Compare commits
1 Commits
957b6dab43
...
589a63f63c
| Author | SHA1 | Date | |
|---|---|---|---|
| 589a63f63c |
@@ -54,10 +54,7 @@ spec:
|
||||
"gemma3-4b",
|
||||
"gemma3-4b-novision",
|
||||
"Qwen3-4B-Thinking-2507",
|
||||
"Qwen3-4B-Thinking-2507-long-ctx",
|
||||
"Qwen2.5-VL-7B-Instruct-GGUF",
|
||||
"Qwen2.5-VL-32B-Instruct-GGUF-IQ1_S",
|
||||
"Qwen2.5-VL-32B-Instruct-GGUF-Q2_K_L"
|
||||
"Qwen3-4B-Thinking-2507-long-ctx"
|
||||
]
|
||||
titleConvo: true
|
||||
titleModel: "gemma3-4b-novision"
|
||||
|
||||
@@ -175,42 +175,3 @@ models:
|
||||
--flash-attn
|
||||
--cache-type-k q8_0 --cache-type-v q8_0
|
||||
--port ${PORT}
|
||||
"Qwen2.5-VL-32B-Instruct-GGUF-IQ1_S":
|
||||
ttl: 600
|
||||
cmd: |
|
||||
/app/llama-server
|
||||
-hf unsloth/Qwen2.5-VL-32B-Instruct-GGUF:IQ1_S
|
||||
-ngl 99 -c 16384 --predict 8192
|
||||
--temp 0.7
|
||||
--min-p 0.00
|
||||
--top-p 0.8
|
||||
--top-k 20
|
||||
--repeat-penalty 1.0
|
||||
--no-warmup
|
||||
--port ${PORT}
|
||||
"Qwen2.5-VL-32B-Instruct-GGUF-Q2_K_L":
|
||||
ttl: 600
|
||||
cmd: |
|
||||
/app/llama-server
|
||||
-hf unsloth/Qwen2.5-VL-32B-Instruct-GGUF:Q2_K_L
|
||||
-ngl 99 -c 16384 --predict 8192
|
||||
--temp 0.7
|
||||
--min-p 0.00
|
||||
--top-p 0.8
|
||||
--top-k 20
|
||||
--repeat-penalty 1.0
|
||||
--no-warmup
|
||||
--port ${PORT}
|
||||
"Qwen2.5-VL-7B-Instruct-GGUF":
|
||||
ttl: 600
|
||||
cmd: |
|
||||
/app/llama-server
|
||||
-hf unsloth/Qwen2.5-VL-7B-Instruct-GGUF:Q4_K_M
|
||||
-ngl 37 -c 16384 --predict 8192
|
||||
--temp 0.7
|
||||
--min-p 0.00
|
||||
--top-p 0.8
|
||||
--top-k 20
|
||||
--repeat-penalty 1.0
|
||||
--no-warmup
|
||||
--port ${PORT}
|
||||
|
||||
@@ -97,7 +97,7 @@ spec:
|
||||
env:
|
||||
- name: GOMEMLIMIT
|
||||
value: 161MiB
|
||||
image: registry.k8s.io/coredns/coredns:v1.13.0
|
||||
image: registry.k8s.io/coredns/coredns:v1.12.4
|
||||
imagePullPolicy: IfNotPresent
|
||||
livenessProbe:
|
||||
failureThreshold: 5
|
||||
|
||||
Reference in New Issue
Block a user