2 Commits

Author SHA1 Message Date
be10ad6ceb Update renovate/renovate Docker tag to v41.76.0 2025-08-17 00:00:50 +00:00
8b06571280 add qwen3-4b-thinking-2507 model 2025-08-16 23:39:35 +02:00
3 changed files with 33 additions and 5 deletions

View File

@@ -72,12 +72,14 @@ spec:
"gemma3-12b-q2", "gemma3-12b-q2",
"gemma3-12b-novision", "gemma3-12b-novision",
"gemma3-4b", "gemma3-4b",
"gemma3-4b-novision" "gemma3-4b-novision",
"Qwen3-4B-Thinking-2507",
"Qwen3-4B-Thinking-2507-long-ctx"
] ]
titleConvo: true titleConvo: true
titleModel: "current_model" titleModel: "gemma3-4b-novision"
summarize: false summarize: false
summaryModel: "current_model" summaryModel: "gemma3-4b-novision"
forcePrompt: false forcePrompt: false
modelDisplayLabel: "Llama.cpp" modelDisplayLabel: "Llama.cpp"
imageVolume: imageVolume:

View File

@@ -26,7 +26,7 @@ models:
--jinja --chat-template-file /config/qwen_nothink_chat_template.jinja --jinja --chat-template-file /config/qwen_nothink_chat_template.jinja
--no-warmup --no-warmup
--port ${PORT} --port ${PORT}
"gemma3n-e3b": "gemma3n-e4b":
ttl: 600 ttl: 600
cmd: | cmd: |
/app/llama-server /app/llama-server
@@ -119,3 +119,29 @@ models:
--no-mmproj --no-mmproj
--no-warmup --no-warmup
--port ${PORT} --port ${PORT}
"Qwen3-4B-Thinking-2507":
ttl: 600
cmd: |
/app/llama-server
-hf unsloth/Qwen3-4B-Thinking-2507-GGUF:Q4_K_M
-ngl 99 -c 16384 --predict 8192
--temp 0.6
--min-p 0.00
--top-p 0.95
--top-k 20
--repeat-penalty 1.0
--no-warmup
--port ${PORT}
"Qwen3-4B-Thinking-2507-long-ctx":
ttl: 600
cmd: |
/app/llama-server
-hf unsloth/Qwen3-4B-Thinking-2507-GGUF:Q4_K_M
-ngl 99 -c 262144 --predict 81920
--temp 0.6
--min-p 0.00
--top-p 0.95
--top-k 20
--repeat-penalty 1.0
--no-warmup
--port ${PORT}

View File

@@ -15,7 +15,7 @@ spec:
- name: renovate - name: renovate
# Update this to the latest available and then enable Renovate on # Update this to the latest available and then enable Renovate on
# the manifest # the manifest
image: renovate/renovate:41.74.2-full image: renovate/renovate:41.76.0-full
envFrom: envFrom:
- secretRef: - secretRef:
name: renovate-gitea-token name: renovate-gitea-token