From 241dce4524d60ca0fc89275dfd9930f0bb242c2b Mon Sep 17 00:00:00 2001 From: Lumpiasty Date: Tue, 29 Jul 2025 02:24:14 +0200 Subject: [PATCH] disable warmups --- apps/llama/configs/config.yaml | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/apps/llama/configs/config.yaml b/apps/llama/configs/config.yaml index 5119de7..2362a8e 100644 --- a/apps/llama/configs/config.yaml +++ b/apps/llama/configs/config.yaml @@ -4,12 +4,14 @@ models: /app/llama-server -hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M -ngl 37 -c 16384 + --no-warmup --port ${PORT} "Qwen3-8B-GGUF": cmd: | /app/llama-server -hf unsloth/Qwen3-8B-GGUF:Q4_K_M -ngl 37 -c 16384 + --no-warmup --port ${PORT} "Qwen3-8B-GGUF-no-thinking": cmd: | @@ -17,6 +19,7 @@ models: -hf unsloth/Qwen3-8B-GGUF:Q4_K_M -ngl 37 -c 16384 --jinja --chat-template-file /config/qwen_nothink_chat_template.jinja + --no-warmup --port ${PORT} "gemma3n-e3b": cmd: | @@ -31,6 +34,7 @@ models: --min-p 0.00 --top-k 64 --top-p 0.95 + --no-warmup --port ${PORT} "gemma3-12b": cmd: | @@ -44,6 +48,7 @@ models: --min-p 0.00 --top-k 64 --top-p 0.95 + --no-warmup --port ${PORT} "gemma3-12b-novision": cmd: | @@ -58,6 +63,7 @@ models: --top-k 64 --top-p 0.95 --no-mmproj + --no-warmup --port ${PORT} "gemma3-12b-q2": cmd: | @@ -71,6 +77,7 @@ models: --min-p 0.00 --top-k 64 --top-p 0.95 + --no-warmup --port ${PORT} "gemma3-4b": cmd: | @@ -84,6 +91,7 @@ models: --min-p 0.00 --top-k 64 --top-p 0.95 + --no-warmup --port ${PORT} "gemma3-4b-novision": cmd: | @@ -98,4 +106,5 @@ models: --top-k 64 --top-p 0.95 --no-mmproj + --no-warmup --port ${PORT}