Text Generation ModelDeprecated

Code Llama

Code Llama is a 34-billion-parameter instruction-tuned model from Meta built for code generation, comprehension, and debugging.

Start Building with Code Llama View All Models

Publisher

Instruction-tuned code generation from Meta

Code Llama is a code-specialized model developed by Meta, built on the LLaMA 2 foundation and fine-tuned specifically for programming tasks. The 34-billion-parameter instruct variant is designed to follow user instructions accurately, making it suited for tasks like generating code from natural language descriptions, explaining existing code, and assisting with debugging across a range of programming languages.

The instruct tuning distinguishes this variant from base Code Llama models by making it more responsive to direct commands and conversational prompts rather than requiring raw completion-style inputs. It operates with a 4,096-token context window and is hosted via DeepInfra on MindStudio. The model is noted as deprecated in the current catalog, so developers building new applications may want to evaluate whether a more recently released variant better fits their needs.

Capabilities

What Code Llama supports

Code Generation

Generates programming code from natural language instructions across multiple languages. The instruct tuning helps it respond accurately to specific coding prompts.

Instruction Following

Fine-tuned to follow explicit user commands rather than relying on completion-style prompting. This makes it suitable for chat-based coding assistants and interactive workflows.

Code Debugging

Assists in identifying and explaining errors in existing code. Users can paste a snippet and ask for diagnosis or a corrected version.

Code Comprehension

Explains what a block of code does in plain language, useful for documentation or onboarding. Supports analysis within the 4,096-token context window.

Text Generation

Operates as a standard chat-style LLM capable of producing natural language responses alongside code. Maximum response size is 2,500 tokens.

Ready to build with Code Llama?

Get Started Free

FAQ

Common questions about Code Llama

What is the context window for Code Llama 34B Instruct?

The model has a context window of 4,096 tokens, which covers both the input prompt and the generated response.

What is the maximum response length?

The maximum response size is 2,500 tokens per generation.

Is this model still actively supported?

This specific variant (codellama-34b-instruct) is marked as deprecated in the MindStudio catalog. Developers starting new projects should check whether a newer model better suits their requirements.

What makes the instruct variant different from the base Code Llama model?

The instruct variant has been fine-tuned to follow explicit user instructions more accurately, making it better suited for conversational and command-driven coding tasks compared to the base completion-style model.

Does this model support image or video inputs?

No. Code Llama 34B Instruct is a text-only model and does not support image or video analysis.

Who published Code Llama and under what license?

Code Llama was published by Meta. It is released under Meta's custom Llama 2 community license, which permits commercial use subject to certain conditions outlined in the license agreement.

Resources