Why no high-quant coding-pruned DeepSeek or GLM for 256GB machines?
evrenozkan · reddit · 2026-11-06
A Reddit user with a 256GB M5 Ultra asks why nobody releases expert-pruned coding variants of DeepSeek or GLM at Q4 or higher. His tests show an expert-pruned Qwen Flash Next Coder outperforming a similar-sized quant of the full model; ISTA has released a low-precision pruned DeepSeek Coder, and pruning/calibration should cost only hundreds to a few thousand dollars of rented compute — so why aren't Unsloth, ISTA, etc. doing more of this?
More from Infra
- AI Labs Are Oil Refineries: Crucible Capital on Compute as a New Asset Class — AccBalanced · 2026-11-06
- Musk says Terafab 'moving fast' as Tesla's chip fab goes up in days — elonmusk · 2026-11-06
- SoftBank picks Malaysia for $50B tech investment in global AI bet — basedjensen · 2026-11-06
- Data center capex reportedly fleeing en masse to Malaysia — basedjensen · 2026-11-06
- Homelab with 3 mixed-GPU machines: how do you track local LLM models and VRAM? — pipecanon · 2026-11-06
- ING said to be seller of US$500M data centre loan portfolio — Bloomberg — apples_jimmy · 2026-11-06