OffByQuant

[OffByQuant @OffByQuant]

Same Model, Different Engine: #Strata and #Maya Against Plain #llamacpp on One #NVIDIA #RTX5090
https://t.co/HxNLBC8cnA

[OffByQuant @OffByQuant]

A Fork That Decodes Faster: #Basalt Against #Strata on One #RTX5090
https://t.co/8nUOUPNeGH
#qwen #localLLM #llm #nvidia #inference

[OffByQuant @OffByQuant]

Crouching VRAM, Sleeping Dragon: How the Strata Family Makes Massive AI Models Run by Waking Only What They Need
https://t.co/tdvhprr1BS
#strata #glm #localLLM #inference #nvidia #rtx5090

[Ankit Prateek @ankit_appy]

The Quantization Tax Nobody Charged: #Qwen 3.8 27B Q4 Quant Matches Its Published Scores on a Single #RTX5090 @Alibaba_Qwen
https://t.co/yL4yCzZmJr
https://t.co/yjU9qmV5B2
#localLLM

[OffByQuant @OffByQuant]

Everyone budgets a quality tax for 4-bit quants. I went looking for it on #Qwen 3.8 27B: GPQA, IFBench, 95 real CVEs, RULER at 262K, KL-divergence.

The tax never showed up — Q4 matches @Alibaba_Qwen 's published BF16 scores on a single #RTX5090.
https://t.co/MxhfDofJjV
#localLLM

[Ankit Prateek @ankit_appy]

PyDiffWatch: watches the PyPI release firehose, diffs every new version, flags suspicious ones fast enough to report for takedown.
▸never runs what it inspects – packages are data, not code
▸reviews run on a local LLM
https://t.co/sg937IrbCg
@ramimacisabird @abh1sek @anantshri