Skip to content

Release v0.8.64 - #117

Merged
nyo16 merged 2 commits into
masterfrom
release-v0.8.64
Nov 1, 2026
Merged

nyo16 merged 2 commits into
masterfrom
release-v0.8.64

Conversation

@nyo16

@nyo16 nyo16 commented Oct 31, 2026

Copy link
Copy Markdown
Owner

Release v0.8.64. llama.cpp stays at 418f905a4.

Ships what has been merged since v0.8.63 (#113, #114): LlamaCppEx.set_log_level/1 / config :llama_cpp_ex, log_level: with llama.cpp DEBUG filtered by default, :tools in CI, elixir_make ~> 0.10.1, req ~> 0.7, and lock updates (decimal 3.2.0, ecto 3.14.2). #115 is docs only and not in the Hex package.

Also fixes a test flake that failed #115's Inference smoke (seed 489492): ServerSmokeTest's telemetry collector was a global handler and took the previous test's server's :done (39 prompt tokens, max_tokens), which that server emits after replying and after its test has already exited. The collectors in server_smoke_test.exs and the telemetry test in llama_cpp_ex_test.exs now pin metadata.server. Test only.

Local matrix (M1 Max, Metal) is in the CHANGELOG entry: default 436; release-guide set 585; CI's set 601; embeddinggemma 447; Clef 440; :tools Qwen3.5 447 (Metal and CPU), Llama-3.2 10/11 tool tests (known); mtp_sidecar 442 + 6 skipped; Hex source build on CPU compiles at 418f905a4.

nyo16 added 2 commits October 31, 2026 18:56
A telemetry handler is global. ServerSmokeTest's collector took the first
:done from any server, and the previous test's server, still alive after
its test exited, emitted :done for its last request after replying. CI
(seed 489492) failed 'halting a stream early' on that stale event.
@nyo16
nyo16 merged commit 4f47d41 into master Nov 1, 2026
9 checks passed
@nyo16
nyo16 deleted the release-v0.8.64 branch November 1, 2026 00:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant