Pre-configured Models
Overview
Jan provides various pre-configured AI models with different capabilities. Please see the following list for details.
| Model | Description |
|---|---|
| Mistral Instruct 7B Q4 | A model designed for a comprehensive understanding through training on extensive internet data |
| OpenHermes Neural 7B Q4 | A merged model using the TIES method. It performs well in various benchmarks |
| Stealth 7B Q4 | This is a new experimental family designed to enhance Mathematical and Logical abilities |
| Trinity-v1.2 7B Q4 | An experimental model merge using the Slerp method |
| Openchat-3.5 7B Q4 | An open-source model that has a performance that surpasses that of ChatGPT-3.5 and Grok-1 across various benchmarks |
| Wizard Coder Python 13B Q5 | A Python coding model that demonstrates high proficiency in specific domains like coding and mathematics |
| OpenAI GPT 3.5 Turbo | The latest GPT-3.5 Turbo model with higher accuracy at responding in requested formats and a fix for a bug that caused a text encoding issue for non-English language function calls |
| OpenAI GPT 3.5 Turbo 16k 0613 | A Snapshot model of gpt-3.5-16k-turbo from June 13th 2023 |
| OpenAI GPT 4 | The latest GPT-4 model intended to reduce cases of “laziness” where the model doesn't complete a task |
| TinyLlama Chat 1.1B Q4 | A tiny model with only 1.1B. It's a good model for less powerful computers |
| Deepseek Coder 1.3B Q8 | A model that excelled in project-level code completion with advanced capabilities across multiple programming languages |
| Phi-2 3B Q8 | a 2.7B model, excelling in common sense and logical reasoning benchmarks, trained with synthetic texts and filtered websites |
| Llama 2 Chat 7B Q4 | A model that is specifically designed for a comprehensive understanding through training on extensive internet data |
| CodeNinja 7B Q4 | A model that is good for coding tasks and can handle various languages, including Python, C, C++, Rust, Java, JavaScript, and more |
| Noromaid 7B Q5 | A model designed for role-playing with human-like behavior. |
| Starling alpha 7B Q4 | An upgrade of Openchat 3.5 using RLAIF, is good at various benchmarks, especially with GPT-4 judging its performance |
| Yarn Mistral 7B Q4 | A language model for long context and supports a 128k token context window |
| LlaVa 1.5 7B Q5 K | A model can bring vision understanding to Jan |
| BakLlava 1 | A model can bring vision understanding to Jan |
| Solar Slerp 10.7B Q4 | A model that uses the Slerp merge method from SOLAR Instruct and Pandora-v1 |
| LlaVa 1.5 13B Q5 K | A model can bring vision understanding to Jan |
| Deepseek Coder 33B Q5 | A model that excelled in project-level code completion with advanced capabilities across multiple programming languages |
| Phind 34B Q5 | A multi-lingual model that is fine-tuned on 1.5B tokens of high-quality programming data, excels in various programming languages, and is designed to be steerable and user-friendly |
| Yi 34B Q5 | A specialized chat model is known for its diverse and creative responses and excels across various NLP tasks and benchmarks |
| Capybara 200k 34B Q5 | A long context length model that supports 200K tokens |
| Dolphin 8x7B Q4 | An uncensored model built on Mixtral-8x7b and it is good at programming tasks |
| Mixtral 8x7B Instruct Q4 | A pre-trained generative Sparse Mixture of Experts, which outperforms 70B models on most benchmarks |
| Tulu 2 70B Q4 | A strong model alternative to Llama 2 70b Chat to act as helpful assistants |
| Llama 2 Chat 70B Q4 | A model that is specifically designed for a comprehensive understanding through training on extensive internet data |
note
OpenAI GPT models require a subscription to use them further. To learn more, click here.
Model details
| Model | Author | Model ID | Format | Size |
|---|---|---|---|---|
| Mistral Instruct 7B Q4 | MistralAI, The Bloke | mistral-ins-7b-q4 | GGUF | 4.07GB |
| OpenHermes Neural 7B Q4 | Intel, Jan | openhermes-neural-7b | GGUF | 4.07GB |
| Stealth 7B Q4 | Jan | stealth-v1.2-7b | GGUF | 4.07GB |
| Trinity-v1.2 7B Q4 | Jan | trinity-v1.2-7b | GGUF | 4.07GB |
| Openchat-3.5 7B Q4 | Openchat | openchat-3.5-7b | GGUF | 4.07GB |
| Wizard Coder Python 13B Q5 | WizardLM, The Bloke | wizardcoder-13b | GGUF | 7.33GB |
| OpenAI GPT 3.5 Turbo | OpenAI | gpt-3.5-turbo | GGUF | - |
| OpenAI GPT 3.5 Turbo 16k 0613 | OpenAI | gpt-3.5-turbo-16k-0613 | GGUF | - |
| OpenAI GPT 4 | OpenAI | gpt-4 | GGUF | - |
| TinyLlama Chat 1.1B Q4 | TinyLlama | tinyllama-1.1b | GGUF | 638.01MB |
| Deepseek Coder 1.3B Q8 | Deepseek, The Bloke | deepseek-coder-1.3b | GGUF | 1.33GB |
| Phi-2 3B Q8 | Microsoft | phi-2-3b | GGUF | 2.76GB |
| Llama 2 Chat 7B Q4 | MetaAI, The Bloke | llama2-chat-7b-q4 | GGUF | 3.80GB |
| CodeNinja 7B Q4 | Beowolx | codeninja-1.0-7b | GGUF | 4.07GB |
| Noromaid 7B Q5 | NeverSleep | noromaid-7b | GGUF | 4.07GB |
| Starling alpha 7B Q4 | Berkeley-nest, The Bloke | starling-7b | GGUF | 4.07GB |
| Yarn Mistral 7B Q4 | NousResearch, The Bloke | yarn-mistral-7b | GGUF | 4.07GB |
| LlaVa 1.5 7B Q5 K | Mys | llava-1.5-7b-q5 | GGUF | 5.03GB |
| BakLlava 1 | Mys | bakllava-1 | GGUF | 5.36GB |