update acc_strategy & fix citest (#2583)

Author: Jintao-Huang

README.md

@@ -255,8 +255,8 @@ pip install ms-swift -U
 - Method 2: Install SWIFT through source code (convenient for running training and inference scripts), please run the following commands:
 
 ```shell
-git clone https://github.com/modelscope/swift.git
-cd swift
+git clone https://github.com/modelscope/ms-swift.git
+cd ms-swift
 pip install -e '.[llm]'
 ```
 

README_CN.md

@@ -256,8 +256,8 @@ pip install ms-swift -U
 - 方法2：通过源代码安装SWIFT（方便运行训练推理脚本），请运行以下命令：
 
 ```shell
-git clone https://github.com/modelscope/swift.git
-cd swift
+git clone https://github.com/modelscope/ms-swift.git
+cd ms-swift
 pip install -e '.[llm]'
 ```
 

docs/source/GetStarted/快速开始.md

@@ -46,6 +46,7 @@ swift sft \
               swift/self-cognition#500 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
     --learning_rate 1e-4 \
     --lora_rank 8 \
     --lora_alpha 32 \
@@ -56,8 +57,11 @@ swift sft \
     --save_total_limit 2 \
     --logging_steps 5 \
     --max_length 2048 \
-    --model_author swift \
     --output_dir output \
+    --system 'You are a helpful assistant.' \
+    --warmup_ratio 0.05 \
+    --dataloader_num_workers 4 \
+    --model_author swift \
     --model_name swift-robot
 ```
 

docs/source/Instruction/命令行参数.md

@@ -292,7 +292,7 @@ Vera使用`target_modules`, `target_regex`, `modules_to_save`三个参数.
 - packing: 是否使用packing，默认为False
 - 🔥lazy_tokenize: 是否使用lazy_tokenize，在LLM训练中默认False，MLLM训练中默认True
 
-- acc_strategy: 训练acc的策略，可以为`sentence`和`token`级别的acc，默认为`token`
+- acc_strategy: 训练acc的策略，可以为`seq`和`token`级别的acc，默认为`token`
 - max_new_tokens: predict_with_generate=True时的最大生成token数量，默认64
 - temperature: predict_with_generate=True时的temperature，默认0
 - optimizer: plugin的自定义optimizer名称

docs/source/Instruction/评测.md

@@ -50,8 +50,8 @@ pip install ms-swift[eval] -U
 或从源代码安装：
 
 ```shell
-git clone https://github.com/modelscope/swift.git
-cd swift
+git clone https://github.com/modelscope/ms-swift.git
+cd ms-swift
 pip install -e '.[eval]'
 ```
 

docs/source_en/GetStarted/Quick-start.md

@@ -46,6 +46,7 @@ swift sft \
               swift/self-cognition#500 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
     --learning_rate 1e-4 \
     --lora_rank 8 \
     --lora_alpha 32 \
@@ -56,8 +57,11 @@ swift sft \
     --save_total_limit 2 \
     --logging_steps 5 \
     --max_length 2048 \
-    --model_author swift \
     --output_dir output \
+    --system 'You are a helpful assistant.' \
+    --warmup_ratio 0.05 \
+    --dataloader_num_workers 4 \
+    --model_author swift \
     --model_name swift-robot
 ```
 

docs/source_en/Instruction/Command-line-parameters.md

@@ -297,7 +297,7 @@ Training arguments include the [base arguments](#base-arguments), [Seq2SeqTraine
 - packing: Whether to use packing, default is False.
 - 🔥lazy_tokenize: Whether to use lazy_tokenize, default is False during LLM training, default is True during MLLM training.
 
-- acc_strategy: Strategy for training accuracy, can be `sentence` or `token` level accuracy, default is `token`.
+- acc_strategy: Strategy for training accuracy, can be `seq` or `token` level accuracy, default is `token`.
 - max_new_tokens: Maximum generated token count when `predict_with_generate=True`, default 64.
 - temperature: Temperature when `predict_with_generate=True`, default 0.
 - optimizer: Custom optimizer name for plugin.

docs/source_en/Instruction/Evaluation.md

@@ -50,8 +50,8 @@ pip install ms-swift[eval] -U
 Or install from the source code:
 
 ```shell
-git clone https://github.com/modelscope/swift.git
-cd swift
+git clone https://github.com/modelscope/ms-swift.git
+cd ms-swift
 pip install -e '.[eval]'
 ```
 

examples/custom/infer.sh

@@ -0,0 +1,9 @@
+# sh examples/custom/infer.sh
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir output/vx-xxx/checkpoint-xxx \
+    --load_dataset_config true \
+    --infer_backend pt \
+    --max_batch_size 16 \
+    --max_new_tokens 256 \
+    --temperature 0

examples/custom/sft.sh

@@ -0,0 +1,24 @@
+# sh examples/custom/sft.sh
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+    --custom_register_path examples/custom/dataset.py \
+                           examples/custom/model.py \
+    --model AI-ModelScope/Nemotron-Mini-4B-Instruct \
+    --train_type lora \
+    --dataset swift/stsb \
+    --num_train_epochs 3 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
+    --learning_rate 1e-4 \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --gradient_accumulation_steps 16 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 5 \
+    --warmup_ratio 0.05 \
+    --dataloader_num_workers 4 \
+    --max_length 2048 \
+    --output_dir output