EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Lobsters·Jul 27·5 min read
Seriously, what is the large code-model even for?
TL;DR
The large code model (mcmodel=large) enables the creation of massive binaries without size or distance assumptions, but performance issues hinder its practical use. Despite its theoretical advantages, developers often avoid it due to significant performance penalties.
✦ Why It Matters
Evaluate your current projects to see if the large code model could be beneficial despite performance concerns.
Key Takeaways
- Understand the limitations of mcmodel=large for TLS.
- Explore alternative methods for managing large binaries.
- Test binary sizes beyond 2GiB to identify potential issues.
- Engage in discussions about improving ABI specifications.
How It Works
The large code model allows for larger binaries by avoiding assumptions about address sizes. It replaces 32-bit references with 64-bit ones for certain data types, enabling larger addressable space.
⚠ The Catch
Despite its capabilities, mcmodel=large cannot handle thread-local storage effectively due to the reliance on 32-bit relocation types, which limits its practical application.