Redditor builds 1.75TB local AI archive to hedge against cloud restrictions
nfored · reddit · 2026-11-20
A Reddit user spent days assembling a 1.75TB preservation archive of local AI models and software, aiming to keep the ability to run AI independently of cloud providers if models get restricted or pulled.
The model collection spans LLMs (Qwen3.6-35B-A3B, DeepSeek V4 Flash, Gemma 4), coding models (Qwen3-Coder, CodeLlama), speech recognition and synthesis (Whisper, Qwen ASR/TTS, Fish Audio Drama 3), image/video generation (HiDream, Wan 2.2, LTX 2.5), plus vision, OCR, translation and forecasting models.
On the software side it archives llama.cpp, vLLM, ComfyUI, PyTorch, Transformers, CUDA 12.8–13.5, ROCm 7.x–10.0, Flash Attention, FFmpeg and Ubuntu ISOs — deliberately supporting both NVIDIA and AMD hardware so the archive isn't tied to one GPU architecture. The author frames it as long-term digital preservation of a personally maintainable AI capability set, not air-gapping, and asks which models others consider irreplaceable.
More from Infra
- Geoffrey Huntley's six takes: compute crunch lasts 6-10 years, open weights trail by 3 months — sebpaquet · 2026-11-20
- Qwen4 Max lands on Together AI at 30% off: 2.4T-param MoE with 1M context — togethercompute · 2026-11-20
- LexiEngine: Rust inference engine for AMD runs 27B at full 262k context 47% faster than llama.cpp — W61k3r · 2026-11-20
- Anthropic would need $602B annual revenue—17x growth—to match tech peers' returns — GaryMarcus · 2026-11-20
- Cost chart: Claude Fable 5.2 burns $54 per task, OpenAI a fraction — haider1 · 2026-11-20
- Basalt 0.5.0 runs Tencent Hy4 Flash locally at 172 tok/s on a 5090 with 1M context — jesdga95 · 2026-11-20