From 76d699dc09c4f9996d50491ccba8e9a3ef110d1e Mon Sep 17 00:00:00 2001
From: Disty0 <disty@disty.xyz>
Date: Fri, 31 Oct 2025 00:21:54 +0300
Subject: [PATCH] SDNQ add common keys

---
 modules/sdnq/common.py    | 31 +++++++++++++++++++++++++++++--
 modules/sdnq/quantizer.py | 10 +++++++---
 2 files changed, 36 insertions(+), 5 deletions(-)

diff --git a/modules/sdnq/common.py b/modules/sdnq/common.py
index 2f8904fa0..e83af542b 100644
--- a/modules/sdnq/common.py
+++ b/modules/sdnq/common.py
@@ -82,6 +82,28 @@ else:
         return fn
 
 
+common_skip_keys = (
+    ".lm_head",
+    ".patch_embedding",
+    ".time_embed",
+    ".time_text_embed",
+    ".context_embedder",
+    ".condition_embedder",
+    ".x_embedder",
+    ".emb_in",
+    ".txt_in",
+    ".img_in",
+    ".vid_in",
+    ".proj_out",
+    ".norm_out",
+    ".emb_out",
+    ".txt_out",
+    ".img_out",
+    ".vid_out",
+    ".final_layer",
+    "pos_embed",
+)
+
 module_skip_keys_dict = {
     "FluxTransformer2DModel": [
         ["single_transformer_blocks.0.norm.linear.weight", ".time_text_embed", ".context_embedder", ".x_embedder", ".proj_out", ".norm_out", "pos_embed"],
@@ -99,12 +121,16 @@ module_skip_keys_dict = {
         ["scale_shift_table", ".rope", ".patch_embedding", ".condition_embedder", ".proj_out", ".norm_out", "pos_embed"],
         {}
     ],
+    "Gemma3nForCausalLM": [
+        [".lm_head", "correction_coefs", "prediction_coefs", "embedding_projection"],
+        {}
+    ],
     "HunyuanImage3ForCausalMM": [
-        [".patch_embed", ".time_embed", ".time_embed_2", ".final_layer", ".model.wte", ".model.ln_f", ".timestep_emb", ".vae", ".vision_aligner", ".vision_model.head", ".vision_model.post_layernorm", ".vision_model.embeddings", ".lm_head"],
+        [".lm_head", ".patch_embed", ".time_embed", ".time_embed_2", ".final_layer", ".model.wte", ".model.ln_f", ".timestep_emb", ".vae", ".vision_aligner", ".vision_model.head", ".vision_model.post_layernorm", ".vision_model.embeddings"],
         {}
     ],
     "Emu3ForCausalLM": [
-        [".lm_head", ".vq_model", ".tokenizer", ".model.embed_tokens", ".model.norm"],
+        [".lm_head", ".vq_model", ".tokenizer"],
         {}
     ],
     "NaDiT": [
@@ -114,4 +140,5 @@ module_skip_keys_dict = {
 }
 
 module_skip_keys_dict["ChronoEditTransformer3DModel"] = module_skip_keys_dict["WanTransformer3DModel"]
+module_skip_keys_dict["Gemma3nForConditionalGeneration"] = module_skip_keys_dict["Gemma3nForCausalLM"]
 module_skip_keys_dict["NaDiTUpscaler"] = module_skip_keys_dict["NaDiT"]
diff --git a/modules/sdnq/quantizer.py b/modules/sdnq/quantizer.py
index dec88e3a9..14ecfcc92 100644
--- a/modules/sdnq/quantizer.py
+++ b/modules/sdnq/quantizer.py
@@ -16,7 +16,7 @@ from accelerate import init_empty_weights
 from accelerate.utils import set_module_tensor_to_device
 
 from modules import devices, shared
-from .common import dtype_dict, module_skip_keys_dict, accepted_weights, use_tensorwise_fp8_matmul, allowed_types, conv_types, conv_transpose_types, use_contiguous_mm
+from .common import dtype_dict, common_skip_keys, module_skip_keys_dict, accepted_weights, use_tensorwise_fp8_matmul, allowed_types, conv_types, conv_transpose_types, use_contiguous_mm
 from .dequantizer import dequantizer_dict, dequantize_sdnq_model
 from .forward import get_forward_func
 
@@ -119,6 +119,8 @@ def add_module_skip_keys(model, modules_to_not_convert: List[str] = None, module
         modules_dtype_dict = {}
     if getattr(model, "_keep_in_fp32_modules", None) is not None:
         modules_to_not_convert.extend(model._keep_in_fp32_modules) # pylint: disable=protected-access
+    if getattr(model, "_tied_weights_keys", None) is not None:
+        modules_to_not_convert.extend(model._tied_weights_keys) # pylint: disable=protected-access
 
     skip_key_list = module_skip_keys_dict.get(model.__class__.__name__, None)
     if skip_key_list is not None:
@@ -128,8 +130,10 @@ def add_module_skip_keys(model, modules_to_not_convert: List[str] = None, module
                 modules_dtype_dict[key].extend(value)
             else:
                 modules_dtype_dict[key] = value
-    elif getattr(model, "_skip_layerwise_casting_patterns", None) is not None:
-        modules_to_not_convert.extend(model._skip_layerwise_casting_patterns) # pylint: disable=protected-access
+    else:
+        modules_to_not_convert.extend(common_skip_keys)
+        if getattr(model, "_skip_layerwise_casting_patterns", None) is not None:
+            modules_to_not_convert.extend(model._skip_layerwise_casting_patterns) # pylint: disable=protected-access
 
     # dedupe
     modules_to_not_convert = list(set(modules_to_not_convert))