server : fix crash when using verbose output with input tokens that a…

too many issues without BOS (+1 squashed commits)

too many issues without BOS

Update build.yml for Windows Vulkan builder to use Vulkan 1.4.304 SDK…

only print bos alert in debug

streamline output console log (+1 squashed commits)

streamline output console log

gemma3 now works correctly

sd lora from url

gemma3 vision works, but is using more tokens than expected - may nee…

llama.swiftui : fix xcframework dir in README [no ci] (ggml-org#12353)

sycl : variable sg_size support for mmvq kernels (ggml-org#12336)

add more perf stats

CUDA/HIP: refractor mmqv to unify the calculation of nwarps and rows …

fixed ggml common path in metal build

added draft results

updated lite

Merge branch 'upstream' into concedo_experimental

server : add speculative decoding presets for FIM (ggml-org#12287)

remove some unhelpful warnings

allow remote saving to google drive

add GGML_HIP_ROCWMMA_FATTN

honor add bos token settings from metadata

reenable k80

use actual null instead of string "null" for finish_reason in openai …

Merge branch 'upstream' into concedo_experimental

update function-calling.md w/ template override for functionary-small…

ci : add fetch-depth to xcframework upload (ggml-org#12195)

try to avoid fp16 for anything below cc 610