ql390962

Results 4 comments of ql390962

建个群抱团取暖~

> @Exuan148 扫描二维码入群 > > ![SCR-20240420-odtl](https://private-user-images.githubusercontent.com/155712877/324163641-90d19f24-e09f-4baa-ba15-afb3535c1f41.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MjI4NDg1MzcsIm5iZiI6MTcyMjg0ODIzNywicGF0aCI6Ii8xNTU3MTI4NzcvMzI0MTYzNjQxLTkwZDE5ZjI0LWUwOWYtNGJhYS1iYTE1LWFmYjM1MzVjMWY0MS5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwODA1JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDgwNVQwODU3MTdaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT0zOTE0ZDQwYTc4MzQ4MWU1MzNiZThiMmMxODhlYzg5NjMwNmEwYzI5MTRlMDFlMDNiNTEyZWFkZDdiOTJjZDRlJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.LIRKxtMzDMP3VyPydqwkjKVwiU74FNVv2PbbCpxT8_w) 大哥能不能再给个码，想进群学习。

Token repetition issue with Qwen2.5-VL

I have encountered the same problem, do you have a solution?

NPU out of memory

> 我在昇腾910B上部署完成，python inference/cli_demo.py 时报错： `hidden_states = F.scaled_dot_product_attention( RuntimeError: NPU out of memory. Tried to allocate 35.31 GiB (NPU 0; 60.97 GiB total capacity; 26.13 GiB already allocated; 26.13 GiB current...

不用flash attention推理结果异常

我也遇到了类似问题，输出全是[!!!!!!!!!!!!!!!!]