ValueError: The model's max seq len (4096) is larger than the maximum number of tokens that can be stored in KV cache (3792). Try increasing `gpu_memory_utilization` or decreasing `max_model_len` when initializing the engine.

[Hzzhang-nlp]

是否已有关于该错误的issue或讨论？ | Is there an existing issue / discussion for this?

• [X] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答？ | Is there an existing answer for this in FAQ?

• [X] 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

Python环境安装，运行bash scripts/run_for_7B_in_Linux_or_WSL.sh，报错： ValueError: The model's max seq len (4096) is larger than the maximum number of tokens that can be stored in KV cache (3792). Try increasing gpu_memory_utilization or decreasing max_model_len when initializing the engine.

期望行为 | Expected Behavior

No response

运行环境 | Environment

- OS:ubuntu22.04
- NVIDIA Driver:
- CUDA:12.2
- docker:
- docker-compose:
- NVIDIA GPU:RTX 4090
- NVIDIA GPU Memory:24GB

QAnything日志 | QAnything logs

No response

复现方法 | Steps To Reproduce

No response

备注 | Anything else?

No response

[huweibin1983]

bash ./run.sh -c local -i 0 -b vllm -m Qwen-7B-QAnything -t qwen-7b-qanything -p 1 -r 0.85 将后面的0.85改成0.95试一下

[Ericocococo]

问题解决了吗？为也遇到类似问题

[hgsw]

4090 24G 显卡 纯python部署 代码分支 qanything-python 搜索 gpu_memory_utilization 字段， 修改文件qanything_kernel/utils/general_utils.py 205行 gpu_memory_utilization = 0.89； 208行7B模型修改也修改0.89； 这里字段gpu_memory_in_GB打印应该为24G

希望能帮助后面的人。

[mrhan1993]

把 round(8 / gpu_memory_in_GB, 2) 中的 8 改成 10