Architecture Decision Records

This directory contains Architecture Decision Records (ADRs) for the LLM project.

What is an ADR?

An Architecture Decision Record (ADR) is a document that captures an important architectural decision made along with its context and consequences.

Format

Each ADR follows this structure:

# [Number]. [Title]

Date: YYYY-MM-DD

## Status

[Proposed | Accepted | Deprecated | Superseded]

## Context

What is the issue that we're seeing that is motivating this decision or change?

## Decision

What is the change that we're proposing and/or doing?

## Consequences

What becomes easier or more difficult to do because of this change?

Index

• ADR-001: Use Grouped Query Attention (GQA)
• ADR-002: Use SwiGLU Activation Function
• ADR-003: Use Pre-Allocated KV Cache
• ADR-004: Paged Attention Serving Integration
• ADR-005: Export Registry Parity with BACKEND_REGISTRY
• ADR-006: Checkpoint Format Unification
• ADR-007: GPTQ Integration Architecture
• ADR-008: Mixed-Precision GPTQ via Atomic LayerQuantPolicy
• ADR-009: AWQ Integration
• ADR-010: SmoothQuant Integration
• ADR-011: GGUF Export Format
• ADR-012: Use QLoRA for Memory-Efficient Fine-Tuning

Creating a New ADR

1. Copy the template: cp template.md 00X-title.md
2. Fill in the sections
3. Update this index
4. Submit a PR