Codegemma 2B is an open-source language model by google. Features: 2b LLM, VRAM: 5.1GB, Context: 8K, License: gemma, LLM Explorer Score: 0.21, Arc: 25.7, HellaSwag: 39.6, MMLU: 28.3, GSM8K: 4.6, HumanEval: 27.3.
Codegemma 2B Parameters and Internals
| Model Type | | text-to-text, text-to-code |
|
| Use Cases |
| Areas: | | Generative AI, Code-related tasks |
|
| Applications: | | IDE extension for code, Interactive code learning, Code conversation |
|
| Primary Use Cases: | | Code completion, Code generation, Instruction following |
|
| Limitations: | | Intrinsic LLM limitations |
|
| Considerations: | | Refer to the Gemma model card for evaluation results. |
|
|
| Additional Notes | |
| Supported Languages | | English (high proficiency, specifically for code-related tasks) |
|
| Training Details |
| Data Sources: | | publicly available code repositories, open source mathematics datasets, synthetically generated code |
|
| Data Volume: | |
| Methodology: | | FIM tasks with 80% FIM rate, 50-50 PSM/SPM mode |
|
| Hardware Used: | |
| Model Architecture: | |
|
| Safety Evaluation |
| Methodologies: | | Structured evaluations, Internal red-teaming testing |
|
| Findings: | | Acceptable thresholds for child safety, content safety, representational harms, memorization, large-scale harms |
|
| Risk Categories: | | Representational harms, Content safety |
|
| Ethical Considerations: | | Deferred to Gemma model card details |
|
|
| Responsible Ai Considerations |
| Fairness: | | Evaluated through human evaluation on prompts |
|
| Transparency: | | Detailed in Gemma model card |
|
| Accountability: | |
| Mitigation Strategies: | | Various evaluations and policy adherence as seen in Gemma model details |
|
|
| Input Output |
| Input Format: | | code prefix/suffix for pretrained, text for instruction-tuned |
|
| Accepted Modalities: | |
| Output Format: | | code completion or generation |
|
| Performance Tips: | | Avoid extra spaces around tokens for completion. |
|
|
| Release Notes |
| Version: | |
| Notes: | | Fast code completion variant. |
|
| Version: | |
| Notes: | | Specialized in code completion and generation. |
|
| Version: | |
| Notes: | | Instruction tuned for chat and instruction-following. |
|
|
|
Quantized Models of the Codegemma 2B
Best Alternatives to Codegemma 2B
Note: green Score (e.g. "73.2") means that the model is better than google/codegemma-2b.