From 5470d55b79e59b28a24216dec992e9db3c278cc5 Mon Sep 17 00:00:00 2001
From: Story Crater Bot <19826264+Riotpiaole@users.noreply.github.com>
Date: Wed, 19 Aug 2026 18:23:15 -0700
Subject: [PATCH] reasoning: fix quantization flag mismatch
 (compressed-tensors, not awq_marlin)

cpatonn's "AWQ-4bit" repo is actually quantized via llm-compressor --
config.json declares compressed-tensors. Passing awq_marlin explicitly
conflicted with the checkpoint's own declared format and 400d at
config-validation time.
---
 k8s/apps/llm-serving/reasoning.yaml | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/k8s/apps/llm-serving/reasoning.yaml b/k8s/apps/llm-serving/reasoning.yaml
index 8aab953..4b45795 100644
--- a/k8s/apps/llm-serving/reasoning.yaml
+++ b/k8s/apps/llm-serving/reasoning.yaml
@@ -31,7 +31,12 @@ spec:
       # on-the-fly bnb gamble a second time.
       - --model=cpatonn/Qwen3-30B-A3B-Thinking-2507-AWQ-4bit
       - --served-model-name=reasoning
-      - --quantization=awq_marlin
+      # Named "AWQ" but actually quantized via llm-compressor -- config.json
+      # declares compressed-tensors, not classic AWQ. vLLM auto-detects this
+      # from the checkpoint; passing awq_marlin explicitly mismatches and
+      # 400s at config-validation time. Letting vLLM read it from the
+      # checkpoint instead of asserting the wrong format.
+      - --quantization=compressed-tensors
       - --dtype=float16
       - --kv-cache-dtype=auto
       - --tensor-parallel-size=1