Ternary-Bonsai-27B Local Inference Tested
pbaylies · x · 2026-07-15
A repost showcases the performance of Ternary-Bonsai-27B running on a single RTX 3090, focusing on quantization and inference efficiency:
- Under llama-bench, throughput at various context lengths (32k / 64k / 128k) is approximately 956 / 830 / 547 t/s, respectively.
- Using llama-server and dspark speculative decoding, the maximum context reaches 132k.
- VRAM usage is around 23.3GB, with a decoding speed of roughly 44 t/s, marking about a 1.5× improvement over the baseline.
- The original post also mentions that this series includes 1-bit / ternary versions, a 262K token context window, and smaller companion models aimed at low-VRAM deployments.
Overall, this is a local inference release/test focusing on quantization compression, long context, and inference acceleration.
Related event: Bonsai 27B: The first 27B model that runs on phones(15 posts)→
More from Infra
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11