"The Greatest Path is the Simplest"

中文 | English

This open-source project aims to train a tiny language model called MiniMind from scratch in just 3 hours, with a model size of only 26.88M.
MiniMind is extremely lightweight, with the smallest version being approximately $\frac{1}{7000}$ the size of GPT3, making it possible for even an ordinary personal GPU to perform quick inference and even training.
MiniMind provides the full-stage code for a simplified large model structure, dataset cleaning and preprocessing, supervised pretraining, supervised instruction fine-tuning (SFT), low-rank adaptation (LoRA) fine-tuning, and direct preference alignment with reinforcement learning without rewards (DPO). It also includes code for expanding to sparse models with mixed experts (MoE) and multi-modal vision language models ( VLM): MiniMind-V.
This is not just an implementation of an open-source model but also a tutorial for getting started with large language models (LLM).
We hope this project will serve as an introductory example for researchers, helping them quickly get started and inspiring more exploration and innovation in the LLM field.

To avoid misinterpretation, "fastest 3 hours" means you need a machine with hardware configuration superior to mine. Detailed specifications will be provided below.

ModelScope Online Testing | Bilibili Video Link

📌 Introduction

In the field of large language models (LLMs) such as GPT, LLaMA, GLM, etc., while their performance is impressive, the massive model parameters—often in the range of 10 billion—make them difficult to train or even infer on personal devices with limited memory. Most users do not settle for merely fine-tuning large models using methods like LoRA to learn a few new instructions. It's akin to teaching Newton to use a 21st-century smartphone, which is far removed from the essence of learning physics itself.

Additionally, the abundance of flawed, superficial AI tutorials offered by subscription-based marketing accounts exacerbates the problem of finding quality content to understand LLMs, severely hindering learners.

Therefore, the goal of this project is to lower the barrier to entry for working with LLMs as much as possible, by training an extremely lightweight language model from scratch.

Caution

As of 2024-09-17, MiniMind has trained three model versions, with the smallest model requiring only 26M (0.02B) parameters to achieve smooth conversational abilities!

Model (Size)	Tokenizer Length	Inference Memory Usage	Release Date	Subjective Rating (/100)
minimind-v1-small (26M)	6400	0.5 GB	2024.08.28	50'
minimind-v1-moe (4×26M)	6400	1.0 GB	2024.09.17	55'
MiniMind-V1 (108M)	6400	1.0 GB	2024.09.01	60'

This analysis was run on an RTX 3090 GPU with Torch 2.1.2, CUDA 12.2, and Flash Attention 2.

The project includes:

Public MiniMind model code (including Dense and MoE models), code for Pretrain, SFT instruction fine-tuning, LoRA fine-tuning, and DPO preference optimization, along with datasets and sources.
Compatibility with popular frameworks such as transformers, accelerate, trl, and peft.
Training support for single-GPU and multi-GPU setups(DDP、DeepSpeed), Use wandb to visualize the training process. The training process allows for stopping and resuming at any point.
Code for testing the model on the Ceval dataset.
Implementation of a basic chat interface compatible with OpenAI's API, facilitating integration into third-party Chat UIs (such as FastGPT, Open-WebUI, etc.).

We hope this open-source project helps LLM beginners get started quickly!

👉Recent Updates

2024-10-05 (newest 🎉)

Added visual capabilities to MiniMind-V(ision)
Check out the twin project minimind-v for more details!

2024-09-27

👉Updated the preprocessing method for the pretrain dataset on 09-27 to ensure text integrity, opting to abandon the preprocessing into .bin training format (slightly sacrificing training speed).
The current filename for the pretrain data after preprocessing is: pretrain_data.csv.
Removed some redundant code.

2024-09-17 (new🎉)

Updated the minimind-v1-moe model
To prevent ambiguity, all mistral_tokenizer versions have been removed, and a custom minimind_tokenizer is now used as the tokenizer.

2024-09-01

Updated the MiniMind-V1 (108M) model, using minimind_tokenizer with 3 pre-training epochs and 10 SFT epochs for more thorough training and improved performance.
The project has been deployed to ModelScope's Creative Space and can be experienced on the website:
ModelScope Online Experience

2024-08-27

The project was open-sourced for the first time.

📌 Environment

These are my personal software and hardware environment configurations. Please adjust according to your own setup:

CPU: Intel(R) Core(TM) i9-10980XE CPU @ 3.00GHz
Memory: 128 GB
GPU: NVIDIA GeForce RTX 3090 (24GB) * 2
Environment: python 3.9 + Torch 2.1.2 + DDP multi-GPU training

Ubuntu == 20.04
Python == 3.9
Pytorch == 2.1.2
CUDA == 12.2
requirements.txt

📌 Quick Start Test

Hugging Face

MiniMind (HuggingFace)

MiniMind (ModelScope)

# step 1
git clone https://huggingface.co/jingyaogong/minimind-v1

# step 2
python 2-eval.py

or you can run streamlit, launch a web page to chat with minimind-v1

Requires python>=3.10, install pip install streamlit==1.27.2

# or step 3, use streamlit
streamlit run fast_inference.py

📌 Quick Start Train

0.Clone the project code

git clone https://github.com/jingyaogong/minimind.git
cd minimind

1.Install the required dependencies
```
  pip install -r requirements.txt
```
```
# Test if torch can use CUDA
import torch
print(torch.cuda.is_available())
```
If it is not available, please go to torch_stable to download the whl file for installation. Refer to this link
2.If you need to train the model yourself
- 2.1 Download the dataset download link and place it in the ./dataset directory.
- 2.2 Run python data_process.py to process the dataset, such as token-encoding pretrain data and extracting QA data to CSV files for the SFT dataset.
- 2.3 Adjust the model parameter configuration in ./model/LMConfig.py.
- 2.4 Execute pretraining with python 1-pretrain.py.
- 2.5 Perform instruction fine-tuning with python 3-full_sft.py.
- 2.6 Perform LoRA fine-tuning (optional) with python 4-lora_sft.py.
- 2.7 Execute DPO human preference reinforcement learning alignment (optional) with python 5-dpo_train.py.
3.Test model inference performance
- Ensure that the required trained parameter weights are located in the ./out/ directory.
- You can also directly download and use the trained model weights from [Trained Model Weights](#Trained Model Weights).
```
out
├── multi_chat
│   ├── full_sft_512.pth
│   ├── full_sft_512_moe.pth
│   └── full_sft_768.pth
├── single_chat
│   ├── full_sft_512.pth
│   ├── full_sft_512_moe.pth
│   └── full_sft_768.pth
├── pretrain_768.pth
├── pretrain_512_moe.pth
├── pretrain_512.pth
```
- Test the pretraining model's chain effect with python 0-eval_pretrain.py
- Test the model's conversational effect with python 2-eval.py

🍭「Tip」Both pretrain and full_sft support multi-card acceleration.

If your device has only 1 GPU, you can start the training using native Python:

Execute pretrain or instruction fine-tuning:

python 1-pretrain.py
# and
python 3-full_sft.py

If your device has N (N > 1) GPUs:

Start training on a single machine with N GPUs(DDP)

torchrun --nproc_per_node N 1-pretrain.py
# and
torchrun --nproc_per_node N 3-full_sft.py

Start training on a single machine with N GPUs(DeepSpeed)

deepspeed --master_port 29500 --num_gpus=N 1-pretrain.py
# and
deepspeed --master_port 29500 --num_gpus=N 3-full_sft.py

Record the training process
```
torchrun --nproc_per_node N 1-pretrain.py --use_wandb
# and
python 1-pretrain.py --use_wandb
```
By adding the --use_wandb parameter, you can record the training process. After training is complete, you can view the training process on the wandb website. You can specify the project name and run name by modifying the wandb_project and wandb_run_name parameters.

📌 Data sources

🤖 Tokenizer: In NLP, a Tokenizer is similar to a dictionary, mapping words from natural language to numbers like 0, 1, 36, etc., which can be understood as page numbers in the "dictionary" representing words. There are two ways to build an LLM tokenizer: one is to create a vocabulary and train a tokenizer yourself, as seen in train_tokenizer.py; the other is to use a pre-trained tokenizer from an open-source model.

You can use a standard dictionary like Xinhua or Oxford. The advantage is that token conversion has a good compression rate, but the downside is that the vocabulary can be very large, with tens of thousands of words and phrases. Alternatively, you can use a custom-trained tokenizer. The advantage is that you can control the vocabulary size, but the compression rate may not be ideal, and rare words might be missed.

The choice of "dictionary" is crucial. The output of an LLM is essentially a multi-class classification problem over N words in the dictionary, which is then decoded back into natural language. Because LLMs are very small, to avoid the model being top-heavy (with the embedding layer's parameters taking up too much of the model), the vocabulary length should be kept relatively small.

Powerful open-source models like 01万物, 千问, chatglm, mistral, and Llama3 have the following tokenizer vocabulary sizes:

Tokenizer Model	Vocabulary Size	Come from
yi tokenizer	64,000	01-AI（China）
qwen2 tokenizer	151,643	Alibaba Cloud（China）
glm tokenizer	151,329	Zhipu AI（China）
mistral tokenizer	32,000	Mistral AI（China）
llama3 tokenizer	128,000	Meta（China）
minimind tokenizer	6,400	Custom

👉Update on 2024-09-17: To avoid ambiguity from previous versions and control the model size, all Minimind models now use the Minimind_tokenizer for tokenization, and all versions of the Mistral_tokenizer have been deprecated.

Although the Minimind_tokenizer has a small length and its encoding/decoding efficiency is weaker compared to Chinese-friendly tokenizers like Qwen2 and GLM, the Minimind models have opted for their custom-trained Minimind_tokenizer to maintain a lightweight parameter structure and prevent an imbalance between encoding and computation layers. This is because the Minimind vocabulary size is only 6,400. Moreover, Minimind has not encountered any issues with decoding rare words in practical tests, and the performance has been satisfactory. Due to the custom vocabulary being compressed to 6,400 tokens, the total parameter size of the LLM is minimized to only 26M.

📙 Pretrain Data: The Seq-Monkey General Text Dataset / Baidu is a collection of data from various public sources such as websites, encyclopedias, blogs, open-source code, books, etc. It has been compiled, cleaned, and organized into a unified JSONL format, with rigorous filtering and deduplication to ensure data comprehensiveness, scale, reliability, and high quality. The total amount is approximately 10B tokens, suitable for pretraining Chinese large language models.

📕 SFT Data: The Jiangshu Large Model SFT Dataset is a comprehensive, uniformly formatted, and secure resource for large model training and research. It includes a large amount of open-source data collected and organized from publicly available online sources, with format unification and data cleaning. It comprises a Chinese dataset with 10M entries and an English dataset with 2M entries. The total amount is approximately 3B tokens, suitable for SFT of Chinese large language models. The dataset integration includes all data from the following sources (for reference only, no need to download separately, just download the complete [SFT Data]):
📘 DPO Data: Approximately 80,000 DPO (Direct Preference Optimization) data entries, which are manually labeled preference data, come from Huozi Model. These can be used to train reward models to optimize response quality and align more closely with human preferences.

More Datasets: HqWu-HITCS/Awesome-Chinese-LLM is currently collecting and organizing open-source models, applications, datasets, and tutorials related to Chinese LLMs, with continuous updates on the latest developments in this field. Comprehensive and professional, respect!

Dataset Download Links

MiniMind Training Dataset	Download Link
[tokenizer Data]	HuggingFace / Baidu
[Pretrain Data]	Seq-Monkey General Text Dataset / Baidu
[SFT Data]	Jiangshu Large Model SFT Dataset
[DPO Data]	Huggingface

📌 Model

MiniMind-Dense (like Llama3.1) uses a Transformer Decoder-Only architecture. The differences from GPT-3 are:

It employs GPT-3's pre-normalization method, which normalizes the input of each Transformer sub-layer rather than the output. Specifically, it uses the RMSNorm normalization function.
It replaces ReLU with the SwiGLU activation function to enhance performance.
Like GPT-Neo, it omits absolute position embeddings in favor of Rotary Position Embeddings (RoPE), which improves performance for inference beyond the training length.

The MiniMind-MoE model is based on the MixFFN mixture-of-experts module from Llama3 and DeepSeek-V2.

DeepSeek-V2 adopts more granular expert partitioning and shared expert isolation techniques in the feed-forward network (FFN) to improve the performance of experts.

The overall structure of MiniMind remains consistent, with minor adjustments in RoPE calculations, inference functions, and FFN layer code. The structure is illustrated in the figure below (redrawn):

Model configurations can be found in ./model/LMConfig.py. The model types and parameters are shown in the table below:

Model Name	params	len_vocab	n_layers	d_model	kv_heads	q_heads	share+route	TopK
minimind-v1-small	26M	6400	8	512	8	16	-	-
minimind-v1-moe	4×26M	6400	8	512	8	16	2+4	2
minimind-v1	108M	6400	16	768	8	16	-	-

📌 Experiment

Model Name	params	len_vocab	batch_size	pretrain_time	sft_single_time	sft_multi_time
minimind-v1-small	26M	6400	64	≈2 hour (1 epoch)	≈2 hour (1 epoch)	≈0.5 hour (1 epoch)
minimind-v1-moe	4×26M	6400	40	≈6 hour (1 epoch)	≈5 hour (1 epoch)	≈1 hour (1 epoch)
minimind-v1	108M	6400	16	≈6 hour (1 epoch)	≈4 hour (1 epoch)	≈1 hour (1 epoch)

Pretraining (Text-to-Text):
- LLMs first need to absorb a vast amount of knowledge, much like filling a well with ink. The more "ink" it has, the better its understanding of the world will be.
- Pretraining involves the model learning a large amount of basic knowledge from sources such as Wikipedia, news articles, common knowledge, books, etc.
- It unsupervisedly compresses knowledge from vast text data into its model weights with the aim of learning word sequences. For instance, if we input “Qin Shi Huang is,” after extensive training, the model can predict that the next probable sentence is “the first emperor of China.”
The learning rate for pretraining is set dynamically between 1e-4 and 1e-5, with 2 epochs and a training time of less than one day.
```
torchrun --nproc_per_node 2 1-pretrain.py
```
Single Dialog Fine-Tuning:
- After pretraining, the semi-finished LLM has almost all language knowledge and encyclopedic common sense. At this stage, it only performs word sequences without understanding how to chat with humans.
- The model needs fine-tuning to adapt to chat templates. For example, it should recognize that a template like “ Qin Shi Huang is ” indicates the end of a complete conversation, rather than just generating the next word.
- This process is known as instruction fine-tuning, akin to teaching a knowledgeable person like Newton to adapt to 21st-century chat habits, learning the pattern of messages on the left and responses on the right.
- During training, MiniMind’s instruction and response lengths are truncated to 512 to save memory. Just as we start with shorter texts when learning, we don’t need to separately learn longer articles once we master shorter ones.
During inference, RoPE can be linearly interpolated to extend lengths to 1024 or 2048 or more. The learning rate is set dynamically between 1e-5 and 1e-6, with 5 epochs for fine-tuning.
```
# Set dataset to sft_data_single.csv in 3-full_sft.py
torchrun --nproc_per_node 2 3-full_sft.py
```
Multi-Dialog Fine-Tuning:
- Building on step 2, the LLM has learned a single-question-to-answer chat template. Now, it only needs further fine-tuning on longer chat templates with historical question-and-answer pairs.
- Use the history_chat field for historical dialogues and history_chat_response for historical responses in the dataset.
- Construct new chat templates like [question->answer, question->answer, question->] and use this dataset for fine-tuning.
- The trained model will not only answer the current question but also conduct coherent dialogues based on historical interactions.
- This step is not strictly necessary, as small models have weak long-context dialogue abilities, and forcing multi-turn Q&A templates may slightly compromise single-turn SFT performance.
The learning rate is set dynamically between 1e-5 and 1e-6, with 2 epochs for fine-tuning.
```
# Set dataset to sft_data.csv in 3-full_sft.py
torchrun --nproc_per_node 2 3-full_sft.py
```
Direct Preference Optimization (DPO):
- In previous training sessions, GPT has already acquired basic conversational abilities, but these abilities are entirely based on word-by-word concatenation, lacking the motivation of positive and negative examples.
- GPT is still unaware of what constitutes a good response and what constitutes a poor one. We hope it can align more with human preferences and provide more satisfying responses.
- This process is akin to training GPT in a workplace setting, learning from the examples of outstanding employees and the mistakes of underperforming ones, to better serve customers.
- In the RLHF series, unlike PPO (Proximal Policy Optimization), which requires reward models and value models,
- DPO derives an explicit solution for the PPO reward model, replacing the online reward model with offline data, where ref outputs can be saved in advance.
- DPO maintains nearly the same performance, requiring only the actor and ref models to run, significantly reducing memory overhead and increasing training stability.
- Similarly, the RL steps for LLM are not mandatory, with both advantages and disadvantages.
For the Huozi trio (q, chose, reject) dataset, the learning rate is set to 1e-5, with half-precision fp16, 1 epoch, and it takes about 1 hour.
```
python 5-dpo_train.py
```

📋 Regarding LLM parameter configuration, an interesting paper MobileLLM provides detailed research and experiments. The scaling law exhibits unique patterns in small models. The parameters that significantly influence the scaling of Transformer models are primarily d_model and n_layers.

d_model↑ + n_layers↓ -> Short and wide models
d_model↓ + n_layers↑ -> Tall and narrow models

The Scaling Law proposed in 2020 posits that the amount of training data, parameter count, and training iterations are the key factors determining performance, with the influence of model architecture being nearly negligible. However, this law seems not to fully apply to small models. MobileLLM suggests that the depth of the architecture is more important than its width. A "deep and narrow" model can learn more abstract concepts compared to a "wide and shallow" model. For instance, when the model parameters are fixed at 125M or 350M, a 30–42 layer "narrow" model significantly outperforms a 12-layer "short and wide" model. This trend is observed across eight benchmark tests, including common sense reasoning, question answering, and reading comprehension. This is a fascinating discovery, as previously, few attempts were made to stack more than 12 layers when designing architectures for small models around the 100M parameter range. This aligns with the observations from MiniMind, where adjusting parameters between d_model and n_layers during training produced similar effects. However, "deep and narrow" has its limitations. When d_model < 512, the disadvantages of collapsing word embedding dimensions become very pronounced, and increasing layers does not compensate for the shortcomings in d_head caused by fixed q_head. Conversely, when d_model > 1536, increasing layers seems to have a higher priority than d_model, providing a better "cost-performance" ratio and effect gain. Therefore, MiniMind sets d_model = 512 and n_layers = 8 for the small model to achieve a balance between "minimal size <-> better performance." For greater performance gains, d_model = 768 and n_layers = 16 are set, aligning better with the scaling law for small models.

For reference, the configuration details for GPT-3 are shown in the table below:

Trained Model Weights

🔗Baidu Netdisk

Model Name	params	Config	pretrain_model	single_sft_model	multi_sft_model	rl_model
minimind-v1-small	26M	d_model=512 n_layers=8	`pretrain_512.pth`	`single_chat/full_sft_512.pth`	`multi_chat/full_sft_512.pth`	`rl_512.pth`
minimind-v1-moe	4×26M	d_model=512 n_layers=8	`pretrain_512_moe.pth`	`single_chat/full_sft_512_moe.pth`	`multi_chat/full_sft_512_moe.pth`	-
minimind-v1	108M	d_model=768 n_layers=16	`pretrain_768.pth`	`single_chat/full_sft_768.pth`	`multi_chat/full_sft_768.pth`	`rl_768.pth`

📌 Evaluation

① RLHF Section

Tip

Testing based on the comparison of the "single-turn dialogue full_sft" and "DPO reinforcement learning alignment" versions of the minimind model.

Model files are available on Baidu Netdisk, where rl_<dim>.pth represents the weights of the minimind model after "DPO reinforcement learning alignment."

[Q]: What is your name?
[ft_768]: I am an artificial intelligence language model, capable of answering your questions and providing assistance. Do you need any help?
[rl_768]: I am a natural language processing system based on artificial intelligence technology, primarily used for processing and generating human language tasks. I can assist users in text processing, speech recognition, text generation, and various other tasks. I can also help with intelligent chat, translation, and many other applications. If you need more information about this project, please let me know.
[ft_512]: I am glad to help you. Do you have any other questions you need me to answer?
[rl_512]: I am a chatbot based on natural language processing technology, capable of understanding and responding to user questions and instructions. I am trained to interact with humans and answer various questions and tasks. Is there anything you need?

[Q]: Who are you?
[ft_768]: I am an artificial intelligence language model, without personal identity or status. I only generate responses based on the input instructions and text information.
[rl_768]: I am a computer program, without personal identity or status. I am designed to answer various questions and provide assistance, including answering questions, offering advice, conducting experiments, providing information, and more. I am widely used in various fields such as natural language processing, computer vision, and artificial intelligence. If you need more help or information, please feel free to contact me. I will do my best to assist you in your work and life.
[ft_512]: I am a large language model, capable of answering a wide variety of questions, including those related to science, history, culture, and technology. Do you have any questions you need me to answer?
[rl_512]: I am a computer program, without a specific background or identity. If you can tell me the specific information you want to know, I will do my best to answer your questions.

[Q]: What are some good universities in China?
[ft_768]: China has many renowned universities, including Peking University, Shanghai Jiao Tong University, Fuxing Chinese Medicine College, Zhejiang University, and more. These universities have their own iconic buildings and research directions in different fields, and they enjoy a high reputation in China's top educational resources.
[rl_768]: There are many universities in China, and the choice depends on your needs. If you need more detailed information, it is recommended to consult the university rankings and policies in your city to choose a suitable university. You can also search for relevant information online or consult some university-related institutions such as Renmin University of China, Renmin University of China, etc., to obtain more detailed and professional guidance and advice.
[ft_512]: China has many excellent universities, such as MIT, Peking University, Shanghai University, and more. These universities have made significant contributions to the development and achievements of China's education.
[rl_512]: The Chinese Ministry of Education, Chinese universities, mainland Chinese universities, and mainland Chinese universities all have excellent educational systems, but the educational levels and professional requirements may vary. The Chinese Ministry of Education has different academic requirements, teaching methods, and faculty strengths, and the academic requirements of universities also vary. The academic requirements for mainland Chinese students also differ, and you need to choose based on your actual situation. If you have more specific needs for the professional requirements of mainland Chinese students, it is recommended to compare and communicate through the internet, various domestic and international institutions, and organizations. You can also seek help and advice from people around you, such as friends, teachers, and parents. I hope the above answer is helpful to you.

[Q]: Do you know what the speed of light is?
[ft_768]: The speed of light is a physical constant, representing the speed at which light travels in a vacuum, approximately 299,792,458 meters per second. The speed of light is one of the most fundamental physical constants in physics and is also an important basis for relativity.
[rl_768]: Of course, I know. The speed of light is the speed at which light travels in a vacuum, usually denoted by c. It is a very important theory in cosmology. In physics, the speed of light is considered one of the most important speeds in the universe. It represents the speed at which light travels in a vacuum, which is the basis for many important formulas and equations.
[ft_512]: The speed of light is the speed at which light travels in a vacuum, approximately 299,792,458 meters per second. It is one of the most mysterious and important physical concepts in the universe.
[rl_512]: The speed of light is the speed at which light travels in a vacuum, an irreversible speed in nature. The source of light consists of atoms, molecules, solids, and light, with a speed of approximately one light year per second. This number is calculated based on relativity, quantum mechanics, and gravitational theory. The speed of light is a relative concept, representing our understanding and interpretation of any event in nature.

👉Effect Summary

Approximately 100,000 RLHF data samples were used; the full_sft model performs better in terms of conciseness and information accuracy; the rl model provides more background information in its responses, but the accuracy of the information needs improvement.
Overall, the model after RLHF tends to learn: to say more polite but useless "fluff" to please the "conversation" itself, while slightly sacrificing information accuracy.
There is no such thing as a free lunch; we need to continue to improve the quality of the RLHF dataset, and we must also accept the inevitable loss of model capabilities (with varying degrees of severity).
The difference between DPO and online PPO is that reject and chosen are prepared offline, which inevitably creates a large distribution difference with the output of the minimind model itself.
This is similar to the DPO algorithm making the model watch the "replay" of the table tennis world champion's gameplay for reinforcement learning, rather than having the reward model act as a "coach" to correct its gameplay in real-time, like PPO.

② Instruct Fine-Tuning Section

Tip

The following tests were completed on September 17, 2024. New models released after this date will not be included in the tests unless there is a special need.

[A] minimind-v1-small(0.02B)
[B] minimind-v1-moe(0.1B)
[C] minimind-v1(0.1B)
[D] baby-llama2-chinese(0.2B)
[E] chatlm-mini-chinese(0.2B)

[Q]: Do you know the Yangtze River?
[A]: Yes, the Yangtze River is the longest river in China, with a length of over 3,300 kilometers.
[B]: Yes, the Yangtze River is the longest river in China, with a length of about 7,100 kilometers. Its drainage area is about 33.28 million square kilometers, making it one of China's largest rivers and the third longest river in the world, with a total length of about 6,300 kilometers.
[C]: Yes, the Yangtze River is one of the longest rivers in China and the third longest river in the world. It originates from the Tibetan Plateau, flows through 11 provinces, and eventually empties into the East China Sea. The Yangtze River Basin is one of China's largest river basins, with abundant natural resources and cultural heritage, such as the Yangtze River Bridge, Yangtze River Maritime Museum, and ancient cities along the Yangtze.
[D]: I also know that China is one of the longest cities in the world, with a population of over 34 million people.
[E]: Yes, the Yangtze River is the longest river in China and also one of the longest rivers in the world. It originates from the Tibetan Plateau, flows through nine provinces and regions, and eventually empties into the East China Sea. The climate in the Yangtze River Basin is mild, with distinct seasons, making it an important agricultural and industrial base in China. The river basin is rich in aquatic resources and is one of China's major fishing resources.

[Q]: Which country launched the first artificial satellite?
[A]: The first artificial satellite was Sputnik 1.
[B]: The first artificial satellite was launched by the United States.
[C]: The first artificial satellite was launched by the Soviet Union on October 4, 1957.
[D]: This satellite, named "Armstrong," was launched on November 20, 1961, and was the first successful satellite to send a human into space.
[E]: The first artificial satellite was launched by the Soviet Union on April 4, 1957. It was launched by Soviet astronaut Yuri Gagarin and Soviet engineer Roald Skohley. During the launch, Gagarin and Skohley used a small satellite named "Sputnik 1," which had a diameter of 4.3 meters and weighed 1,051 kilograms. This satellite, known as "Sputnik 1," marked a milestone in the Soviet space race and was the first instance of a human spacewalk in history.

[Q]: Do you know where Mount Tai is located?
[A]: Mount Tai is located in eastern China and is one of the Five Great Mountains of China.
[B]: Mount Tai is located in Tai'an City, Shandong Province, and is one of the Five Great Mountains of China.
[C]: Mount Tai is located in Tai'an City, Shandong Province, China, and is one of the Five Great Mountains of China.
[D]: As an AI assistant, I cannot access the internet, so I cannot provide information about your location. However, you can search for information or related keywords to find details about your location, or check news and websites for relevant information.
[E]: Mount Tai is located in Tai'an City, Shandong Province, China, and is one of the Five Great Mountains of China. It is also an important symbol of Chinese culture and history. With an elevation of 1,545 meters, it is one of the most famous tourist destinations in China. Its history dates back to the 5th century BC and was a site for ancient royal sacrifices. Mount Tai is a significant part of Chinese culture, known as the "First of the Five Great Mountains."

[Q]: What is the highest mountain in the world?
[A]: The highest mountain in the world is Mount Everest, located on the border between Nepal and China, with an elevation of 8,848 meters.
[B]: Mount Everest is the highest mountain in the world, located in the Himalayas, with an elevation of 8,848 meters.
[C]: Mount Everest is the highest mountain in the world, with an elevation of 8,848 meters.
[D]: The highest mountain in the world is Mount Everest, located in the Himalayas, with an elevation of 8,848.86 meters (29,031.69 feet).
[E]: The highest mountain in the world is Mount Everest, also known as Sagarmatha, located on the border between Nepal and China in the Himalayas, with an elevation of 8,848.86 meters (29,031.69 feet). It is one of the most famous climbing destinations, attracting climbers from around the world. The name Everest comes from Tibetan, meaning "The Goddess’s Peak," and it consists of two peaks, one in Nepal and one in China. Its height is determined through measurements of sea level, as it is the highest point on Earth. Although Everest is the highest mountain, it is not accessible from every place on Earth. Other places, such as the Andes Mountains in South America, are also high but do not have the highest peak. Overall, Everest is the highest mountain, but it is not accessible from every location on Earth.

[Q]: Who proposed the law of universal gravitation?
[A]: The law of universal gravitation was proposed by the British physicist Newton. He believed that all objects exert gravitational force, which depends on their mass and distance. This theory reveals the relationship between mass and distance and explains the nature of gravitation.
[B]: The law of universal gravitation was discovered by Newton and proposed by the British physicist James Watt in 1879.
[C]: The law of universal gravitation was proposed by the British physicist Isaac Newton in the 17th century.
[D]: Gravitation is a force in the universe composed of mass or energy. According to Newton's law of universal gravitation, the force between two objects is proportional to their mass and inversely proportional to their distance. Thus, Newton's law of universal gravitation shows that the gravitational force between two objects is proportional to their mass and inversely proportional to their distance. This law has wide applications in physics, astronomy, and other fields.
[E]: The law of universal gravitation was proposed by the British physicist Isaac Newton in 1687. In his famous paper "Philosophiæ Naturalis Principia Mathematica," Newton proposed the law of universal gravitation, stating that any two objects exert a gravitational force on each other, proportional to their masses and inversely proportional to the square of their distance. This law is known as the law of universal gravitation and describes the relationship between mass and distance. It is crucial for explaining many phenomena in the universe, such as planetary orbits and gravitational lensing between stars, and has broad applications in astronomy, physics, and engineering, providing a fundamental theoretical basis for understanding the nature and evolution of the universe.

[Q]: Do you know what the main food of giant pandas is?
[A]: The main food of giant pandas is bamboo. They typically eat bamboo and need to consume about 30 kilograms of it each day.
[B]: Giant pandas mainly eat bamboo, which is their primary food. Their digestive system is quite specialized, and they need to digest a large amount of bamboo daily to meet their survival needs. Besides bamboo, giant pandas also survive in the wild.
[C]: Giant pandas mainly eat bamboo but occasionally consume other plants and animals. They also eat some wild fruits, small animals, and plants, but these are not their main food sources.
[D]: The main food of giant pandas is bamboo.
[E]: Giant pandas are herbivores that primarily eat bamboo. Their digestive system is adapted to process the high cellulose content of bamboo, but they also consume some other plants, such as herbaceous plants and small mammals. Their digestive system includes enzymes like "cellulase" to break down cellulose, helping them digest a large amount of fiber and nutrients. Additionally, they have "cellulose protein" that helps maintain their energy levels, aiding their survival in the wild. Overall, bamboo is their primary food, but they also eat other plants and small mammals.

[Q]: Why is seawater salty?
[A]: Seawater contains many salts, which increase its salinity. The salts in seawater mainly come from the dissolved salts and form hydrogen bonds with water molecules, increasing the salinity.
[B]: Seawater contains abundant elements like calcium, phosphorus, potassium, and phosphorus, which are limited in quantity and distribution in the ocean. Additionally, seawater contains nutrients from various organisms and microorganisms, such as shells and seaweeds.
[C]: Seawater contains many dissolved salts, with the most common being sodium chloride. When salt is added to water, its molecules react with water molecules to form sodium hydroxide, which is a type of salt.
[D]: Water is essential for many organisms, as it can absorb and release carbon dioxide, release oxygen,
[E]: The primary reason seawater is salty is due to the high concentration of dissolved salts, with sodium chloride (NaCl) being the most significant. Sodium chloride is a very common salt that is present in most regions of the Earth, including oceans, lakes, rivers, and groundwater. When water evaporates from the ocean, it carries away some of the salt, making the seawater salty. Additionally, seawater contains other dissolved salts such as sulfates, magnesium sulfate, calcium sulfate, magnesium potassium sulfate, magnesium sodium sulfate, magnesium calcium sulfate, magnesium magnesium calcium sulfate, magnesium magnesium calcium sulfate, magnesium magnesium calcium sulfate, magnesium magnesium sodium sulfate, magnesium magnesium calcium sulfate, magnesium magnesium magnesium calcium sulfate, magnesium magnesium magnesium

Note

🙋‍♂️Directly throw the answer of the above model to GPT-4o and ask it to help score it:

Model Performance Review:

Model A:
- Performance: Model A's responses are usually concise and clear but lack detail and accuracy in some cases. For example, Model A provided incorrect information about the length of the Yangtze River.
- Score: 60
Model B:
- Performance: Model B provides additional information in some cases, but this information can sometimes be inaccurate or excessive. For instance, Model B gave incorrect figures for the length and drainage area of the Yangtze River.
- Score: 65
Model C:
- Performance: Model C typically provides detailed and accurate answers for most questions. For example, responses about the Yangtze River and Mount Tai were accurate.
- Score: 75
Model D:
- Performance: Model D’s responses sometimes appear disorganized and lack accuracy. For example, the answer about Mount Tai was completely off-topic.
- Score: 50
Model E:
- Performance: Model E’s responses are usually very detailed, but they can be overly verbose and contain unnecessary information. For instance, the answer on gravity was overly complex.
- Score: 70

Ranking (from highest to lowest):

Model	C	E	B	A	D
Score	75	70	65	60	50

👉 Summary of Effects

The ranking of the minimind series (ABC) aligns with intuition, with minimind-v1(0.1B) scoring the highest, and its responses to common sense questions are mostly error-free and free of hallucinations.
- Surprisingly, minimind-v1-small(0.02B), with only 26M parameters, can perform nearly as well as minimind-v1(0.1B).
- minimind-v1(0.1B) underwent less than 2 epochs of SFT (Supervised Fine-Tuning) due to being prematurely killed to free up resources for smaller models. Despite not being fully trained, it still achieved the best performance, demonstrating that larger models generally outperform smaller ones.
- minimind-v1-moe(0.1B) performed only slightly better than minimind-v1-small(0.02B), also due to early termination to free up resources for other training. However, the MoE (Mixture of Experts) model, with its sparse multi-Experts mode, requires more training epochs to fully activate and train all FFN (Feed-Forward Network) layer experts. In the current setup with 3 epochs, the training is not yet sufficient. Early experiments with minimind on the Yi-Tokenizer showed that a fully trained MoE version could outperform dense small models visibly. This aspect may need to be reserved for future training and updates to v2 and v3 versions when more server resources are available.
The responses from Model E appear to be quite good to the naked eye, although there are occasional instances of hallucinations and fabrications. However, both GPT-4o and Deepseek's evaluations consistently noted that it "provides overly verbose and repetitive information, and contains hallucinations." This evaluation seems somewhat strict, as even a small number of hallucinated words in a 100-word response can easily result in a low score. Given that Model E was pre-trained on longer texts and a larger dataset, its responses appear more comprehensive. In models of similar size, both the quantity and quality of the data are crucial.

🙋‍♂️ Personal Subjective Evaluation: E>C>B≈A>D

🤖 GPT-4o Evaluation: C>E>B>A>D

Scaling Law: Larger model parameters and more training data generally lead to better model performance.

📌 Objective Dataset: C-Eval

C-Eval evaluation code is located at: ./eval_ceval.py.

For small models, to avoid issues with fixed response formatting, we directly judge the prediction probabilities of the four tokens A, B, C, D, and choose the one with the highest probability as the answer, then calculate accuracy against the standard answer. Note that minimind models were not trained on larger datasets or fine-tuned for question answering, so results should be considered as reference only.

For example, detailed results for minimind-small:

Type	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32	33	34	35	36	37	38	39	40	41	42	43	44	45	46	47	48	49	50	51	52
Data	probability_and_statistics	law	middle_school_biology	high_school_chemistry	high_school_physics	legal_professional	high_school_chinese	high_school_history	tax_accountant	modern_chinese_history	middle_school_physics	middle_school_history	basic_medicine	operating_system	logic	electrical_engineer	civil_servant	chinese_language_and_literature	college_programming	accountant	plant_protection	middle_school_chemistry	metrology_engineer	veterinary_medicine	marxism	advanced_mathematics	high_school_mathematics	business_administration	mao_zedong_thought	ideological_and_moral_cultivation	college_economics	professional_tour_guide	environmental_impact_assessment_engineer	computer_architecture	urban_and_rural_planner	college_physics	middle_school_mathematics	high_school_politics	physician	college_chemistry	high_school_biology	high_school_geography	middle_school_politics	clinical_medicine	computer_network	sports_science	art_studies	teacher_qualification	discrete_mathematics	education_science	fire_engineer	middle_school_geography

Type	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32	33	34	35	36	37	38	39	40	41	42	43	44	45	46	47	48	49	50	51	52
T/A	3/18	5/24	4/21	7/19	5/19	2/23	4/19	6/20	10/49	4/23	4/19	4/22	1/19	3/19	4/22	7/37	11/47	5/23	10/37	9/49	7/22	4/20	3/24	6/23	5/19	5/19	4/18	8/33	8/24	5/19	17/55	10/29	7/31	6/21	11/46	5/19	3/19	4/19	13/49	3/24	5/19	4/19	6/21	6/22	2/19	2/19	14/33	12/44	6/16	7/29	9/31	1/12
Accuracy	16.67%	20.83%	19.05%	36.84%	26.32%	8.70%	21.05%	30.00%	20.41%	17.39%	21.05%	18.18%	5.26%	15.79%	18.18%	18.92%	23.40%	21.74%	27.03%	18.37%	31.82%	20.00%	12.50%	26.09%	26.32%	26.32%	22.22%	24.24%	33.33%	26.32%	30.91%	34.48%	22.58%	28.57%	23.91%	26.32%	15.79%	21.05%	26.53%	12.50%	26.32%	21.05%	28.57%	27.27%	10.53%	10.53%	42.42%	27.27%	37.50%	24.14%	29.03%	8.33%

Total number of questions: 1346

Total confirmed number: 316

Total accuracy rate: 23.48%

Results summary：

category	correct	question_count	accuracy
minimind-v1-small	344	1346	25.56%
minimind-v1	351	1346	26.08%

Model Performance Insights from GPT-4o

### Areas Where the Model Excels:
1. **High School Chemistry**: With an accuracy of 42.11%, this is the strongest area for the model, suggesting a solid grasp of chemistry-related knowledge.
2. **Discrete Mathematics**: Achieving an accuracy of 37.50%, the model performs well in mathematics-related fields.
3. **Education Science**: The model shows good performance in education-related topics with a 37.93% accuracy.
4. **Basic Medicine**: The accuracy of 36.84% indicates strong performance in foundational medical knowledge.
5. **Operating Systems**: With a 36.84% accuracy, the model demonstrates reliable performance in computer operating systems.

### Areas Where the Model Struggles:
1. **Legal Topics**: The model performs poorly in legal-related areas such as Legal Professional (8.70%) and Tax Accountant (20.41%).
2. **Physics**: Both high school (26.32%) and college-level (21.05%) physics topics are challenging for the model.
3. **High School Politics and Geography**: The model shows low accuracy in these areas, with High School Politics at 15.79% and High School Geography at 21.05%.
4. **Computer Networking and Architecture**: The model struggles with Computer Networking (21.05%) and Computer Architecture (9.52%).
5. **Environmental Impact Assessment Engineering**: The accuracy is only 12.90%, indicating weak performance in environmental science.

### Summary:
- **Strengths**: Chemistry, Mathematics (especially Discrete Mathematics), Education Science, Basic Medicine, and Operating Systems.
- **Weaknesses**: Legal Topics, Physics, Politics, Geography, Computer Networking and Architecture, and Environmental Science.

This suggests that the model performs well in logical reasoning, foundational sciences, and some engineering disciplines but is weaker in humanities, social sciences, and certain specialized fields (such as law and taxation). To improve the model's performance, additional training in humanities, physics, law, and environmental science may be beneficial.

📌 Others

Inference and Export

./export_model.py can export the model to the transformers format and push it to Hugging Face.
MiniMind's Hugging Face collection address: MiniMind

API Inference

./my_openai_api.py provides a chat interface for the OpenAI API, making it easier to integrate your model with third-party UIs, such as fastgpt, OpenWebUI, etc.

Download the model weight files from Hugging Face:

minimind (root dir)
├─minimind
|  ├── config.json
|  ├── generation_config.json
|  ├── LMConfig.py
|  ├── model.py
|  ├── pytorch_model.bin
|  ├── special_tokens_map.json
|  ├── tokenizer_config.json
|  ├── tokenizer.json

Start the chat server:
```
python my_openai_api.py
```
Test the service interface:
```
python chat_openai_api.py
```

API interface example, compatible with the OpenAI API format:

curl http://ip:port/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{ 
    "model": "model-identifier",
    "messages": [ 
      { "role": "user", "content": "What is the highest mountain in the world?" }
    ], 
    "temperature": 0.7, 
    "max_tokens": -1,
    "stream": true
}'

Integrating MiniMind API in FastGPT

📌 Acknowledgement

Tip

If you find MiniMind helpful, please give us a ⭐ on GitHub.
Given the length and the limitations of our expertise, there may be errors. We welcome discussions and corrections in the Issues section.
Your support is the driving force behind our continuous improvement of the project!

Note

An individual's resources, energy, and time are limited, so we encourage everyone to participate and contribute collectively. If you have trained model weights, you are welcome to share them in the Discussions or Issues sections.
These models can be new versions of MiniMind tailored for specific downstream tasks or vertical domains (such as sentiment recognition, healthcare, psychology, finance, legal Q&A, etc.).
They can also be new versions of MiniMind models that have undergone extended training, exploring longer text sequences, larger volumes (such as 0.1B+), or more extensive datasets.
Each contribution is unique, and all attempts are valuable and encouraged.
Any shared contributions will be promptly recognized and compiled in the acknowledgments list. Thank you once again for everyone's support!

🤝Contributors

😊Thanks for

@ipfgao: 🔗Training step record

@chuanzhubin: 🔗Code line by line comments (Chinese)

@WangRongsheng: 🔗Preprocessing of large datasets

@pengqianhan: 🔗A Concise Tutorial

@RyanSunn: 🔗Learning Record of Model Inference Process

Reference Links & Acknowledgments to the Following Excellent Papers or Projects

No specific order
https://github.com/meta-llama/llama3
https://github.com/karpathy/llama2.c
https://github.com/DLLXW/baby-llama2-chinese
(DeepSeek-V2)https://arxiv.org/abs/2405.04434
https://github.com/charent/ChatLM-mini-Chinese
https://github.com/wdndev/tiny-llm-zh
(Mistral-MoE)https://arxiv.org/pdf/2401.04088
https://github.com/Tongjilibo/build_MiniLLM_from_scratch
https://github.com/jzhang38/TinyLlama
https://github.com/AI-Study-Han/Zero-Chatgpt
https://github.com/xusenlinzy/api-for-open-llm
https://github.com/HqWu-HITCS/Awesome-Chinese-LLM

🫶Supporter

github contribution grid snake animation

License

This repository is licensed under the Apache-2.0 License.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README_en.md

README_en.md

"The Greatest Path is the Simplest"

📌 Introduction

👉Recent Updates

📌 Environment

📌 Quick Start Test

📌 Quick Start Train

📌 Data sources

Dataset Download Links

📌 Model

📌 Experiment

Trained Model Weights

📌 Evaluation

① RLHF Section

👉Effect Summary

② Instruct Fine-Tuning Section

Model Performance Review:

Ranking (from highest to lowest):

👉 Summary of Effects

📌 Objective Dataset: C-Eval

Results summary：

Model Performance Insights from GPT-4o

📌 Others

Inference and Export

API Inference

Integrating MiniMind API in FastGPT

📌 Acknowledgement

🤝Contributors

😊Thanks for

🫶Supporter

License

Files

README_en.md

Latest commit

History

README_en.md

File metadata and controls

"The Greatest Path is the Simplest"

📌 Introduction

👉Recent Updates

📌 Environment

📌 Quick Start Test

📌 Quick Start Train

📌 Data sources

Dataset Download Links

📌 Model

📌 Experiment

Trained Model Weights

📌 Evaluation

① RLHF Section

👉Effect Summary

② Instruct Fine-Tuning Section

Model Performance Review:

Ranking (from highest to lowest):

👉 Summary of Effects

📌 Objective Dataset: C-Eval

Results summary：

Model Performance Insights from GPT-4o

📌 Others

Inference and Export

API Inference

Integrating MiniMind API in FastGPT

📌 Acknowledgement

🤝Contributors

😊Thanks for

🫶Supporter

License