Re-trigger HF build (transient pip network error on prior run)

Update data/lexicons.json

Add batch /v1/audio/transcriptions endpoint to async apps

Fix context-echo leak: check raw (pre-romanize) text + catch mid-context runs

Auto enforce_eager on pre-Ampere GPUs (T4 engine-core init fix)

Dockerfile: single-line model-download RUNs (ACR dep-scanner compatibility)

Update data/lexicons.json

Update app_en_async.py

Auto-select dtype per GPU (fix T4 engine-init crash)

Clean Space to async-only + rewrite README

Deploy combined-async: Dockerfile CMD -> main_async.py

Port /en to async + combine both async apps under main_async.py

Spike: async streaming for Chughtai on vllm.AsyncLLM

Serialize all vLLM access with a shared model_lock

Fix Chughtai WS stall: unsupported language crashed the stream

Update shared_model.py

Fix context-prompt echoed as transcript on silent utterances

Add generic doctor↔patient medical context to English/Malay app

Drop flash-attn source build (OOM-kills the HF Spaces builder)

Split into shared-model combined server (English/Malay + Chughtai)

Update app_chughtai.py

Update data/lexicons.json

Merge HF Space commit aa425ba (GPU_MEMORY_UTILIZATION=0.80)

Remove dev/tooling files from HF Space deploy

Update app_chughtai.py

Deploy: run app_chughtai.py (lexicon) on HF Space; keep app.py/app-2.py

docs: add uv setup, .env config, and server/UI run instructions to README

Initial commit: Qwen3-ASR vLLM server (Chughtai variant), lexicons and project config