ReferenceFull Library ReferenceNemo AutomodelNemo AutomodelComponentsModelsGlm5 Next

nemo_automodel.components.models.glm5_next.model

View as Markdown

Native AutoModel implementation of GLM-5.3-Flash.

Module Contents

Classes

NameDescription
Glm5NextForConditionalGenerationTrainable GLM-5.3 VLM with EP and contiguous packed CP support.
Glm5NextModelCheckpoint-layout container for visual and language_model.
Glm5NextTextModelEmbedding, mHC decoder stack, mean stream collapse and final RMSNorm.

Functions

NameDescription
_packed_context_from_inputsBuild one global document map for a non-CP forward.
build_glm5_next_moe_configTranslate the GLM router/expert contract to AutoModel’s grouped MoE.

Data

ModelClass

__all__

API

class nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration(
    config: nemo_automodel.components.models.glm5_next.config.Glm5NextConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig | None = None,
    backend: nemo_automodel.components.models.common.BackendConfig | None = None,
    kwargs: typing.Any = {}
)

Bases: HFCheckpointingMixin, Module, MoEFSDPSyncMixin

Trainable GLM-5.3 VLM with EP and contiguous packed CP support.

_keep_in_fp32_modules_strict
_packed_cp_attn_backends
= ('sdpa', 'cudnn')
backend
language_model
Glm5NextTextModel

Expose the text module through the multimodal discovery protocol.

lm_head
model
state_dict_adapter
tie_word_embeddings_support
TieSupport = TieSupport.UNTIED_ONLY
vocab_size
= text_config.vocab_size
nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration._embed_and_splice(
    input_ids: torch.Tensor,
    pixel_values: torch.Tensor | None,
    image_grid_thw: torch.Tensor | None
) -> torch.Tensor

Embed the full sequence and replace image placeholder positions.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.forward(
    input_ids: torch.Tensor | None = None,
    attention_mask: torch.Tensor | None = None,
    position_ids: torch.Tensor | None = None,
    inputs_embeds: torch.Tensor | None = None,
    padding_mask: torch.Tensor | None = None,
    pixel_values: torch.Tensor | None = None,
    image_grid_thw: torch.Tensor | None = None,
    pixel_values_videos: torch.Tensor | None = None,
    logits_to_keep: int | torch.Tensor = 0,
    output_hidden_states: bool | None = None,
    kwargs: typing.Any = {}
) -> transformers.modeling_outputs.CausalLMOutputWithPast

Run image splice, contiguous CP slicing, text decoding and lm head.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.from_config(
    config: nemo_automodel.components.models.glm5_next.config.Glm5NextConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig | None = None,
    backend: nemo_automodel.components.models.common.BackendConfig | None = None,
    kwargs: typing.Any = {}
) -> 'Glm5NextForConditionalGeneration'
classmethod

Construct from an already resolved native config.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.from_pretrained(
    pretrained_model_name_or_path: str,
    model_args: typing.Any = (),
    kwargs: typing.Any = {}
) -> 'Glm5NextForConditionalGeneration'
classmethod

Resolve the local config; checkpoint loading is owned by AutoModel.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.get_image_features(
    pixel_values: torch.Tensor,
    image_grid_thw: torch.Tensor
) -> nemo_automodel.components.models.glm5_next.vision.Glm5NextVisionOutput

Return raw and merged features for flattened image patches.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.get_input_embeddings() -> torch.nn.Module

Return the token embedding table.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.get_output_embeddings() -> torch.nn.Module

Return the untied language-model head.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.initialize_weights(
    buffer_device: torch.device | None = None,
    dtype: torch.dtype = torch.bfloat16
) -> None

Initialize all tensors for checkpoint-free construction.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.prepare_model_inputs_for_cp(
    batch: dict[str, typing.Any],
    num_chunks: int = 1
) -> dict[str, typing.Any]

Install GLM’s contiguous packed sharder while leaving media and ids global.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.set_input_embeddings(
    value: torch.nn.Module
) -> None

Replace the token embedding table.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.set_output_embeddings(
    value: torch.nn.Module
) -> None

Replace the language-model head.

nemo_automodel.components.models.glm5_next.model.Glm5NextForConditionalGeneration.update_moe_gate_bias() -> None

Update no-aux-loss router correction biases after an optimizer step.

class nemo_automodel.components.models.glm5_next.model.Glm5NextModel(
    config: nemo_automodel.components.models.glm5_next.config.Glm5NextConfig,
    backend: nemo_automodel.components.models.common.BackendConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig | None = None,
    moe_overrides: dict[str, typing.Any] | None = None
)

Bases: Module

Checkpoint-layout container for visual and language_model.

language_model
visual
= Glm5NextVisionModel(config.vision_config)
nemo_automodel.components.models.glm5_next.model.Glm5NextModel.get_image_features(
    pixel_values: torch.Tensor,
    image_grid_thw: torch.Tensor
) -> nemo_automodel.components.models.glm5_next.vision.Glm5NextVisionOutput

Encode image patches and split-free concatenated features.

class nemo_automodel.components.models.glm5_next.model.Glm5NextTextModel(
    config: nemo_automodel.components.models.glm5_next.config.Glm5NextTextConfig,
    backend: nemo_automodel.components.models.common.BackendConfig,
    moe_config: nemo_automodel.components.moe.config.MoEConfig | None = None,
    moe_overrides: dict[str, typing.Any] | None = None
)

Bases: Module

Embedding, mHC decoder stack, mean stream collapse and final RMSNorm.

embed_tokens
layers
moe_config
norm
padding_idx
= config.pad_token_id
nemo_automodel.components.models.glm5_next.model.Glm5NextTextModel.forward(
    input_ids: torch.Tensor | None = None,
    inputs_embeds: torch.Tensor | None = None,
    glm5_next_packed_context: nemo_automodel.components.models.glm5_next.cp.Glm5NextPackedContext,
    padding_mask: torch.Tensor | None = None,
    kwargs: typing.Any = {}
) -> torch.Tensor

Run [batch, local_sequence] ids/embeddings through the text model.

nemo_automodel.components.models.glm5_next.model.Glm5NextTextModel.init_weights(
    buffer_device: torch.device
) -> None

Initialize a checkpoint-free text model on buffer_device.

nemo_automodel.components.models.glm5_next.model.Glm5NextTextModel.update_moe_gate_bias() -> None

Update every sparse layer’s no-aux-loss routing correction bias.

nemo_automodel.components.models.glm5_next.model._packed_context_from_inputs(
    input_ids: torch.Tensor,
    attention_mask: torch.Tensor | None,
    padding_mask: torch.Tensor | None,
    cu_seqlens: torch.Tensor | None,
    packed_seq_ids: torch.Tensor | None
) -> nemo_automodel.components.models.glm5_next.cp.Glm5NextPackedContext

Build one global document map for a non-CP forward.

nemo_automodel.components.models.glm5_next.model.build_glm5_next_moe_config(
    config: nemo_automodel.components.models.glm5_next.config.Glm5NextTextConfig,
    dtype: torch.dtype,
    overrides: dict[str, typing.Any] | None = None
) -> nemo_automodel.components.moe.config.MoEConfig

Translate the GLM router/expert contract to AutoModel’s grouped MoE.

nemo_automodel.components.models.glm5_next.model.ModelClass = Glm5NextForConditionalGeneration
 
nemo_automodel.components.models.glm5_next.model.__all__ = ['Glm5NextForConditionalGeneration', 'Glm5NextModel', 'Glm5NextTextModel', 'buil...