UPSTREAM PR #18276: tools : use common_log_pause to fix fit-params output race

[loci-dev]

Mirrored from ggml-org/llama.cpp#18276

Replaced the std::this_thread::sleep_for(10ms) hack with common_log_pause(common_log_main()) in llama-fit-params.

Reasoning:
The previous 10ms delay was intended to let the async logger drain its queue to stderr before the main thread printed the result to stdout. However, on high-performance or loaded systems, 10ms is not guaranteed to be enough, leading to interleaved output (see Issue #18085).

common_log_pause explicitly waits for the background worker to drain the log queue, ensuring deterministic separation between logs and the final result.

Fixes #18085

[loci-review]

Explore the complete analysis inside the Version Insights

Performance Analysis Summary - PR #669

Analysis Overview

PR #669 modifies a single line in tools/fit-params/fit-params.cpp, replacing a 10ms sleep with common_log_flush(common_log_main()) to fix output race conditions. Performance analysis shows no measurable impact across all analyzed binaries and functions.

Performance Metrics:

• Response Time change: 0% across all functions
• Throughput Time change: 0% across all functions
• Power consumption change: -0.0001% system-wide (maximum -1 nJ in llama-tts)

Code Modification Scope:

• 1 file changed
• 1 line modified
• 0 performance-critical functions affected
• Change isolated to utility tool, not inference path

The modification addresses correctness without affecting computational performance. All core libraries and inference functions remain unchanged.