Compare commits
1 Commits
renovate/c
...
b7d739bb6e
| Author | SHA1 | Date | |
|---|---|---|---|
| b7d739bb6e |
@@ -15,8 +15,8 @@ spec:
|
|||||||
spec:
|
spec:
|
||||||
containers:
|
containers:
|
||||||
- name: crawl4ai-proxy
|
- name: crawl4ai-proxy
|
||||||
image: gitea.lumpiasty.xyz/lumpiasty/crawl4ai-proxy-fit:latest
|
image: ghcr.io/lennyerik/crawl4ai-proxy:latest
|
||||||
imagePullPolicy: Always
|
imagePullPolicy: IfNotPresent
|
||||||
env:
|
env:
|
||||||
- name: LISTEN_PORT
|
- name: LISTEN_PORT
|
||||||
value: "8000"
|
value: "8000"
|
||||||
|
|||||||
@@ -18,7 +18,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: immich
|
chart: immich
|
||||||
version: 1.2.2
|
version: 1.1.1
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: secustor
|
name: secustor
|
||||||
|
|||||||
@@ -4,13 +4,10 @@ logToStdout: "both" # proxy and upstream
|
|||||||
|
|
||||||
macros:
|
macros:
|
||||||
base_args: "--no-warmup --port ${PORT}"
|
base_args: "--no-warmup --port ${PORT}"
|
||||||
common_args: "--fit-target 1536 --no-warmup --port ${PORT}"
|
common_args: "--fit-target 1536 --fit-ctx 65536 --no-warmup --port ${PORT}"
|
||||||
gemma3_ctx_128k: "--ctx-size 131072"
|
|
||||||
qwen35_ctx_128k: "--ctx-size 131072"
|
|
||||||
qwen35_ctx_256k: "--ctx-size 262144"
|
|
||||||
gemma_sampling: "--prio 2 --temp 1.0 --repeat-penalty 1.0 --min-p 0.00 --top-k 64 --top-p 0.95"
|
gemma_sampling: "--prio 2 --temp 1.0 --repeat-penalty 1.0 --min-p 0.00 --top-k 64 --top-p 0.95"
|
||||||
qwen35_sampling: "--temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -ctk q8_0 -ctv q8_0"
|
qwen35_sampling: "--temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -ctk q4_0 -ctv q4_0"
|
||||||
qwen35_35b_args: "--temp 1.0 --min-p 0.00 --top-p 0.95 --top-k 20 -ctk q8_0 -ctv q8_0"
|
qwen35_35b_args: "--temp 1.0 --min-p 0.00 --top-p 0.95 --top-k 20 -ctk q4_0 -ctv q4_0"
|
||||||
qwen35_35b_heretic_mmproj: "--mmproj-url https://huggingface.co/unsloth/Qwen3.5-35B-A3B-GGUF/resolve/main/mmproj-F16.gguf --mmproj /root/.cache/llama.cpp/unsloth_Qwen3.5-35B-A3B-GGUF_mmproj-F16.gguf"
|
qwen35_35b_heretic_mmproj: "--mmproj-url https://huggingface.co/unsloth/Qwen3.5-35B-A3B-GGUF/resolve/main/mmproj-F16.gguf --mmproj /root/.cache/llama.cpp/unsloth_Qwen3.5-35B-A3B-GGUF_mmproj-F16.gguf"
|
||||||
qwen35_4b_heretic_mmproj: "--mmproj-url https://huggingface.co/unsloth/Qwen3.5-4B-GGUF/resolve/main/mmproj-F16.gguf --mmproj /root/.cache/llama.cpp/unsloth_Qwen3.5-4B-GGUF_mmproj-F16.gguf"
|
qwen35_4b_heretic_mmproj: "--mmproj-url https://huggingface.co/unsloth/Qwen3.5-4B-GGUF/resolve/main/mmproj-F16.gguf --mmproj /root/.cache/llama.cpp/unsloth_Qwen3.5-4B-GGUF_mmproj-F16.gguf"
|
||||||
glm47_flash_args: "--temp 0.7 --top-p 1.0 --min-p 0.01 --repeat-penalty 1.0"
|
glm47_flash_args: "--temp 0.7 --top-p 1.0 --min-p 0.01 --repeat-penalty 1.0"
|
||||||
@@ -42,7 +39,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/gemma-3-12b-it-GGUF:Q4_K_M
|
-hf unsloth/gemma-3-12b-it-GGUF:Q4_K_M
|
||||||
${gemma3_ctx_128k}
|
|
||||||
${gemma_sampling}
|
${gemma_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
|
|
||||||
@@ -50,7 +46,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/gemma-3-12b-it-GGUF:Q4_K_M
|
-hf unsloth/gemma-3-12b-it-GGUF:Q4_K_M
|
||||||
${gemma3_ctx_128k}
|
|
||||||
${gemma_sampling}
|
${gemma_sampling}
|
||||||
--no-mmproj
|
--no-mmproj
|
||||||
${common_args}
|
${common_args}
|
||||||
@@ -59,7 +54,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/gemma-3-4b-it-GGUF:Q4_K_M
|
-hf unsloth/gemma-3-4b-it-GGUF:Q4_K_M
|
||||||
${gemma3_ctx_128k}
|
|
||||||
${gemma_sampling}
|
${gemma_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
|
|
||||||
@@ -67,7 +61,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/gemma-3-4b-it-GGUF:Q4_K_M
|
-hf unsloth/gemma-3-4b-it-GGUF:Q4_K_M
|
||||||
${gemma3_ctx_128k}
|
|
||||||
${gemma_sampling}
|
${gemma_sampling}
|
||||||
--no-mmproj
|
--no-mmproj
|
||||||
${common_args}
|
${common_args}
|
||||||
@@ -83,14 +76,13 @@ models:
|
|||||||
--top-p 0.95
|
--top-p 0.95
|
||||||
--top-k 40
|
--top-k 40
|
||||||
--repeat-penalty 1.0
|
--repeat-penalty 1.0
|
||||||
-ctk q8_0 -ctv q8_0
|
-ctk q4_0 -ctv q4_0
|
||||||
${common_args}
|
${common_args}
|
||||||
|
|
||||||
"Qwen3.5-35B-A3B-GGUF:Q4_K_M":
|
"Qwen3.5-35B-A3B-GGUF:Q4_K_M":
|
||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-35B-A3B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-35B-A3B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_35b_args}
|
${qwen35_35b_args}
|
||||||
${common_args}
|
${common_args}
|
||||||
|
|
||||||
@@ -98,7 +90,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-35B-A3B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-35B-A3B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_35b_args}
|
${qwen35_35b_args}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
@@ -110,7 +101,6 @@ models:
|
|||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf mradermacher/Qwen3.5-35B-A3B-heretic-GGUF:Q4_K_M
|
-hf mradermacher/Qwen3.5-35B-A3B-heretic-GGUF:Q4_K_M
|
||||||
${qwen35_35b_heretic_mmproj}
|
${qwen35_35b_heretic_mmproj}
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_35b_args}
|
${qwen35_35b_args}
|
||||||
${common_args}
|
${common_args}
|
||||||
|
|
||||||
@@ -119,7 +109,6 @@ models:
|
|||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf mradermacher/Qwen3.5-35B-A3B-heretic-GGUF:Q4_K_M
|
-hf mradermacher/Qwen3.5-35B-A3B-heretic-GGUF:Q4_K_M
|
||||||
${qwen35_35b_heretic_mmproj}
|
${qwen35_35b_heretic_mmproj}
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_35b_args}
|
${qwen35_35b_args}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
@@ -128,7 +117,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-0.8B-GGUF:Q4_K_XL
|
-hf unsloth/Qwen3.5-0.8B-GGUF:Q4_K_XL
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${base_args}
|
${base_args}
|
||||||
${thinking_on}
|
${thinking_on}
|
||||||
@@ -146,7 +134,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-2B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-2B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_on}
|
${thinking_on}
|
||||||
@@ -155,7 +142,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-2B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-2B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
@@ -164,7 +150,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_128k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_on}
|
${thinking_on}
|
||||||
@@ -173,7 +158,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_128k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
@@ -183,7 +167,6 @@ models:
|
|||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf mradermacher/Qwen3.5-4B-heretic-GGUF:Q4_K_M
|
-hf mradermacher/Qwen3.5-4B-heretic-GGUF:Q4_K_M
|
||||||
${qwen35_4b_heretic_mmproj}
|
${qwen35_4b_heretic_mmproj}
|
||||||
${qwen35_ctx_128k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_on}
|
${thinking_on}
|
||||||
@@ -193,7 +176,6 @@ models:
|
|||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf mradermacher/Qwen3.5-4B-heretic-GGUF:Q4_K_M
|
-hf mradermacher/Qwen3.5-4B-heretic-GGUF:Q4_K_M
|
||||||
${qwen35_4b_heretic_mmproj}
|
${qwen35_4b_heretic_mmproj}
|
||||||
${qwen35_ctx_128k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
@@ -202,7 +184,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-9B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-9B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_on}
|
${thinking_on}
|
||||||
@@ -211,7 +192,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-9B-GGUF:Q4_K_M
|
-hf unsloth/Qwen3.5-9B-GGUF:Q4_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
@@ -220,7 +200,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-9B-GGUF:Q3_K_M
|
-hf unsloth/Qwen3.5-9B-GGUF:Q3_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_on}
|
${thinking_on}
|
||||||
@@ -229,7 +208,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-9B-GGUF:Q3_K_M
|
-hf unsloth/Qwen3.5-9B-GGUF:Q3_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
@@ -238,7 +216,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-27B-GGUF:Q3_K_M
|
-hf unsloth/Qwen3.5-27B-GGUF:Q3_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_on}
|
${thinking_on}
|
||||||
@@ -247,7 +224,6 @@ models:
|
|||||||
cmd: |
|
cmd: |
|
||||||
/app/llama-server
|
/app/llama-server
|
||||||
-hf unsloth/Qwen3.5-27B-GGUF:Q3_K_M
|
-hf unsloth/Qwen3.5-27B-GGUF:Q3_K_M
|
||||||
${qwen35_ctx_256k}
|
|
||||||
${qwen35_sampling}
|
${qwen35_sampling}
|
||||||
${common_args}
|
${common_args}
|
||||||
${thinking_off}
|
${thinking_off}
|
||||||
|
|||||||
@@ -18,7 +18,7 @@ spec:
|
|||||||
spec:
|
spec:
|
||||||
containers:
|
containers:
|
||||||
- name: llama-swap
|
- name: llama-swap
|
||||||
image: ghcr.io/mostlygeek/llama-swap:v199-vulkan-b8576
|
image: ghcr.io/mostlygeek/llama-swap:v198-vulkan-b8352
|
||||||
imagePullPolicy: IfNotPresent
|
imagePullPolicy: IfNotPresent
|
||||||
command:
|
command:
|
||||||
- /app/llama-swap
|
- /app/llama-swap
|
||||||
|
|||||||
@@ -18,7 +18,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: open-webui
|
chart: open-webui
|
||||||
version: 12.13.0
|
version: 12.10.0
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: open-webui
|
name: open-webui
|
||||||
|
|||||||
@@ -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:43.95.0-full
|
image: renovate/renovate:43.76.5-full
|
||||||
envFrom:
|
envFrom:
|
||||||
- secretRef:
|
- secretRef:
|
||||||
name: renovate-gitea-token
|
name: renovate-gitea-token
|
||||||
|
|||||||
@@ -18,7 +18,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: cert-manager-webhook-ovh
|
chart: cert-manager-webhook-ovh
|
||||||
version: 0.9.5
|
version: 0.9.4
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: cert-manager-webhook-ovh
|
name: cert-manager-webhook-ovh
|
||||||
|
|||||||
@@ -23,7 +23,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: cert-manager
|
chart: cert-manager
|
||||||
version: v1.20.1
|
version: v1.20.0
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: cert-manager
|
name: cert-manager
|
||||||
|
|||||||
@@ -23,7 +23,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: cilium
|
chart: cilium
|
||||||
version: 1.19.2
|
version: 1.19.1
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: cilium
|
name: cilium
|
||||||
|
|||||||
@@ -23,7 +23,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: cloudnative-pg
|
chart: cloudnative-pg
|
||||||
version: 0.28.0
|
version: 0.27.1
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: cnpg
|
name: cnpg
|
||||||
|
|||||||
@@ -23,7 +23,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: k8up
|
chart: k8up
|
||||||
version: 4.9.0
|
version: 4.8.6
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: k8up-io
|
name: k8up-io
|
||||||
|
|||||||
@@ -23,7 +23,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: ingress-nginx
|
chart: ingress-nginx
|
||||||
version: 4.15.1
|
version: 4.15.0
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: ingress-nginx
|
name: ingress-nginx
|
||||||
|
|||||||
@@ -23,7 +23,7 @@ spec:
|
|||||||
chart:
|
chart:
|
||||||
spec:
|
spec:
|
||||||
chart: openbao
|
chart: openbao
|
||||||
version: 0.26.2
|
version: 0.26.0
|
||||||
sourceRef:
|
sourceRef:
|
||||||
kind: HelmRepository
|
kind: HelmRepository
|
||||||
name: openbao
|
name: openbao
|
||||||
|
|||||||
Reference in New Issue
Block a user