feat(distributed): multi-host capability exchange + distributed speculative decoding (rebase of #105 onto main)#157

Merged

cursor[bot] merged 10 commits into

mainfrom

AgentMemory/distributed-spec-decode-rebased-2815

Jun 19, 2026

Commits on Jun 10, 2026

v0.5-M1: capability exchange platform + distributed spec decode (ADR 0009)

cursoragent
and
FluffyAIcode
committed
demo: default to no-thinking template for echo-style spec decode demo

cursoragent
and
FluffyAIcode
committed

Commits on Jun 19, 2026

Merge remote-tracking branch 'origin/AgentMemory/v05-mlx-distributed-capability-exchange-b876' into AgentMemory/distributed-spec-decode-rebased-2815

cursoragent
and
FluffyAIcode
committed
fix(proto): regenerate distributed gRPC stub with current grpcio-tools (1.81.0->1.81.1)

cursoragent
and
FluffyAIcode
committed
test(distributed): add run_distributed_demo.sh + mlx-distributed-spec-decode-demo bridge preset

cursoragent
and
FluffyAIcode
committed
fix(distributed-demo): make apply_chat_template token-id coercion transformers 4.x AND 5.x safe

cursoragent
and
FluffyAIcode
committed
fix(ci): integration workflow installs requirements.txt, not 'pip install -e .'

cursoragent
and
FluffyAIcode
committed
fix(distributed): use apply_chat_template(tokenize=True, return_dict=False) for cross-version token ids

cursoragent
and
FluffyAIcode
committed
perf(distributed): add bench (throughput / bounded-KV / gRPC RTT) + mlx-distributed-spec-decode-bench preset

cursoragent
and
FluffyAIcode
committed
fix(bridge): register int:rtt_samples param kind for the distributed bench preset

cursoragent
and
FluffyAIcode
committed