Granite

Generates syntactically correct and semantically meaningful code across 116 programming languages by leveraging a unified decoder-only transformer architecture trained on 3-4 trillion tokens of language-agnostic code data during Phase 1, followed by mixed code-language training in Phase 2. The model learns cross-language patterns and idioms through exposure to diverse codebases, enabling it to generate contextually appropriate code regardless of target language without language-specific tokenizers or specialized heads.

Fine-tunes base models on instruction datasets derived from Git commits paired with human-written instructions and synthetically generated code instruction data, enabling the model to follow natural language directives for code modification tasks. The instruction tuning process leverages commit messages as implicit task descriptions and diffs as ground-truth code transformations, teaching the model to understand intent-driven code changes rather than just pattern completion.

Performs targeted code edits and refactoring operations (e.g., extract function, rename variables, restructure logic) while preserving code semantics and functionality. The model understands code structure and intent well enough to make surgical edits without breaking functionality, leveraging semantic understanding developed during training on diverse codebases.

Generates contextually appropriate code completions by leveraging surrounding code context and, within context window limits, multi-file context to understand project structure and dependencies. The model uses attention mechanisms to identify relevant code patterns from the context window and generate completions that align with existing code style, naming conventions, and architectural patterns.

Implements a multi-stage data processing pipeline that filters, deduplicates, and sanitizes code training data through exact and fuzzy deduplication, PII redaction (replacing sensitive information with tokens), ClamAV malware scanning, and content filtering to reduce harmful code generation. This pipeline ensures training data complies with enterprise security and compliance requirements while maintaining code quality and diversity.

Provides four parameter-size variants (3B, 8B, 20B, 34B) each with configurable context windows (2K, 4K, 8K tokens), enabling deployment across diverse hardware constraints from edge devices to data centers. The model family uses a unified architecture with consistent tokenization and training methodology, allowing seamless model swapping without retraining or prompt engineering changes.

Generates natural language explanations of code functionality, purpose, and behavior by leveraging the model's understanding of code semantics learned during Phase 2 training (80% code + 20% language mixture). The model can produce docstrings, comments, and high-level summaries by conditioning on code input and generating corresponding natural language output.

Identifies and fixes common code bugs by leveraging semantic understanding of code patterns learned during training on diverse codebases. The model can detect logical errors, missing error handling, type mismatches, and resource leaks by conditioning on buggy code and generating corrected versions, without explicit bug detection rules or static analysis.

Translates code from one programming language to another while preserving logic and intent by leveraging cross-language patterns learned during training on 116 languages. The model maps language-specific idioms, APIs, and syntax to equivalent constructs in the target language, enabling semantic-preserving code translation without explicit language-to-language mapping rules.

Supports fine-tuning of base models on custom code datasets to specialize the model for domain-specific code patterns, internal coding standards, or proprietary languages. The fine-tuning process leverages the pre-trained weights as initialization, enabling efficient adaptation to new domains with limited computational overhead compared to training from scratch.

Released under Apache 2.0 license with full model weights available for download, enabling unrestricted commercial and research use without API dependencies or vendor lock-in. Organizations can deploy models on-premises, in private clouds, or on any infrastructure without licensing restrictions or usage monitoring.

Developed according to IBM's AI Ethics principles with explicit focus on reducing harmful code generation, bias in recommendations, and ensuring responsible AI deployment. The training data curation pipeline includes content filtering to reduce harmful code patterns and PII redaction to prevent sensitive information leakage, embedding ethical considerations into the model architecture rather than as post-hoc guardrails.