{"apiVersion":"v1","taxonomyVersion":1,"generatedAt":"2026-10-29T15:49:54.852Z","data":{"recordId":"wsa_vhj7jt9fvbgf","score":0.875,"confidence":"high","capped":false,"capReason":null,"sources":[{"n":1,"url":"https://huggingface.co/prism-ml/Ternary-Bonsai-2-2B-gguf","publisher":"PrismML","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-27T01:40:35.745Z","detectedPublishedAt":"2026-09-23T05:51:42.000Z","rights":"original"},{"n":2,"url":"https://huggingface.co/prism-ml/Ternary-Bonsai-2-2B-mlx-2bit","publisher":"PrismML","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-27T01:40:35.745Z","detectedPublishedAt":"2026-10-01T17:06:42.000Z","rights":"original"},{"n":3,"url":"https://prismml.com/news/bonsai-2-family","publisher":"PrismML","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-27T01:40:35.745Z","detectedPublishedAt":"2026-10-15T00:00:00.000Z","rights":"original"},{"n":4,"url":"https://prismml.com/news/prismml-expands-bonsai-2-family-with-2b-4b-and-9b-models-for-on-device-ai","publisher":"PrismML","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-27T01:40:35.745Z","detectedPublishedAt":"2026-10-15T00:00:00.000Z","rights":"original"}],"claims":[{"id":"996cbe0d-14f4-4d6b-9888-4c486bc06b40","kind":"capability","statement":"The model supports reasoning, visual understanding, and tool-driven workflows.","citations":[{"source":3,"supportProbability":0.65,"supports":false,"quote":null},{"source":4,"supportProbability":0.68,"supports":false,"quote":{"text":"Building on the first-generation Bonsai 1.7B, 4B and 8B models, the new releases combine improved reasoning, visual understanding, and support for tool-driven workflows.","rights":"original"}}]},{"id":"181f66e1-475d-479b-a6c8-7dab242fa2de","kind":"availability","statement":"The model is downloadable starting October 15, 2026.","citations":[{"source":4,"supportProbability":0.85,"supports":true,"quote":{"text":"Developers, researchers, and other users can download the new Bonsai 2 models under the Apache 2.0 license for free starting October 15, 2026.","rights":"original"}}]},{"id":"8a8e9790-fb59-4888-91a1-640f66f02d65","kind":"deployment","statement":"The model is optimized for low-latency inference on consumer-grade CPUs and edge GPUs.","citations":[{"source":4,"supportProbability":0.87,"supports":true,"quote":{"text":"They have been designed for seamless integration with existing AI workflows and are optimized for low-latency inference on consumer-grade CPUs, and edge GPUs.","rights":"original"}}]},{"id":"5945ca2d-a34a-4ef2-be9b-95018c89d9bd","kind":"capability","statement":"The model accepts vision input through the official vision tower.","citations":[{"source":2,"supportProbability":0.97,"supports":true,"quote":{"text":"- **Vision input** through the official vision tower, carried unrotated in FP16.","rights":"original"}}]},{"id":"1a7b03eb-51ca-4e3a-95b9-1b10a0bf49f1","kind":"availability","statement":"The model is downloadable from Hugging Face.","citations":[{"source":2,"supportProbability":0.42,"supports":false,"quote":null}]},{"id":"866c8887-face-4f96-9e1c-57172a12a875","kind":"deployment","statement":"An OpenAI-compatible server can be run with mlx-vlm.","citations":[{"source":2,"supportProbability":0.96,"supports":true,"quote":{"text":"For an OpenAI-compatible server, run `python -m mlx_vlm.server --model bonsai2-2b-mlx` and send the sampling settings in each request.","rights":"original"}}]},{"id":"337057af-2931-4e91-94b0-01efb54e368b","kind":"capability","statement":"Its weights are ternary values stored in MLX affine 2-bit format.","citations":[{"source":2,"supportProbability":0.97,"supports":true,"quote":{"text":"- **Ternary language weights** stored in MLX's 2-bit grouped format (group 128).","rights":"original"}}]},{"id":"699262cc-adf9-40e2-9a59-4f19279bd88c","kind":"capability","statement":"The model accepts vision input through a separate mmproj pack.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"- **Vision input** through a separate mmproj pack (Q8_0, or BF16 for reference precision), loaded only when an image arrives.","rights":"original"}}]},{"id":"017557f1-0ab5-432b-adf1-2b941ce229b9","kind":"capability","statement":"The model has a 262,144-token context length.","citations":[{"source":1,"supportProbability":0.99,"supports":true,"quote":{"text":"| Context length | 262,144 tokens (inherited from the base model) |","rights":"original"}},{"source":2,"supportProbability":0.76,"supports":true,"quote":null}]},{"id":"cb8688ee-990b-40b6-a139-fd42b02fb2fb","kind":"availability","statement":"Three GGUF language-model packings are provided: PTQ1_0, PQ2_0, and Q2_0.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"- **Three GGUF packings** of the same weights, consumed directly by custom kernels and never expanded back to FP16: **PTQ1_0** (dense trits, 0.43 GB), **PQ2_0** (2-bit slots, 0.52 GB) and **Q2_0** (0.55 GB).","rights":"original"}}]},{"id":"b6b93a45-829c-4287-bec8-061546ee06d7","kind":"deployment","statement":"The model runs with the PrismML llama.cpp build on CUDA, Metal, and CPU.","citations":[{"source":1,"supportProbability":0.92,"supports":true,"quote":{"text":"A 2B-class vision-language model with ternary transformer weights, for llama.cpp (CUDA, Metal, CPU).","rights":"original"}}]},{"id":"9edf6328-2cb4-4178-bd29-1f1a90197c3c","kind":"recommendation-relevant","statement":"The model is licensed under Apache 2.0.","citations":[{"source":1,"supportProbability":0.99,"supports":true,"quote":null},{"source":2,"supportProbability":0.98,"supports":true,"quote":null}]}]},"nextCursor":null,"license":{"catalog":{"name":"CC BY 4.0","spdx":"CC-BY-4.0","url":"https://creativecommons.org/licenses/by/4.0/","attribution":"wowsuchai","attributionUrl":"https://wowsuchai.withglint.app","covers":["catalog metadata","classifications","generated summaries","feed event metadata"]},"excluded":{"covers":["source quotations","linked source content"],"terms":"Not licensed by wowsuchai. Source quotations and linked source content keep their original rights."}}}