Maintainer: microsoft

Total Score


Last updated 6/5/2024


Model LinkView on HuggingFace
API SpecView on HuggingFace
Github LinkNo Github link provided
Paper LinkNo paper link provided

Get summaries of the top AI models delivered straight to your inbox:

Model Overview

wavecoder-ultra-6.7b is a large language model (LLM) developed by Microsoft for the coding domain. It is part of the WaveCoder series of models designed to solve relevant problems in the field of code through instruction-following learning. The model was trained on a dataset generated from a subset of the Code Search Net data using a generator-discriminator framework based on LLMs, covering four general code-related tasks: code generation, code summary, code translation, and code repair.

The wavecoder-ultra-6.7b model demonstrates strong performance on benchmarks like HumanEval, MBPP, and HumanEval Fix and Explain tasks, outperforming GPT-4 in some cases. It is a larger and more capable variant in the WaveCoder series, with the base wavecoder-ds-6.7b and wavecoder-pro-6.7b models also available.

Similar models such as deepseek-coder-6.7b-instruct, Magicoder-S-DS-6.7B, and deepseek-coder-6.7b-base also focus on code-related tasks, with various approaches and training data.

Model Inputs and Outputs

The wavecoder-ultra-6.7b model is a text-to-text transformer that can be used for a variety of code-related tasks. It takes natural language instructions or prompts as input and generates the corresponding code or code-related output.


  • Natural language instructions or prompts related to code generation, code summarization, code translation, or code repair.


  • Generated code or code-related text, such as:
    • Code snippets
    • Code summaries
    • Translated code
    • Code fixes or repairs


The wavecoder-ultra-6.7b model is capable of performing a wide range of code-related tasks, including:

  • Code Generation: Given a natural language prompt, the model can generate relevant code snippets in various programming languages.
  • Code Summarization: The model can summarize the functionality of a given code snippet in natural language.
  • Code Translation: The model can translate code from one programming language to another.
  • Code Repair: The model can identify and fix bugs or errors in a given code snippet.

These capabilities are demonstrated by the model's strong performance on benchmarks like HumanEval, MBPP, and HumanEval Fix and Explain tasks.

What Can I Use It For?

The wavecoder-ultra-6.7b model can be useful for a variety of applications and use cases in the software development and programming domains, such as:

  • Automated Code Generation: Generating code snippets from natural language descriptions, which can assist developers in rapid prototyping or coding tasks.
  • Code Documentation and Summarization: Automatically summarizing the functionality of code segments, which can improve code readability and maintainability.
  • Code Translation: Translating code between different programming languages, which can facilitate cross-team collaboration or porting of projects.
  • Code Repair and Debugging: Identifying and fixing bugs or errors in code, which can streamline the debugging process.

These capabilities can be leveraged in tools, services, or applications that require strong code-related AI capabilities, such as code editors, IDEs, developer productivity tools, or even low-code/no-code platforms.

Things to Try

Here are some ideas for things to try with the wavecoder-ultra-6.7b model:

  • Generating Code from Natural Language Prompts: Try providing the model with natural language descriptions of programming tasks or algorithms, and see how it generates the corresponding code.
  • Summarizing Code Functionality: Take a code snippet, provide it as input to the model, and see how it summarizes the functionality of the code in natural language.
  • Translating Code between Languages: Experiment with providing the model with code in one programming language and see how it translates it to another language.
  • Fixing Code Bugs: Give the model a code snippet with known bugs or errors, and observe how it identifies and repairs the issues.

By experimenting with these capabilities, you can gain a deeper understanding of the model's strengths and limitations, and explore how it can be integrated into your own projects or workflows.

This summary was produced with help from an AI and may contain inaccuracies - check out the links to read the original source documents!

Related Models




Total Score


The WizardCoder-15B-V1.0 model is a large language model (LLM) developed by the WizardLM Team that has been fine-tuned specifically for coding tasks using their Evol-Instruct method. This method involves automatically generating a diverse set of code-related instructions to further train the model on instruction-following capabilities. Compared to similar open-source models like CodeGen-16B-Multi, LLaMA-33B, and StarCoder-15B, the WizardCoder-15B-V1.0 model exhibits significantly higher performance on the HumanEval benchmark, achieving a pass@1 score of 57.3 compared to the 18.3-37.8 range of the other models. Model inputs and outputs Inputs Natural language instructions**: The model takes in natural language prompts that describe coding tasks or problems to be solved. Outputs Generated code**: The model outputs code in a variety of programming languages (e.g. Python, Java, etc.) that attempts to solve the given problem or complete the requested task. Capabilities The WizardCoder-15B-V1.0 model has been specifically trained to excel at following code-related instructions and generating functional code to solve a wide range of programming problems. It is capable of tasks such as writing simple algorithms, fixing bugs in existing code, and even generating complex programs from high-level descriptions. What can I use it for? The WizardCoder-15B-V1.0 model could be a valuable tool for developers, students, and anyone working on code-related projects. Some potential use cases include: Prototyping and rapid development of new software features Automating repetitive coding tasks Helping to explain programming concepts by generating sample code Tutoring and teaching programming by providing step-by-step solutions Things to try One interesting thing to try with the WizardCoder-15B-V1.0 model is to provide it with vague or open-ended prompts and see how it interprets and responds to them. For example, you could ask it to "Write a Python program that analyzes stock market data" and see the creative and functional solutions it comes up with. Another idea is to give the model increasingly complex or challenging coding problems, like those found on programming challenge websites, and test its ability to solve them. This can help uncover the model's strengths and limitations when it comes to more advanced programming tasks.

Read more

Updated Invalid Date




Total Score


WizardCoder-33B-V1.1 is a large language model (LLM) developed by the WizardLM team that is trained to excel at code-related tasks. It is based on the DeepSeek-Coder-33B-base model and has been further fine-tuned using the Evol-Instruct method to improve its code generation and understanding capabilities. Compared to previous versions, WizardCoder-33B-V1.1 achieves state-of-the-art performance on several industry-standard benchmarks, outperforming models like ChatGPT 3.5, Gemini Pro, and DeepSeek-Coder-33B-instruct. Model inputs and outputs Inputs Natural language instructions**: The model accepts natural language descriptions of coding tasks or problems that it should solve. Outputs Generated code**: The model's primary output is Python, Java, or other programming language code that attempts to fulfill the given instruction or solve the provided problem. Capabilities WizardCoder-33B-V1.1 demonstrates impressive abilities in generating functional code to solve a wide variety of programming tasks. It achieves 79.9 pass@1 on the HumanEval benchmark, 73.2 pass@1 on HumanEval-Plus, 78.9 pass@1 on MBPP, and 66.9 pass@1 on MBPP-Plus. These results show the model's strong performance compared to other code LLMs, making it a valuable tool for developers and programmers. What can I use it for? The WizardCoder-33B-V1.1 model can be utilized in a range of applications that involve code generation or understanding, such as: Automated code completion and suggestions to assist developers Prototyping and building initial versions of software applications Translating natural language descriptions into working code Educational tools for teaching programming concepts and skills Augmenting human programming workflows to boost productivity Things to try One interesting aspect of WizardCoder-33B-V1.1 is its ability to handle complex, multi-part instructions and generate code that addresses all the requirements. You could try providing the model with detailed prompts involving various coding tasks and see how it responds. Additionally, experimenting with different decoding strategies, such as adjusting the temperature or number of samples, may uncover further nuances in the model's capabilities.

Read more

Updated Invalid Date

AI model preview image



Total Score


WizardCoder-33B-V1.1 is an AI model developed by lucataco that is part of the WizardCoder family. It is an improvement upon the earlier WizardCoder-15B-V1.0 and WizardCoder-34B-V1.0 models, achieving better performance on benchmarks like HumanEval and MBPP. The model is designed to empower code-oriented large language models with the Evol-Instruct technique. Model inputs and outputs WizardCoder-33B-V1.1 takes in a text prompt as input, which can include an instruction for the model to carry out. The model then generates a text response that completes the requested task. Inputs Prompt**: The instruction or text the model should use to generate a response. System Prompt**: A default prompt that helps guide the model's behavior. Temperature**: A parameter that controls how "creative" the model's response will be. Repeat Penalty**: A parameter that discourages the model from repeating itself too much. Max New Tokens**: The maximum number of new tokens the model should generate. Outputs Text Response**: The model's generated text completing the requested task. Capabilities WizardCoder-33B-V1.1 has been shown to outperform several closed-source and open-source models on programming-related benchmarks like HumanEval and MBPP. It can generate original code to complete a wide variety of coding tasks, from writing a simple snake game in Python to more complex programming challenges. What can I use it for? WizardCoder-33B-V1.1 could be used for a range of applications involving code generation, such as aiding software developers, automating certain programming tasks, or even as a starting point for building custom AI applications. The model's strong performance on benchmarks suggests it may be particularly useful for tasks like prototyping, debugging, or generating boilerplate code. Things to try One interesting thing to try with WizardCoder-33B-V1.1 would be to give it increasingly complex or open-ended coding challenges to see how it performs. You could also experiment with adjusting the temperature and repeat penalty parameters to find the sweet spot for your specific use case. Additionally, comparing the model's outputs to those of other code-oriented language models could yield interesting insights.

Read more

Updated Invalid Date




Total Score


WizardLM-70B-V1.0 is a large language model developed by the WizardLM Team. It is part of the WizardLM family of models, which also includes the WizardCoder and WizardMath models. The WizardLM-70B-V1.0 model was trained to follow complex instructions and demonstrates strong performance on tasks like open-ended conversation, reasoning, and math problem-solving. Compared to similar large language models, the WizardLM-70B-V1.0 exhibits several key capabilities. It outperforms some closed-source models like ChatGPT 3.5, Claude Instant 1, and PaLM 2 540B on the GSM8K benchmark, achieving an 81.6 pass@1 score, which is 24.8 points higher than the current SOTA open-source LLM. Additionally, the model achieves a 22.7 pass@1 score on the MATH benchmark, 9.2 points above the SOTA open-source LLM. Model inputs and outputs Inputs Natural language instructions and prompts**: The model is designed to accept a wide range of natural language inputs, from open-ended conversation to specific task descriptions. Outputs Natural language responses**: The model generates coherent and contextually appropriate responses to the given inputs. This can include answers to questions, elaborations on ideas, and solutions to problems. Code generation**: The WizardLM-70B-V1.0 model has also been shown to excel at code generation, with its WizardCoder variant achieving state-of-the-art performance on benchmarks like HumanEval. Capabilities The WizardLM-70B-V1.0 model demonstrates impressive capabilities across a range of tasks. It is able to engage in open-ended conversation, providing helpful and detailed responses. The model also excels at reasoning and problem-solving, as evidenced by its strong performance on the GSM8K and MATH benchmarks. One key strength of the WizardLM-70B-V1.0 is its ability to follow complex instructions and tackle multi-step problems. Unlike some language models that struggle with tasks requiring sequential reasoning, this model is able to break down instructions, generate relevant outputs, and provide step-by-step solutions. What can I use it for? The WizardLM-70B-V1.0 model has a wide range of potential applications. It could be used to power conversational AI assistants, provide tutoring and educational support, assist with research and analysis tasks, or even help with creative writing and ideation. The model's strong performance on math and coding tasks also makes it well-suited for use in STEM education, programming tools, and scientific computing applications. Developers could leverage the WizardCoder variant to build intelligent code generation and autocomplete tools. Things to try One interesting aspect of the WizardLM-70B-V1.0 model is its ability to engage in multi-turn conversations and follow up on previous context. Try providing the model with a series of related prompts and see how it maintains coherence and builds upon the discussion. You could also experiment with the model's reasoning and problem-solving capabilities by presenting it with complex, multi-step instructions or math problems. Observe how the model breaks down the task, generates intermediate steps, and arrives at a final solution. Another area to explore is the model's versatility across different domains. Test its performance on a variety of tasks, from open-ended conversation to specialized technical queries, to understand the breadth of its capabilities.

Read more

Updated Invalid Date