Compare commits

..
Author SHA1 Message Date
hermes fab69b521e fix: desactivar el thinking de Qwen3 (SYSTEM /no_think)
Qwen3 4B tiene el thinking mode ON por defecto: para 'Say hi in one
word' generaba 232 tokens de razonamiento interno antes de responder
(45+ s en una GTX 1660). Con /no_think: 7 tokens.

- SYSTEM /no_think en el Modelfile (import HF y re-create num_ctx)
- el re-create pasa SIEMPRE al validar el modelo ya presente (asi el
  tag existente del usuario hereda el fix sin re-descargar)
2026-09-12 22:38:08 +02:00
2 changed files with 2 additions and 2 deletions

No files matched your search

+1 -1
View File
@@ -48,7 +48,7 @@ services:
ENABLE_WEB_SEARCH: "false"
ENABLE_DIRECT_CONNECTIONS: "false"
CHAT_RESPONSE_MAX_TOOL_CALL_ITERATIONS: "1"
DEFAULT_MODEL_PARAMS: '{"function_calling": "legacy", "think": false}'
DEFAULT_MODEL_PARAMS: '{"function_calling": "legacy"}'
depends_on:
ollama:
condition: service_healthy
+1 -1
View File
@@ -78,7 +78,7 @@ services:
WEBUI_AUTH: "false"
DEFAULT_LOCALE: "es"
CHAT_RESPONSE_MAX_TOOL_CALL_ITERATIONS: "1"
DEFAULT_MODEL_PARAMS: '{"function_calling": "legacy", "think": false}'
DEFAULT_MODEL_PARAMS: '{"function_calling": "legacy"}'
depends_on:
ollama:
condition: service_healthy