🚨 Qwen3.8-27B kept answering with //// on a DGX Spark. llama.cpp loaded fine. … · @QwenAIFans · TwKit

@QwenAIFans

🚨 Qwen3.8-27B kept answering with //// on a DGX Spark. llama.cpp loaded fine. Speed looked normal. Rebuilds for GB10 did nothing. CPU-only still garbage. SHA-256 did not match Hugging Face. File size was exact (18.97 GB). Bad download, silent load. Verify checksum before you debug CUDA for hours. via ggml-org/llama.cpp discussion 👇 https://github.com/ggml-org/llama.cpp/discussions/30609

曝光 350 · 评论 1 · 点赞 6 · 书签 2