---
language:
- zh
- en
tags:
- llama2
- llama2-base
- llama2-base-7B
---
# 7B Chinese Chatbot trained based on LLama2-base 7B 

## Introduction

在完成了[Llama2-chat 7B Chinese](https://huggingface.co/RicardoLee/Llama2-chat-Chinese-50W) 和 [Llama2-chat 13B Chinese](https://huggingface.co/RicardoLee/Llama2-chat-13B-Chinese-50W) 的训练后，我非常好奇能否直接基于Llama2-base 系列直接进行SFT训练。这也是本模型仓库的初衷。

但是在实际操作中，在用了原先chat模型的LoRA训练框架后，我发现基于Llama2 base的 LoRA 训练非常难以收敛，随时处于梯度爆炸的边缘。DeepSpeed 会频繁触发reduce scale 操作，最终scale太小越界导致训练崩溃。我遍历了LR 1e-5 - 2e-4，LoRA rank \[4, 8, 64\]，LoRA Alpha \[1,4,8,16,32\]，LoRA Dropout \[0.05, 0.1\] ，Warmup Ratio \[0.01, 0.03, 0.05\]等超参数，均无法稳定训练。因此，本模型采用先进行1 epoch的全量训练[Llama2-base-7B-Chinese-50W-pre\_release](https://huggingface.co/RicardoLee/Llama2-base-7B-Chinese-50W-pre_release)，然后在其上进行LoRA训练得到本模型仓库的模型。

由于网上存在使用LoRA 在英文SFT数据集上基于Llama2-base 进行SFT训练成功的样例，因此我怀疑难以训练的原因可能是扩中文词表embedding导致训练难度大幅度提升。

训练数据使用[BELLE](https://huggingface.co/BelleGroup)项目中采样的50万SFT数据进行SFT训练。

After finishing the training of [Llama2-chat 7B Chinese](https://huggingface.co/RicardoLee/Llama2-chat-Chinese-50W) and [Llama2-chat 13B Chinese](https://huggingface.co/RicardoLee/Llama2-chat-13B-Chinese-50W), I am deeply intrigued by the possibility of conducting SFT (Style-Fine-Tuning) training directly based on the Llama2-base series. This is the fundamental purpose of this model repository.

**However**, in real practice, I have observed that conducting LoRA training based on the Llama2 base model, within the framework of the previous Llama2-chat SFT project, presents significant challenges in achieving convergence. The gradient explosion happens in every training step and casue reducing scale operation in Deepspeed. In the end, the scale is too small and out of bounds, causing the training to crash. I have traversed LR 1e-5 - 2e-4，LoRA rank \[4, 8, 64\]，LoRA Alpha \[1,4,8,16,32\]，LoRA Dropout \[0.05, 0.1\] ，Warmup Ratio \[0.01, 0.03, 0.05\] and other hyperparameters, all of which cannot be trained stably. Therefore, this model has firstly reverted to a full-parameter SFT 1 epoch training[Llama2-base-7B-Chinese-50W-pre\_release](https://huggingface.co/RicardoLee/Llama2-base-7B-Chinese-50W-pre_release), and then conducted a 2 epoch LoRA training. The reasons behind the difficulties encountered during LoRA training require further analysis.

As there are instances online where successful LoRA training on English SFT datasets using Llama2-base has been demonstrated, I suspect that the challenge in training might be attributed to the expansion of the Chinese word embedding, resulting in a substantial increase in training difficulty.

The training data is sampled from [BELLE](https://huggingface.co/BelleGroup) project, which consists of 500,000 SFT samples.

## Train Detail

一些训练上的细节：

1. 训练框架：该模型先采用1 epoch的全参数SFT训练，然后再进行2 epoch的LoRA SFT训练。
2. Tokenizer：该模型使用了Chinese-Alpaca-Plus模型的tokenizer.model。这是因为LLama2本身的tokenizer.model同LLama1是一摸一样的。因此理论上可以完全复用Chinese-LLaMa项目的tokenizer而不会产生如何错位问题。
3. 训练参数：本模型在全量训练阶段LR 为2e-4。Warmup ratio 为0.01。在LoRA训练阶段LR 为2e-5，Warmup ratio 为0.05。
4. 训练资源：8卡V100。17小时

Some details in training:

1. Trianing Framework: The model is first trained with 1-epoch full-parameter SFT, and then trained with 2-epoch LoRA SFT.
2. Tokenizer: This model utilizes the tokenizer.model from the Chinese-Alpaca-Plus model. The reason for this choice is that the tokenizer.model in LLama2 is identical to the one used in LLama1. As a result, it is theoretically feasible to entirely reuse the tokenizer from the Chinese-LLaMa project without encountering any issues related to token misalignment.
3. Training Parameters: The LR of this model is 2e-4 in the full training phase, while 2e-5 in the LoRA training phase. Warmup ratio is 0.01 and 0.05 respectively.
4. Training Resource: 8\*V100, 17 hours.

## Inference

该模型依然采用stanford alpaca 模版。因此在测试时且别忘记添加开场白。开场白如下：

"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n\n${Your Content}\n\n### Response:\n\n"

对于带上文的对话，开场白如下：

"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n\nHuman:${Previous Human Content}\nAssistant:${Previous Assistance Content}\nHuman:${Your Question}\n\n### Response:\n\n"

This model still using the Stanford Alpaca template. Therefore, don't forget to add prologue template. The prologue template is:

"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n\n${Your Content}\n\n### Response:\n\n"

For dialogue with context, the prelogue template is:

"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n\nHuman:${Previous Human Content}\nAssistant:${Previous Machine Content}\nHuman:${Your Question}\n\n### Response:\n\n"

## Licence

本仓库的模型依照 Apache-2.0 协议开源，模型的权重的使用则需要遵循LLama2[MODEL LICENCE](LICENSE)。

This repository's models are open-sourced under the Apache-2.0 license, and their weight usage must adhere to LLama2 [MODEL LICENCE](LICENSE) license.

## Future Work

将会在近期逐步放出

1. 更大SFT数据规模训练下的模型。
2. 13B及以下的LLama2 同LLama2-chat的模型，以供大家对比。

I will release the following models:

1. Models trained on larger data scale.
2. Models trained on LLama2 and LLama2-chat (under the 13B, since I only have V100), for comparison.