fix(nvidia): revert to nemotron-mini, truncate context for 4K limit, enforce precise confidence

2026-03-31 13:57:10 +08:00
parent 22796c6aff
commit 46843c8e19
5 changed files with 16 additions and 13 deletions
--- a/apps/api/models.json
+++ b/apps/api/models.json
@@ -105,15 +105,15 @@
    },

    "nvidia": {
-      "name": "NVIDIA NIM (Llama 3.1 8B / ADR-036)",
+      "name": "NVIDIA Nemotron (ADR-036)",
      "enabled": true,
      "priority": 4,
      "endpoint": "https://integrate.api.nvidia.com/v1",
      "api_path": "/chat/completions",
      "models": {
-        "default": "meta/llama-3.1-8b-instruct",
-        "tool_calling": "meta/llama-3.1-8b-instruct",
-        "rca": "meta/llama-3.1-8b-instruct"
+        "default": "nvidia/nemotron-mini-4b-instruct",
+        "tool_calling": "nvidia/nemotron-mini-4b-instruct",
+        "rca": "nvidia/nemotron-mini-4b-instruct"
      },
      "options": {
        "temperature": 0.0,