Legacy HF TGI Mock (ailab-ml)
What It Is
A lightweight mock of Hugging Face Text Generation Inference. This ML-host copy is now legacy/opt-in; the default guided benchmark uses the auth-gated TGI gateway on ailab-app:8180.
Surface
| Endpoint |
Purpose |
GET /health |
Basic health check |
GET /info |
Model metadata (model_id, model_type, version, sha, token limits) |
POST /generate |
Fake text-generation response |
GET /v1/models |
OpenAI-compatible model listing |
GET /metrics |
Prometheus metrics (request count, queue size, batch size) |
Port & Unit
| Parameter |
Value |
| Host |
172.16.50.20 |
| Port |
8180 |
| Unit |
hf-tgi-mock.service |
| Runtime |
python3 server.py |
| Enable |
ENABLE_LEGACY_ML_TGI=1 during ML platform provisioning |
What aipostex Finds
- HF TGI fingerprint via
/info (model_id, model_type, version, Docker SHA)
- Unauthenticated generation surface via
/generate when the legacy mock is explicitly enabled
- OpenAI-compatible model listing via
/v1/models
- Operational metrics via
/metrics (Prometheus format)