TitanPythons/Qwen3.5-VL-9B-8bit-MLX-CRACK

Add Ko-fi support section

Add vMLX promotional banner to README

Add vMLX promotional banner to README

Upload model.safetensors.index.json with huggingface_hub

Upload tokenizer_config.json with huggingface_hub

Update: increase repetition_penalty to 1.3, add presence_penalty 1.5 per Qwen recommendation

Fix: disable thinking by default to prevent repetition loops on small models

Fix: disable thinking by default to prevent repetition loops on small models

Fix enable_thinking in tokenizer_config: match Qwen original (thinking ON by default for 4B+)

Fix enable_thinking: match Qwen original (thinking ON by default for 4B+)

Fix: inject chat_template with enable_thinking support into tokenizer_config.json

Fix standalone/spec-decoding positioning in README

Fix README: clarify standalone vs spec-decode role, remove redundant phrasing

Fix: remove incorrect draft model references (9B is standalone)

Fix: default thinking OFF to prevent CoT loop on CRACK models (F13-DENSE)

Upload generation_config.json with huggingface_hub

Upload config.json with huggingface_hub

Upload folder using huggingface_hub

initial commit