jose-compu / jose-compu/funes.cpp
[Tests] Add performance benchmarks for memory operations
- Dominant language
- C++
- Stars
- 1
- Forks
- 0
- PR merge metrics
- No merged PRs in 30d
Description
## Summary
Add performance benchmarks to detect regressions in memory-critical operations.
## Current State
- No dedicated performance tests for memory operations
- Performance issues only caught manually
## Benchmark Categories
1. **Allocation Performance**
- Graph allocation time vs graph size
- Buffer reuse efficiency
- Multi-buffer allocation overhead
2. **Memory State Operations**
- `seq_rm`, `seq_cp`, `seq_add` performance vs sequence length
- State serialization/deserialization speed
- Cache clearing performance
3. **Semantic Memory**
- Embedding generation throughput
- HNSW search latency vs index size
- Memory file load/save performance
## Implementation
- Use `tests/test-quantize-perf.cpp` as template
- Add to CI with performance regression detection
- Store baseline results in `tests/baselines/`
## Acceptance Criteria
- [ ] Create `tests/test-memory-perf.cpp`
- [ ] Benchmark allocation performance
- [ ] Benchmark state operations
- [ ] Add semantic memory benchmarks
- [ ] Document performance characteristics
## Related Files
- `tests/test-quantize-perf.cpp`
- `src/llama-memory*.cpp`
- `tools/longmemeval-mini/`
Contributor guide
Assessment
This issue has not been assessed yet.