x

app/model_catalog.py

@@ -110,6 +110,7 @@ def resolve_model_profile(
         "served_model_name": profile.get("served_model_name", model_key),
         "dtype": _to_str(profile.get("dtype")),
         "quantization": _to_str(profile.get("quantization")),
+        "model_impl": _to_str(profile.get("model_impl")),
         "reasoning_parser": _to_str(profile.get("reasoning_parser")),
         "max_model_len": _to_int(profile.get("ctx"), 8192),
         "max_num_seqs": _to_int(profile.get("max_num_seqs"), 64),

app/start_openai.py

@@ -25,6 +25,7 @@ def build_command() -> list[str]:
     api_key = runtime["api_key"] or ""
     dtype = str(updates["dtype"] or runtime["dtype"])
     quantization = str(updates["quantization"] or "").strip()
+    model_impl = str(updates["model_impl"] or "").strip()
     reasoning_parser = str(updates["reasoning_parser"] or "").strip()
     revision = runtime["revision"] or ""
     cmd = [
@@ -68,6 +69,8 @@ def build_command() -> list[str]:
         cmd.extend(["--reasoning-parser", reasoning_parser])
     if quantization:
         cmd.extend(["--quantization", quantization])
+    if model_impl:
+        cmd.extend(["--model-impl", model_impl])
     if revision:
         cmd.extend(["--revision", revision])
     if api_key: