Merge f5028e058b into 3212710a4a

studio/backend/hub/tests/test_model_services.py

@@ -85,6 +85,11 @@ class TestExtractQuantToken:
     def test_ud_prefix_preserved(self):
         assert gguf.extract_quant_token("Foo-BF16-UD-Q4_K_XL.gguf") == "UD-Q4_K_XL"
 
+    def test_packed_and_grouped_quant_variants_do_not_collapse(self):
+        assert gguf.extract_quant_token("Ternary-Bonsai-1.7B-PQ2_0.gguf") == "PQ2_0"
+        assert gguf.extract_quant_token("Ternary-Bonsai-1.7B-Q2_0.gguf") == "Q2_0"
+        assert gguf.extract_quant_token("Ternary-Bonsai-1.7B-Q2_0_g64.gguf") == "Q2_0_g64"
+
     def test_precision_infix_variants_do_not_collapse(self):
         labels = {
             gguf.extract_quant_label("Foo-BF16-Q4_K_M.gguf"),
@@ -1086,6 +1091,21 @@ def test_gguf_variant_requirements_include_split_files_and_preferred_mmproj():
     )
 
 
+def test_gguf_variant_requirements_keep_packed_q2_files_separate():
+    requirements = gguf_variants._build_gguf_variant_requirements(
+        [
+            _sibling("Ternary-Bonsai-1.7B-PQ2_0.gguf", 10, "pq"),
+            _sibling("Ternary-Bonsai-1.7B-Q2_0.gguf", 20, "q2"),
+            _sibling("Ternary-Bonsai-1.7B-Q2_0_g64.gguf", 30, "q2g64"),
+        ]
+    )
+
+    assert set(requirements) == {"pq2_0", "q2_0", "q2_0_g64"}
+    assert requirements["pq2_0"].target_filenames == ("Ternary-Bonsai-1.7B-PQ2_0.gguf",)
+    assert requirements["q2_0"].target_filenames == ("Ternary-Bonsai-1.7B-Q2_0.gguf",)
+    assert requirements["q2_0_g64"].target_filenames == ("Ternary-Bonsai-1.7B-Q2_0_g64.gguf",)
+
+
 def test_gguf_variant_requirements_skip_big_endian_sibling():
     requirements = gguf_variants._build_gguf_variant_requirements(
         [

studio/backend/hub/utils/gguf.py

@@ -72,7 +72,7 @@ _GGUF_QUANT_RE = re.compile(
     r"|IQ[0-9]+_[A-Z]+(?:_[A-Z0-9]+)?"
     r"|TQ[0-9]+_[0-9]+"
     r"|Q[0-9]+_K_[A-Z]+"
-    r"|Q[0-9]+_[0-9]+"
+    r"|P?Q[0-9]+_[0-9]+(?:_G[0-9]+)?"
     r"|Q[0-9]+_K"
     r"|BF16|F16|F32)",
     re.IGNORECASE,

studio/backend/utils/models/model_config.py

@@ -1630,7 +1630,7 @@ def _extract_quant_label(filename: str) -> str:
         r"|IQ[0-9]+_[A-Z]+(?:_[A-Z0-9]+)?"  # IQ variants: IQ4_XS, IQ4_NL, IQ1_S
         r"|TQ[0-9]+_[0-9]+"  # Ternary quant: TQ1_0, TQ2_0
         r"|Q[0-9]+_K_[A-Z]+"  # K-quant: Q4_K_M, Q3_K_S
-        r"|Q[0-9]+_[0-9]+"  # Standard: Q8_0, Q5_1
+        r"|P?Q[0-9]+_[0-9]+(?:_G[0-9]+)?"  # Standard/Packed: Q8_0, PQ2_0, Q2_0_g64
         r"|Q[0-9]+_K"  # Short K-quant: Q6_K
         r"|BF16|F16|F32)"  # Full precision
         # Optional bits-per-weight modifier so repos that ship multiple