AITOPAI热报
ModelsSelected

d3-ultra multimodal model released with Apache-2.0 weights

官方账号vLLM@vllm_project
Post
View the post
Why we picked it

d3-ultra being trained and served entirely on AMD Instinct GPUs shows an alternative to the Nvidia-dominated inference stack, a meaningful signal for teams weighing hardware costs. Developers building routing or decision systems can inspect the per-answer probability outputs and deploy via vLLM immediately using the open weights.

The vLLM Semantic Router team and AMD released d3-ultra, an open multimodal decision-making model. It provides per-answer probabilities, visual grounding and video tracking capabilities. The model was trained and served on AMD Instinct GPUs. Apache-2.0 licensed weights are available on Hugging Face.

Original · vLLM

Great work from @AMD and the vLLM Semantic Router team on d3-ultra! Open multimodal decision-making with per-answer probabilities, visual grounding, and video tracking—trained and served on AMD Instinct GPUs. Apache-2.0 weights on @huggingface.

#d3-ultra#vLLM#AMD Instinct#open-source models#multimodal

Keep reading

Recent stories with the same tag or section
ModelsBuildersSelected09:20
ProtAgent-8B Matches Frontier Models on Protein Tool Tasks

Interaction-level credit assignment, not raw model scale, is what closes the gap with frontier systems here — a 671B model loses by roughly 20 points under the same protocol. Researchers building tool-using agents in specialized domains can apply the same step-level reward scheme without frontier API budgets. The authors' caveat that annotation coverage does not prove biological correctness is worth keeping in mind when comparing reported scores.

#ProtAgent#DeepSeek#agents#fine-tuning
arXiv: DeepSeek@Yuxiang Wang et al. (6)Source