Fixed the bug in flow matching (#68)

* fix a bug in flow matching

* fixed the bugs in flow matching

* fixed format

---------

Co-authored-by: Geoffrey19 <sympathischmann35@gmail.com>

wall_x/model/qwen2_5_based/modeling_qwen2_5_vl_act.py

@@ -2046,7 +2046,7 @@ class Qwen2_5_VLMoEForAction(Qwen2_5_VLForConditionalGeneration):
             times = torch.linspace(
                 0,
                 1,
-                num_inference_timesteps,
+                num_inference_timesteps + 1,
                 device=inputs_embeds.device,
                 dtype=inputs_embeds.dtype,
             )