nemo_automodel.components.models.gpt_oss.layers

Module Contents

Classes

Name	Description
`GptOssAttention`	-

API

class nemo_automodel.components.models.gpt_oss.layers.GptOssAttention(
    config: transformers.models.gpt_oss.configuration_gpt_oss.GptOssConfig,
    backend: nemo_automodel.components.models.common.BackendConfig,
    use_sliding_attention: bool = False
)

Bases: Module

head_dim

= config.head_dim

hidden_size

= config.hidden_size

k_proj

num_attention_heads

= config.num_attention_heads

num_key_value_heads

= config.num_key_value_heads

o_proj

q_proj

sinks

sliding_window

softmax_scale

= self.head_dim ** -0.5

v_proj

yarn_concentration

= yarn_get_mscale(config.rope_scaling['factor'])

nemo_automodel.components.models.gpt_oss.layers.GptOssAttention.forward(
    x: torch.Tensor,
    freqs_cis: torch.Tensor,
    attention_mask: torch.Tensor | None = None,
    attn_kwargs: typing.Any = {}
) -> torch.Tensor

nemo_automodel.components.models.gpt_oss.layers.GptOssAttention.init_weights(
    buffer_device: torch.device,
    init_std: float = 0.02
)