Which current local models that can run within 128GB generate the best SVG pelicans?
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
一位开发者正在对128GB Apple Silicon上运行的本地大型语言模型进行基准测试，特别关注它们生成SVG鹈鹕的能力，测试提示为“骑自行车的鹈鹕”。测试重点是Qwen3.8 27B量化，其中4位量化表现良好，而1位量化则崩溃。该开发者还使用了Terminal-Bench 2.1、GPQA Diamond和IFBench进行评估，并质疑Qwen3.6 27B量化是否会破坏鹈鹕的生成效果。
I used a famous Simon Willison's pelican riding a bicycle prompt on the biggest local LLMs that can run on 128GB Apple Silicon. U used quantizations by Unsloth.
Qwen3.8 Flash-Next gives a lot of details. DeepSeek V4 Flash is strangely underwhelming. Qwen3.8 27B still rocks, and I like its consistent minimalism.
Is Qwen3.8 27B still large at 31GB? It is! But for this tasks 2-bit quantizations (at around 12GB) will give the same results. For more complicated coding, 4-bit are more than enough. RTX cards are well enough!
- Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses - Terminal-Bench 2.1, GPQA Diamond and IFBench