diff --git a/infra/helm/stonks-oracle/values-beta.yaml b/infra/helm/stonks-oracle/values-beta.yaml index 181f21e..4ea13ff 100644 --- a/infra/helm/stonks-oracle/values-beta.yaml +++ b/infra/helm/stonks-oracle/values-beta.yaml @@ -43,7 +43,7 @@ config: OLLAMA_BASE_URL: "http://192.168.42.254:11434" OLLAMA_MODEL: "qwen3.6" VLLM_BASE_URL: "http://192.168.42.254:8000" - VLLM_MODEL: "RedHatAI/Qwen3.6-35B-A3B-NVFP4" + VLLM_MODEL: "AxionML/Qwen3.5-9B-NVFP4" VLLM_TIMEOUT: "120" VLLM_MAX_RETRIES: "2" VLLM_TEMPERATURE: "0.7" diff --git a/infra/helm/stonks-oracle/values.yaml b/infra/helm/stonks-oracle/values.yaml index 21d1643..8b38a04 100644 --- a/infra/helm/stonks-oracle/values.yaml +++ b/infra/helm/stonks-oracle/values.yaml @@ -182,7 +182,7 @@ config: OLLAMA_RETRY_MAX_DELAY: "10.0" OLLAMA_RETRY_BACKOFF_MULTIPLIER: "2.0" VLLM_BASE_URL: "http://192.168.42.254:8000" - VLLM_MODEL: "RedHatAI/Qwen3.6-35B-A3B-NVFP4" + VLLM_MODEL: "AxionML/Qwen3.5-9B-NVFP4" VLLM_TIMEOUT: "120" VLLM_MAX_RETRIES: "2" VLLM_TEMPERATURE: "0.7"