Skip to content

pos_cis已经改成freqs_cos和freqs_sin #477

@lbgitjp

Description

@lbgitjp

在很多train_*.py文件中都有下面的语句:

model._ddp_params_and_buffers_to_ignore = {"pos_cis"}

现在pos_cis已经改成了freqs_cos和freqs_sin.

freqs_cos, freqs_sin = precompute_freqs_cis(dim=config.hidden_size // config.num_attention_heads,
end=config.max_position_embeddings, theta=config.rope_theta)
self.register_buffer("freqs_cos", freqs_cos, persistent=False)
self.register_buffer("freqs_sin", freqs_sin, persistent=False)

所以train_*.py文件中的相应语句是不是也需要修改一下?
model._ddp_params_and_buffers_to_ignore = {"freqs_cos", "freqs_sin"}

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions