Ops Console
FishID Ops
Machines
↻ Refresh all
Basic Mode · Always On
—
loading…
Specs
Pipeline
Tunnel
High-Performance Mode · Opt-In
—
loading…
Specs
Pipeline
Tunnel
Fallback Behaviour
Requests default to Basic (CPU, always on). Users can opt-in to High-Performance — if the GPU is stopped, switching prompts to start it. If both machines are unreachable, the edge function falls back to Emergency mode (VLM only, no segmentation mask). The VLM model and prompt are configured in the Config tab.
Engine
Pick which identify engine's config to edit · V2 and V3 share
v2_config KV
GroundingDINO
Segment pipeline · bbox detection · single fixed model
Detection Model
—
loading…
SAM Segmentation
Segment pipeline · bbox → mask · reload ~30s
CPU Segmentation Model
—
loading…
VLM Model
V1 only · provider · browser-local
QWEN Default
qwen-vl-max · Alibaba DashScope · V1 default
Doubao
doubao-seed-2-0-pro-260215 · Volcengine ARK
VLM Prompt
V1 · server-side · activate ⇒ instant
Version
System Message
User Prompt
Max Tokens
Save writes to the selected version. Set Active makes it the one the API uses.
Reset all versions to default…
Tuning
V2 / V3 · candidate recall + thresholds + defaults
Max Candidates
Low Confidence Threshold
Default Model
Default Cutout Mode
Stage A · Visual Feature Extraction
V2 / V3 · per-version prompt · activate ⇒ instant
Version
System Prompt
User Prompt
Max Tokens
Stage C · Fine-grained Pick from Candidates
V2 / V3 · runs only when Stage A returns matched=false
Version
System Prompt
User Prompt (candidates JSON appended at runtime)
Max Tokens
Identifications
All V1/V2/V3 calls log to D1 · model prefix marks version
Stage A 准确率（命中样本中，对照真实物种）
Loading…
Events (last 7 days)
Billing
Account Balance
—
—
Last 7 Days Spend
—
CPU ¥0.44/h (c4il.xlarge) · GPU N/A