github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #6953] AMD ROCm Card can not use flash attention #30161

New Issue
Open
opened 2026-04-22 09:39:37 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-22 09:39:37 -05:00
Owner
Copy Link

Originally created by @superligen on GitHub (Sep 25, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6953

What is the issue?

My cards is w7900, and rocm driver is 6.3 , I found the llama-cpp server started by Ollama always without -fa flag.

I check the code , found :

           // only cuda (compute capability 7+) and metal support flash attention
	if g.Library != "metal" && (g.Library != "cuda" || g.DriverMajor < 7) {
		flashAttnEnabled = false
	}

This code sames wrong.

Ref: https://github.com/Dao-AILab/flash-attention/pull/1010 Support for RoCM has been added tk flash attention 2

OS

Linux

GPU

AMD

CPU

Intel

Ollama version

No response

Originally created by @superligen on GitHub (Sep 25, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6953 ### What is the issue? My cards is w7900, and rocm driver is 6.3 , I found the llama-cpp server started by Ollama always without -fa flag. I check the code , found : // only cuda (compute capability 7+) and metal support flash attention if g.Library != "metal" && (g.Library != "cuda" || g.DriverMajor < 7) { flashAttnEnabled = false } This code sames wrong. Ref: https://github.com/Dao-AILab/flash-attention/pull/1010 Support for RoCM has been added tk flash attention 2 ### OS Linux ### GPU AMD ### CPU Intel ### Ollama version _No response_
GiteaMirror added the feature requestamd labels 2026-04-22 09:39:37 -05:00
GiteaMirror commented 2026-04-22 09:39:39 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Sep 25, 2024):

It looks like support may be limited to the MI200 and MI300 GPUs - https://github.com/ROCm/flash-attention?tab=readme-ov-file#amd-rocm-support

<!-- gh-comment-id:2375148194 --> @dhiltgen commented on GitHub (Sep 25, 2024): It looks like support may be limited to the MI200 and MI300 GPUs - https://github.com/ROCm/flash-attention?tab=readme-ov-file#amd-rocm-support
GiteaMirror commented 2026-04-22 09:39:40 -05:00
Author
Owner
Copy Link

@awz commented on GitHub (Sep 30, 2024):

I would love to see this work with multiple w7900s. I have 3x w7900 + 1x 6800 working without flash attention. I want 4x w7900 working with flash attention.

<!-- gh-comment-id:2381858784 --> @awz commented on GitHub (Sep 30, 2024): I would love to see this work with multiple w7900s. I have 3x w7900 + 1x 6800 working without flash attention. I want 4x w7900 working with flash attention.
GiteaMirror commented 2026-04-22 09:39:40 -05:00
Author
Owner
Copy Link

@LunNova commented on GitHub (Oct 9, 2024):

May be fixed when/if https://github.com/Dao-AILab/flash-attention/pull/1203 gets merged.

<!-- gh-comment-id:2403381031 --> @LunNova commented on GitHub (Oct 9, 2024): May be fixed when/if https://github.com/Dao-AILab/flash-attention/pull/1203 gets merged.
GiteaMirror commented 2026-04-22 09:39:41 -05:00
Author
Owner
Copy Link

@Crandel commented on GitHub (Dec 19, 2024):

May be fixed when/if Dao-AILab/flash-attention#1203 gets merged.

It was merged. Would it be possible to have it in Ollama?

<!-- gh-comment-id:2555631069 --> @Crandel commented on GitHub (Dec 19, 2024): > May be fixed when/if [Dao-AILab/flash-attention#1203](https://github.com/Dao-AILab/flash-attention/pull/1203) gets merged. It was merged. Would it be possible to have it in Ollama?
GiteaMirror commented 2026-04-22 09:39:41 -05:00
Author
Owner
Copy Link

@rtaic-coder commented on GitHub (Mar 28, 2025):

I would like this to be added as well.

<!-- gh-comment-id:2762757042 --> @rtaic-coder commented on GitHub (Mar 28, 2025): I would like this to be added as well.
GiteaMirror commented 2026-04-22 09:39:42 -05:00
Author
Owner
Copy Link

@xNefas commented on GitHub (Aug 7, 2025):

Friendly reminder, if you find this issue important, leave an upvote, not a comment! Numbers can go a long way in open source motivation!

<!-- gh-comment-id:3162150905 --> @xNefas commented on GitHub (Aug 7, 2025): Friendly reminder, if you find this issue important, leave an upvote, not a comment! Numbers can go a long way in open source motivation!
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label amd feature request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#30161
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?