Rwkv 4 169m Pile

🚀 RWKV-4 | 基于Pile数据集训练的1.69亿参数模型卡片

RWKV是一个由Bo Peng领导的项目。你可以通过Johan Wind的博客文章这里和这里了解更多关于该模型架构的信息。你还可以通过加入RWKV Discord服务器来深入了解这个项目。

🚀 快速开始

模型简述

以下是来自原仓库的描述：

RWKV是一种具有Transformer级大语言模型性能的循环神经网络（RNN）。它可以像GPT一样直接进行训练（可并行化）。它结合了RNN和Transformer的优点——性能出色、推理速度快、节省显存、训练速度快、具有“无限”上下文长度，并且能免费获得句子嵌入。

✨ 主要特性

• 数据集：使用了EleutherAI/pile数据集进行训练。

📚 详细文档

模型细节

模型架构的详细信息可以在上述博客文章以及Hugging Face关于该模型集成的博客文章中找到。

模型使用

将原始权重转换为Hugging Face格式

你可以使用convert_rwkv_checkpoint_to_hf.py脚本，通过指定原始权重的仓库ID、文件名和输出目录来进行转换。你还可以选择通过传递--push_to_hub标志和--model_name参数，将转换后的模型直接推送到Hugging Face Hub上。

python convert_rwkv_checkpoint_to_hf.py --repo_id RAW_HUB_REPO --checkpoint_file RAW_FILE --output_dir OUTPUT_DIR --push_to_hub --model_name dummy_user/converted-rwkv

文本生成

你可以使用AutoModelForCausalLM和AutoTokenizer类从模型中生成文本。展开以下部分，了解如何在不同场景下运行该模型：

💻 使用示例

基础用法

在CPU上运行模型

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("RWKV/rwkv-4-169m-pile")
tokenizer = AutoTokenizer.from_pretrained("RWKV/rwkv-4-169m-pile")

prompt = "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese."

inputs = tokenizer(prompt, return_tensors="pt")
output = model.generate(inputs["input_ids"], max_new_tokens=40)
print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))

高级用法

在单个GPU上运行模型

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("RWKV/rwkv-4-169m-pile").to(0)
tokenizer = AutoTokenizer.from_pretrained("RWKV/rwkv-4-169m-pile")

prompt = "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese."

inputs = tokenizer(prompt, return_tensors="pt").to(0)
output = model.generate(inputs["input_ids"], max_new_tokens=40)
print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))

在GPU上以半精度运行模型

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("RWKV/rwkv-4-169m-pile", torch_dtype=torch.float16).to(0)
tokenizer = AutoTokenizer.from_pretrained("RWKV/rwkv-4-169m-pile")

prompt = "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese."

inputs = tokenizer(prompt, return_tensors="pt").to(0)
output = model.generate(inputs["input_ids"], max_new_tokens=40)
print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))

在多个GPU上运行模型

# pip install accelerate
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("RWKV/rwkv-4-169m-pile", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("RWKV/rwkv-4-169m-pile")

prompt = "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese."

inputs = tokenizer(prompt, return_tensors="pt").to(0)
output = model.generate(inputs["input_ids"], max_new_tokens=40)
print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))

📄 许可证

如果你使用此模型，请考虑引用原仓库此处的原始工作。