Server & Testing Environment Variables
Server
Read on uvicorn boot by server/app.py + server/engine.py + server/scheduler.py.
| Variable | Required | Default | Effect |
|---|
TINYLOOP_MODEL_PATH | yes | — | Path to .tinyloop model. |
TINYLOOP_MAX_SEQ_LEN | no | 2048 | Max prompt + generated length. |
TINYLOOP_MAX_QUEUE_DEPTH | no | 128 | Max queued requests before HTTP 503. |
TINYLOOP_DEFAULT_SLO_MS | no | unset | Default SLO for requests without one. |
TINYLOOP_PREFIX_POOL_CAPACITY | no | 16 | Shared PrefixPool LRU capacity. |
TINYLOOP_BUILD_DIR | no | ./build | Path to tinyloop_py.so. |
TINYLOOP_COMMIT | no | null | Version string in model_info response. |
HF_HOME | no | default | HuggingFace tokenizer cache. |
Testing
| Variable | Effect |
|---|
TINYLOOP_TEST_MODEL_PATH | Path to real .tinyloop artifact. Tests skip (exit 77) when absent. |
TINYLOOP_TEST_BINARY | Path to built tinyloop CLI binary. |
TINYLOOP_TEST_BUILD_DIR | Path to directory containing tinyloop_py.*.so. |
See β-Informed Deployment Features for full details.
| Variable | Value | Effect |
|---|
TINYLOOP_BETA_MONITOR | 1 | Enable live entropy telemetry. |
TINYLOOP_BETA_MON_FILE | path | JSON output file. |
TINYLOOP_CTX_TIER | 1 | Context-length-aware quantization tier policy. |
TINYLOOP_CTX_TIER_SHORT_T | 512 | SHORT tier boundary. |
TINYLOOP_CTX_TIER_LONG_T | 2048 | LONG tier boundary. |