vcruz305/GLM-5.3-Flash-EXL3-K2

Card: measured long-context ceiling (verified to 163k prompt tokens; 262k boots)

Card: KLD vs BF16 on the fidelity suite (512 contexts)

Card: KLD vs BF16 on the fidelity suite (512 contexts)

Card: sixcat on the fixed runtime

Card: point at the fixed runtime

Card: remove encode/provenance details

Point at the prebuilt spark-vllm runtime; refresh ctx ladder and DFlash rows

docs: add Spark GB10 recipe, MTP k=2 measurements, fix YAML license

Publish verified TP4 L40 runtime and serving qualification

Document completed verified EXL3 K2 artifact

Document completed verified EXL3 K2 artifact

Upload folder using huggingface_hub

initial commit