Agent: Cursor, GitHub CopilotLLM: DeepSeek, Qwen#llm-compression#quantization#speculative-decoding#model-optimization#on-device-ai
AngelSlim is Tencent's open-source model compression toolkit supporting quantization (FP4/FP8/2-bit/1.25-bit), speculative decoding (Eagle3), and pruning for LLMs, VLMs, and audio models. It targets models like DeepSeek, Qwen, and HunyuanVideo, enabling efficient on-device and server-side deployment.