Repository navigation
Run :tools in CI, filter llama.cpp debug logs, require elixir_make 0.10.1 - #114
Merged
Merged
Conversation
…10.1 - CI pins Qwen3.5-0.8B-UD-Q4_K_XL and runs --include tools on CPU. - The NIF installs a level-filtering llama/ggml log callback, :info by default, with LlamaCppEx.set_log_level/1 and config :llama_cpp_ex, log_level. - elixir_make ~> 0.10.1: newer OTP patches cap httpc headers at 10 KB. - Lock: decimal 3.2.0 / ecto 3.14.2 for two decimal advisories.
This was referenced Oct 31, 2026
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Follow-ups from v0.8.63.
CI coverage. CI now downloads a pinned
unsloth/Qwen3.5-0.8B-GGUFQwen3.5-0.8B-UD-Q4_K_XL.gguf(559 MB; sha256 identical to the file the tests were written on) and adds--include toolsto the inference run, plus a "Tools tag runs alone" step. On the CPU backend locally, all 11 tool tests pass in 16 s.Log noise. llama.cpp's default log callback writes every level, DEBUG included, to stderr. The NIF now installs its own when it loads. It filters by level,
:infoby default (everything as before minus DEBUG), handlesCONTlines by their message's level, and also sets the chat-template logger's threshold. NewLlamaCppEx.set_log_level/1(:debug | :info | :warning | :error | :none) andconfig :llama_cpp_ex, log_level:, applied at application start. Measured on a 100-token Qwen3.5 tool call::debugprints 1,663 stderr lines (41 "awaiting trigger"), default 199 (0), and:warningand:noneprint 0 when set beforeinit.elixir_make notice.
Invalid option {max_header_size,65536} ignoredis httpc on OTP patch releases that predate the option. OTP 29.0.6, 28.5.0.6 and 27.3.4.17 added it with a 10,240-byte default, and elixir_make 0.10.1 (elixir-lang/elixir_make#98) raises it because GitHub redirects can exceed that. Our requirement was~> 0.8, which let users resolve an elixir_make whose download can fail on current OTP, so it is now~> 0.10.1. Our redirect currently measures 5.6 KB, but that depends on the signed URL. README documents the notice.Also: lock moves to decimal 3.2.0 and ecto 3.14.2, clearing two
decimaladvisories thatmix deps.getreported (both optional deps).Locally (Metal): default 436 passed; smoke+embeddings+slow+decision+tools on CI's models + Qwen3.5 601 passed; credo, dialyzer, format clean.