Redditor suspects Hugging Face throttles downloads of niche models, urges backing up weights
silenceimpaired · reddit · 2026-11-16
A Reddit user reports a pattern of Hugging Face downloads of lesser-known models running 3-4x slower (or stalling for over an hour), speeding up instantly after switching to a popular model. He stresses it may just be load balancing, not malice, but suspects deliberate rate-limiting that steers users toward mainstream models — and asks which model weights people should back up locally in case HF's behavior or availability changes. He frames it as speculation backed by multiple observed cases.
More from Infra
- EE asks Claude to price 1,000 human brains in silicon: 39 sq-mile data center, 27x global nuclear output — Recent-Day3062 · 2026-11-16
- Intel 170hx cards unlocked to 80GB VRAM, price jumps from hundreds to $4k — GabGarrett · 2026-11-16
- Track cost per finished task, not token price: the agent metric that matters — arpit_bhayani · 2026-11-16
- Nebius and CoreWeave already used customer prepayments to secure GPUs — context WSJ missed — firstadopter · 2026-11-16
- Anthropic models reportedly optimize kernels for NVIDIA, TPU, Trainium — but not Apple — GabGarrett · 2026-11-16
- Bittensor-backed Engy claims 50–95% cheaper inference, GPU and storage vs CoreWeave and S3 — markjeffrey · 2026-11-16