master

cann-robotHunyuanVideo sparse attention svg算法 sample_mse性能优化

3cf067a4创建于 2 天前历史提交

文件	最后提交记录	最后更新时间
blockwise_sparse	HunyuanVideo sparse attention svg算法 sample_mse性能优化	2 天前
dit_cache	[refactor] Unified inference script for multimodal models	1 个月前
fa_quant	[feat] HunyuanVideo support fa/matmul quant and uaa	2 个月前
quantization	[feat]swiglu group quant	3 天前
unified_sp	[feat] HunyuanVideo support fa/matmul quant and uaa	2 个月前
fuse_moe_gmm.py	[feat] step3p7_flash: 8 卡 / 16 rank NPU 推理适配 + 端到端优化（Decode ~5.6×）+ 图像输入	7 天前
linear.py	fix: shard VocabParallelEmbedding weight at init	23 天前
utils.py	【feat】add DeepSeek-V3.2-Exp model	8 个月前
vae_patch_parallel.py	[feat] support hunyuanImage-3.0 model inference	5 个月前