Fix finetune supervised dataset issue

.gitignore

@@ -0,0 +1,4 @@
+*__pycache__*
+*.pyc
+finetune/output/*
+wip.*

finetune/README.md

@@ -64,11 +64,6 @@ pip install -r requirements.txt
 
 ## 数据集格式示例
 
-这里以 AdvertiseGen 数据集为例,
-您可以从 [Google Drive](https://drive.google.com/file/d/13_vf0xRTQsyneRKdD1bZIr93vBGOczrk/view?usp=sharing)
-或者 [Tsinghua Cloud](https://cloud.tsinghua.edu.cn/f/b3f119a008264b1cabd1/?dl=1) 下载 AdvertiseGen 数据集。
-将解压后的 AdvertiseGen 目录放到 `data` 目录下并自行转换为如下格式数据集。
-
 > 请注意，现在的微调代码中加入了验证集，因此，对于一组完整的微调数据集，必须包含训练数据集和验证数据集，测试数据集可以不填写。或者直接用验证数据集代替。
 
 ```

finetune/README_en.md

@@ -66,11 +66,6 @@ For the data file, the example uses the following format
 
 ## Dataset Format Example
 
-Here, taking the AdvertiseGen dataset as an example,
-you can download the AdvertiseGen dataset from [Google Drive](https://drive.google.com/file/d/13_vf0xRTQsyneRKdD1bZIr93vBGOczrk/view?usp=sharing)
-or [Tsinghua Cloud](https://cloud.tsinghua.edu.cn/f/b3f119a008264b1cabd1/?dl=1) . After extracting the AdvertiseGen directory, place it in the `data` directory and convert it into the following format dataset.
-
-
 > Please note, the fine-tuning code now includes a validation set, so for a complete set of fine-tuning datasets, it must contain training and validation datasets, while the test dataset is optional. Or, you can use the validation dataset in place of it.
 
 ```

finetune/finetune.py

@@ -73,7 +73,7 @@ class SupervisedDataset(Dataset):
 
     def preprocessing(self, example):
         input_ids = [self.tokenizer.bos_token_id]
-        label_ids = []
+        label_ids = [self.ignore_index]
 
         for message in example["messages"]:
             role = message["role"]
@@ -92,17 +92,22 @@ class SupervisedDataset(Dataset):
                     + content_ids
                 )
 
+        input_ids.append(self.tokenizer.eos_token_id)
+        label_ids.append(self.tokenizer.eos_token_id)
+        # truncate to max len
         input_ids = input_ids[: self.model_max_length]
         label_ids = label_ids[: self.model_max_length]
-        # input_ids += [self.tokenizer.eos_token_id] * (len(label_ids) - len(input_ids))
+        attention_mask = [1] * len(input_ids)
+        # pad to max len
         input_ids += [self.tokenizer.eos_token_id] * (
             self.model_max_length - len(input_ids)
         )
         label_ids += [self.ignore_index] * (self.model_max_length - len(label_ids))
+        attention_mask += [0] * (self.model_max_length - len(attention_mask))
+        # convert to pt tensor
         input_ids = torch.LongTensor(input_ids)
         label_ids = torch.LongTensor(label_ids)
-        # print(f"len input_ids: {len(input_ids)}, len label_ids: {len(label_ids)}")
+        attention_mask = torch.LongTensor(attention_mask)
-        attention_mask = input_ids.ne(self.tokenizer.eos_token_id)
         return {
             "input_ids": input_ids,
             "label_ids": label_ids,
@@ -158,7 +163,6 @@ def load_model_and_tokenizer(
 
 if __name__ == "__main__":
     model_path = "/mnt/data/user/tc_agi/yh/models/MiniCPM"
-    max_length = 512
     parser = transformers.HfArgumentParser(
         (ModelArguments, DataArguments, TrainingArguments)
     )

finetune/lora_finetune.ipynb

@@ -4,11 +4,13 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# MiniCPM-2B 参数高效微调（LoRA）消费级单卡示例\n",
+    "# MiniCPM-2B 参数高效微调（LoRA）A100 80G 单卡示例\n",
+    "\n",
+    "显存更小的显卡可用 batch size 和 grad_accum 间时间换空间\n",
     "\n",
     "本 notebook 是一个使用 `AdvertiseGen` 数据集对 MiniCPM-2B 进行 LoRA 微调，使其具备专业的广告生成能力的代码示例。\n",
     "\n",
-    "## 硬件需求\n",
+    "## 最低硬件需求\n",
     "- 显存：12GB\n",
     "- 显卡架构：安培架构（推荐）\n",
     "- 内存：16GB"
@@ -20,31 +22,7 @@
    "source": [
     "## 1. 准备数据集\n",
     "\n",
-    "下载 AdvertiseGen 数据集\n",
+    "将数据集转换为更通用的格式\n"
-    "- [Google Drive](https://drive.google.com/file/d/13_vf0xRTQsyneRKdD1bZIr93vBGOczrk/view?usp=sharing)\n",
-    "- [Tsinghua Cloud](https://cloud.tsinghua.edu.cn/f/b3f119a008264b1cabd1/?dl=1)\n",
-    "\n",
-    "下载后的数据集格式为 `.tar.gz` 的压缩格式，接下来的操作中，假设该压缩包被置于 `finetune/data/`。\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 校验文件完整性\n",
-    "!md5sum data/AdvertiseGen.tar.gz "
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# 解压数据集\n",
-    "!tar xvf data/AdvertiseGen.tar.gz "
    ]
   },
   {
@@ -103,6 +81,47 @@
    "source": [
     "!bash lora_finetune.sh"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 推理验证"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "from tqdm import tqdm\n",
+    "from transformers import AutoModelForCausalLM, AutoTokenizer"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "path = \"output/AdvertiseGenLoRA/20240315224356/checkpoint-3000\"\n",
+    "tokenizer = AutoTokenizer.from_pretrained(path)\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    path, torch_dtype=torch.bfloat16, device_map=\"cuda\", trust_remote_code=True\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "res, history = model.chat(tokenizer, query=\"<用户>类型#上衣*材质#牛仔布*颜色#白色*风格#简约*图案#刺绣*衣样式#外套*衣款式#破洞<AI>\", max_length=80, top_p=0.5)\n",
+    "res, history"
+   ]
   }
  ],
  "metadata": {
@@ -121,7 +140,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.13"
+   "version": "3.10.12"
   }
  },
  "nbformat": 4,

finetune/lora_finetune.sh

@@ -2,13 +2,13 @@ formatted_time=$(date +"%Y%m%d%H%M%S")
 echo $formatted_time
 
 
-deepspeed --include localhost:0 finetune.py \
+deepspeed --include localhost:1 finetune.py \
-    --model_name_or_path <your_model_name_or_path> \
+    --model_name_or_path MiniCPM-2B-sft-bf16 \
     --output_dir output/AdvertiseGenLoRA/$formatted_time/ \
     --train_data_path data/AdvertiseGenChatML/train.json \
     --eval_data_path data/AdvertiseGenChatML/dev.json \
-    --learning_rate 1e-3 --per_device_train_batch_size 1 \
+    --learning_rate 5e-5 --per_device_train_batch_size 32 \
-    --per_device_eval_batch_size 1 --fp16 --use_lora \
+    --per_device_eval_batch_size 64  --model_max_length 384 --bf16 --use_lora \
     --gradient_accumulation_steps 1 --warmup_steps 100 \
     --max_steps 3000 --weight_decay 0.01 \
     --evaluation_strategy steps --eval_steps 500 \

finetune/lora_finetune_ocnli.sh

@@ -0,0 +1,17 @@
+formatted_time=$(date +"%Y%m%d%H%M%S")
+echo $formatted_time
+
+
+deepspeed --include localhost:1 --master_port 19888 finetune.py \
+    --model_name_or_path MiniCPM-2B-sft-bf16 \
+    --output_dir output/OCNLILoRA/$formatted_time/ \
+    --train_data_path data/ocnli_public_chatml/train.json \
+    --eval_data_path data/ocnli_public_chatml/dev.json \
+    --learning_rate 5e-5 --per_device_train_batch_size 80 \
+    --per_device_eval_batch_size 128 --model_max_length 128 --bf16 --use_lora \
+    --gradient_accumulation_steps 1 --warmup_steps 100 \
+    --max_steps 1000 --weight_decay 0.01 \
+    --evaluation_strategy steps --eval_steps 500 \
+    --save_strategy steps --save_steps 500 --seed 42 \
+    --log_level info --logging_strategy steps --logging_steps 10 \
+    --deepspeed configs/ds_config_zero3_offload.json

finetune/sft_finetune.sh

@@ -2,16 +2,16 @@ formatted_time=$(date +"%Y%m%d%H%M%S")
 echo $formatted_time
 
 
-deepspeed --include localhost:0,1,2,3 finetune.py \
+deepspeed --include localhost:0,1 finetune.py \
-    --model_name_or_path <your_model_name_or_path> \
+    --model_name_or_path MiniCPM-2B-sft-bf16 \
     --output_dir output/AdvertiseGenSFT/$formatted_time/ \
     --train_data_path data/AdvertiseGenChatML/train.json \
     --eval_data_path data/AdvertiseGenChatML/dev.json \
-    --learning_rate 1e-3 --per_device_train_batch_size 1 \
+    --learning_rate 5e-5 --per_device_train_batch_size 14 \
-    --per_device_eval_batch_size 4 --bf16 \
+    --per_device_eval_batch_size 32 --bf16 \
-    --gradient_accumulation_steps 8 --warmup_steps 100 \
+    --gradient_accumulation_steps 2 --warmup_steps 100 \
     --max_steps 3000 --weight_decay 0.01 \
-    --evaluation_strategy steps --eval_steps 500 \
+    --evaluation_strategy steps --eval_steps 100 \
     --save_strategy steps --save_steps 500 --seed 42 \
     --log_level info --logging_strategy steps --logging_steps 10 \
-    --deepspeed configs/ds_config_zero3_offload.json
+    --deepspeed configs/ds_config_zero2.json