Zero One EverythingAnnounced todayOpen Source Yi-Coder series models,This model is one of the Yi series modelsProgramming Assistant.
The Yi-Coder series of models are designed for coding tasks.Two parameters are available: 1.5B and 9B.Among them, Yi-Coder-9B’s performance is said to be “better than other models with parameters below 10B”, such as CodeQwen1.5 7B and CodeGeex4 9B, and can even be “comparable to DeepSeek-Coder 33B”.
It is reported that Yi-Coder can handle up to 128K tokens of context,Suitable for understanding and generating complex project-level codes. Yi-Coder also supports 52 major programming languages.
'java', 'markdown', 'python', 'php', 'java', 'c++', 'c#', 'c', 'type', 'html', 'go', 'java_server_pages', 'dart ', 'objective-c', 'kotlin', 'tex', 'swift', 'ruby', 'sql', 'rust', 'css', 'yaml', 'matlab', 'lua', 'json ', 'shell', 'visual_basic', 'scala', 'rmarkdown', 'pascal', 'fortran', 'haskell', 'assembly', 'perl', 'julia', 'cmake', 'groovy', 'ocaml', 'powershell', 'elixir', 'clojure', 'makefile', 'coffee', 'erlang', 'lisp', 'toml', 'batchfile', 'cobol', 'dockerfile', 'r', 'prolog', 'verilog '
According to the official test of Zero One Everything, the pass rates of Yi-Coder-9B-Chat in HumanEval and MBPP reached 85.4% and 73.8% respectively. In addition, the accuracy of Yi-Coder 9B in CRUXEval-O reached more than 50% for the first time among open source Coder models.
Yi-Coder is also strong in code editing and completion capabilities, and achieved excellent results in the CodeEditorBench evaluation. In terms of long sequence modeling and mathematical reasoning capabilities, Yi-Coder also demonstrated strong performance, whether in long text understanding or mathematical problem solving.
Project address:https://github.com/01-ai/Yi-Coder