deepseek-v4.1IMAGE

Running quantized DeepSeek V4.1 Flash locally on a 3090

Runs a quantized deepseek-v4.1-flash-1017 for Project Maya on a 3090 with 64GB RAM; slow, but the author says low-bit quants of big models still deliver usable results.

@ItsmeAjayKVArchived 11 days agoOriginally posted 2026.10.21

Running quantized DeepSeek V4.1 Flash locally on a 3090 (image 1 of 2)

Running quantized DeepSeek V4.1 Flash locally on a 3090 (image 2 of 2)

AJ @ItsmeAjayKVOriginally posted 2026.10.21

Open original on X

Project Maya is awesome. It's deepseek-v4.1-flash-1017 that am running on my 3090 + 64GB RAM. Although speed is very slow, am really looking forward to the result. I had also prev run ds-v4-flash, antirez quant, it was about as slow as this, but i ended up still getting good result with iq1 quant. Lower quant's of bigger models can still do some quality works.

This post quotes another post: “Running V4.1 Flash Maya-S quant on a 3090 with 64GB RAM”

x.com/itsmeajaykv/status/2112684034501271815Archived 2026.10.21 12:07