add ideogram prequant and set default schedule

Signed-off-by: Vladimir Mandic <mandic00@live.com>

data/reference-quant.json

@@ -290,7 +290,7 @@
     "date": "2026 May"
   },
   "Ideogram 4 sdnq-hadamard-uint4": {
-    "path": "vladmandic/Ideogram-4-sdnq-uint4-hadamard",
+    "path": "Disty0/Ideogram-4-SDNQ-4bit-dynamic-hadamard",
     "desc": "Ideogram 4 is Ideogram's first open-weight text-to-image model: a two 9.3B flow-matching DiTs that uses a Qwen3-VL vision-language model as its text encoder, with strong in-image text rendering. Requires structured JSON-caption prompts; prompt-enhance (on by default) rewrites a plain prompt into one.",
     "skip": true,
     "extras": "sampler: Default, cfg_scale: 7.0, steps: 20, width: 1024, height: 1024",

modules/sd_hijack_schedulers.py

@@ -73,8 +73,7 @@ def _patch_scheduler_set_timesteps(scheduler_cls):
                 self,
                 timesteps,
             )
-            _assign_custom_schedule(self, num_inference_steps, timesteps_array, sigmas_array, device)
-            return
+            return _assign_custom_schedule(self, num_inference_steps, timesteps_array, sigmas_array, device)
 
         if sigmas is not None:
             if "sigmas" in set(inspect.signature(scheduler_cls.original_set_timesteps).parameters.keys()):
@@ -88,8 +87,9 @@ def _patch_scheduler_set_timesteps(scheduler_cls):
                     **kwargs,
                 )
             num_inference_steps, timesteps_array, sigmas_array = _prepare_custom_schedule_from_sigmas(self, sigmas, mu=mu)
-            _assign_custom_schedule(self, num_inference_steps, timesteps_array, sigmas_array, device)
-            return
+            return _assign_custom_schedule(self, num_inference_steps, timesteps_array, sigmas_array, device)
+
+        return None
 
     scheduler_cls.set_timesteps = set_timesteps
     _patched_schedulers.add(scheduler_cls)
@@ -152,7 +152,7 @@ def _compute_timesteps_from_sigmas(scheduler, sigmas_array: np.ndarray) -> np.nd
     base_sigmas = _get_base_sigmas(scheduler)
     log_sigmas = np.log(base_sigmas)
     if hasattr(scheduler, "_sigma_to_t"):
-        sigma_to_t = scheduler._sigma_to_t
+        sigma_to_t = scheduler._sigma_to_t # pylint: disable=protected-access
         parameters = list(inspect.signature(sigma_to_t).parameters)
         if len(parameters) == 2:
             results = []
@@ -260,8 +260,8 @@ def _assign_custom_schedule(scheduler, num_inference_steps, timesteps, sigmas, d
         scheduler.model_outputs = [None] * scheduler.config.solver_order
         scheduler.lower_order_nums = 0
 
-    scheduler._step_index = None
-    scheduler._begin_index = None
+    scheduler._step_index = None # pylint: disable=protected-access
+    scheduler._begin_index = None # pylint: disable=protected-access
     scheduler.sigmas = scheduler.sigmas.to("cpu")
 
 

pipelines/generic_shared.py

@@ -101,7 +101,7 @@ shared_te_map = {
     'Qwen3-VL 8B SDNQ-UInt4': {
         'cls': transformers.Qwen3VLModel,
         'identifier': 'uint4',
-        'target_repo': 'vladmandic/Ideogram-4-sdnq-uint4-hadamard',
+        'target_repo': 'Disty0/Ideogram-4-SDNQ-4bit-dynamic-hadamard',
         'target_subfolder': 'text_encoder',
     },
     'Qwen3-VL 8B Base': {

pipelines/ideogram/ideogram4.py

@@ -647,6 +647,8 @@ class Ideogram4Pipeline(DiffusionPipeline):
         # use the provided `guidance_schedule`, validated by `check_inputs`) and the tensor of per-step weights `gw`.
         if guidance_scale is not None:
             guidance_schedule = [float(guidance_scale)] * num_inference_steps
+        if guidance_schedule is None:
+            guidance_schedule = [7.0] * int(num_inference_steps * 0.9) + [3.0] * (num_inference_steps - int(num_inference_steps * 0.9)) # 90% of steps with 7.0, last 10% with 3.0
         gw = torch.as_tensor(guidance_schedule, dtype=torch.float32, device=device)
 
         # 6. Prepare latents in the packed (B, num_image_tokens, latent_dim) layout.