Update README.md

README.md

@@ -41,7 +41,6 @@ llava-Qwen2-7B-Instruct-Chinese-CLIP-v3 = Qwen/Qwen2-7B-Instruct + multi_modal_p
 - **数据集**: REILX/chinese-meme-description-dataset、SWHL/ChineseOCRBench、priyank-m/chinese_text_recognition、fly0331/ChineseTest、liuhaotian/LLaVA-Pretrain、Lin-Chen/ShareGPT4V
 - **微调参数**:
   - lora_r=32, lora_alpha=64, num_train_epochs=2, per_device_train_batch_size=1, gradient_accumulation_steps=8, high_lr=1e-3, low_lr=2e-5, model_max_length=2048
-- **设备**: 8 * A800
 - **训练时长**: 84小时02分钟
 
 ### 阶段2:
@@ -49,7 +48,6 @@ llava-Qwen2-7B-Instruct-Chinese-CLIP-v3 = Qwen/Qwen2-7B-Instruct + multi_modal_p
 - **数据集**: REILX/Chinese-Image-Text-Corpus-dataset
 - **微调参数**:
   - lora_r=32, lora_alpha=64, num_train_epochs=3, per_device_train_batch_size=1, gradient_accumulation_steps=8, high_lr=5e-4, low_lr=1e-5, model_max_length=2048
-- **设备**: 8 * A800
 - **训练时长**: 36小时56分钟
 
 ### 阶段3:
@@ -57,7 +55,6 @@ llava-Qwen2-7B-Instruct-Chinese-CLIP-v3 = Qwen/Qwen2-7B-Instruct + multi_modal_p
 - **数据集**: REILX/chinese-meme-description-dataset 中的 ChineseBQB-Claude-3-5-sonnet-20240620.jsonl 和 emo-visual-data-Claude-3-5-sonnet-20240620.jsonl，仅使用质量最高的 Claude-3-5-sonnet-20240620 模型输出进行最后的微调
 - **微调参数**:
   - lora_r=32, lora_alpha=64, num_train_epochs=3, per_device_train_batch_size=1, gradient_accumulation_steps=8, high_lr=5e-4, low_lr=1e-5, model_max_length=2048
-- **设备**: 8 * A800
 - **训练时长**: 1小时04分钟
 
 **3阶段共耗时**: 122小时