Webtagr - Technology News Summarizer

Qwen2.5-VL-32B: A 32B Parameter Visual-Language Model That's More Human-Friendly

2025-03-24

Following the widespread acclaim of the Qwen2.5-VL series, we've open-sourced the new 32-billion parameter visual-language model, Qwen2.5-VL-32B-Instruct. This model boasts significant improvements in mathematical reasoning, fine-grained image understanding, and alignment with human preferences. Benchmarking reveals its superiority over comparable models in multimodal tasks (like MMMU, MMMU-Pro, and MathVista), even outperforming the larger 72-billion parameter Qwen2-VL-72B-Instruct. It also achieves top-tier performance in pure text capabilities at its scale.

(qwenlm.github.io)

AI visual language model

QwQ-32B: Scaling RL for Enhanced Reasoning in LLMs

2025-03-05

Researchers have achieved a breakthrough in scaling reinforcement learning (RL) for large language models (LLMs). Their 32-billion parameter QwQ-32B model demonstrates performance comparable to the 671-billion parameter DeepSeek-R1 (with 37 billion activated parameters), highlighting the effectiveness of RL applied to robust foundation models. QwQ-32B, open-sourced on Hugging Face and ModelScope under the Apache 2.0 license, excels in math reasoning, coding, and general problem-solving. Future work focuses on integrating agents with RL for long-horizon reasoning, pushing the boundaries towards Artificial General Intelligence (AGI).

(qwenlm.github.io)

AI

Alibaba Unveils Qwen2.5-Max: A Massive MoE Language Model

2025-01-28

Alibaba has released Qwen2.5-Max, a large-scale Mixture-of-Experts (MoE) model pre-trained on over 20 trillion tokens and further refined with supervised fine-tuning and reinforcement learning from human feedback. Benchmarks like MMLU-Pro, LiveCodeBench, LiveBench, and Arena-Hard show Qwen2.5-Max outperforming models such as DeepSeek V3. The model is accessible via Qwen Chat and an Alibaba Cloud API. This release represents a significant advancement in scaling large language models and paves the way for future improvements in model intelligence.

(qwenlm.github.io)

AI MoE Model Alibaba

Qwen2.5-1M: Open-Source LLMs with 1 Million Token Context Length

2025-01-26

The Qwen team released Qwen2.5-1M, open-source large language models supporting up to one million tokens of context length, in 7B and 14B parameter versions. These models significantly outperform their 128K counterparts on long-context tasks, even surpassing GPT-4o-mini in some cases. An open-sourced inference framework based on vLLM, leveraging sparse attention for a 3x to 7x speed boost, is also provided for efficient deployment. Qwen2.5-1M's training employed a progressive approach, incorporating Dual Chunk Attention (DCA) and sparse attention techniques for effective long-context handling.

(qwenlm.github.io)

AI