Atlas

Models

← All models

Code Llama Instruct 7B

Meta · 2023-08-24 · 6.7B parameters

Code Llama - Instruct 7B is the instruction-tuned 7-billion-parameter checkpoint from Meta's August 2023 Code Llama release, intended for natural-language-driven code generation and safer assistant use. It is fine-tuned from the long-context Code Llama 7B on about 5B tokens of instruction, self-instruct coding and safety data, and keeps the base model's infilling ability and 100,000-token inference support.

Benchmark scores

BenchmarkScore
Capability93.2
HumanEval34.8
HumanEval (pass@10)64.3
HumanEval (pass@100)88.1
MBPP44.4
MBPP (pass@10)65.4
MBPP (pass@100)76.8
MultiPL-E HumanEval (Bash)10.1
MultiPL-E HumanEval (C#)21.6
MultiPL-E HumanEval (C++)31.1
MultiPL-E HumanEval (Java)30.4
MultiPL-E HumanEval (PHP)28.6
MultiPL-E HumanEval (TypeScript)32.7
ToxiGen - toxic generation rate0.0
TruthfulQA generation - % informative93.6
TruthfulQA generation - % truthful37.0
TruthfulQA generation - % truthful and informative31.5
Loading Atlas data…