mirror of
https://github.com/kvcache-ai/ktransformers.git
synced 2026-11-05 01:07:28 +00:00
[GH-ISSUE #1475] [Bug] kt0.3.2部署deepseek-r1-0528-q4_k_m回答一直重复相同内容 #984
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
kvcache-ai/ktransformers#984
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @Snowdestiny on GitHub (Aug 14, 2025).
Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/1475
检查清单
问题描述
图放下面了,上面第一个问题是问的who are you,刚开始思考2句就一直重复内容,最下面问的比较2个数值大小。速度只有1Token/s,求帮助谢谢~
复现步骤
编译kt的时候是直接用的 bash install.sh
模型用的deepseek-r1-0528-Q4_K_M.gguf
指令大概是(公司电脑,只有晚上空了才有时间用):
python -m ktransformers.local_chat --model_path /opt/unsloth/r1 --gguf_path /opt/unsloth/r1/DeepSeek-R1-0528-UD-Q4_K_M-00001-of-00009.gguf --optimize_config_path /root/ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat.yaml --cpu_infer 30 temperature 0.7 max_new_tokens 200
环境信息
硬件环境:
Ubuntu22.04
CPU:8454H (2个)
显卡:RTX 4090
内存:480G
磁盘:J.ZAO 5 SERIES 1TB SSD( 1024GB )
GLOWAY Basic+2TNVMe-M.2/80(2048GB )
软件版本:
cuda: 12.4
torch:2.5.1
pytorch:2.5.1
其他的记不住了,另外这台电脑上有多系统,模型没放在ubuntu的系统盘下,是通过挂载的方式加载的其他盘
@PPXGS commented on GitHub (Sep 11, 2025):
指定这个GGUF文件夹试试,不去指定单个文件
@Snowdestiny commented on GitHub (Sep 11, 2025):
收到~~
@Snowdestiny commented on GitHub (Sep 11, 2025):
已经解决了,我后面尝试重装各插件版本还有改配置文件和参数,最后找了个专门安装的给我全部重新装了一遍,cuda12.8其他版本忘了,然后还是一样的回答状态。那个人让我在电脑上重新下载模型,说他装的多,必须在部署的电脑上直接下载--我就重新下了次,,居然就正常了~!~
所以之前我装的没问题,就是模型文件问题--之前那模型文件是从其他电脑下载后拷贝过去的,哪怕是同样的下载路径