fix import error (#573)

README_zh.md

@@ -292,4 +292,3 @@ pipe.generate(
 <div align="center">
 由 LightX2V 团队用 ❤️ 构建
 </div>
-

lightx2v/models/input_encoders/hf/qwen25/qwen25_vlforconditionalgeneration.py

@@ -3,7 +3,12 @@ import math
 import os
 
 import torch
-from transformers import Qwen2Tokenizer, Qwen2_5_VLForConditionalGeneration
+
+try:
+    from transformers import Qwen2Tokenizer, Qwen2_5_VLForConditionalGeneration
+except ImportError:
+    Qwen2Tokenizer = None
+    Qwen2_5_VLForConditionalGeneration = None
 
 from lightx2v_platform.base.global_var import AI_DEVICE
 

lightx2v/models/input_encoders/hf/wan/t5/model.py

@@ -24,6 +24,7 @@ from lightx2v.models.input_encoders.hf.q_linear import (  # noqa E402
     SglQuantLinearFp8,  # noqa E402
     TorchaoQuantLinearInt8,  # noqa E402
     VllmQuantLinearInt8,  # noqa E402,
+    VllmQuantLinearFp8,  # noqa E402
 )
 from lightx2v_platform.ops.mm.cambricon_mlu.q_linear import MluQuantLinearInt8  # noqa E402
 from lightx2v.models.input_encoders.hf.wan.t5.tokenizer import HuggingfaceTokenizer  # noqa E402
@@ -195,6 +196,8 @@ class T5Attention(nn.Module):
                 linear_cls = VllmQuantLinearInt8
             elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
+            elif quant_scheme == "fp8-vllm":
+                linear_cls = VllmQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
             elif quant_scheme == "int8-q8f":
@@ -268,6 +271,8 @@ class T5FeedForward(nn.Module):
                 linear_cls = VllmQuantLinearInt8
             elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
+            elif quant_scheme == "fp8-vllm":
+                linear_cls = VllmQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
             elif quant_scheme == "int8-q8f":

lightx2v/models/input_encoders/hf/wan/xlm_roberta/model.py

@@ -10,7 +10,7 @@ from loguru import logger
 
 # from lightx2v.attentions import attention
 from lightx2v.common.ops.attn import TorchSDPAWeight
-from lightx2v.models.input_encoders.hf.q_linear import Q8FQuantLinearFp8, Q8FQuantLinearInt8, SglQuantLinearFp8, TorchaoQuantLinearInt8, VllmQuantLinearInt8
+from lightx2v.models.input_encoders.hf.q_linear import Q8FQuantLinearFp8, Q8FQuantLinearInt8, SglQuantLinearFp8, TorchaoQuantLinearInt8, VllmQuantLinearFp8, VllmQuantLinearInt8
 from lightx2v.utils.utils import load_weights
 from lightx2v_platform.base.global_var import AI_DEVICE
 from lightx2v_platform.ops.mm.cambricon_mlu.q_linear import MluQuantLinearInt8
@@ -65,6 +65,8 @@ class SelfAttention(nn.Module):
                 linear_cls = VllmQuantLinearInt8
             elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
+            elif quant_scheme == "fp8-vllm":
+                linear_cls = VllmQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
             elif quant_scheme == "int8-q8f":
@@ -147,6 +149,8 @@ class AttentionBlock(nn.Module):
                 linear_cls = VllmQuantLinearInt8
             elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
+            elif quant_scheme == "fp8-vllm":
+                linear_cls = VllmQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
             elif quant_scheme == "int8-q8f":

lightx2v/models/networks/wan/infer/matrix_game2/transformer_infer.py

@@ -8,10 +8,14 @@ try:
 
     FLASH_ATTN_3_AVAILABLE = True
 except ImportError:
+    try:
         from flash_attn import flash_attn_func
 
         FLASH_ATTN_3_AVAILABLE = False
 
+    except ImportError:
+        FLASH_ATTN_3_AVAILABLE = False
+
 
 from lightx2v.models.networks.wan.infer.matrix_game2.posemb_layers import apply_rotary_emb, get_nd_rotary_pos_embed
 from lightx2v.models.networks.wan.infer.self_forcing.transformer_infer import WanSFTransformerInfer, causal_rope_apply

lightx2v/pipeline.py

@@ -113,12 +113,14 @@ class LightX2VPipeline:
         boundary_step_index=2,
         denoising_step_list=[1000, 750, 500, 250],
         config_json=None,
+        rope_type="torch",
     ):
         if config_json is not None:
             self.set_infer_config_json(config_json)
         else:
             self.set_infer_config(
                 attn_mode,
+                rope_type,
                 infer_steps,
                 num_frames,
                 height,
@@ -142,6 +144,7 @@ class LightX2VPipeline:
     def set_infer_config(
         self,
         attn_mode,
+        rope_type,
         infer_steps,
         num_frames,
         height,
@@ -164,7 +167,7 @@ class LightX2VPipeline:
             self.enable_cfg = False
         else:
             self.enable_cfg = True
-
+        self.rope_type = rope_type
         self.fps = fps
         self.aspect_ratio = aspect_ratio
         self.boundary = boundary