Base URL: /api
Endpoints:
GET /api/health → server status
GET /api/models → list loaded models
POST /api/models/load → {"source":"org/model","quant":"int8","tag":"main"}
POST /api/models/unload?tag= → unload a model
GET /api/models/{tag}/info → model info
GET /api/models/{tag}/tensors?limit=100&offset=0 → tensor list
POST /api/models/{tag}/tensor → {"tensor_name":"..."} → tensor stats
GET /api/models/{tag}/benchmark?n=10 → latency benchmark
POST /api/generate → {"prompt":"...","width":256,"height":256,"steps":20,"seed":42}
→ returns PNG image
POST /api/generate/base64 → same but returns {"image_base64":"..."}
POST /api/compress → {"source":"...","output":"...","quant":"int8","sparsity":0}
GET /api/inspect?path=... → inspect .nanomodel file
Interactive docs:
/api/docs
Python usage:
import requests
r = requests.post("http://localhost:7860/api/models/load",
json={"source":"org/model","quant":"int4"})
print(r.json())
r = requests.post("http://localhost:7860/api/generate",
json={"prompt":"a sunset","width":256,"height":256})
with open("out.png","wb") as f:
f.write(r.content)