README.md

---
title: HiTech VLM
emoji: 🔭
colorFrom: blue
colorTo: indigo
sdk: gradio
python_version: "3.12"
app_file: app.py
pinned: false
license: apache-2.0
---

Multimodal model Space for Hi-Tech AI Platform — handles **vision and text**
(drawings, scanned POs, photos, summaries, Q&A, classification, drafting).

Model: `Qwen/Qwen3-VL-8B-Instruct` (dense, bf16, ~16 GB)
Hardware: ZeroGPU `large` (48 GB); `@spaces.GPU(duration=90)`

Moved off `Qwen3.6-35B-A3B-FP8`: its fine-grained-FP8 inference is broken on the
current ZeroGPU image (deep-gemm kernel `RuntimeError: Unknown recipe`). bf16
needs no FP8/AWQ kernels.

API contract (called by `core.clients`):
`predict(prompt: str, schema_json: str, image_path: str | None) -> str` (JSON),
at `api_name="/predict"`.