chore(config): switch model to Qwen3.8-27B Q4_K_P with FastMTP 32K draft

scripts/ollama-proxy.py

@@ -353,7 +353,7 @@ def _model_entry(tag):
             "family": "qwen35",
             "families": ["qwen35", "qwen2", "clip"],
             "parameter_size": "27.0B",
-            "quantization_level": "Q5_K_P",
+            "quantization_level": "Q4_K_P",
             "context_length": ctx,
         },
         "capabilities": ["completion", "chat", "tools", "thinking", "vision", "multimodal"],
@@ -376,7 +376,7 @@ def ollama_ps():
                     "family": "qwen35",
                     "families": ["qwen35", "qwen2", "clip"],
                     "parameter_size": "27.0B",
-                    "quantization_level": "Q5_K_P",
+                    "quantization_level": "Q4_K_P",
                     "context_length": ctx,
                 },
                 "expires_at": time.strftime("%Y-%m-%dT%H:%M:%S", time.gmtime(time.time() + 86400)) + ".000000Z",
@@ -401,7 +401,7 @@ def ollama_show(payload=None):
             "family": "qwen35",
             "families": ["qwen35", "qwen2", "clip"],
             "parameter_size": "27.0B",
-            "quantization_level": "Q5_K_P",
+            "quantization_level": "Q4_K_P",
             "context_length": ctx,
         },
         "model_info": {

scripts/start-server.sh

@@ -17,7 +17,7 @@ RTX_DEV=$(/opt/llama.cpp-xrip/build-nccl/bin/llama-cli --list-devices | grep -i
 RTX_DEV=${RTX_DEV:-CUDA0}
 
 exec /opt/llama.cpp-xrip/build-nccl/bin/llama-server \
-  -m /opt/models/gguf/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q5_K_P.gguf \
+  -m /opt/models/gguf/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q4_K_P.gguf \
   --spec-draft-model /opt/models/gguf/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-FastMTP-32K.gguf \
   --spec-draft-device "$RTX_DEV" \
   --spec-draft-ngl all \