TheQweaker/mdlm-owt-noflash

docs: link candle-mi (github) at its first mention too

docs: clarify no tokenizer shipped (+gpt2 link), framework-agnostic note, align param count

Card: determinism note, official-noflash disambiguation, candle-mi cross-check, torch>=2.3

Pin SDPA math backend (bit-reproducible fp32 oracle); device-aware rotary cache; drop dead dropout

Fix non-embedding param count (~92M) and drop unsourced token-count claim

Upload folder using huggingface_hub

initial commit