← 研踪 PaperTrailarXiv:2611.25521 · v1
DRVEXO-4B-Unified: Design and Evaluation of a Modular Speech-Vision System
首次提交 2026-11-19 13:23
版本更新 2026-11-19 13:23
雷达观测 2026-11-20 19:58
摘要
We present DRVEXO-4B-Unified, a modular multimodal assistant built around a 4B language backbone. A self-contained checkpoint combines text, image, and speech inputs with reference-conditioned speech output and structured interaction. A temporal residual connector adapts acoustic sequences to the language interface; explicit text boundaries support incremental speech synthesis; and three task-conditioned scoring heads rank supplied action candidates alongside generative tool calls. These interfaces connect conversation to application-defined choices while leaving action authorization to the application. The system obtains 74.22% on all 900 MMMU validation questions, exceeding the published validation scores of Gemma 3 4B IT and InternVL3.5-4B without test-time scaling. Strict prompt-level IFEval accuracy is 82.26% and MATH-500 accuracy is 89.40%, above the original Qwen3-4B non-thinking report's 81.2% and 84.8%, respectively. On NVIDIA GB10, the model achieves a speech-synthesis first-chunk latency of 390 ms. For concurrent text generation, a growth-aware admission policy completes a 128-answer batch at 594.41 output tokens/s, including filling and draining, without preemptions. We plan to release the model weights and accompanying code under Apache License 2.0, retaining applicable third-party terms.
作者备注:23 pages, 5 figures
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-19 13:23 · 最新观测版本
DRVEXO-4B-Unified: Design and Evaluation of a Modular Speech-Vision System
We present DRVEXO-4B-Unified, a modular multimodal assistant built around a 4B language backbone. A self-contained checkpoint combines text, image, and speech inputs with reference-conditioned speech output and structured interaction. A temporal residual connector adapts acoustic sequences to the language interface; explicit text boundaries support incremental speech synthesis; and three task-conditioned scoring heads rank supplied action candidates alongside generative tool calls. These interfaces connect conversation to application-defined choices while leaving action authorization to the application. The system obtains 74.22% on all 900 MMMU validation questions, exceeding the published validation scores of Gemma 3 4B IT and InternVL3.5-4B without test-time scaling. Strict prompt-level IFEval accuracy is 82.26% and MATH-500 accuracy is 89.40%, above the original Qwen3-4B non-thinking report's 81.2% and 84.8%, respectively. On NVIDIA GB10, the model achieves a speech-synthesis first-chunk latency of 390 ms. For concurrent text generation, a growth-aware admission policy completes a 128-answer batch at 594.41 output tokens/s, including filling and draining, without preemptions. We plan to release the model weights and accompanying code under Apache License 2.0, retaining applicable third-party terms.
查看此版本解读 →查看此版本原文 ↗