GLM-OCR · Bench
zai-org/GLM-OCR · vLLM · modal A10G

Optical test bench

Send an image.
Read it back.

Drop a document photo or scan. The extracted fields come back structured on the right.

Endpoint

The full chat-completions path of your Modal deployment.

Optional on ocr.vec.run — the server injects it. Only needed when calling Modal directly.

Document

prompt contract: the model was trained on “Text Recognition:” — any other instruction degrades it into empty output.
engine: vLLM bf16 · A10G · memory-snapshot cold starts

Readings

idle

No document loaded

Drop an image on the left to see what the model reads out of it.