aifollow.news
← TrendingLatest development Nov 21 11:24
STORY IN FOCUS

Tencent Hunyuan says small Hy models are on its roadmap, with a 24GB VRAM model planned

In a Reddit Q&A, the Tencent Hunyuan team said small Hy models are on its roadmap and that it plans to release a model for 24GB VRAM. Hy4 flash is already available in WorkBuddy for tasks such as editing spreadsheets, documents and presentations. In the team's internal assessment, its average cost per task was about one-twentieth that of the flagship Hy4.

1 developments2 reports2 sources

Event developments

Coverage of the same occurrence shares a node; subsequent developments have their own nodes.

Tencent Hunyuan team plans larger models and further post-training work on Hy4

In a Reddit Q&A, the Tencent Hunyuan team said Hy4 still has considerable room for improvement. It is investing in large-scale reinforcement learning, continuing to improve user experience and model safety, and planning to train larger models. The team said it is considering a dense model with 24 billion to 31 billion parameters.

View source reports · 2 sources · 2 reports
All reports · 2

By original publication time, with each report's bookmarks and feedback preserved.

量子位 AI Machine translation

Tencent Hunyuan says small Hy models are on its roadmap, with a 24GB VRAM model planned

In a Reddit Q&A, the Tencent Hunyuan team said small Hy models are on its roadmap and that it plans to release a model for 24GB VRAM. Hy4 flash is already available in WorkBuddy for tasks such as editing spreadsheets, documents and presentations. In the team's internal assessment, its average cost per task was about one-twentieth that of the flagship Hy4.

IT之家 科技新闻 Machine translation

Tencent Hunyuan team plans larger models and further post-training work on Hy4

In a Reddit Q&A, the Tencent Hunyuan team said Hy4 still has considerable room for improvement. It is investing in large-scale reinforcement learning, continuing to improve user experience and model safety, and planning to train larger models. The team said it is considering a dense model with 24 billion to 31 billion parameters.