[BugFix] Disable fp8 kv-cache by default for DeepSeek V3.2 (#27121)

Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

vllm/model_executor/models/config.py

@@ -481,12 +481,9 @@ class DeepseekV32ForCausalLM(VerifyAndUpdateConfig):
         is_v32 = hasattr(hf_config, "index_topk")
         assert is_v32
 
-        # For DeepSeekV3.2, we use a custom fp8 format as default (i.e.
-        #   "auto")
+        # For DeepSeekV3.2, a custom fp8 format is used when fp8 kv-cache is enabled.
         cache_config = vllm_config.cache_config
-        if cache_config.cache_dtype == "auto" or cache_config.cache_dtype.startswith(
-            "fp8"
-        ):
+        if cache_config.cache_dtype.startswith("fp8"):
             cache_config.cache_dtype = "fp8_ds_mla"
             logger.info("Using custom fp8 kv-cache format for DeepSeekV3.2")
         if cache_config.cache_dtype == "bfloat16":