Eval bug: hy_v3 loads exp_probs_b with no .bias suffix — existing Hy3 GGUFs (AngelSlim, pre-merge quants) fail with “wrong number of tensors

用户在 llama.cpp b10000 版本(含)之后,使用 llama-server 或 llama-cli 加载 Tencent 量化团队 AngelSlim 发布的 Hy3 GGUF 模型(例如 Hy3-IQ1_M-mtp.gguf 、 Hy3-Q4_K_M-mtp.gguf )时触发。也适








![[Question]: How to obtain the original quoted text through HTTP or Python API?](https://www.chat-gpts.plus/wp-content/uploads/2026/07/8788-18831989-768x403.jpg)