---

base_model: prism-ml/Bonsai-27B-unpacked

license: apache-2.0

library_name: gguf

tags:

• gguf
• llama.cpp
• multimodal
• qwen35
• bonsai
• 1-bit

---

Bonsai 27B GGUF

Private LM Studio conversion of prism-ml/Bonsai-27B-unpacked.

Files

| File | Purpose | Size |

| --- | --- | ---: |

| Bonsai-27B-Q1_0.gguf | Pure Q1_0 language model | 3.80 GB |

| mmproj-Bonsai-27B-BF16.gguf | BF16 vision projector | 0.93 GB |

The language model was converted with llama.cpp commit 0e4a0362239713ea95a6864a17a8de4b0ad90d62 using --no-mtp, then quantized with llama-quantize --pure ... Q1_0. Excluding MTP is required because the source configuration advertises one MTP layer while the checkpoint contains language blocks 0 through 63 only.

This repository intentionally does not include Prism’s separately packaged DSpark draft model. It therefore indexes as one 27B target plus one vision adapter in LM Studio.

Validation

Validated with llama-server and LM Studio/llmster:

• qwen35.block_count = 64; no qwen35.nextn_predict_layers
• Text and vision inference
• Parallel tool calls
• A three-stage tool workflow with nested JSON arguments and tool-result synthesis

Checksums

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