Skip to content

some bugs when enable LLama2-7b with PaddleNLP #2627

@fmiao2372

Description

@fmiao2372
  1. 分布式环境的初始化
    Image
  2. AutoTokenizer的选择和get_pad_id的返回
    Image
  3. worker最好添加一个参数传给PredictorArgument,同时load完模型后初始化kv cache
    Image

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions