Model Releases
We made inspecting models tensors on the Hugging Face Hub up to 20× faster ⚡ DeepSeek-V3.2: 9.2s → 0.45s GLM-5.3: 4.0s → 0.57s Qwen2.5-72B: …
In October 2026, Victor M announced a performance boost for inspecting model tensors on the Hugging Face Hub, achieving up to a 20× speed increase. Sample results include DeepSeek‑V3.2 dropping from 9
In October 2026, Victor M announced a performance boost for inspecting model tensors on the Hugging Face Hub, achieving up to a 20× speed increase. Sample results include DeepSeek‑V3.2 dropping from 9.2 s to 0.45 s, GLM‑5.3 from 4.0 s to 0.57 s, and Qwen2.5‑72B from 1.2 s to 0.27 s. The update was highlighted in a tweet that garnered over 5,300 views.
Related
- The week just gets better the mad men from China do it again China is hot Deepseek V4 Pro Let’s see how it is https://huggingface.co/deepsee…
- DeepSeek-V4 just dropped on Hugging Face https://huggingface.co/collections/deepseek-ai/deepseek-v4
- 🤗 DeepSeek V4 is now live on @huggingface — supported by Novita 1M context. Massive-scale MoE. Pro or Flash — pick your tradeoff.
Source: Clem Delangue (X) | 2026-10-15