L

Llm Compiler 13b

Developed by facebook
LLM Compiler is an advanced LLM based on Code Llama, specifically designed for code optimization and compiler reasoning tasks
Downloads 107
Release Time : 6/4/2024

Model Overview

A language model focused on compiler optimization, capable of predicting optimization effects, automatically tuning code size, and disassembling into intermediate representations

Model Features

Compiler Behavior Simulation
Perfectly simulates compiler behavior in 20% of cases, predicting optimization effects
Code Size Optimization
Achieves 5.24% improvement in code size compared to GPT-4 Turbo
Disassembly Capability
Decompiles assembly code into LLVM-IR with a BLEU score of 0.96, far surpassing GPT-4 Turbo's 0.43
Multi-Architecture Support
Supports multiple assembly languages including LLVM-IR, x86_64, ARM, and CUDA
Long Context Handling
16,000 token context window, suitable for processing complex code

Model Capabilities

Compiler optimization prediction
Automatic code tuning
Assembly decompilation
LLVM-IR generation
Multi-architecture code analysis

Use Cases

Compiler Development
Optimization Pass Prediction
Predicts the best combination of optimization passes for the LLVM optimizer
Significantly reduces manual tuning time
Compiler Behavior Simulation
Simulates compiler responses to different optimization passes
Perfectly matches real compiler output in 20% of cases
Code Optimization
Code Size Optimization
Automatically finds optimization solutions to minimize code size
5.24% improvement over GPT-4 Turbo
Reverse Engineering
Disassembly
Converts x86_64 or ARM assembly into LLVM-IR
BLEU score of 0.96, far surpassing GPT-4 Turbo
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase