Qwen3Next compatibilty issue with flash-linear-attention

[Tcc0403]

System Info

• transformers version: 4.57.1
• Platform: Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.35
• Python version: 3.10.12
• Huggingface_hub version: 0.35.3
• Safetensors version: 0.6.2
• Accelerate version: not installed
• Accelerate config: not found
• DeepSpeed version: not installed
• PyTorch version (accelerator?): 2.7.1+cu126 (CUDA)
• Tensorflow version (GPU?): not installed (NA)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using distributed or parallel set-up in script?:
• Using GPU in script?:
• GPU type: NVIDIA GeForce RTX 3080

Who can help?

@ArthurZucker @Cyrilvallez

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

1. Install flash-linear-attention

2. Run the following code:

from transformers import AutoTokenizer
from transformers import Qwen3NextConfig
from transformers import Qwen3NextForCausalLM

config = Qwen3NextConfig(
    vocab_size=32000,
    hidden_size=896,
    intermediate_size=4864,
    num_hidden_layers=4,
    num_attention_heads=8,
    num_key_value_heads=2,
    hidden_act="silu",
    max_position_embeddings=32768,
    initializer_range=0.02,
    rms_norm_eps=1e-6,
    use_cache=True,
    tie_word_embeddings=False,
    rope_theta=10000.0,
    rope_scaling=None,
    attention_bias=False,
    use_sliding_window=False,
    sliding_window=4096,
    max_window_layers=28,
    attention_dropout=0.0,
    decoder_sparse_step=1,
    moe_intermediate_size=768,
    num_experts_per_tok=2,
    num_experts=8,
    norm_topk_prob=False,
    output_router_logits=False,
    router_aux_loss_coef=0.001,
    # dtype=torch.bfloat16,  <---- No issue if dtype is given
)
model = Qwen3NextForCausalLM(config)
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-Next-80B-A3B-Instruct")

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]

error output:

AttributeError: module 'torch' has no attribute 'get_current_dtype'

Expected behavior

transformers/src/transformers/models/qwen3_next/modeling_qwen3_next.py

Line 613 in 8cb5963

dtype=config.dtype if config.dtype is not None else torch.get_current_dtype(),

I guess it's meant to call torch.get_default_dtype()?

Also, setting config.dtype=torch.float32 won't work with flash-linear-attention either since fla's ChunkGatedDeltaRuleFunction does not support float32 (torch's does support it if fla is installed). Should we raise the error during model initialization instead?

AssertionError: ChunkGatedDeltaRuleFunction does not support float32. Please use bfloat16.

[Rocketknight1]

Yeah, this function doesn't exist! I think the fix at #41735 is good once the unrelated changes are removed