From 9ba6ef8522a3a75619972b5ee54bed00f4dd3dc3 Mon Sep 17 00:00:00 2001 From: Russell Ballestrini Date: Sat, 11 Oct 2025 09:39:41 -0400 Subject: [PATCH] set qwen context window to 32K (practical limit for 24GB VRAM) --- src/config.js | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/config.js b/src/config.js index a49d62f..3be7494 100644 --- a/src/config.js +++ b/src/config.js @@ -72,8 +72,9 @@ export const VLLM_ENDPOINTS = [ id: "qwen.ai.unturf.com", url: "https://qwen.ai.unturf.com/v1", // Ollama endpoints don't return max_model_len, so we specify it here + // Note: Model supports 262144 but limited by GPU VRAM (24GB) modelContextWindows: { - 'hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M': 262144, + 'hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M': 32768, } }, ];