Enable KV cache and inline inference settings in examples

README.md

@@ -88,7 +88,8 @@ Text-to-image with the official Qwen model-card example prompt, BF16, batch size
 on one NVIDIA H100 80GB. Median of 3 requests after one warmup per configuration.
 Includes prompt encoding, denoising and decoding; excludes PNG saving, model
 loading and warmup. Base: 40 steps with KV cache on; P-Turbo: 5 or 8 steps with
-KV cache off, following the respective model-card settings. LoRAs were unmerged;
+KV cache off, as configured for this benchmark. The examples below enable KV caching for
+P-Turbo; the chart has not been remeasured with that setting. LoRAs were unmerged;
 no CFG, compilation or CPU offload. These timings do not imply equal image quality.
 
 ## Quickstart
@@ -128,13 +129,6 @@ pipe.scheduler = FlowMatchEulerDiscreteScheduler.from_config(
     shift=1.0,
     shift_terminal=None,
 )
-
-settings = dict(
-    num_inference_steps=STEPS,
-    sigmas=SIGMAS,
-    true_cfg_scale=1.0,
-    use_kv_cache=False,
-)
 ```
 
 ## For Text-to-Image
@@ -151,9 +145,11 @@ image = pipe(
     width=1024,
     height=1024,
     generator=torch.Generator("cuda").manual_seed(42),
-    **settings,
+    num_inference_steps=STEPS,
+    sigmas=SIGMAS,
+    true_cfg_scale=1.0,
+    use_kv_cache=True,
 ).images[0]
-image.save("t2i_example.png")
 ```
 
 ### For Image Editing
@@ -163,9 +159,11 @@ image = pipe(
     prompt="Change the background to a sunset beach",
     image=Image.open("input.png").convert("RGB"),
     generator=torch.Generator("cuda").manual_seed(42),
-    **settings,
+    num_inference_steps=STEPS,
+    sigmas=SIGMAS,
+    true_cfg_scale=1.0,
+    use_kv_cache=True,
 ).images[0]
-image.save("edit_example.png")
 ```
 
 ## Recommended settings