llm360 / amber-train

Pre-training code for Amber 7B LLM

Amber: the first model of LLM360

---

🤗 [Amber Download] • 🤗 [AmberChat Download] • 📈 [Analysis and Results] • 📗 Pretraining Dataset

About LLM360

LLM360 is an initiative for comprehensive and fully open-sourced LLMs, where all training details, model checkpoints, intermediate results, and additional analyses are made available to the community. Our goal is to advance the field by inviting the community to deepen the understanding of LLMs together. As the first step of the project LLM360, we release all intermediate model checkpoints, our fully-prepared pre-training dataset, all source code and configurations, and training details. We are committed to continually pushing the boundaries of LLMs through this open-source effort.

Get access now at LLM360 site

Model Description

Amber is the first model in the LLM360 family. Amber is an 7B English language model with the LLaMA architecture.

• Model type: Language model with the same architecture as LLaMA-7B
• Language(s) (NLP): English
• License: Apache 2.0
• Resources for more information:
  • Training Code
  • Data Preparation
  • Metrics
  • Fully processed Amber pretraining data

Loading Amber

from transformers import LlamaTokenizer, LlamaForCausalLM

tokenizer = LlamaTokenizer.from_pretrained("LLM360/Amber", revision="ckpt_356")
model = LlamaForCausalLM.from_pretrained("LLM360/Amber", revision="ckpt_356")

input_text = "translate English to German: How old are you?"
input_ids = tokenizer(input_text, return_tensors="pt").input_ids

outputs = model.generate(input_ids)
print(tokenizer.decode(outputs[0]))

Amber Training Details

DataMix

Subset	Tokens (Billion)
Arxiv	30.00
Book	28.86
C4	197.67
Refined-Web	665.01
StarCoder	291.92
StackExchange	21.75
Wikipedia	23.90
Total	1259.13

Hyperparameters

Hyperparameter	Value
Total Parameters	6.7B
Hidden Size	4096
Intermediate Size (MLPs)	11008
Number of Attention Heads	32
Number of Hidden Layers	32
RMSNorm ɛ	1e^-6
Max Seq Length	2048
Vocab Size	32000

Training Loss

Evaluation

ARC	HellaSwag

MMLU	TruthfulQA

Citation

BibTeX:

@misc{liu2023llm360,
      title={LLM360: Towards Fully Transparent Open-Source LLMs}, 
      author={Zhengzhong Liu and Aurick Qiao and Willie Neiswanger and Hongyi Wang and Bowen Tan and Tianhua Tao and Junbo Li and Yuqi Wang and Suqi Sun and Omkar Pangarkar and Richard Fan and Yi Gu and Victor Miller and Yonghao Zhuang and Guowei He and Haonan Li and Fajri Koto and Liping Tang and Nikhil Ranjan and Zhiqiang Shen and Xuguang Ren and Roberto Iriondo and Cun Mu and Zhiting Hu and Mark Schulze and Preslav Nakov and Tim Baldwin and Eric P. Xing},
      year={2023},
      eprint={2312.06550},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

amber-train's Issues

Aligning logits with labels through two shifts？

在 main.py中数据准备时：

def collate_fn(examples, device):
    token_ids = torch.tensor(
        [example['token_ids'] for example in examples], device=device)
    return **{'input_ids': token_ids[:, :-1], 'labels': token_ids[:, 1:]}**

def train_chunk(.......):
..........
batch = collate_fn(
            examples=examples[i:i+per_device_batch_size], device=fabric.device)
input_ids, labels = batch['input_ids'], batch['labels']

在 modeling_llama.py 中loss计算时：

class LlamaForCausalLM(LlamaPreTrainedModel):
....................
        if labels is not None:
            # Shift so that tokens < n predict n
            **shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()**

为什么在模型数据sample输入时进行了预测和真实值之间的位移对齐，在模型中loss计算时还进行了一次位移对齐？

Where did you train the model?

Did you use Deepspeed to train the model?

Please share requirements of the project

Thanks for this great work and open sourcing it.

I'm trying to execute the training code but due to the lack of requirements, I failed. Could you share the requirements needed for the project? By requirement I mean the version of Python library packages, cuda version, etc.

Thanks,

Request for AmberChat and AmberSafe Implementation Details

Truly grateful for your exceptional work in the open source community. Could you possibly share the implementation details of AmberChat and AmberSafe?"

Question about pretraining.

I attached my training loss below， the data we are using refers to LLM360's paper, we use less data starcode.
For each training epoch our data contains 30B arxiv , Book 57B, C4 197.67B, Refined-Web 665.01, StarCoder 150B, StackExchange 21.75B, Wikipedia 23.90B.
And the hyperparameter we are using the same as LLM360 demonstrated. And the max_seq_len is 4096 instead of 2048, tokenizer is gpt tokenzier.
We are using an opensource repo to run the experiment on H100 Node with 2048 global bsize.
Currently our model can only achieve around 10.5 PPL on the falcon dataset. which is much worse than LLM360 amber model (Around 8 PPL) and llama-2 (Around 8 PPL).
Just wondering what would be the possible reason that our model perform much worse?