← All models

Qwen2.5 14B Instruct 1M

Alibaba · 2025-01-27 · 14.8B parameters

Qwen2.5-14B-Instruct-1M is Alibaba's long-context version of Qwen2.5-14B-Instruct, released in January 2025 with the Qwen2.5-1M technical report. It was trained with progressive long-context pretraining and multi-stage fine-tuning and supports contexts of up to 1,010,000 tokens using dual chunk attention and sparse attention in Qwen's vLLM-based inference framework.

Benchmark scores