[V1] Support any head size for FlexAttention backend (#20467)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
2025-07-07 00:54:36 +08:00
parent e202dd2736
commit 9fb52e523a
20 changed files with 202 additions and 118 deletions
--- a/tests/models/quantization/test_gguf.py
+++ b/tests/models/quantization/test_gguf.py
@@ -83,7 +83,7 @@ MODELS = [
    QWEN2_CONFIG,
    PHI3_CONFIG,
    GPT2_CONFIG,
-    # STABLELM_CONFIG,  # enable this when v1 support head_size=80
+    STABLELM_CONFIG,
    DOLPHIN_CONFIG,
    # STARCODER_CONFIG, # broken
 ]