Lumpiasty/klaster
0
0
Fork 0
Code Issues Pull Requests 4 Packages Projects Releases Wiki Activity

disable warmups

Browse Source
This commit is contained in:
Lumpiasty
2025-07-29 02:24:14 +02:00
parent 17805e6b31
commit 241dce4524
1 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
apps/llama/configs/config.yaml
View File

@@ -4,12 +4,14 @@ models:
/app/llama-server /app/llama-server
-hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M -hf unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF:Q4_K_M
-ngl 37 -c 16384 -ngl 37 -c 16384
--no-warmup
--port ${PORT} --port ${PORT}
"Qwen3-8B-GGUF": "Qwen3-8B-GGUF":
cmd: | cmd: |
/app/llama-server /app/llama-server
-hf unsloth/Qwen3-8B-GGUF:Q4_K_M -hf unsloth/Qwen3-8B-GGUF:Q4_K_M
-ngl 37 -c 16384 -ngl 37 -c 16384
--no-warmup
--port ${PORT} --port ${PORT}
"Qwen3-8B-GGUF-no-thinking": "Qwen3-8B-GGUF-no-thinking":
cmd: | cmd: |
@@ -17,6 +19,7 @@ models:
-hf unsloth/Qwen3-8B-GGUF:Q4_K_M -hf unsloth/Qwen3-8B-GGUF:Q4_K_M
-ngl 37 -c 16384 -ngl 37 -c 16384
--jinja --chat-template-file /config/qwen_nothink_chat_template.jinja --jinja --chat-template-file /config/qwen_nothink_chat_template.jinja
--no-warmup
--port ${PORT} --port ${PORT}
"gemma3n-e3b": "gemma3n-e3b":
cmd: | cmd: |
@@ -31,6 +34,7 @@ models:
--min-p 0.00 --min-p 0.00
--top-k 64 --top-k 64
--top-p 0.95 --top-p 0.95
--no-warmup
--port ${PORT} --port ${PORT}
"gemma3-12b": "gemma3-12b":
cmd: | cmd: |
@@ -44,6 +48,7 @@ models:
--min-p 0.00 --min-p 0.00
--top-k 64 --top-k 64
--top-p 0.95 --top-p 0.95
--no-warmup
--port ${PORT} --port ${PORT}
"gemma3-12b-novision": "gemma3-12b-novision":
cmd: | cmd: |
@@ -58,6 +63,7 @@ models:
--top-k 64 --top-k 64
--top-p 0.95 --top-p 0.95
--no-mmproj --no-mmproj
--no-warmup
--port ${PORT} --port ${PORT}
"gemma3-12b-q2": "gemma3-12b-q2":
cmd: | cmd: |
@@ -71,6 +77,7 @@ models:
--min-p 0.00 --min-p 0.00
--top-k 64 --top-k 64
--top-p 0.95 --top-p 0.95
--no-warmup
--port ${PORT} --port ${PORT}
"gemma3-4b": "gemma3-4b":
cmd: | cmd: |
@@ -84,6 +91,7 @@ models:
--min-p 0.00 --min-p 0.00
--top-k 64 --top-k 64
--top-p 0.95 --top-p 0.95
--no-warmup
--port ${PORT} --port ${PORT}
"gemma3-4b-novision": "gemma3-4b-novision":
cmd: | cmd: |
@@ -98,4 +106,5 @@ models:
--top-k 64 --top-k 64
--top-p 0.95 --top-p 0.95
--no-mmproj --no-mmproj
--no-warmup
--port ${PORT} --port ${PORT}