[Docs] Change tables to list items (#10656)

Author: hanlintang

llm/docs/finetune_tutorial.md

@@ -7,12 +7,10 @@
 ## 1. 什么是精调 （Fine-tuning）
 经过预训练，我们拥有了经过预训练之后的基础模型，但是这个模型如果直接应用在某一个特定领域，效果可能并不太好。这是由于我们的预训练语料没有针对特定的任务场景进行特化。但是大模型经过预训练已经拥有了很强的通用能力，只需要少量特定数据，就可以大幅提升大模型在特定领域的能力。
 
-我们用一个表格来简单对比下预训练与精调：
-| 对比项  | 预训练      | 精调        |
-| ---- | -------- | --------- |
-| 数据量  | 海量通用文本   | 少量特定任务数据  |
-| 目标   | 学习通用语言知识 | 优化具体任务表现  |
-| 训练时间 | 很长   | 较短 |
+我们简单对比下预训练与精调：
+* 数据量:（预训练）海量通用文本/(精调)少量特定任务数据
+* 训练目标：（预训练）学习通用语言知识/(精调)优化具体任务表现
+* 训练时间：（预训练）很长/(精调)较短
 
 ## 2. 精调数据
 与预训练的无监督数据不同，精调使用的数据是有监督数据，一般是一个指令（向大模型说的话）和一个输出（大模型应该的回应）。常用的数据格式是[Alpaca](https://crfm.stanford.edu/2023/03/13/alpaca.html)，我们用一个简单的例子来展示一下其基本组成：
@@ -108,11 +106,7 @@ python run_finetune.py ./config/qwen/pt_argument_0p5b.json
 <img src=https://github.com/PaddlePaddle/PaddleNLP/assets/37530985/63d56558-247a-4a8d-a6ca-121c820f7534 width=30% height=30% />
 </div>
 
-LoRA(Low-rank Adaptation)方法向模型插入少量可训练的低秩矩阵，在不改变原模型参数的前提下，完成微调任务。参考图里面的示例，一个很大的灰色参数矩阵可以近似分解为两个较小的矩阵 A 与 B 相乘：
-
-$$\Delta W \approx A \cdot B$$
-
-这样一个大矩阵就被分解为了两个小矩阵，需要微调的参数量大大减少。在训练完成后，在推理的过程中将原本的参数矩阵与微调的参数矩阵相加，既不影响原本的预训练参数，也可以实现微调效果，这样的优点也让 LoRA 成为了大模型微调的热门方法。
+LoRA(Low-rank Adaptation)方法向模型插入少量可训练的低秩矩阵，在不改变原模型参数的前提下，完成微调任务。参考图里面的示例，一个很大的灰色参数矩阵可以近似分解为两个较小的矩阵 A 与 B 相乘。这样一个大矩阵就被分解为了两个小矩阵，需要微调的参数量大大减少。在训练完成后，在推理的过程中将原本的参数矩阵与微调的参数矩阵相加，既不影响原本的预训练参数，也可以实现微调效果，这样的优点也让 LoRA 成为了大模型微调的热门方法。
 
 
 ```python

llm/docs/pretrain_tutorial.md

@@ -80,8 +80,8 @@ python -u ~/PaddleNLP/llm/run_pretrain.py ~/PaddleNLP/llm/config/qwen/pretrain_a
 
 看到下面的提示说明已经开始进行训练了：
 ```
-[2025-05-18 10:13:15,353] [    INFO] - loss: 12.0635252, learning_rate: 2e-06, global_step: 1, current_memory_allocated: 7.549170255661011, current_memory_reserved: 7.753237724304199, max_memory_allocated: 7.549170255661011, max_memory_reserved: 7.753237724304199, interval_runtime: 1.1518, interval_samples_per_second: 0.8682, interval_tokens_per_second_per_device: 889.0448, interval_hardware_tflops_per_device: 2.77, interval_steps_per_second: 0.8682, progress_or_epoch: 0.0
-[2025-05-18 10:13:15,607] [    INFO] - loss: 12.05887604, learning_rate: 3e-06, global_step: 2, current_memory_allocated: 7.549170255661011, current_memory_reserved: 12.4834623336792, max_memory_allocated: 12.307440280914307, max_memory_reserved: 12.4834623336792, interval_runtime: 0.2556, interval_samples_per_second: 3.913, interval_tokens_per_second_per_device: 4006.9441, interval_hardware_tflops_per_device: 12.49, interval_steps_per_second: 3.913, progress_or_epoch: 0.0
+[    INFO] - loss: 12.0635252, learning_rate: 2e-06, global_step: 1, current_memory_allocated: 7.549170255661011, current_memory_reserved: 7.753237724304199, max_memory_allocated: 7.549170255661011, max_memory_reserved: 7.753237724304199, interval_runtime: 1.1518, interval_samples_per_second: 0.8682, interval_tokens_per_second_per_device: 889.0448, interval_hardware_tflops_per_device: 2.77, interval_steps_per_second: 0.8682, progress_or_epoch: 0.0
+[    INFO] - loss: 12.05887604, learning_rate: 3e-06, global_step: 2, current_memory_allocated: 7.549170255661011, current_memory_reserved: 12.4834623336792, max_memory_allocated: 12.307440280914307, max_memory_reserved: 12.4834623336792, interval_runtime: 0.2556, interval_samples_per_second: 3.913, interval_tokens_per_second_per_device: 4006.9441, interval_hardware_tflops_per_device: 12.49, interval_steps_per_second: 3.913, progress_or_epoch: 0.0
 ```
 
 ### FAQ1：显存不足
@@ -131,29 +131,29 @@ python -u -m paddle.distributed.launch --devices "0,1,2,3,4,5,6,7"  --master=192
 ### 3.3 训练结果
 当训练结束时，可以看到下面的输出：
 ```
-[2025-05-18 10:49:13,774] [    INFO] - Saving model checkpoint to ./checkpoints/pretrain_ckpts
-[2025-05-18 10:49:13,775] [    INFO] - tokenizer config file saved in ./checkpoints/pretrain_ckpts/tokenizer_config.json
-[2025-05-18 10:49:13,776] [    INFO] - Special tokens file saved in ./checkpoints/pretrain_ckpts/special_tokens_map.json
-[2025-05-18 10:49:13,776] [    INFO] - added tokens file saved in ./checkpoints/pretrain_ckpts/added_tokens.json
-[2025-05-18 10:49:13,803] [ WARNING] - Asynchronous saving is not supported for single card environment currently.
-[2025-05-18 10:49:15,691] [    INFO] - Configuration saved in ./checkpoints/pretrain_ckpts/config.json
-[2025-05-18 10:49:15,694] [    INFO] - Configuration saved in ./checkpoints/pretrain_ckpts/generation_config.json
-[2025-05-18 10:49:15,695] [    INFO] - ***** train metrics *****
-[2025-05-18 10:49:15,695] [    INFO] -   progress_or_epoch        =     0.0868
-[2025-05-18 10:49:15,695] [    INFO] -   train_loss               =     5.4334
-[2025-05-18 10:49:15,695] [    INFO] -   train_runtime            = 0:35:59.57
-[2025-05-18 10:49:15,696] [    INFO] -   train_samples_per_second =     4.6305
-[2025-05-18 10:49:15,696] [    INFO] -   train_steps_per_second   =     4.6305
-[2025-05-18 10:49:15,697] [    INFO] - ***** Running Prediction *****
-[2025-05-18 10:49:15,697] [    INFO] -   Num examples = 258
-[2025-05-18 10:49:15,697] [    INFO] -   Total prediction steps = 129
-[2025-05-18 10:49:15,697] [    INFO] -   Pre device batch size = 2
-[2025-05-18 10:49:15,697] [    INFO] -   Total Batch size = 2
-[2025-05-18 10:49:28,420] [    INFO] - ***** test metrics *****
-[2025-05-18 10:49:28,420] [    INFO] -   test_loss               =     4.8691
-[2025-05-18 10:49:28,420] [    INFO] -   test_runtime            = 0:00:12.72
-[2025-05-18 10:49:28,420] [    INFO] -   test_samples_per_second =    20.2781
-[2025-05-18 10:49:28,420] [    INFO] -   test_steps_per_second   =    10.1391
+[    INFO] - Saving model checkpoint to ./checkpoints/pretrain_ckpts
+[    INFO] - tokenizer config file saved in ./checkpoints/pretrain_ckpts/tokenizer_config.json
+[    INFO] - Special tokens file saved in ./checkpoints/pretrain_ckpts/special_tokens_map.json
+[    INFO] - added tokens file saved in ./checkpoints/pretrain_ckpts/added_tokens.json
+[ WARNING] - Asynchronous saving is not supported for single card environment currently.
+[    INFO] - Configuration saved in ./checkpoints/pretrain_ckpts/config.json
+[    INFO] - Configuration saved in ./checkpoints/pretrain_ckpts/generation_config.json
+[    INFO] - ***** train metrics *****
+[    INFO] -   progress_or_epoch        =     0.0868
+[    INFO] -   train_loss               =     5.4334
+[    INFO] -   train_runtime            = 0:35:59.57
+[    INFO] -   train_samples_per_second =     4.6305
+[    INFO] -   train_steps_per_second   =     4.6305
+[    INFO] - ***** Running Prediction *****
+[    INFO] -   Num examples = 258
+[    INFO] -   Total prediction steps = 129
+[    INFO] -   Pre device batch size = 2
+[    INFO] -   Total Batch size = 2
+[    INFO] - ***** test metrics *****
+[    INFO] -   test_loss               =     4.8691
+[    INFO] -   test_runtime            = 0:00:12.72
+[    INFO] -   test_samples_per_second =    20.2781
+[    INFO] -   test_steps_per_second   =    10.1391
 Effective Tokens per second: 4741.68
 ips: 4741.68 tokens/s
 ```