Use the size label when detecting the model.

gguf-py splits a model name such as "Qwen3.8-27B-MyTune" into the basename
"Qwen3.8" and the size label "27B". The basename alone is then matched by
the shorter name "qwen3", the wrong converter is picked, and the run dies
with KeyError: 'blk.0.attn_gate.weight' unless -f names the model. The same
happens to a "Qwen3.5" basename of any size. Try "basename-size_label"
before the bare basename.

A renamed fine-tune ("MyModel-4B") matches no name at all and is rejected
as unsupported. Fall back to the GGUF architecture joined with the size
label, which the table already spells "qwen35-4B" for the Qwen3.5 sizes,
and add "qwen35-27B" for the 27B model. A renamed Qwen3.5-27B or
Qwen3.6-27B file takes that name too: the three share the architecture,
and their published configs give the same layers, sizes and head counts
(compared 2026-10-11). Under their own names those two are still claimed
by "qwen3", as before. A size label computed from the parameter count
("4.7B") matches nothing and still needs -f.

A file without a basename no longer dies with an AttributeError: it goes
through the same fallback, and to the intended "Unsupported model
architecture" error when that finds nothing.

Not upstream as of 2026-10-11.

--- a/q4nx/constants.py
+++ b/q4nx/constants.py
@@ -33,7 +33,11 @@
     ModelArch.QWEN35_OMNI: ["qwen35-omni","qwen3.5-omni"],
     ModelArch.QWEN35_A3B: ["qwen35moe"],
     # Listed before QWEN3: names are matched by prefix and "qwen3" would claim it.
-    ModelArch.QWEN38_27B: ["qwen38-27B", "qwen3.8-27B"],
+    # "qwen35-27B" is the GGUF architecture plus the size label, which is all a
+    # renamed fine-tune keeps. A renamed Qwen3.5-27B or Qwen3.6-27B file matches
+    # it too: their published configs give the same layers, sizes and head
+    # counts (compared 2026-10-11).
+    ModelArch.QWEN38_27B: ["qwen38-27B", "qwen3.8-27B", "qwen35-27B"],
     ModelArch.QWEN3VL: ["qwen3vl", "qwen3-vl"],
     ModelArch.QWEN2VL: ["qwen2.5-Vl"],    
     ModelArch.QWEN2:   ["qwen2"],
--- a/q4nx/model_converter.py
+++ b/q4nx/model_converter.py
@@ -1121,11 +1121,14 @@
     # The architecture is typically stored in the 'general.architecture' field
     arch_str:str|None = None
     basename_str:str|None = None
+    size_label_str:str|None = None
     for field in reader.fields.values():
         if field.name == 'general.architecture':
             arch_str = str(field.parts[field.data[0]], encoding='utf-8') if field.data else None
         elif field.name == 'general.basename':
             basename_str = str(field.parts[field.data[0]], encoding='utf-8') if field.data else None
+        elif field.name == 'general.size_label':
+            size_label_str = str(field.parts[field.data[0]], encoding='utf-8') if field.data else None
     
 
     # Map the architecture string to ModelArch enum
@@ -1135,10 +1138,31 @@
                 return arch_enum
     
 
-    for arch_enum, arch_names in ModelArchNames.items():
-        for arch_name in arch_names:
-            if basename_str.lower().startswith(arch_name.lower()):
-                return arch_enum
+    # gguf-py splits a model name such as "Qwen3.8-27B-MyTune" into the basename
+    # "Qwen3.8" and the size label "27B". Match the two joined first: the bare
+    # basename would be claimed by a shorter name such as "qwen3".
+    basenames = []
+    if basename_str and size_label_str:
+        basenames.append(f"{basename_str}-{size_label_str}")
+    if basename_str:
+        basenames.append(basename_str)
+
+    for basename in basenames:
+        for arch_enum, arch_names in ModelArchNames.items():
+            for arch_name in arch_names:
+                if basename.lower().startswith(arch_name.lower()):
+                    return arch_enum
+
+    # A renamed fine-tune ("MyModel-4B") keeps only the architecture and the
+    # size label of its base. Accept the names spelled "<architecture>-<size>",
+    # such as "qwen35-4B", and no shorter one: "gemma4-assistant" is not "gemma4".
+    if size_label_str:
+        arch_prefix = f"{arch_str}-".lower()
+        arch_size = f"{arch_str}-{size_label_str}".lower()
+        for arch_enum, arch_names in ModelArchNames.items():
+            for arch_name in arch_names:
+                if arch_name.lower().startswith(arch_prefix) and arch_size.startswith(arch_name.lower()):
+                    return arch_enum
 
 
     raise ValueError(f"Unsupported model architecture: {arch_str}")