pipeline_tag: 文本生成 inference: false license: apache-2.0 library_name: transformers model-index:

• name: ibm/PowerLM-3b results:
  • task: type: 文本生成 dataset: type: lm-eval-harness name: ARC metrics:
    • name: 标准化准确率 type: accuracy-norm value: 60.5 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: BoolQ metrics:
    • name: 准确率 type: accuracy value: 72.0 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: Hellaswag metrics:
    • name: 标准化准确率 type: accuracy-norm value: 74.6 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: OpenBookQA metrics:
    • name: 标准化准确率 type: accuracy-norm value: 43.6 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: PIQA metrics:
    • name: 标准化准确率 type: accuracy-norm value: 79.9 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: Winogrande metrics:
    • name: 标准化准确率 type: accuracy-norm value: 70.0 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: MMLU (5样本) metrics:
    • name: 准确率 type: accuracy value: 49.2 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: GSM8k (5样本) metrics:
    • name: 准确率 type: accuracy value: 34.9 verified: false
  • task: type: 文本生成 dataset: type: lm-eval-harness name: 数学 (4样本) metrics:
    • name: 准确率 type: accuracy value: 15.2 verified: false
  • task: type: 文本生成 dataset: type: bigcode-eval name: humaneval metrics:
    • name: pass@1 type: pass@1 value: 26.8 verified: false
  • task: type: 文本生成 dataset: type: bigcode-eval name: MBPP metrics:
    • name: pass@1 type: pass@1 value: 33.6 verified: false

模型概述

PowerLM-3B是一个采用Power学习率调度器训练的30亿参数先进小型语言模型。该模型基于开源和专有数据集的混合训练而成。在自然语言多选、代码生成和数学推理等多个基准测试中，PowerLM-3B相较于同规模级别的其他模型展现出显著优势。 论文链接：https://arxiv.org/abs/2408.13359

使用说明

注意：需从源码安装HF transformers库

生成示例

以下是使用PowerLM-3b模型的基础示例：

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # 或"cpu"
model_path = "ibm/PowerLM-3b"
tokenizer = AutoTokenizer.from_pretrained(model_path)
# 若在CPU运行请移除device_map参数
model = AutoModelForCausalLM.from_pretrained(model_path, device_map=device)
model.eval()
# 可自定义输入文本
prompt = "编写查找数字列表中最大值的代码。"
# 文本分词
input_tokens = tokenizer(prompt, return_tensors="pt")
# 将分词结果传输至设备
for i in input_tokens:
    input_tokens[i] = input_tokens[i].to(device)
# 生成输出token
output = model.generate(**input_tokens, max_new_tokens=100)
# 解码输出token为文本
output = tokenizer.batch_decode(output)
# 遍历批次输出（本例批次大小为1）
for i in output:
    print(i)