Apply rotary embeddings to input tensors using the given frequency tensor. This function applies rotary embeddings to the given query 'xq' a

根据模型类型、是否 distill 和模型类别获取对应的 Hugging Face 仓库 ID

检查模块是否已安装

Patchify audio latents from [C, T, F] to [T, D].

Convert VAE decoder output to ComfyUI Image format (inplace operation) Args: vae_output: VAE decoder output tensor, typically in ran

Expose inlined Matrix-Game-3 helpers through a module-like interface.

This matches the implementation in Denoising Diffusion Probabilistic Models: Create sinusoidal timestep embeddings. Args timesteps (t

Lock the dictionary. When recursive=True, also recursively locks nested LockableDict objects.

Create a 2D convolution layer that can be either causal or non-causal. Args: in_channels: Number of input channels out_channe

Perform 2D max pooling on the input array. Convenience function for 2D max pooling, commonly used in computer vision and image processin

All-gather tensor along sequence dimension (dim=1).

q: [B, Lq, Nq, C1]. k: [B, Lk, Nk, C1]. v: [B, Lk, Nk, C2]. Nq must be divisible by Nk. q_lens

获取 LoRA 模型可选项，从 model_path/loras 目录检索 Args: model_path: 模型根路径 Returns: list: LoRA 文件列表，如果没有则返回 [""]

检查GPU是否支持fp8

Prepare position IDs for LongCat. Args: modality_id: 0 for text, 1 for image type: "text" or "image" start: Starting posi

Quantize input tensor to FP4 and return quantized tensor and scale. This function quantizes the last dimension of the given tensor `input`.