nemo_automodel.components.models.gpt_oss.model

Module Contents

Classes

Name	Description
`Block`	-
`GptOssForCausalLM`	-
`GptOssModel`	-

Data

ModelClass

logger

API

class nemo_automodel.components.models.gpt_oss.model.Block(
    layer_idx: int,
    config: transformers.models.gpt_oss.configuration_gpt_oss.GptOssConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig,
    backend: nemo_automodel.components.models.common.BackendConfig
)

Bases: Module

input_layernorm

mlp

= MoE(moe_config, backend)

post_attention_layernorm

self_attn

nemo_automodel.components.models.gpt_oss.model.Block._mlp(
    x: torch.Tensor,
    padding_mask: torch.Tensor | None
) -> torch.Tensor

nemo_automodel.components.models.gpt_oss.model.Block.forward(
    x: torch.Tensor,
    freqs_cis: torch.Tensor,
    attention_mask: torch.Tensor | None = None,
    padding_mask: torch.Tensor | None = None,
    attn_kwargs: typing.Any = {}
) -> tuple[torch.Tensor, torch.Tensor | None]

nemo_automodel.components.models.gpt_oss.model.Block.init_weights(
    buffer_device: torch.device
)

class nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM(
    config: transformers.models.gpt_oss.configuration_gpt_oss.GptOssConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig | None = None,
    backend: nemo_automodel.components.models.common.BackendConfig | None = None,
    kwargs = {}
)

Bases: HFCheckpointingMixin, Module, MoEFSDPSyncMixin

backend

= backend or BackendConfig(attn='flex')

lm_head

model

state_dict_adapter

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.forward(
    input_ids: torch.Tensor,
    position_ids: torch.Tensor | None = None,
    attention_mask: torch.Tensor | None = None,
    padding_mask: torch.Tensor | None = None,
    logits_to_keep: typing.Union[int, torch.Tensor] = 0,
    output_hidden_states: typing.Optional[bool] = None,
    attn_kwargs: typing.Any = {}
) -> transformers.modeling_outputs.CausalLMOutputWithPast

Forward pass returning :class:~transformers.modeling_outputs.CausalLMOutputWithPast.

Parameters:

input_ids

torch.Tensor

Token IDs. BSHD: [B, S]; THD: [1, T] (squeezed internally).

position_ids

torch.Tensor | NoneDefaults to None

Optional position indices.

attention_mask

torch.Tensor | NoneDefaults to None

Optional attention mask.

padding_mask

torch.Tensor | NoneDefaults to None

Optional padding mask.

logits_to_keep

Union[int, torch.Tensor]Defaults to 0

If 0 (default) compute logits for all positions; otherwise compute logits only for the last logits_to_keep positions (used by memory-efficient fused cross-entropy / generation).

output_hidden_states

Optional[bool]Defaults to None

When truthy, the returned output carries the final hidden states (the input to lm_head) so the recipe can run fused cross-entropy.

**attn_kwargs

AnyDefaults to {}

Additional attention kwargs forwarded to the base model (e.g. qkv_format, cu_seqlens, seq_idx, CP kwargs).

Returns: CausalLMOutputWithPast

CausalLMOutputWithPast with logits and optional hidden_states.

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.from_config(
    config: transformers.models.gpt_oss.configuration_gpt_oss.GptOssConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig | None = None,
    backend: nemo_automodel.components.models.common.BackendConfig | None = None,
    kwargs = {}
)

classmethod

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.from_pretrained(
    pretrained_model_name_or_path: str,
    model_args = (),
    kwargs = {}
)

classmethod

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.get_input_embeddings()

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.get_output_embeddings()

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.initialize_weights(
    buffer_device: torch.device | None = None,
    dtype: torch.dtype = torch.bfloat16
) -> None

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.set_input_embeddings(
    value
)

nemo_automodel.components.models.gpt_oss.model.GptOssForCausalLM.set_output_embeddings(
    new_embeddings
)

class nemo_automodel.components.models.gpt_oss.model.GptOssModel(
    config: transformers.models.gpt_oss.configuration_gpt_oss.GptOssConfig,
    backend: nemo_automodel.components.models.common.BackendConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig | None = None,
    moe_overrides: dict | None = None
)

Bases: Module

embed_tokens

head_dim

layers

= torch.nn.ModuleDict()

max_seq_len

= config.max_position_embeddings

moe_config

= moe_config or MoEConfig(**moe_defaults)

norm

rotary_emb

nemo_automodel.components.models.gpt_oss.model.GptOssModel.forward(
    input_ids: torch.Tensor,
    position_ids: torch.Tensor | None = None,
    attention_mask: torch.Tensor | None = None,
    padding_mask: torch.Tensor | None = None,
    attn_kwargs: typing.Any = {}
) -> torch.Tensor

nemo_automodel.components.models.gpt_oss.model.GptOssModel.init_weights(
    buffer_device: torch.device | None = None
) -> None

nemo_automodel.components.models.gpt_oss.model.ModelClass = GptOssForCausalLM

nemo_automodel.components.models.gpt_oss.model.logger = logging.getLogger(__name__)