[GGUF] Support non-standard quant types with prefix (e.g. UD-IQ1_S) (#39471)

Signed-off-by: Injae Ryou <injaeryou@gmail.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>

tests/transformers_utils/test_utils.py

@@ -81,6 +81,25 @@ class TestIsRemoteGGUF:
         assert not is_remote_gguf("repo/model:INVALID_M")
         assert not is_remote_gguf("repo/model:Q9_K_M")
 
+    def test_is_remote_gguf_nonstandard_quant_type(self):
+        """Test is_remote_gguf with non-standard quant types containing
+        a known GGML type."""
+        # Non-standard quant types with known GGML type after prefix
+        assert is_remote_gguf("unsloth/Qwen3.5-35B-A3B-GGUF:UD-Q4_K_XL")
+        assert is_remote_gguf("user/Model:UD-Q4_K_M")
+        assert is_remote_gguf("user/SomeModel:Custom-Q8_0")
+
+        # Exact GGML type after prefix (no suffix stripping needed)
+        assert is_remote_gguf("user/Model-GGUF:UD-IQ4_NL")
+        assert is_remote_gguf("user/Model-GGUF:UD-Q8_0")
+
+        # Completely unknown quant types should still fail
+        assert not is_remote_gguf("repo/model:TOTALLY-RANDOM")
+        assert not is_remote_gguf("user/Model:UD-INVALID")
+
+        # No dash separator → not recognized as prefixed
+        assert not is_remote_gguf("repo/model:UDIQ4NL")
+
     def test_is_remote_gguf_without_colon(self):
         """Test is_remote_gguf without colon."""
         assert not is_remote_gguf("repo/model")
@@ -143,6 +162,14 @@ class TestSplitRemoteGGUF:
         assert repo_id == "repo/model"
         assert quant_type == "Q3_K_S"
 
+    def test_split_remote_gguf_nonstandard_quant_type(self):
+        """Test split_remote_gguf with non-standard quant types in GGUF repos."""
+        repo_id, quant_type = split_remote_gguf(
+            "unsloth/Qwen3.5-35B-A3B-GGUF:UD-Q4_K_XL"
+        )
+        assert repo_id == "unsloth/Qwen3.5-35B-A3B-GGUF"
+        assert quant_type == "UD-Q4_K_XL"
+
     def test_split_remote_gguf_with_path_object(self):
         """Test split_remote_gguf with Path object."""
         repo_id, quant_type = split_remote_gguf(Path("unsloth/Qwen3-0.6B-GGUF:IQ1_S"))

vllm/transformers_utils/gguf_utils.py

@@ -40,15 +40,48 @@ def check_gguf_file(model: str | PathLike) -> bool:
 
 @cache
 def is_remote_gguf(model: str | Path) -> bool:
-    """Check if the model is a remote GGUF model."""
+    """Check if the model is a remote GGUF model.
+
+    Recognizes two forms:
+    1. Standard: ``repo_id:quant_type`` where *quant_type* is a known
+       GGML quantization type (e.g. ``Q4_K_M``).
+    2. Non-standard: ``repo_id:quant_type`` where *quant_type* contains
+       a known GGML type with extra prefixes (e.g. ``UD-Q4_K_XL``).
+       A warning is logged and actual file existence is validated later
+       during download.
+    """
     pattern = r"^[a-zA-Z0-9][a-zA-Z0-9._-]*/[a-zA-Z0-9][a-zA-Z0-9._-]*:[A-Za-z0-9_+-]+$"
     model = str(model)
     if re.fullmatch(pattern, model):
         _, quant_type = model.rsplit(":", 1)
-        return is_valid_gguf_quant_type(quant_type)
+        if is_valid_gguf_quant_type(quant_type):
+            return True
+        if is_nonstandard_gguf_quant_type(quant_type):
+            logger.warning(
+                "Non-standard GGUF quant type '%s' detected.",
+                quant_type,
+            )
+            return True
     return False
 
 
+def is_nonstandard_gguf_quant_type(quant_type: str) -> bool:
+    """Check if a non-standard quant type contains a known GGML type.
+
+    Splits the quant type by the last ``-`` and checks whether the
+    trailing part is a standard GGML type.  For example::
+
+        UD-Q4_K_XL      → rsplit → ["UD", "Q4_K_XL"]      → Q4_K_XL valid ✓
+        UD-IQ4_NL       → rsplit → ["UD", "IQ4_NL"]       → IQ4_NL  valid ✓
+        Custom-UD-Q4_K  → rsplit → ["Custom-UD", "Q4_K"]  → Q4_K    valid ✓
+        RANDOM          → no "-" → False
+    """
+    if "-" not in quant_type:
+        return False
+    _, remainder = quant_type.rsplit("-", 1)
+    return is_valid_gguf_quant_type(remainder)
+
+
 # Common suffixes used in GGUF file naming conventions
 # e.g., Q4_K_M, Q3_K_S, Q5_K_L, Q2_K_XL
 _GGUF_QUANT_SUFFIXES = ("_M", "_S", "_L", "_XL", "_XS", "_XXS")
@@ -84,7 +117,9 @@ def split_remote_gguf(model: str | Path) -> tuple[str, str]:
         f"Wrong GGUF model or invalid GGUF quant type: {model}.\n"
         "- It should be in repo_id:quant_type format.\n"
         f"- Valid base quant types: {GGMLQuantizationType._member_names_}\n"
-        f"- Extended suffixes also supported: {_GGUF_QUANT_SUFFIXES}",
+        f"- Extended suffixes also supported: {_GGUF_QUANT_SUFFIXES}\n"
+        "- Non-standard GGUF quant types also supported: "
+        "dash-separated prefixes (e.g. UD-Q4_K_XL, Custom-Q8_0)",
     )