catmind/dx-brian-v1
The catmind/dx-brian-v1 is a 5.1 billion parameter language model, finetuned and converted to GGUF format by catmind using Unsloth. This model is available in both text-only and multimodal (vision-capable) variants, supporting a 32768 token context length. It is optimized for efficient deployment and usage, particularly for applications requiring either pure text generation or combined text and vision processing.
Loading preview...
dx-brian-v1 Overview
The catmind/dx-brian-v1 model is a 5.1 billion parameter language model, finetuned and converted to the GGUF format. This model was developed by catmind, leveraging the Unsloth framework for efficient training and conversion, reportedly achieving 2x faster training speeds.
Key Capabilities
- Text Generation: Supports standard text-only large language model operations.
- Multimodal Processing: Includes a vision-capable variant (
gemma-4-e2b-it.BF16-mmproj.gguf) for applications requiring image understanding alongside text. - Efficient Deployment: Provided in GGUF format, suitable for local inference with tools like
llama-cliandollama. - Extended Context: Features a substantial 32768 token context length, allowing for processing longer inputs and generating more coherent, extended outputs.
Usage and Deployment Notes
This model offers separate GGUF files for text-only and multimodal use. For multimodal applications with Ollama, users need to create a unified bf16 model by placing a Modelfile alongside the merged bf16 model, as Ollama currently does not support separate mmproj files for vision models. Example command-line usage is provided for both text-only and multimodal scenarios, utilizing the --jinja flag for prompt templating.