M4 Pro
A professional-grade chip in Apple's self-designed M series, featuring a unified memory architecture and an advanced TSMC process, targeting professional workloads such as video rendering and 3D modeling.
Timeline (last 90 days)
The GGUF Q4_K_M quantized version of Kev 4B scored 72/82 on an M4 Pro / 24GB Mac, matching the cloud version's score, with a median latency of 0.173 seconds—approximately 9 times faster than calling the cloud version (1.534 seconds) and at zero cost.
In actual benchmark threads for llama.cpp, the 2021 M1 Max achieved a decode write speed of 61 tokens/s, higher than the 2024 M4 Pro's 51 tokens/s.
Running via llama.cpp on a MacBook Pro with an M4 Pro chip and 24GB unified memory, the text decoding speed is 10.13 tokens/second, with each complete task taking 4 to 7 minutes.
Ling 3.0 Tiny achieves an inference speed of approximately 86–90 tokens per second on a MacBook with M4 Pro, with a peak memory usage of approximately 8.34 GiB under an 8K context.
With 64 GB of memory, models with 30B to 34B parameters can run smoothly under Q4 quantization, while a 70B model is barely feasible under Q4 but with a noticeably slower response speed.
For M4 Pro 64GB users pursuing fast visual response, Qwen2.5-VL-7B or Llama 3.2 Vision 11B is recommended.
All Facts (6)
Ling 3.0 Tiny achieves an inference speed of approximately 86–90 tokens per second on a MacBook with M4 Pro, with a peak memory usage of approximately 8.34 GiB under an 8K context.
60%UnverifiedThe GGUF Q4_K_M quantized version of Kev 4B scored 72/82 on an M4 Pro / 24GB Mac, matching the cloud version's score, with a median latency of 0.173 seconds—approximately 9 times faster than calling the cloud version (1.534 seconds) and at zero cost.
50%UnverifiedIn actual benchmark threads for llama.cpp, the 2021 M1 Max achieved a decode write speed of 61 tokens/s, higher than the 2024 M4 Pro's 51 tokens/s.
50%UnverifiedRunning via llama.cpp on a MacBook Pro with an M4 Pro chip and 24GB unified memory, the text decoding speed is 10.13 tokens/second, with each complete task taking 4 to 7 minutes.
50%UnverifiedWith 64 GB of memory, models with 30B to 34B parameters can run smoothly under Q4 quantization, while a 70B model is barely feasible under Q4 but with a noticeably slower response speed.
50%UnverifiedFor M4 Pro 64GB users pursuing fast visual response, Qwen2.5-VL-7B or Llama 3.2 Vision 11B is recommended.
50%