modeling_qwen3¶

Classes¶

`Qwen3PreTrainedModel`
`Qwen3Model`
`Qwen3ForCausalLM`
`Qwen3ForSequenceClassification`
`Qwen3ForTokenClassification`
`Qwen3ForQuestionAnswering`

Module Contents¶

class modeling_qwen3.Qwen3PreTrainedModel¶

Bases: transformers.modeling_utils.PreTrainedModel

config: transformers.models.qwen3.configuration_qwen3.Qwen3Config¶

base_model_prefix = 'model'¶

supports_gradient_checkpointing = True¶

class modeling_qwen3.Qwen3Model(config)¶

Bases: Qwen3PreTrainedModel

Parameters:: config (transformers.models.qwen3.configuration_qwen3.Qwen3Config)

padding_idx¶

vocab_size¶

embed_tokens¶

layers¶

norm¶

rotary_emb¶

gradient_checkpointing = False¶

has_sliding_layers¶

sin_embedding_input_qdq¶

cos_embedding_input_qdq¶

norm_input_qdq¶

convert_rope_for_deploy()¶

forward(input_ids=None, attention_mask=None, position_ids=None, past_key_values=None, inputs_embeds=None, use_cache=None, cache_position=None, **kwargs)¶

Parameters:

input_ids (Optional[torch.LongTensor])
attention_mask (Optional[torch.Tensor])
position_ids (Optional[torch.LongTensor])
past_key_values (Optional[transformers.cache_utils.Cache])
inputs_embeds (Optional[torch.FloatTensor])
use_cache (Optional[bool])
cache_position (Optional[torch.LongTensor])
kwargs (transformers.processing_utils.Unpack[transformers.utils.TransformersKwargs])

Return type:

transformers.modeling_outputs.BaseModelOutputWithPast

class modeling_qwen3.Qwen3ForCausalLM(config)¶

Bases: Qwen3PreTrainedModel, transformers.generation.GenerationMixin

model¶

vocab_size¶

lm_head¶

mllm_qualcomm_max_length = None¶

lm_head_input_qdq¶

lm_head_output_qdq¶

forward(input_ids=None, attention_mask=None, position_ids=None, past_key_values=None, inputs_embeds=None, labels=None, use_cache=None, cache_position=None, logits_to_keep=0, **kwargs)¶

labels (torch.LongTensor of shape (batch_size, sequence_length), optional):: Labels for computing the masked language modeling loss. Indices should either be in [0, …, config.vocab_size] or -100 (see input_ids docstring). Tokens with indices set to -100 are ignored (masked), the loss is only computed for the tokens with labels in [0, …, config.vocab_size].

Example:

```python >>> from transformers import AutoTokenizer, Qwen3ForCausalLM

>>> model = Qwen3ForCausalLM.from_pretrained("Qwen/Qwen3-8B")
>>> tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")

>>> prompt = "Hey, are you conscious? Can you talk to me?"
>>> inputs = tokenizer(prompt, return_tensors="pt")

>>> # Generate
>>> generate_ids = model.generate(inputs.input_ids, max_length=30)
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
"Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
```

Parameters:

input_ids (Optional[torch.LongTensor])
attention_mask (Optional[torch.Tensor])
position_ids (Optional[torch.LongTensor])
past_key_values (Optional[transformers.cache_utils.Cache])
inputs_embeds (Optional[torch.FloatTensor])
labels (Optional[torch.LongTensor])
use_cache (Optional[bool])
cache_position (Optional[torch.LongTensor])
logits_to_keep (Union[int, torch.Tensor])
kwargs (transformers.processing_utils.Unpack[transformers.utils.TransformersKwargs])

Return type:

transformers.modeling_outputs.CausalLMOutputWithPast

class modeling_qwen3.Qwen3ForSequenceClassification¶: Bases: transformers.modeling_layers.GenericForSequenceClassification, Qwen3PreTrainedModel

class modeling_qwen3.Qwen3ForTokenClassification¶: Bases: transformers.modeling_layers.GenericForTokenClassification, Qwen3PreTrainedModel

class modeling_qwen3.Qwen3ForQuestionAnswering¶

Bases: transformers.modeling_layers.GenericForQuestionAnswering, Qwen3PreTrainedModel

base_model_prefix = 'transformer'¶