One MacBook vs 2× DGX Spark: DeepSeek-V4-Flash scored 54% vs 52% on Terminal-Bench 2.1
英文标题：One MacBook vs 2× DGX Spark: DeepSeek-V4-Flash scored 54% vs 52% on Terminal-Bench 2.1
TL;DR: I ran DeepSeek-V4-Flash through the same 89-task Terminal-Bench 2.1 suite on two very different local setups: an aggressively quantized 80.8 GiB GGUF on one 128 GB M5 Max MacBook; the native mixed FP8/FP4 checkpoint with DSpark speculative decoding on 2× DGX Spark. The Mac finished at 47/87
打开原文相关 2 · 机会 0
中文
TL;DR: I ran DeepSeek-V4-Flash through the same 89-task Terminal-Bench 2.1 suite on two very different local setups: an aggressively quantized 80.8 GiB GGUF on one 128 GB M5 Max MacBook; the native mixed FP8/FP4 checkpoint with DSpark speculative decoding on 2× DGX Spark. The Mac finished at 47/87
中文全文翻译正在补齐，可先查看下方英文原文。
English / 原文
TL;DR: I ran DeepSeek-V4-Flash through the same 89-task Terminal-Bench 2.1 suite on two very different local setups: an aggressively quantized 80.8 GiB GGUF on one 128 GB M5 Max MacBook; the native mixed FP8/FP4 checkpoint with DSpark speculative decoding on 2× DGX Spark. The Mac finished at 47/87