Seeing everyone doing this pelican riding a bicycle, I also did it using the same prompt words and using deepseek v4 flash to see the effect.
"Local LLM now" DGX Spark x 2 + DeepSeek V4 Flash Vision Exp (FP8) achieves this speed.
By the way, it's OpenCode2.
I've reached the point where I can get up to 50-60 tok/s with DGX Spark x 2 + DeepSeek V4 Flash Vision Exp.
I want one more voice. Since this is a review, the context is over 128K.
You have reached the end of the archive
All of deepseek