update examples

examples/README.md

@@ -3,41 +3,42 @@ We provide diverse examples about fine-tuning LLMs.
 ```
 examples/
 ├── lora_single_gpu/
-│   ├── pretrain.sh: Do pre-training
+│   ├── pretrain.sh: Do pre-training using LoRA
-│   ├── sft.sh: Do supervised fine-tuning
+│   ├── sft.sh: Do supervised fine-tuning using LoRA
-│   ├── reward.sh: Do reward modeling
+│   ├── reward.sh: Do reward modeling using LoRA
-│   ├── ppo.sh: Do PPO training
+│   ├── ppo.sh: Do PPO training using LoRA
-│   ├── dpo.sh: Do DPO training
+│   ├── dpo.sh: Do DPO training using LoRA
-│   ├── orpo.sh: Do ORPO training
+│   ├── orpo.sh: Do ORPO training using LoRA
 │   ├── prepare.sh: Save tokenized dataset
-│   └── predict.sh: Do batch predict
+│   └── predict.sh: Do batch predict and compute BLEU and ROUGE scores after LoRA tuning
 ├── qlora_single_gpu/
-│   ├── bitsandbytes.sh: Fine-tune 4/8-bit BNB models
+│   ├── bitsandbytes.sh: Fine-tune 4/8-bit BNB models using QLoRA
-│   ├── gptq.sh: Fine-tune 4/8-bit GPTQ models
+│   ├── gptq.sh: Fine-tune 4/8-bit GPTQ models using QLoRA
-│   ├── awq.sh: Fine-tune 4-bit AWQ models
+│   ├── awq.sh: Fine-tune 4-bit AWQ models using QLoRA
-│   └── aqlm.sh: Fine-tune 2-bit AQLM models
+│   └── aqlm.sh: Fine-tune 2-bit AQLM models using QLoRA
 ├── lora_multi_gpu/
-│   ├── single_node.sh: Fine-tune model with Accelerate on single node
+│   ├── single_node.sh: Fine-tune model with Accelerate on single node using LoRA
-│   └── multi_node.sh: Fine-tune model with Accelerate on multiple nodes
+│   └── multi_node.sh: Fine-tune model with Accelerate on multiple nodes using LoRA
 ├── full_multi_gpu/
-│   ├── single_node.sh: Fine-tune model with DeepSpeed on single node
+│   ├── single_node.sh: Full fine-tune model with DeepSpeed on single node
-│   └── multi_node.sh: Fine-tune model with DeepSpeed on multiple nodes
+│   ├── multi_node.sh: Full fine-tune model with DeepSpeed on multiple nodes
+│   └── predict.sh: Do batch predict and compute BLEU and ROUGE scores after full tuning
 ├── merge_lora/
 │   ├── merge.sh: Merge LoRA weights into the pre-trained models
-│   └── quantize.sh: Quantize fine-tuned model with AutoGPTQ
+│   └── quantize.sh: Quantize the fine-tuned model with AutoGPTQ
 ├── inference/
-│   ├── cli_demo.sh: Launch a command line interface
+│   ├── cli_demo.sh: Launch a command line interface with LoRA adapters
-│   ├── api_demo.sh: Launch an OpenAI-style API
+│   ├── api_demo.sh: Launch an OpenAI-style API with LoRA adapters
-│   ├── web_demo.sh: Launch a web interface
+│   ├── web_demo.sh: Launch a web interface with LoRA adapters
-│   └── evaluate.sh: Evaluate model on the MMLU benchmark
+│   └── evaluate.sh: Evaluate model on the MMLU/CMMLU/C-Eval benchmarks with LoRA adapters
 └── extras/
     ├── galore/
     │   └── sft.sh: Fine-tune model with GaLore
     ├── loraplus/
-    │   └── sft.sh: Fine-tune model with LoRA+
+    │   └── sft.sh: Fine-tune model using LoRA+
     ├── llama_pro/
     │   ├── expand.sh: Expand layers in the model
-    │   └── sft.sh: Fine-tune expanded model
+    │   └── sft.sh: Fine-tune the expanded model
     └── fsdp_qlora/
-        └── sft.sh: Fine-tune quantized model with FSDP
+        └── sft.sh: Fine-tune quantized model with FSDP+QLoRA
 ```

examples/README_zh.md

@@ -1,36 +1,36 @@
-我们提供了多样化的示例脚本。
+我们提供了多样化的大模型微调示例脚本。
 
 ```
 examples/
 ├── lora_single_gpu/
-│   ├── pretrain.sh: 进行预训练
+│   ├── pretrain.sh: 基于 LoRA 进行预训练
-│   ├── sft.sh: 进行指令监督微调
+│   ├── sft.sh: 基于 LoRA 进行指令监督微调
-│   ├── reward.sh: 进行奖励模型训练
+│   ├── reward.sh: 基于 LoRA 进行奖励模型训练
-│   ├── ppo.sh: 进行 PPO 训练
+│   ├── ppo.sh: 基于 LoRA 进行 PPO 训练
-│   ├── dpo.sh: 进行 DPO 训练
+│   ├── dpo.sh: 基于 LoRA 进行 DPO 训练
-│   ├── orpo.sh: 进行 ORPO 训练
+│   ├── orpo.sh: 基于 LoRA 进行 ORPO 训练
 │   ├── prepare.sh: 保存预处理后的数据集
-│   └── predict.sh: 进行批量预测
+│   └── predict.sh: 基于 LoRA 进行批量预测并计算 BLEU 和 ROUGE 分数
 ├── qlora_single_gpu/
-│   ├── bitsandbytes.sh: 微调 4/8 比特 BNB 模型
+│   ├── bitsandbytes.sh: 基于 QLoRA 微调 4/8 比特 BNB 模型
-│   ├── gptq.sh: 微调 4/8 比特 GPTQ 模型
+│   ├── gptq.sh: 基于 QLoRA 微调 4/8 比特 GPTQ 模型
-│   ├── awq.sh: 微调 4 比特 AWQ 模型
+│   ├── awq.sh: 基于 QLoRA 微调 4 比特 AWQ 模型
-│   └── aqlm.sh: 微调 2 比特 AQLM 模型
+│   └── aqlm.sh: 基于 QLoRA 微调 2 比特 AQLM 模型
 ├── lora_multi_gpu/
-│   ├── single_node.sh: 使用 Accelerate 进行单节点训练
+│   ├── single_node.sh: 使用 Accelerate 进行单节点 LoRA 训练
-│   └── multi_node.sh: 使用 Accelerate 进行多节点训练
+│   └── multi_node.sh: 使用 Accelerate 进行多节点 LoRA 训练
 ├── full_multi_gpu/
-│   ├── single_node.sh: 使用 DeepSpeed 进行单节点训练
+│   ├── single_node.sh: 使用 DeepSpeed 进行单节点全量训练
-│   └── multi_node.sh: 使用 DeepSpeed 进行多节点训练
+│   ├── multi_node.sh: 使用 DeepSpeed 进行多节点全量训练
-|   └── predict.sh: 使用单卡做全参批量预测
+│   └── predict.sh: 基于全量训练进行批量预测并计算 BLEU 和 ROUGE 分数
 ├── merge_lora/
 │   ├── merge.sh: 将 LoRA 权重合并到预训练模型中
-│   └── quantize.sh: 使用 AutoGPTQ 量化模型
+│   └── quantize.sh: 使用 AutoGPTQ 量化微调后的模型
 ├── inference/
-│   ├── cli_demo.sh: 启动命令行推理接口
+│   ├── cli_demo.sh: 启动 LoRA 模型的命令行推理接口
-│   ├── api_demo.sh: 启动 OpenAI 风格 API
+│   ├── api_demo.sh: 启动 LoRA 模型的 OpenAI 风格 API
-│   ├── web_demo.sh: 启动浏览器推理接口
+│   ├── web_demo.sh: 启动 LoRA 模型的浏览器推理接口
-│   └── evaluate.sh: 在 MMLU 数据集上评测模型
+│   └── evaluate.sh: 在 MMLU/CMMLU/C-Eval 数据集上评测 LoRA 模型
 └── extras/
     ├── galore/
     │   └── sft.sh: 使用 GaLore 训练模型
@@ -40,5 +40,5 @@ examples/
     │   ├── expand.sh: 扩展模型中的层
     │   └── sft.sh: 训练扩展后的模型
     └── fsdp_qlora/
-        └── sft.sh: 使用 FSDP 微调量化模型
+        └── sft.sh: 使用 FSDP+QLoRA 微调量化模型
 ```

examples/extras/loraplus/sft.sh

@@ -9,6 +9,7 @@ CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
     --template default \
     --finetuning_type lora \
     --lora_target q_proj,v_proj \
+    --loraplus_lr_ratio 16.0 \
     --output_dir ../../saves/LLaMA2-7B/loraplus/sft \
     --overwrite_cache \
     --overwrite_output_dir \
@@ -29,5 +30,4 @@ CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
     --max_samples 3000 \
     --val_size 0.1 \
     --plot_loss \
-    --fp16 \
+    --fp16
-    --loraplus_lr_ratio 16.0

examples/full_multi_gpu/predict.sh

@@ -3,7 +3,7 @@
 CUDA_VISIBLE_DEVICES=0 python ../../src/train_bash.py \
     --stage sft \
     --do_predict \
-    --model_name_or_path meta-llama/Llama-2-7b-hf \
+    --model_name_or_path ../../saves/LLaMA2-7B/full/sft \
     --dataset alpaca_gpt4_en,glaive_toolcall \
     --dataset_dir ../../data \
     --template default \

examples/merge_lora/merge.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 # DO NOT use quantized model or quantization_bit when merging lora weights
 
-CUDA_VISIBLE_DEVICES= python ../../src/export_model.py \
+CUDA_VISIBLE_DEVICES=0 python ../../src/export_model.py \
     --model_name_or_path meta-llama/Llama-2-7b-hf \
     --adapter_name_or_path ../../saves/LLaMA2-7B/lora/sft \
     --template default \

src/llmtuner/chat/vllm_engine.py

@@ -1,8 +1,6 @@
 import uuid
 from typing import TYPE_CHECKING, AsyncGenerator, AsyncIterator, Dict, List, Optional, Sequence
 
-from transformers.utils.versions import require_version
-
 from ..data import get_template_and_fix_tokenizer
 from ..extras.misc import get_device_count
 from ..extras.packages import is_vllm_available
@@ -25,7 +23,6 @@ class VllmEngine(BaseEngine):
         finetuning_args: "FinetuningArguments",
         generating_args: "GeneratingArguments",
     ) -> None:
-        require_version("vllm>=0.3.3", "To fix: pip install vllm>=0.3.3")
         self.can_generate = finetuning_args.stage == "sft"
         engine_args = AsyncEngineArgs(
             model=model_args.model_name_or_path,

src/llmtuner/extras/packages.py

@@ -49,10 +49,6 @@ def is_starlette_available():
     return _is_package_available("sse_starlette")
 
 
-def is_unsloth_available():
-    return _is_package_available("unsloth")
-
-
 def is_uvicorn_available():
     return _is_package_available("uvicorn")
 

src/llmtuner/hparams/parser.py

@@ -8,10 +8,10 @@ import transformers
 from transformers import HfArgumentParser, Seq2SeqTrainingArguments
 from transformers.trainer_utils import get_last_checkpoint
 from transformers.utils import is_torch_bf16_gpu_available
+from transformers.utils.versions import require_version
 
 from ..extras.logging import get_logger
 from ..extras.misc import check_dependencies, get_current_device
-from ..extras.packages import is_unsloth_available
 from .data_args import DataArguments
 from .evaluation_args import EvaluationArguments
 from .finetuning_args import FinetuningArguments
@@ -74,6 +74,26 @@ def _verify_model_args(model_args: "ModelArguments", finetuning_args: "Finetunin
             raise ValueError("Quantized model only accepts a single adapter. Merge them first.")
 
 
+def _check_extra_dependencies(
+    model_args: "ModelArguments",
+    finetuning_args: "FinetuningArguments",
+    training_args: Optional["Seq2SeqTrainingArguments"] = None,
+) -> None:
+    if model_args.use_unsloth:
+        require_version("unsloth", "Please install unsloth: https://github.com/unslothai/unsloth")
+
+    if model_args.infer_backend == "vllm":
+        require_version("vllm>=0.3.3", "To fix: pip install vllm>=0.3.3")
+
+    if finetuning_args.use_galore:
+        require_version("galore_torch", "To fix: pip install galore_torch")
+
+    if training_args is not None and training_args.predict_with_generate:
+        require_version("jieba", "To fix: pip install jieba")
+        require_version("nltk", "To fix: pip install nltk")
+        require_version("rouge_chinese", "To fix: pip install rouge-chinese")
+
+
 def _parse_train_args(args: Optional[Dict[str, Any]] = None) -> _TRAIN_CLS:
     parser = HfArgumentParser(_TRAIN_ARGS)
     return _parse_args(parser, args)
@@ -131,9 +151,6 @@ def get_train_args(args: Optional[Dict[str, Any]] = None) -> _TRAIN_CLS:
     if training_args.do_train and training_args.predict_with_generate:
         raise ValueError("`predict_with_generate` cannot be set as True while training.")
 
-    if training_args.do_train and model_args.use_unsloth and not is_unsloth_available():
-        raise ValueError("Unsloth was not installed: https://github.com/unslothai/unsloth")
-
     if finetuning_args.use_dora and model_args.use_unsloth:
         raise ValueError("Unsloth does not support DoRA.")
 
@@ -158,6 +175,7 @@ def get_train_args(args: Optional[Dict[str, Any]] = None) -> _TRAIN_CLS:
         raise ValueError("vLLM backend is only available for API, CLI and Web.")
 
     _verify_model_args(model_args, finetuning_args)
+    _check_extra_dependencies(model_args, finetuning_args, training_args)
 
     if (
         training_args.do_train
@@ -277,6 +295,7 @@ def get_infer_args(args: Optional[Dict[str, Any]] = None) -> _INFER_CLS:
             raise ValueError("vLLM engine does not support RoPE scaling.")
 
     _verify_model_args(model_args, finetuning_args)
+    _check_extra_dependencies(model_args, finetuning_args)
 
     if model_args.export_dir is not None:
         model_args.device_map = {"": torch.device(model_args.export_device)}
@@ -298,6 +317,7 @@ def get_eval_args(args: Optional[Dict[str, Any]] = None) -> _EVAL_CLS:
         raise ValueError("vLLM backend is only available for API, CLI and Web.")
 
     _verify_model_args(model_args, finetuning_args)
+    _check_extra_dependencies(model_args, finetuning_args)
 
     model_args.device_map = "auto"
 

src/llmtuner/model/loader.py

@@ -85,7 +85,9 @@ def load_model(
             logger.warning("Unsloth does not support loading adapters.")
 
     if model is None:
-        model = AutoModelForCausalLM.from_pretrained(model_args.model_name_or_path, config=config, **init_kwargs)
+        init_kwargs["config"] = config
+        init_kwargs["pretrained_model_name_or_path"] = model_args.model_name_or_path
+        model: "PreTrainedModel" = AutoModelForCausalLM.from_pretrained(**init_kwargs)
 
     patch_model(model, tokenizer, model_args, is_trainable)
     register_autoclass(config, model, tokenizer)

src/llmtuner/train/sft/metric.py

@@ -2,7 +2,6 @@ from dataclasses import dataclass
 from typing import TYPE_CHECKING, Dict, Sequence, Tuple, Union
 
 import numpy as np
-from transformers.utils.versions import require_version
 
 from ...extras.constants import IGNORE_INDEX
 from ...extras.packages import is_jieba_available, is_nltk_available, is_rouge_available
@@ -33,10 +32,6 @@ class ComputeMetrics:
         r"""
         Uses the model predictions to compute metrics.
         """
-        require_version("jieba", "To fix: pip install jieba")
-        require_version("nltk", "To fix: pip install nltk")
-        require_version("rouge_chinese", "To fix: pip install rouge-chinese")
-
         preds, labels = eval_preds
         score_dict = {"rouge-1": [], "rouge-2": [], "rouge-l": [], "bleu-4": []}
 

src/llmtuner/train/utils.py

@@ -5,7 +5,6 @@ from transformers import Trainer
 from transformers.optimization import get_scheduler
 from transformers.pytorch_utils import ALL_LAYERNORM_LAYERS
 from transformers.trainer_pt_utils import get_parameter_names
-from transformers.utils.versions import require_version
 
 from ..extras.logging import get_logger
 from ..extras.packages import is_galore_available
@@ -168,8 +167,6 @@ def _create_galore_optimizer(
     training_args: "Seq2SeqTrainingArguments",
     finetuning_args: "FinetuningArguments",
 ) -> "torch.optim.Optimizer":
-    require_version("galore_torch", "To fix: pip install galore_torch")
-
     if len(finetuning_args.galore_target) == 1 and finetuning_args.galore_target[0] == "all":
         galore_targets = find_all_linear_modules(model)
     else: