try to avoid fp16 for anything below cc 610

Merge remote-tracking branch 'jg/cuda-fa-mma-17' into debug4

Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	README.md
#	examples/imatrix/README.md
#	scripts/compare-llama-bench.py

trying new ubuntu for ci

ensure scale before rep pen

allow ssl with remote tunnel

fix for chat templates and drafting

fixed another tts bug, clblast selection and quiet mode

do another patch release for the new deepseek models

allow smaller gguf