Releases
v0.10.19
Compare
Sorry, something went wrong.
No results found
Features
agent: report Linux host capacity on the MetalNode (#2541 ) (2a28f21 )
cli: metal token create names its target cluster and warns on a loopback server or a shared namespace (#2555 ) (4b6c2e2 )
export InferenceService concurrency limit, context length, GPUs and config; allow parallelSlots above 64 for vLLM (#2576 ) (ffbdf4f )
metal-agent: start on Linux hosts, experimentally (#2542 ) (c5740dc )
metal: check free disk space before a model download and report the model store size (#2548 ) (3ddfdc1 )
metal: report controller and agent version skew (#2547 ) (c240496 )
router: add spec.backendSelector to ModelRouter (#2572 ) (f07134c )
router: forward other POST /v1 paths to the model the request names (#2544 ) (0b9babc )
router: label per-endpoint metrics with pod and node, and count session bindings created (#2574 ) (79149ea )
router: per-pod KV, slot-utilization, running and waiting histograms, with a failed-scrape counter (#2575 ) (5952b32 )
ship third-party license and NOTICE files with release archives and images (#2568 ) (2ac4c44 )
Bug Fixes
cli: metal token create follow-ups from the #2236 review (#2355 ) (61ba31b )
controller: keep a Metal agent's refusal over a stale slice heartbeat (#2556 ) (6442fd5 )
controller: roll past a waitForIdle service whose pod is crash-looping (#2557 ) (d64655f )
metal: keep the engine while its Model is deleted and re-created (#2546 ) (b33a3e5 )
metal: route the Mac client proxy by the model a request names (#2543 ) (7eda9a7 )
metal: size a downloaded hybrid GGUF from its file's header (#2551 ) (30aad65 )
model: record the GGUF layer layout on the Model status (#2559 ) (e58aaa4 )
use a memory-conscious default context for Llama 3.1 8B (#2461 ) (3f91c65 )
Documentation
restore the standard Apache 2.0 wording in LICENSE clause 9 (#2563 ) (e18be27 )
You can’t perform that action at this time.