Ollama Docker Rocm面临持续的问题,

kulphzqa  于 4个月前  发布在  Docker
关注(0)|答案(6)|浏览(132)

问题是什么?

尽管我在Fedora 40的Rocm安装中有一个一致的工作环境,目标是gfx1030,我也已经在Rocm和Vulkan目标以及甚至在CPU模式下的Ollama中完美地运行了llama.cpp,但是一直不断地在尝试运行ollama:rocm镜像时出现docker: layers from manifest don't match image configuration.错误。参考一下,在CPU模式下Docker中的正常ollama/ollama工作得很好。这个问题几乎普遍存在,在任何其他ollama:x.y.x-rocm标签中也存在。

操作系统:Linux,Docker

GPU:AMD

CPU:AMD

Ollama版本:0.3.6

j2cgzkjk

j2cgzkjk1#

我希望我能卸载pigz,但我在一个不可变的发行版上。

注意,我能够拉取ollama:latest,只是不能拉取:rocm。我也无法在Almada Linux 9机器上拉取它。

r8xiu3jd

r8xiu3jd2#

我正在Fedora 40上遇到相同的问题。对我来说,问题始于0.1.42-rocm之后。

命令输出:

docker pull ollama/ollama:0.3.6-rocm
0.3.6-rocm: Pulling from ollama/ollama
2d473b07cdd5: Pull complete 
b3800af90b90: Pull complete 
35d951731348: Pull complete 
93191fa4e1f8: Pull complete 
ec6b9c70f06a: Pull complete 
4767b6298c18: Pull complete 
bd707b33a96a: Pull complete 
10645e0d079c: Pull complete 
ba086e88d2e3: Pull complete 
f35b80e1d7b6: Pull complete 
13bfe13f7f56: Pull complete 
efc528f475fe: Pull complete 
1db71b1d7c67: Pull complete 
4177e5cd44fe: Pull complete 
f1fe20d85e24: Pull complete 
layers from manifest don't match image configuration

Docker信息:

Client: Docker Engine - Community
 Version:           27.1.2
 API version:       1.46
 Go version:        go1.21.13
 Git commit:        d01f264
 Built:             Mon Aug 12 11:52:28 2024
 OS/Arch:           linux/amd64
 Context:           default

Server: Docker Engine - Community
 Engine:
  Version:          27.1.2
  API version:      1.46 (minimum version 1.24)
  Go version:       go1.21.13
  Git commit:       f9522e5
  Built:            Mon Aug 12 11:50:46 2024
  OS/Arch:          linux/amd64
  Experimental:     false
 containerd:
  Version:          1.7.20
  GitCommit:        8fc6bcff51318944179630522a095cc9dbf9f353
 runc:
  Version:          1.1.13
  GitCommit:        v1.1.13-0-g58aa920
 docker-init:
  Version:          0.19.0
  GitCommit:        de40ad0
2skhul33

2skhul333#

在podman中基本上也存在相同的问题:

Trying to pull docker.io/ollama/ollama:rocm...
Getting image source signatures
Copying blob 4767b6298c18 skipped: already exists  
Copying blob 2d473b07cdd5 skipped: already exists  
Copying blob bd707b33a96a skipped: already exists  
Copying blob b3800af90b90 skipped: already exists  
Copying blob 93191fa4e1f8 skipped: already exists  
Copying blob ec6b9c70f06a skipped: already exists  
Copying blob 10645e0d079c skipped: already exists  
Copying blob f35b80e1d7b6 skipped: already exists  
Copying blob ba086e88d2e3 skipped: already exists  
Copying blob 13bfe13f7f56 skipped: already exists  
Copying blob efc528f475fe skipped: already exists  
Copying blob 3c9d729a222e done   | 
Copying blob 1db71b1d7c67 done   | 
Copying blob c630fdf8b6e2 done   | 
Copying blob 35d951731348 skipped: already exists  
Error: writing blob: storing blob to file "/var/tmp/container_images_storage1414294462/2": pigz: skipping: <stdin>: corrupted -- crc32 mismatch: exit status 1

正在运行的命令是 podman pull ollama/ollama:rocm

Client:       Podman Engine
Version:      5.2.1
API Version:  5.2.1
Go Version:   go1.22.5
Built:        Tue Aug 13 18:00:00 2024
OS/Arch:      linux/amd64
bfnvny8b

bfnvny8b4#

我从3.7.0-rc6-rocm开始,逆序遍历每个版本,发现最近一个正确拉取的版本是0.2.5-rocm。
看起来在0.2.5和0.2.6之间有很多变化:diff。但我能看到的唯一与Docker镜像构建过程相关的更改是从rocm版本6.1.1升级到6.1.2。
这是我的知识允许我调查此问题的范围。

rggaifut

rggaifut5#

fedora pigz包存在问题,如果只是卸载它并拉取docker镜像,应该可以正常工作。此外,这个问题似乎与#5979重复。

9avjhtql

9avjhtql6#

fedora pigz包存在问题,如果你只是卸载它并拉取docker镜像,它应该可以正常工作。此外,这个问题似乎是一个重复的问题#5979
嘿,确实是个问题,我现在已经让它正常工作了。

相关问题