Ternary-Bonsai-27B — tool-eval-bench Results
Published benchmark artifacts scoring Prism ML's Ternary-Bonsai-27B model on tool-eval-bench, a multi-turn tool-calling evaluation, including per-trial Markdown traces and two self-contained interactive HTML dashboards.
This repository holds the raw output of running tool-eval-bench, a multi-turn, mock-tool-calling evaluation with deterministic scenarios and three-tier pass/partial/fail scoring, against Prism ML's Ternary-Bonsai-27B model in its Q2_0 GGUF quantization. The model was run for 8 sequential trials across 84 scenarios, producing a mean score of 85 out of 100 with no variance between trials, a 76.2% pass rate, and a deployability score of 80.
The repository includes a cross-trial summary, eight full per-trial Markdown traces recording every scenario and turn, and two self-contained HTML dashboards - one showing the model's results alone, the other comparing it head-to-head against NVIDIA's Qwen3.6-27B-NVFP4. The evaluation measures agentic tool use specifically - tool selection, multi-step chains, state handling and safety - rather than general knowledge or coding benchmarks.
It is a results artifact rather than a tool: there is no code to run, only the recorded benchmark output for this one model.