use correct dtype

2026-05-11 11:34:39 -05:00
parent 719a6e3d11
commit 08106b6693
1 changed files with 1 additions and 1 deletions
--- a/clusters/k3s-dgx/nim-service/qwen.yaml
+++ b/clusters/k3s-dgx/nim-service/qwen.yaml
@@ -65,7 +65,7 @@ spec:
    - "254"                   # ↑ Allow more concurrent sequences
    - --enable-prefix-caching
    - --dtype
-    - "float8"                # Explicit FP8 encoding
+    - "float"                # Explicit FP8 encoding
    - --quantization
    - "fbgemm-fp8"            # Explicit quantization backend
    - --sv2-transformer-bindings  # Skip SA checks (small speedup)