Update nvidia_llm to use minimaxai/minimax-m2.7 and new sampling params

agents/nvidia_llm.py

@@ -24,7 +24,7 @@ NVIDIA_API_KEY = os.getenv("NVIDIA_API_KEY")
 _QUEUE_MAX_SIZE    = int(os.getenv("NVIDIA_QUEUE_MAX", "8"))
 _REQUEST_TIMEOUT_S = int(os.getenv("NVIDIA_LLM_TIMEOUT", "600"))
 
-DEFAULT_MODEL = "qwen/qwen3-next-80b-a3b-instruct"
+DEFAULT_MODEL = "minimaxai/minimax-m2.7"
 
 # ── Flask app ─────────────────────────────────────────────────────────────────
 app = Flask(__name__)
@@ -54,9 +54,9 @@ def _run_inference(data: dict) -> dict:
 
     prompts = raw_prompt if isinstance(raw_prompt, list) and (len(raw_prompt) == 0 or not isinstance(raw_prompt[0], dict)) else [raw_prompt]
     
-    max_tokens = int(data.get("max_tokens", 4096))
-    temperature = float(data.get("temperature", 0.6))
-    top_p = float(data.get("top_p", 0.7))
+    max_tokens = int(data.get("max_tokens", 8192))
+    temperature = float(data.get("temperature", 1.0))
+    top_p = float(data.get("top_p", 0.95))
     model_name = data.get("model", DEFAULT_MODEL)
 
     choices = []