PanGu-Coder

model paper
Huawei
Code generation models based on PanGu-alpha architecture. PanGu-Coder uses CLM+MLM two-stage training. PanGu-Coder2 (July 2023) applies RRTF on StarCoder-15B, reaching 62.20% pass@1 on HumanEval.
Paper (arXiv)GitHubHuggingFace

Outputs 3

PanGu-Coder

model

Variants

Name Parameters Notes
PanGu-Coder-317M 317M —
PanGu-Coder-2.6B 2.6B —

PanGu-Coder: Program Synthesis with Function-Level Language Modeling

paper
Paper (arXiv)

arXiv: 2207.11280

PanGu-Coder2: Boosting LLMs for Code with Ranking Feedback

paper
Paper (arXiv)

arXiv: 2307.14936

code-generation

Related