ValueError: Could not find either of model.onnx, model_optimized.onnx in /var/folders/zn/0_np64n96r50prt1m5mlm7fr0000gn/T/fastembed_cache/fast-bge-small-en

该报错通常出现在 AutoGen 的 Qdrant RetrieveChat 笔记本示例中,原因是 FastEmbed 依赖下载 ONNX 模型时中断或缓存不完整。优先尝试将 fastembed 升级到最新版,并确认 pyautogen 版本已更新到官方修复版本。
![[Krea-2] `enable_gqa` + `attn_mask` produces high usage of VRAM](https://www.chat-gpts.plus/wp-content/uploads/2026/08/14518-5ee4eb67-768x403.jpg)
![[Bug]: MTP spec decode still advances the grammar matcher after termination when a structural tag is built (residual after #44297)](https://www.chat-gpts.plus/wp-content/uploads/2026/08/52767-d83cfbd1-768x403.jpg)
![[Bug]: Native MTP speculative decoding degenerates into garbage token loops on deep agentic conversations (Qwen3-MoE)](https://www.chat-gpts.plus/wp-content/uploads/2026/08/47087-d5d5041b-768x403.jpg)





