Conversation
tattn
force-pushed
the
fix/ci
branch
2 times, most recently
from
August 25, 2026 14:52
6758eab to
54be778
Compare
…ending tokens in LlamaClient
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
CI has been red on
mainsince the Gemma 4 merge. This fixes the three failing jobs: the Llama tests timed out, the iOS example build had no matching destination, and the Ubuntu build never compiled.Changes
llama backend
n_threads = 1.textStream(from:)discards tokens left pending by a generation that ended early. Appending the next prompt on top of them wrote past the batch allocation (Batch.swift:20: Fatal error: Unexpectedly found nil). Covered bydecodeAfterCancelledGeneration().Package.swift
LocalLLMClientLlamaCnow share their C/C++ settings. The Linux copy was missing-UDEBUG,LLAMA_BUILD_NUMBERand thecommonheader path, soconstexpr bool DEBUGinmtmd-audio.cppfailed to compile.utils.hgets#pragma once. It is both listed in the module map and included byLocalLLMClientLlamaC.h, which definedllm_chat_paramstwice on Linux.Tests
toolCallWithMultipleTools()into one test per behaviour; it packed two tool round-trips into a single time limit.seed: 0,temperature: 0.1) so tool calling is deterministic.Workflow
GGML_METAL_DEVICES=0. A hosted runner only has the Apple Paravirtual device, which reportssimdgroup reduction = false, so llama.cpp runs every matmul on the CPU anyway while still paying for the backend split and the Metal shader compilation.generic/platform=iOS Simulator. The pinnedOS=26.4does not exist on the runner image, which ships 26.4.1..tar.gz; the.zipasset no longer exists. Addedcurl -fso a future rename fails at the download.Test plan
swift:6.1container, since this job sits behindneeds: test-macos.