Bonsai-27B-GGUF

Bonsai-27B-GGUF

A GGUF conversion of the 27B Bonsai model for LM Studio, with a 3.80 GB Q1_0 language file and a 0.93 GB BF16 vision projector. Tested for text, vision and parallel tool calls.

Made by lmstudio-community on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Bonsai-27B-GGUF is a GGUF conversion of Bonsai 27B, made by the LM Studio community account from the unpacked release by Prism ML. The card calls it a private conversion made for LM Studio.

The repository holds two files. One is the language model in pure Q1_0 form, and the other is a BF16 vision projector that lets the model take images. The card says it was tested with llama-server and with LM Studio.

Who it suits: People who run models on their own computer with LM Studio or llama.cpp tools, and who want a 27B model with image input in a small download.

Industry | Who uses it | How they use it | Example
Software | App developers | Test a local model with tools and images | A prototype that calls tools from a local model
Research | Quantization testers | Try a heavy Q1_0 compression | Comparing it with a larger quantization
Hobbyists | Local model users | Chat with a 27B model on a home PC | Asking questions about a photo
Enterprise | Platform teams | Check parallel tool calls offline | A workflow with nested JSON arguments

You can skip it if you need the best output quality, since Q1_0 is a heavy compression and the card gives no quality results.

Where to use it: The card names two places. LM Studio indexes the files as one 27B model plus one vision adapter, and llama-server was also used for checks.

• LM Studio, where the files appear as one model with an adapter
• llama-server from llama.cpp
• Any other tool that loads GGUF files and a vision projector

Use case | Who | What they get back
Text chat | Local model users | Written replies from the 27B model
Image questions | App developers | Answers that use the vision projector
Tool calls | Agent builders | Parallel calls and a three-stage workflow

How to set it up: You need Python for the download tool and a GGUF runtime such as LM Studio. Both files together are about 4.7 GB on disk, going by the sizes on the card.

• Python with pip
• LM Studio or llama.cpp's llama-server
• About 4.7 GB of free disk space

1. Install the Hugging Face hub tools with pip.
2. Download the repository files with the hf command.
3. Check each file against the checksums on the card.
4. Open the files in LM Studio, or point llama-server at them.

pip install -U huggingface_hub
hf download lmstudio-community/Bonsai-27B-GGUF

The card lists these checksums so you can confirm your downloads match.

71d6871836886267f39f1ac69b1b1dd8a783970708d7a189960112ed02978901  Bonsai-27B-Q1_0.gguf
c4c09d300638f6d2759e1ffe94d97cf08be94b8c9dad6920509be1aab11fd8cb  mmproj-Bonsai-27B-BF16.gguf

How to use it: Load the language file together with its vision projector, then send text, or text plus an image. The card shows no run command, so follow the docs of the tool you load it in.

The card lists four checks it ran. They cover reading text, reading images, calling several tools at once, and a chained tool job that ends with an answer drawn from the tool output.

Example: You load both files in LM Studio, attach a photo and ask what is in it. The projector file is what should let the model read the picture.

For newcomers and experts: If you are new, download both files together. The vision file is an add-on, so it needs the language file next to it.

If you know the field, the language file was converted at a pinned llama.cpp commit with the no-mtp option and then quantized with the pure Q1_0 setting. The card says the source config lists one MTP layer, but the checkpoint holds language blocks 0 through 63 only.

Prism ML unpacked model
Converted with llama.cpp, no MTP
Quantized to pure Q1_0
Two GGUF files
Loaded in LM Studio

Good to know: The card leaves out Prism's separate DSpark draft model on purpose. The chart shows the two file sizes the card lists.

Title: File size, GB
Language model, Q1_0: 3.80 GB
Vision projector, BF16: 0.93 GB

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the prompts and images you give it, and the files you download
• Not for: draft-model speedups, since the card leaves out Prism's DSpark draft model

On Hugging Face, 14 November 2026

• Library: gguf
• Parameters: 26.9B
• Weights: about 13.4 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 247,968
• Likes: 39
• License: apache-2.0
• Based on: prism-ml/Bonsai-27B-unpacked
• Tags: gguf, llama.cpp, multimodal, qwen35, bonsai, 1-bit, conversational

Links

• Model: https://huggingface.co/lmstudio-community/Bonsai-27B-GGUF
• Files: https://huggingface.co/lmstudio-community/Bonsai-27B-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download lmstudio-community/Bonsai-27B-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.