microPost: Coding Models at Ollama.
Or: Listing the Tools for Automated Code Generation.
This is a μPost (microPost) lists the coding models, available from Ollama.ai, that I will evaluate in 2024 as I look to define a Full Stack App Development Process.
| Install Command | Description | License |
ollama run codellama | 7B: A large language model that can use text prompts to generate and discuss code. | Llama 2 Community License Agreement. |
ollama run wizardcoder | 13B: Llama-based code generation model focused on Python. | Llama 2 Community License Agreement. |
ollama run codeup | 13B: Great code generation model based on Llama2. | Llama 2 Community License Agreement. |
ollama run open-orca-platypus2 | 13B: Merge of the Open Orca OpenChat model and the Garage-bAInd Platypus 2 model. Designed for chat and code generation. | Llama 2 Community License Agreement. |
ollama run phind-codellama | 34B: Code generation model based on CodeLlama. | Llama 2 Community License Agreement. |
ollama run codebooga | 34B: A high-performing code instruct model created by merging two existing code models. | Llama 2 Community License Agreement. |
| . | ||
ollama run dolphin-mistral | 7B: The uncensored Dolphin model based on Mistral that excels at coding tasks. Updated to version 2.6. | Apache License 2.0. |
ollama run openhermes2.5-mistral | 7B: OpenHermes 2.5 Mistral is a Mistral 7B fine-tune, a continuation of OpenHermes 2 model, which trained on additional code datasets. | Apache License 2.0. |
ollama run sqlcoder | 15B: SQLCoder is a code completion model fined-tuned on StarCoder for SQL generation tasks. | Apache License 2.0. |
ollama run starcoder | 15.5B: StarCoder is a code generation model trained on 80+ programming languages. | Apache License 2.0. |
ollama run dolphin-mixtral | 56B: An uncensored, fine-tuned model based on the Mixtral mixture of experts model that excels at coding tasks. Created by Eric Hartford. | Apache License 2.0. |
ollama run deepseek-coder | 6.7B: DeepSeek Coder is trained from scratch on both 87% code and 13% natural language in English and Chinese. Each of the models are pre-trained on 2 trillion tokens. | MIT License. |
ollama run magicoder | 7B: The Magicoder is a family of models trained on 75K synthetic instruction data using OSS-Instruct, a novel approach to enlightening LLMs with open-source code snippets. | MIT License. |
Ollama.ai also hosts chat models, medical models, and others.