README.md

---
license: mit
language: en
tags:
- LLM
- XVERSE-13B-Chat
---
## Model Card for lyraXVERSE

lyraXVERSE is currently the **fastest XVERSE-13b** available. The inference speed of lyraXVERSE has achieved up to **3900+ tokens/s** on A100, up to **2.7x** acceleration upon the torch version.

Among its main features are:
- device: Nvidia GPU with Amperer architecture or Volta architecture (A100 or higher, V100).
- batch_size: compiled with dynamic batch size, maximum depends on device. 
- MEMOPT mode: significantly optimized VRAM usage and increased speed

We use the XVERSE-13B-Chat model for measurement, but this optimized inference is also applicable to XVERSE-13B model.

## Speed

* Evaluated at tokens/s
* test on A100 40G
* MEMOPT mode

### XVERSE-13B-Chat

| Version | Batch Size 1 | Batch Size 8 | Batch Size 16 | Batch Size 32 | Batch Size 64 |
| --- | --- | --- | --- | --- | --- |
| Torch | 34.8 | 249.2 | 470.1 | 878.6 | 1478.9 |
| lyraXVERSE MEMOPT | 96.6 | 725.5 | 1359.3 | 2415.6 | 3923.2 |

## Docker Environment Recommendation

- For Cuda 11.X: we recommend ```nvcr.io/nvidia/pytorch:22.12-py3```
- For Cuda 12.0: we recommend ```nvcr.io/nvidia/pytorch:23.02-py3```

```bash
docker pull nvcr.io/nvidia/pytorch:23.02-py3
docker run --rm -it --gpus all -v ./:/lyraXVERSE nvcr.io/nvidia/pytorch:23.02-py3

pip install -r requirements.txt
python demo.py
```

## Uses

```python
from lyra_xverse import lyraXVERSE

model_path = "./models/"
tokenizer_path = "./models/"
inference_dtype = 'fp16'
prompt = "讲个故事:"
memopt_mode = 1
max_output_length = 512
arch = "Ampere" # Ampere or Volta
cuda_version = 12 # cuda version, we currently support 11 and 12

model = lyraXVERSE(model_path, 
                   tokenizer_path = tokenizer_path, 
                   dtype = inference_dtype,
                   memopt_mode = memopt_mode,
                   arch = arch,
                   cuda_version = cuda_version)

bs = 1
prompts = [prompt, ] * bs
output_texts = model.generate(
        prompts, output_length=max_output_length,
        top_k=30, top_p=0.85, temperature=1.0, repetition_penalty=1.0, do_sample=False)

print(output_texts)

```

## Demo Outputs

### XVERSE-13B-Chat
#### input

讲个故事:

#### output

有一天,一位年轻的画家来到了一个偏远的村庄。他以其超凡的绘画技巧,为村民画了一幅美丽的图画。图画里,村庄的周围是翠绿的森林,清澈的溪流在其中流淌,村民们正在劳作,孩子们在田野里嬉戏。村民们看着这幅画,都对这位画家赞不绝口。

村庄的领袖看到了这幅画,他想:“这幅画将会让我们的村庄更加美丽,我们应该让村民们知道这幅画。”于是,他带着画家去村庄的各个角落,让每一个村民都看到了这幅画。

画家看着村民们看画的眼神,他意识到了自己的价值。他意识到,他不仅仅是一个画家,他也是一个能让人们看见希望的人。他的画不仅仅是艺术品,它是连接人们与希望的一座桥梁。

这个故事告诉我们,画家的价值不只是他们的绘画技巧,而是他们的画作带给人们的感动和希望。画家的价值并不在于他们的画有多么昂贵,有多么独特,而在于他们能用画作打开人们的心扉,让人们看见希望,看见生活的美好。

## TODO
1. Support for int4
2. Inference for longer context situations
3. Streaming inference mode.

## Citation
``` bibtex
@Misc{lyraXVERSE2023,
  author =       {Haoxiong Su, Kangjian Wu, Zhengtao Wang, Yibo Lu, Bin Wu},
  title =        {lyraXVERSE: Accelerating XVERSE-13B-Chat(fp16) to 3000+ tokens/s},
  howpublished = {\url{https://huggingface.co/TMElyralab/lyraXVERSE}},
  year =         {2023}
}
```

## Report bug
- start a discussion to report any bugs!--> https://huggingface.co/TMElyralab/lyraXVERSE
- report bug with a `[bug]` mark in the title.