fix(config): Ollama 超时与 num_predict 按 ARM 实测调整
- 实测 1024px 帧视觉编码 ~36s/帧 + ~5 tok/s 生成 - num_predict 500→60: 500 会单帧跑数分钟必触发 240s 超时 - vlm_visual/model timeout 240→600, vlm_fusion 120→300, overall 600→1800
This commit is contained in:
@@ -27,12 +27,14 @@ video:
|
|||||||
max_long_edge: 1024
|
max_long_edge: 1024
|
||||||
|
|
||||||
# 超时(秒)
|
# 超时(秒)
|
||||||
|
# vlm_visual 实测: 1024px 帧视觉编码 ~36s/帧 + 生成 ~12s/60token (Oracle ARM CPU)
|
||||||
|
# 30min 视频 12 帧 × ~50s ≈ 600s,超时需覆盖最坏情况
|
||||||
timeout:
|
timeout:
|
||||||
download: 60
|
download: 60
|
||||||
vlm_visual: 240
|
vlm_visual: 600
|
||||||
vlm_fusion: 120
|
vlm_fusion: 300
|
||||||
callback: 30
|
callback: 30
|
||||||
overall: 600
|
overall: 1800
|
||||||
|
|
||||||
# 模型清单
|
# 模型清单
|
||||||
models:
|
models:
|
||||||
@@ -40,8 +42,9 @@ models:
|
|||||||
enabled: true
|
enabled: true
|
||||||
model_name: "llava-phi3"
|
model_name: "llava-phi3"
|
||||||
base_url: "http://localhost:11434"
|
base_url: "http://localhost:11434"
|
||||||
timeout: 240
|
timeout: 600
|
||||||
num_predict: 500
|
# num_predict 必须小: ARM CPU ~5 tok/s,500 会单帧跑数分钟触发超时
|
||||||
|
num_predict: 60
|
||||||
circuit_breaker:
|
circuit_breaker:
|
||||||
enabled: false
|
enabled: false
|
||||||
threshold: 5
|
threshold: 5
|
||||||
|
|||||||
Reference in New Issue
Block a user