feat(api): add feature flag for single-tile panorama highres

api/onnx_web/chain/source_txt2img.py

@@ -79,11 +79,11 @@ class SourceTxt2ImgStage(BaseStage):
             latents = get_tile_latents(latents, int(params.seed), latent_size, dims)
 
         # reseed latents as needed
-        reseed_rng = np.random.default_rng(params.seed)
+        reseed_rng = np.random.RandomState(params.seed)
         prompt, reseed = parse_reseed(prompt)
         for top, left, bottom, right, region_seed in reseed:
             if region_seed == -1:
-                region_seed = reseed_rng.integers(2**32)
+                region_seed = reseed_rng.random_integers(2**32 - 1)
 
             logger.debug(
                 "reseed latent region: [:, :, %s:%s, %s:%s] with %s",

api/onnx_web/diffusers/run.py

@@ -62,7 +62,12 @@ def run_txt2img_pipeline(
     )
 
     # apply upscaling and correction, before highres
-    stage = StageParams(tile_size=params.unet_tile)
+    if params.is_panorama() and server.panorama_tiles:
+        highres_size = tile_size * highres.scale
+    else:
+        highres_size = params.unet_tile
+
+    stage = StageParams(tile_size=highres_size)
     first_upscale, after_upscale = split_upscale(upscale)
     if first_upscale:
         stage_upscale_correction(

api/onnx_web/diffusers/utils.py

@@ -233,9 +233,6 @@ def get_tokens_from_prompt(
     pattern: Pattern,
     parser=parse_float_group,
 ) -> Tuple[str, List[Tuple[str, float]]]:
-    """
-    TODO: replace with Arpeggio
-    """
     remaining_prompt = prompt
 
     tokens = []

api/onnx_web/server/context.py

@@ -10,13 +10,38 @@ from .model_cache import ModelCache
 
 logger = getLogger(__name__)
 
+DEFAULT_ANY_PLATFORM = True
 DEFAULT_CACHE_LIMIT = 5
 DEFAULT_JOB_LIMIT = 10
 DEFAULT_IMAGE_FORMAT = "png"
 DEFAULT_SERVER_VERSION = "v0.10.0"
+DEFAULT_SHOW_PROGRESS = True
+DEFAULT_PANORAMA_TILES = False
+DEFAULT_WORKER_RETRIES = 3
 
 
 class ServerContext:
+    bundle_path: str
+    model_path: str
+    output_path: str
+    params_path: str
+    cors_origin: str
+    any_platform: bool
+    block_platforms: List[str]
+    default_platform: str
+    image_format: str
+    cache_limit: int
+    cache_path: str
+    show_progress: bool
+    optimizations: List[str]
+    extra_models: List[str]
+    job_limit: int
+    memory_limit: int
+    admin_token: str
+    server_version: str
+    worker_retries: int
+    panorama_tiles: bool
+
     def __init__(
         self,
         bundle_path: str = ".",
@@ -24,20 +49,21 @@ class ServerContext:
         output_path: str = ".",
         params_path: str = ".",
         cors_origin: str = "*",
-        any_platform: bool = True,
+        any_platform: bool = DEFAULT_ANY_PLATFORM,
         block_platforms: Optional[List[str]] = None,
         default_platform: Optional[str] = None,
         image_format: str = DEFAULT_IMAGE_FORMAT,
         cache_limit: int = DEFAULT_CACHE_LIMIT,
         cache_path: Optional[str] = None,
-        show_progress: bool = True,
+        show_progress: bool = DEFAULT_SHOW_PROGRESS,
         optimizations: Optional[List[str]] = None,
         extra_models: Optional[List[str]] = None,
         job_limit: int = DEFAULT_JOB_LIMIT,
         memory_limit: Optional[int] = None,
         admin_token: Optional[str] = None,
         server_version: Optional[str] = DEFAULT_SERVER_VERSION,
-        worker_retries: Optional[int] = 3,
+        worker_retries: Optional[int] = DEFAULT_WORKER_RETRIES,
+        panorama_tiles: Optional[bool] = DEFAULT_PANORAMA_TILES,
     ) -> None:
         self.bundle_path = bundle_path
         self.model_path = model_path
@@ -58,6 +84,7 @@ class ServerContext:
         self.admin_token = admin_token or token_urlsafe()
         self.server_version = server_version
         self.worker_retries = worker_retries
+        self.panorama_tiles = panorama_tiles
 
         self.cache = ModelCache(self.cache_limit)
 
@@ -76,12 +103,16 @@ class ServerContext:
             params_path=environ.get("ONNX_WEB_PARAMS_PATH", "."),
             # others
             cors_origin=environ.get("ONNX_WEB_CORS_ORIGIN", "*").split(","),
-            any_platform=get_boolean(environ, "ONNX_WEB_ANY_PLATFORM", True),
+            any_platform=get_boolean(
+                environ, "ONNX_WEB_ANY_PLATFORM", DEFAULT_ANY_PLATFORM
+            ),
             block_platforms=environ.get("ONNX_WEB_BLOCK_PLATFORMS", "").split(","),
             default_platform=environ.get("ONNX_WEB_DEFAULT_PLATFORM", None),
             image_format=environ.get("ONNX_WEB_IMAGE_FORMAT", "png"),
             cache_limit=int(environ.get("ONNX_WEB_CACHE_MODELS", DEFAULT_CACHE_LIMIT)),
-            show_progress=get_boolean(environ, "ONNX_WEB_SHOW_PROGRESS", True),
+            show_progress=get_boolean(
+                environ, "ONNX_WEB_SHOW_PROGRESS", DEFAULT_SHOW_PROGRESS
+            ),
             optimizations=environ.get("ONNX_WEB_OPTIMIZATIONS", "").split(","),
             extra_models=environ.get("ONNX_WEB_EXTRA_MODELS", "").split(","),
             job_limit=int(environ.get("ONNX_WEB_JOB_LIMIT", DEFAULT_JOB_LIMIT)),
@@ -90,7 +121,12 @@ class ServerContext:
             server_version=environ.get(
                 "ONNX_WEB_SERVER_VERSION", DEFAULT_SERVER_VERSION
             ),
-            worker_retries=int(environ.get("ONNX_WEB_WORKER_RETRIES", 3)),
+            worker_retries=int(
+                environ.get("ONNX_WEB_WORKER_RETRIES", DEFAULT_WORKER_RETRIES)
+            ),
+            panorama_tiles=get_boolean(
+                environ, "ONNX_WEB_PANORAMA_TILES", DEFAULT_PANORAMA_TILES
+            ),
         )
 
     def torch_dtype(self):

onnx-web.code-workspace

@@ -35,6 +35,7 @@
 			"ddpm",
 			"deis",
 			"denoise",
+			"denoised",
 			"denoising",
 			"directml",
 			"Dreambooth",