chore: add smoke test and debug logs requested by christian

fix: replace buggy native chat method with custom generate pipeline

fix: switch back to native chat method to avoid generation bugs

fix: logic error in response parsing and add debug logs

fix: remove forced ChatML template and parse native tags

trigger hf space rebuild

fix: set use_cache=False to prevent kv-cache tensor shape mismatch crashes

fix: replace buggy model.chat() regex with custom generation pipeline

fix: update chat template to use ChatML format instead of legacy tags

fix: simplify msgs format for minicpm_v .chat() to prevent empty hyphen outputs

fix: refactor minicpm_v to use official .chat() method to prevent hallucinated OCR text

fix: inject official MiniCPM chat template to prevent garbage hallucinated outputs

fix: switch to official MiniCPM .chat() method to fix gibberish output

fix: use returned input_ids length for slicing position_ids in monkeypatch

fix: monkeypatch MiniCPM prepare_inputs_for_generation to prevent position_ids broadcasting crash

fix: correctly extract input_ids from BatchEncoding UserDict to prevent generate shape crash

fix: pass only input_ids to Core Router generate to prevent PyTorch attention shape mismatch

fix: use eager attention for MiniCPM4.1-8B to prevent SDPA crash, fix Nemotron Parse processor format

fix: add cv2/albumentations for Nemotron Parse, fix Core Router generate kwargs shape error

fix: switch Core Router to AutoModelForCausalLM for generate support, fix Nemotron Parse inference with AutoProcessor

fix: block flash_attn at runtime with sys.modules=None to prevent FA2 on HF ZeroGPU where flash_attn is pre-installed

fix: resolve infinite recursion in flash_attn patch by capturing original get_imports before patching

fix: replace flash_attn mock with get_imports patch to properly disable FA2 for MiniCPM4.1-8B and Nemotron Parse

fix: use correct AutoModelForImageTextToText+AutoProcessor for MiniCPM-V-4.6, eager attn for Core Router

fix: use AutoModelForCausalLM for vision, add sdpa attn for Core Router, add open_clip_torch dep

fix: correct vision model dtype, Nemotron Parse model ID, and patch missing is_torch_fx_available

fix: implement lazy loading for Core Router to prevent Space startup timeout

fix: add tokenizers>=0.21.0 to resolve ModelWrapper deserialization error in MiniCPM tokenizers

fix: correct MiniCPM-V-4.6 model ID and switch Core Router to AutoModel to resolve ModelWrapper crash

fix: resolve Core Router crash by removing attn_implementation override on MiniCPM4.1-8B

feat: add Parallel Vision Engine — MiniCPM-V and Nemotron Parse fire async with confidence-score arbiter

feat: add OpenBMB Core Router (MiniCPM4.1-8B) and wire it to Gradio UI

fix: update chatbot output to use strict Gradio 6 dictionary format

fix: add __spec__ to the mocked flash-attn module

fix: bypass flash-attn compilation by mocking it and using sdpa

fix: add flash-attn and einops to requirements

fix: downgrade transformers to 4.44.2 to fix MiniCPM compatibility

fix: pass HF_TOKEN to access gated MiniCPM-V model

fix: move vision tool import to top level for ZeroGPU startup scan

feat: Integrate MiniCPM-V 2.6 OCR model to vision tool

feat: Fix Gradio warnings and implement ZeroGPU startup hook for vision pipeline

Custom UI

initial commit