arcee-ai/GLM-4-32B-Base-32K

GLM-4-32B-Base-32K is a 32 billion parameter language model developed by arcee-ai, based on THUDM's GLM-4-32B-Base-0414 architecture. This model is specifically engineered for robust performance over an extended 32,000-token context window, significantly improving recall compared to its base model which degraded after 8,192 tokens. It achieves this through targeted long-context training, iterative merging, and short-context distillation, making it ideal for tasks requiring deep understanding and processing of long documents.