[diffusion] fix: avoid illegal memory access in qwen image (#22953)

2026-04-23 12:41:26 +08:00
parent 4d3c7e781a
commit 9716599383
1 changed files with 12 additions and 0 deletions
@@ -247,6 +247,18 @@ class QwenImagePipelineConfig(QwenImageRolloutPipelineMixin, ImagePipelineConfig
        # img_shapes: for global entire image
        img_freqs, txt_freqs = rotary_emb(img_shapes, txt_seq_lens, device=device)
        max_txt_seq_len = max(txt_seq_lens) if txt_seq_lens else 0
        txt_cache_len = int(txt_freqs.shape[0])
        if max_txt_seq_len > txt_cache_len:
            overflow = max_txt_seq_len - txt_cache_len
            raise ValueError(
                "QwenImage RoPE text cache overflow before denoising: "
                f"required_txt_seq_len={max_txt_seq_len}, txt_cache_len={txt_cache_len}, "
                f"overflow={overflow}. "
                "Please reduce the number of input images, shorten the prompt, "
                "or lower the requested resolution."
            )
        # flashinfer RoPE expects a float32 cos/sin cache concatenated on the last dim
        img_cos_half = img_freqs.real.to(dtype=torch.float32).contiguous()
        img_sin_half = img_freqs.imag.to(dtype=torch.float32).contiguous()