, 'learning_rate': 2.8642e-05, 'epoch': 1.92, 'throughput': 9488.29} [INFO|2025-10-20 16:08:51] logging.py:157 >> {'loss': 0.1769, 'learning_rate': 2.8214e-05, 'epoch': 1.93, 'throughput': 9487.56} [INFO|2025-10-20 16:17:39] logging.py:157 >> {'loss': 0.1761, 'learning_rate': 2.7788e-05, 'epoch': 1.94, 'throughput': 9488.08} [INFO|2025-10-20 16:26:37] logging.py:157 >> {'loss': 0.1762, 'learning_rate': 2.7364e-05, 'epoch': 1.95, 'throughput': 9488.56} [INFO|2025-10-20 16:35:36] logging.py:157 >> {'loss': 0.1715, 'learning_rate': 2.6942e-05, 'epoch': 1.96, 'throughput': 9489.02} [INFO|2025-10-20 16:44:11] logging.py:157 >> {'loss': 0.1728, 'learning_rate': 2.6522e-05, 'epoch': 1.97, 'throughput': 9489.67} [INFO|2025-10-20 16:53:22] logging.py:157 >> {'loss': 0.1750, 'learning_rate': 2.6104e-05, 'epoch': 1.98, 'throughput': 9488.69} [INFO|2025-10-20 17:02:13] logging.py:157 >> {'loss': 0.1736, 'learning_rate': 2.5688e-05, 'epoch': 1.98, 'throughput': 9489.47} [INFO|2025-10-20 17:11:21] logging.py:157 >> {'loss': 0.1704, 'learning_rate': 2.5274e-05, 'epoch': 1.99, 'throughput': 9489.29} [INFO|2025-10-20 17:20:16] logging.py:157 >> {'loss': 0.1548, 'learning_rate': 2.4863e-05, 'epoch': 2.00, 'throughput': 9490.07} [INFO|2025-10-20 17:29:16] logging.py:157 >> {'loss': 0.1188, 'learning_rate': 2.4454e-05, 'epoch': 2.01, 'throughput': 9489.70} [INFO|2025-10-20 17:38:27] logging.py:157 >> {'loss': 0.1176, 'learning_rate': 2.4047e-05, 'epoch': 2.02, 'throughput': 9490.21} [INFO|2025-10-20 17:47:24] logging.py:157 >> {'loss': 0.1166, 'learning_rate': 2.3643e-05, 'epoch': 2.03, 'throughput': 9489.93} [INFO|2025-10-20 17:56:19] logging.py:157 >> {'loss': 0.1129, 'learning_rate': 2.3241e-05, 'epoch': 2.04, 'throughput': 9489.27} [INFO|2025-10-20 18:05:04] logging.py:157 >> {'loss': 0.1122, 'learning_rate': 2.2841e-05, 'epoch': 2.05, 'throughput': 9490.08} [INFO|2025-10-20 18:13:56] logging.py:157 >> {'loss': 0.1145, 'learning_rate': 2.2444e-05, 'epoch': 2.06, 'throughput': 9490.09} [INFO|2025-10-20 18:22:51] logging.py:157 >> {'loss': 0.1153, 'learning_rate': 2.2049e-05, 'epoch': 2.07, 'throughput': 9490.25} [INFO|2025-10-20 18:31:59] logging.py:157 >> {'loss': 0.1152, 'learning_rate': 2.1657e-05, 'epoch': 2.08, 'throughput': 9489.42} [INFO|2025-10-20 18:40:42] logging.py:157 >> {'loss': 0.1148, 'learning_rate': 2.1267e-05, 'epoch': 2.08, 'throughput': 9489.77} [INFO|2025-10-20 18:50:03] logging.py:157 >> {'loss': 0.1119, 'learning_rate': 2.0880e-05, 'epoch': 2.09, 'throughput': 9488.55} [INFO|2025-10-20 18:59:08] logging.py:157 >> {'loss': 0.1101, 'learning_rate': 2.0496e-05, 'epoch': 2.10, 'throughput': 9487.59} [INFO|2025-10-20 19:08:09] logging.py:157 >> {'loss': 0.1139, 'learning_rate': 2.0114e-05, 'epoch': 2.11, 'throughput': 9487.45} [INFO|2025-10-20 19:16:41] logging.py:157 >> {'loss': 0.1116, 'learning_rate': 1.9735e-05, 'epoch': 2.12, 'throughput': 9489.64} [INFO|2025-10-20 19:25:49] logging.py:157 >> {'loss': 0.1120, 'learning_rate': 1.9358e-05, 'epoch': 2.13, 'throughput': 9489.09} [INFO|2025-10-20 19:34:50] logging.py:157 >> {'loss': 0.1106, 'learning_rate': 1.8985e-05, 'epoch': 2.14, 'throughput': 9488.78} [INFO|2025-10-20 19:43:50] logging.py:157 >> {'loss': 0.1124, 'learning_rate': 1.8614e-05, 'epoch': 2.15, 'throughput': 9488.59} [INFO|2025-10-20 19:52:41] logging.py:157 >> {'loss': 0.1136, 'learning_rate': 1.8246e-05, 'epoch': 2.16, 'throughput': 9488.83} [INFO|2025-10-20 19:53:28] trainer.py:4289 >> ***** Running Evaluation ***** [INFO|2025-10-20 19:53:28] trainer.py:4291 >> Num examples = 9176 [INFO|2025-10-20 19:53:28] trainer.py:4294 >> Batch size = 2 [INFO|2025-10-20 20:02:26] trainer.py:3966 >> Saving model checkpoint to saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000 [INFO|2025-10-20 20:02:26] configuration_utils.py:423 >> Configuration saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/config.json [INFO|2025-10-20 20:02:26] configuration_utils.py:908 >> Configuration saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/generation_config.json [INFO|2025-10-20 20:09:40] modeling_utils.py:3594 >> The model is bigger than the maximum size per checkpoint (5GB) and is going to be split in 4 checkpoint shards. You can find where each parameters has been saved in the index located at saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/model.safetensors.index.json. [INFO|2025-10-20 20:09:40] tokenization_utils_base.py:2510 >> tokenizer config file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/tokenizer_config.json [INFO|2025-10-20 20:09:40] tokenization_utils_base.py:2519 >> Special tokens file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/special_tokens_map.json [INFO|2025-10-20 20:14:49] image_processing_base.py:261 >> Image processor saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/preprocessor_config.json [INFO|2025-10-20 20:14:49] tokenization_utils_base.py:2510 >> tokenizer config file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/tokenizer_config.json [INFO|2025-10-20 20:14:49] tokenization_utils_base.py:2519 >> Special tokens file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/special_tokens_map.json [INFO|2025-10-20 20:14:49] processing_utils.py:638 >> chat template saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-5000/chat_template.json [INFO|2025-10-20 20:22:41] logging.py:157 >> {'loss': 0.1102, 'learning_rate': 1.7881e-05, 'epoch': 2.17, 'throughput': 9394.81} [INFO|2025-10-20 20:31:44] logging.py:157 >> {'loss': 0.1130, 'learning_rate': 1.7519e-05, 'epoch': 2.18, 'throughput': 9394.59} [INFO|2025-10-20 20:40:35] logging.py:157 >> {'loss': 0.1104, 'learning_rate': 1.7159e-05, 'epoch': 2.18, 'throughput': 9394.84} [INFO|2025-10-20 20:49:22] logging.py:157 >> {'loss': 0.1152, 'learning_rate': 1.6803e-05, 'epoch': 2.19, 'throughput': 9395.21} [INFO|2025-10-20 20:57:59] logging.py:157 >> {'loss': 0.1097, 'learning_rate': 1.6450e-05, 'epoch': 2.20, 'throughput': 9396.50} [INFO|2025-10-20 21:06:50] logging.py:157 >> {'loss': 0.1111, 'learning_rate': 1.6099e-05, 'epoch': 2.21, 'throughput': 9396.54} [INFO|2025-10-20 21:15:45] logging.py:157 >> {'loss': 0.1104, 'learning_rate': 1.5752e-05, 'epoch': 2.22, 'throughput': 9396.81} [INFO|2025-10-20 21:24:38] logging.py:157 >> {'loss': 0.1075, 'learning_rate': 1.5408e-05, 'epoch': 2.23, 'throughput': 9397.25} [INFO|2025-10-20 21:33:34] logging.py:157 >> {'loss': 0.1093, 'learning_rate': 1.5067e-05, 'epoch': 2.24, 'throughput': 9397.23} [INFO|2025-10-20 21:42:37] logging.py:157 >> {'loss': 0.1126, 'learning_rate': 1.4729e-05, 'epoch': 2.25, 'throughput': 9397.67} [INFO|2025-10-20 21:51:20] logging.py:157 >> {'loss': 0.1064, 'learning_rate': 1.4394e-05, 'epoch': 2.26, 'throughput': 9398.27} [INFO|2025-10-20 22:00:12] logging.py:157 >> {'loss': 0.1096, 'learning_rate': 1.4062e-05, 'epoch': 2.27, 'throughput': 9398.52} [INFO|2025-10-20 22:09:09] logging.py:157 >> {'loss': 0.1089, 'learning_rate': 1.3734e-05, 'epoch': 2.27, 'throughput': 9398.21} [INFO|2025-10-20 22:17:32] logging.py:157 >> {'loss': 0.1093, 'learning_rate': 1.3409e-05, 'epoch': 2.28, 'throughput': 9399.94} [INFO|2025-10-20 22:26:21] logging.py:157 >> {'loss': 0.1064, 'learning_rate': 1.3087e-05, 'epoch': 2.29, 'throughput': 9400.36} [INFO|2025-10-20 22:35:22] logging.py:157 >> {'loss': 0.1103, 'learning_rate': 1.2769e-05, 'epoch': 2.30, 'throughput': 9400.86} [INFO|2025-10-20 22:44:04] logging.py:157 >> {'loss': 0.1077, 'learning_rate': 1.2454e-05, 'epoch': 2.31, 'throughput': 9401.43} [INFO|2025-10-20 22:52:55] logging.py:157 >> {'loss': 0.1051, 'learning_rate': 1.2142e-05, 'epoch': 2.32, 'throughput': 9402.42} [INFO|2025-10-20 23:02:06] logging.py:157 >> {'loss': 0.1063, 'learning_rate': 1.1834e-05, 'epoch': 2.33, 'throughput': 9402.19} [INFO|2025-10-20 23:10:53] logging.py:157 >> {'loss': 0.1073, 'learning_rate': 1.1529e-05, 'epoch': 2.34, 'throughput': 9402.65} [INFO|2025-10-20 23:19:56] logging.py:157 >> {'loss': 0.1076, 'learning_rate': 1.1227e-05, 'epoch': 2.35, 'throughput': 9402.88} [INFO|2025-10-20 23:28:32] logging.py:157 >> {'loss': 0.1049, 'learning_rate': 1.0930e-05, 'epoch': 2.36, 'throughput': 9403.97} [INFO|2025-10-20 23:37:11] logging.py:157 >> {'loss': 0.1038, 'learning_rate': 1.0635e-05, 'epoch': 2.37, 'throughput': 9404.57} [INFO|2025-10-20 23:46:14] logging.py:157 >> {'loss': 0.1049, 'learning_rate': 1.0344e-05, 'epoch': 2.37, 'throughput': 9404.62} [INFO|2025-10-20 23:55:18] logging.py:157 >> {'loss': 0.1026, 'learning_rate': 1.0057e-05, 'epoch': 2.38, 'throughput': 9404.53} [INFO|2025-10-21 00:04:01] logging.py:157 >> {'loss': 0.1061, 'learning_rate': 9.7735e-06, 'epoch': 2.39, 'throughput': 9405.00} [INFO|2025-10-21 00:12:43] logging.py:157 >> {'loss': 0.1078, 'learning_rate': 9.4934e-06, 'epoch': 2.40, 'throughput': 9405.66} [INFO|2025-10-21 00:22:04] logging.py:157 >> {'loss': 0.1021, 'learning_rate': 9.2171e-06, 'epoch': 2.41, 'throughput': 9404.52} [INFO|2025-10-21 00:31:08] logging.py:157 >> {'loss': 0.1050, 'learning_rate': 8.9443e-06, 'epoch': 2.42, 'throughput': 9404.23} [INFO|2025-10-21 00:40:06] logging.py:157 >> {'loss': 0.1039, 'learning_rate': 8.6753e-06, 'epoch': 2.43, 'throughput': 9404.10} [INFO|2025-10-21 00:48:47] logging.py:157 >> {'loss': 0.1026, 'learning_rate': 8.4100e-06, 'epoch': 2.44, 'throughput': 9404.46} [INFO|2025-10-21 00:57:41] logging.py:157 >> {'loss': 0.1046, 'learning_rate': 8.1485e-06, 'epoch': 2.45, 'throughput': 9405.18} [INFO|2025-10-21 01:06:27] logging.py:157 >> {'loss': 0.1062, 'learning_rate': 7.8907e-06, 'epoch': 2.46, 'throughput': 9405.68} [INFO|2025-10-21 01:14:52] logging.py:157 >> {'loss': 0.1005, 'learning_rate': 7.6368e-06, 'epoch': 2.47, 'throughput': 9407.14} [INFO|2025-10-21 01:23:49] logging.py:157 >> {'loss': 0.1008, 'learning_rate': 7.3866e-06, 'epoch': 2.47, 'throughput': 9407.69} [INFO|2025-10-21 01:32:54] logging.py:157 >> {'loss': 0.1012, 'learning_rate': 7.1403e-06, 'epoch': 2.48, 'throughput': 9407.74} [INFO|2025-10-21 01:41:45] logging.py:157 >> {'loss': 0.1004, 'learning_rate': 6.8978e-06, 'epoch': 2.49, 'throughput': 9408.87} [INFO|2025-10-21 01:50:52] logging.py:157 >> {'loss': 0.1058, 'learning_rate': 6.6592e-06, 'epoch': 2.50, 'throughput': 9409.14} [INFO|2025-10-21 01:59:29] logging.py:157 >> {'loss': 0.0996, 'learning_rate': 6.4246e-06, 'epoch': 2.51, 'throughput': 9410.56} [INFO|2025-10-21 02:08:18] logging.py:157 >> {'loss': 0.0980, 'learning_rate': 6.1938e-06, 'epoch': 2.52, 'throughput': 9411.15} [INFO|2025-10-21 02:17:08] logging.py:157 >> {'loss': 0.1053, 'learning_rate': 5.9670e-06, 'epoch': 2.53, 'throughput': 9411.73} [INFO|2025-10-21 02:26:00] logging.py:157 >> {'loss': 0.1021, 'learning_rate': 5.7442e-06, 'epoch': 2.54, 'throughput': 9412.46} [INFO|2025-10-21 02:34:44] logging.py:157 >> {'loss': 0.1003, 'learning_rate': 5.5253e-06, 'epoch': 2.55, 'throughput': 9413.49} [INFO|2025-10-21 02:43:21] logging.py:157 >> {'loss': 0.1039, 'learning_rate': 5.3105e-06, 'epoch': 2.56, 'throughput': 9413.93} [INFO|2025-10-21 02:52:07] logging.py:157 >> {'loss': 0.1034, 'learning_rate': 5.0997e-06, 'epoch': 2.56, 'throughput': 9414.81} [INFO|2025-10-21 03:01:16] logging.py:157 >> {'loss': 0.1025, 'learning_rate': 4.8929e-06, 'epoch': 2.57, 'throughput': 9415.33} [INFO|2025-10-21 03:10:08] logging.py:157 >> {'loss': 0.1041, 'learning_rate': 4.6902e-06, 'epoch': 2.58, 'throughput': 9415.61} [INFO|2025-10-21 03:19:13] logging.py:157 >> {'loss': 0.1035, 'learning_rate': 4.4916e-06, 'epoch': 2.59, 'throughput': 9415.24} [INFO|2025-10-21 03:28:15] logging.py:157 >> {'loss': 0.1020, 'learning_rate': 4.2970e-06, 'epoch': 2.60, 'throughput': 9415.55} [INFO|2025-10-21 03:37:01] logging.py:157 >> {'loss': 0.0968, 'learning_rate': 4.1066e-06, 'epoch': 2.61, 'throughput': 9416.74} [INFO|2025-10-21 03:46:20] logging.py:157 >> {'loss': 0.1025, 'learning_rate': 3.9203e-06, 'epoch': 2.62, 'throughput': 9416.50} [INFO|2025-10-21 03:55:19] logging.py:157 >> {'loss': 0.0992, 'learning_rate': 3.7382e-06, 'epoch': 2.63, 'throughput': 9417.01} [INFO|2025-10-21 04:04:01] logging.py:157 >> {'loss': 0.1014, 'learning_rate': 3.5603e-06, 'epoch': 2.64, 'throughput': 9417.49} [INFO|2025-10-21 04:13:07] logging.py:157 >> {'loss': 0.0970, 'learning_rate': 3.3865e-06, 'epoch': 2.65, 'throughput': 9416.80} [INFO|2025-10-21 04:21:34] logging.py:157 >> {'loss': 0.1011, 'learning_rate': 3.2169e-06, 'epoch': 2.66, 'throughput': 9417.61} [INFO|2025-10-21 04:30:30] logging.py:157 >> {'loss': 0.1003, 'learning_rate': 3.0515e-06, 'epoch': 2.66, 'throughput': 9418.35} [INFO|2025-10-21 04:39:21] logging.py:157 >> {'loss': 0.1011, 'learning_rate': 2.8904e-06, 'epoch': 2.67, 'throughput': 9419.24} [INFO|2025-10-21 04:48:13] logging.py:157 >> {'loss': 0.0992, 'learning_rate': 2.7335e-06, 'epoch': 2.68, 'throughput': 9418.86} [INFO|2025-10-21 04:57:13] logging.py:157 >> {'loss': 0.0985, 'learning_rate': 2.5809e-06, 'epoch': 2.69, 'throughput': 9418.72} [INFO|2025-10-21 05:06:03] logging.py:157 >> {'loss': 0.0999, 'learning_rate': 2.4325e-06, 'epoch': 2.70, 'throughput': 9419.01} [INFO|2025-10-21 05:14:56] logging.py:157 >> {'loss': 0.1015, 'learning_rate': 2.2885e-06, 'epoch': 2.71, 'throughput': 9419.26} [INFO|2025-10-21 05:23:49] logging.py:157 >> {'loss': 0.0952, 'learning_rate': 2.1487e-06, 'epoch': 2.72, 'throughput': 9418.88} [INFO|2025-10-21 05:32:51] logging.py:157 >> {'loss': 0.1014, 'learning_rate': 2.0132e-06, 'epoch': 2.73, 'throughput': 9418.79} [INFO|2025-10-21 05:42:05] logging.py:157 >> {'loss': 0.0962, 'learning_rate': 1.8821e-06, 'epoch': 2.74, 'throughput': 9418.29} [INFO|2025-10-21 05:51:14] logging.py:157 >> {'loss': 0.0984, 'learning_rate': 1.7553e-06, 'epoch': 2.75, 'throughput': 9417.96} [INFO|2025-10-21 05:59:46] logging.py:157 >> {'loss': 0.0996, 'learning_rate': 1.6328e-06, 'epoch': 2.76, 'throughput': 9418.90} [INFO|2025-10-21 06:08:27] logging.py:157 >> {'loss': 0.0980, 'learning_rate': 1.5147e-06, 'epoch': 2.76, 'throughput': 9419.80} [INFO|2025-10-21 06:17:49] logging.py:157 >> {'loss': 0.0979, 'learning_rate': 1.4010e-06, 'epoch': 2.77, 'throughput': 9419.54} [INFO|2025-10-21 06:26:53] logging.py:157 >> {'loss': 0.0960, 'learning_rate': 1.2916e-06, 'epoch': 2.78, 'throughput': 9419.30} [INFO|2025-10-21 06:35:34] logging.py:157 >> {'loss': 0.0995, 'learning_rate': 1.1866e-06, 'epoch': 2.79, 'throughput': 9419.37} [INFO|2025-10-21 06:44:22] logging.py:157 >> {'loss': 0.0979, 'learning_rate': 1.0861e-06, 'epoch': 2.80, 'throughput': 9419.58} [INFO|2025-10-21 06:53:12] logging.py:157 >> {'loss': 0.0944, 'learning_rate': 9.8988e-07, 'epoch': 2.81, 'throughput': 9419.64} [INFO|2025-10-21 07:01:56] logging.py:157 >> {'loss': 0.0949, 'learning_rate': 8.9813e-07, 'epoch': 2.82, 'throughput': 9420.47} [INFO|2025-10-21 07:11:06] logging.py:157 >> {'loss': 0.0959, 'learning_rate': 8.1080e-07, 'epoch': 2.83, 'throughput': 9420.22} [INFO|2025-10-21 07:20:02] logging.py:157 >> {'loss': 0.0995, 'learning_rate': 7.2790e-07, 'epoch': 2.84, 'throughput': 9420.29} [INFO|2025-10-21 07:28:41] logging.py:157 >> {'loss': 0.0973, 'learning_rate': 6.4944e-07, 'epoch': 2.85, 'throughput': 9421.16} [INFO|2025-10-21 07:37:33] logging.py:157 >> {'loss': 0.0992, 'learning_rate': 5.7543e-07, 'epoch': 2.85, 'throughput': 9421.50} [INFO|2025-10-21 07:46:42] logging.py:157 >> {'loss': 0.0978, 'learning_rate': 5.0586e-07, 'epoch': 2.86, 'throughput': 9421.29} [INFO|2025-10-21 07:55:32] logging.py:157 >> {'loss': 0.0963, 'learning_rate': 4.4076e-07, 'epoch': 2.87, 'throughput': 9421.46} [INFO|2025-10-21 08:04:53] logging.py:157 >> {'loss': 0.0932, 'learning_rate': 3.8012e-07, 'epoch': 2.88, 'throughput': 9420.75} [INFO|2025-10-21 08:13:35] logging.py:157 >> {'loss': 0.0980, 'learning_rate': 3.2395e-07, 'epoch': 2.89, 'throughput': 9421.03} [INFO|2025-10-21 08:22:02] logging.py:157 >> {'loss': 0.0990, 'learning_rate': 2.7225e-07, 'epoch': 2.90, 'throughput': 9422.17} [INFO|2025-10-21 08:31:18] logging.py:157 >> {'loss': 0.0974, 'learning_rate': 2.2504e-07, 'epoch': 2.91, 'throughput': 9421.85} [INFO|2025-10-21 08:39:56] logging.py:157 >> {'loss': 0.0995, 'learning_rate': 1.8231e-07, 'epoch': 2.92, 'throughput': 9422.72} [INFO|2025-10-21 08:49:10] logging.py:157 >> {'loss': 0.0955, 'learning_rate': 1.4406e-07, 'epoch': 2.93, 'throughput': 9422.15} [INFO|2025-10-21 08:57:45] logging.py:157 >> {'loss': 0.0950, 'learning_rate': 1.1031e-07, 'epoch': 2.94, 'throughput': 9422.79} [INFO|2025-10-21 09:06:43] logging.py:157 >> {'loss': 0.0995, 'learning_rate': 8.1053e-08, 'epoch': 2.95, 'throughput': 9422.90} [INFO|2025-10-21 09:15:28] logging.py:157 >> {'loss': 0.0976, 'learning_rate': 5.6291e-08, 'epoch': 2.95, 'throughput': 9423.33} [INFO|2025-10-21 09:24:09] logging.py:157 >> {'loss': 0.0955, 'learning_rate': 3.6029e-08, 'epoch': 2.96, 'throughput': 9423.84} [INFO|2025-10-21 09:32:53] logging.py:157 >> {'loss': 0.0979, 'learning_rate': 2.0267e-08, 'epoch': 2.97, 'throughput': 9424.10} [INFO|2025-10-21 09:41:43] logging.py:157 >> {'loss': 0.0999, 'learning_rate': 9.0080e-09, 'epoch': 2.98, 'throughput': 9425.00} [INFO|2025-10-21 09:50:56] logging.py:157 >> {'loss': 0.0990, 'learning_rate': 2.2521e-09, 'epoch': 2.99, 'throughput': 9424.34} [INFO|2025-10-21 10:00:15] logging.py:157 >> {'loss': 0.0971, 'learning_rate': 0.0000e+00, 'epoch': 3.00, 'throughput': 9423.59} [INFO|2025-10-21 10:00:23] trainer.py:3966 >> Saving model checkpoint to saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951 [INFO|2025-10-21 10:00:23] configuration_utils.py:423 >> Configuration saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/config.json [INFO|2025-10-21 10:00:23] configuration_utils.py:908 >> Configuration saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/generation_config.json [INFO|2025-10-21 10:07:28] modeling_utils.py:3594 >> The model is bigger than the maximum size per checkpoint (5GB) and is going to be split in 4 checkpoint shards. You can find where each parameters has been saved in the index located at saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/model.safetensors.index.json. [INFO|2025-10-21 10:07:28] tokenization_utils_base.py:2510 >> tokenizer config file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/tokenizer_config.json [INFO|2025-10-21 10:07:28] tokenization_utils_base.py:2519 >> Special tokens file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/special_tokens_map.json [INFO|2025-10-21 10:12:28] image_processing_base.py:261 >> Image processor saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/preprocessor_config.json [INFO|2025-10-21 10:12:28] tokenization_utils_base.py:2510 >> tokenizer config file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/tokenizer_config.json [INFO|2025-10-21 10:12:28] tokenization_utils_base.py:2519 >> Special tokens file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/special_tokens_map.json [INFO|2025-10-21 10:12:29] processing_utils.py:638 >> chat template saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/checkpoint-6951/chat_template.json [INFO|2025-10-21 10:12:29] trainer.py:2665 >> Training completed. Do not forget to share your model on huggingface.co/models =) [INFO|2025-10-21 10:12:29] image_processing_base.py:261 >> Image processor saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/preprocessor_config.json [INFO|2025-10-21 10:12:29] tokenization_utils_base.py:2510 >> tokenizer config file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/tokenizer_config.json [INFO|2025-10-21 10:12:29] tokenization_utils_base.py:2519 >> Special tokens file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/special_tokens_map.json [INFO|2025-10-21 10:12:30] processing_utils.py:638 >> chat template saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/chat_template.json [INFO|2025-10-21 10:12:37] trainer.py:3966 >> Saving model checkpoint to saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123 [INFO|2025-10-21 10:12:37] configuration_utils.py:423 >> Configuration saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/config.json [INFO|2025-10-21 10:12:37] configuration_utils.py:908 >> Configuration saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/generation_config.json [INFO|2025-10-21 10:19:32] modeling_utils.py:3594 >> The model is bigger than the maximum size per checkpoint (5GB) and is going to be split in 4 checkpoint shards. You can find where each parameters has been saved in the index located at saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/model.safetensors.index.json. [INFO|2025-10-21 10:19:32] tokenization_utils_base.py:2510 >> tokenizer config file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/tokenizer_config.json [INFO|2025-10-21 10:19:32] tokenization_utils_base.py:2519 >> Special tokens file saved in saves/Qwen2.5-VL-7B-Instruct/full/all_stage_123/special_tokens_map.json [WARNING|2025-10-21 10:19:34] logging.py:162 >> No metric eval_accuracy to plot. [INFO|2025-10-21 10:19:34] trainer.py:4289 >> ***** Running Evaluation ***** [INFO|2025-10-21 10:19:34] trainer.py:4291 >> Num examples = 9176 [INFO|2025-10-21 10:19:34] trainer.py:4294 >> Batch size = 2 [INFO|2025-10-21 10:28:27] modelcard.py:449 >> Dropping the following result as it does not have all the necessary fields: {'task': {'name': 'Causal Language Modeling', 'type': 'text-generation'}}