philbert440/Qwen3.6-27B-Uncensored-Aggressive-NVFP4

Fix tokenizer: drop leftover truncation block, restore upstream tokenizer.json/tokenizer_config.json (fixes vision >1024 tokens; see discussion #3)

Standardize model card

packed NVFP4 MTP head variant (hedge for SM70)

packed NVFP4 MTP head variant (hedge for SM70)

packed NVFP4 MTP head variant (hedge for SM70)

NVFP4A16 compressed-tensors + bf16 MTP head; thinking-mode calibration

initial commit