exo icon indicating copy to clipboard operation
exo copied to clipboard

Two M3 Ultra are running the model qwen3-coder-480b-a35b-8bit. An error is thrown when using VS Code Client.

Open wangaocheng opened this issue 3 days ago • 24 comments

Hardware: Mac Studio M3 Ultra 512gb x 2

An error is thrown when using VS Code Client, while the normal chat mode works fine.

libc++abi: terminating due to uncaught exception of type std::runtime_error: [METAL] Command buffer execution failed: Caused GPU Timeout Error (00000002:kIOGPUCommandBufferCallbackErrorTimeout)

wangaocheng avatar Dec 21 '25 15:12 wangaocheng

Sounds like the other machine crashed - could be a particularly large prompt if its only happening during agentic workflows. Any more info?

Evanev7 avatar Dec 21 '25 17:12 Evanev7

I've also seen this issue when running DeepSeek V3.1 in pipeline parallel (it does not happen in tensor parallel). Have not encountered this in Qwen Coder, but do want to get to the bottom of this.

rltakashige avatar Dec 21 '25 21:12 rltakashige

@Evanev7

No additional information was available. After the error occurred, the model was unloaded from the memory of both Macs. Qwen3 Coder runs smoothly on LM Studio via the Cline plugin for VS Code. Since the prompt is very long, it seems that LM Studio splits the prompt into two segments before performing inference.

wangaocheng avatar Dec 22 '25 08:12 wangaocheng

ok - i'll assume this is an issue with chunking large prompts for now

Evanev7 avatar Dec 22 '25 11:12 Evanev7

ok - i'll assume this is an issue with chunking large prompts for now

Looking forward to hearing your good news soon. You’ve been working hard.

wangaocheng avatar Dec 22 '25 15:12 wangaocheng

understood appreciate the response

On Sun, Dec 21, 2025 at 3:06 PM wangaocheng @.***> wrote:

wangaocheng created an issue (exo-explore/exo#948) https://github.com/exo-explore/exo/issues/948

An error is thrown when using VS Code Client, while the normal chat mode works fine.

libc++abi: terminating due to uncaught exception of type std::runtime_error: [METAL] Command buffer execution failed: Caused GPU Timeout Error (00000002:kIOGPUCommandBufferCallbackErrorTimeout)

— Reply to this email directly, view it on GitHub https://github.com/exo-explore/exo/issues/948, or unsubscribe https://github.com/notifications/unsubscribe-auth/B3ZXLZLHWOH4BJSBVSS44T34C2ZN3AVCNFSM6AAAAACPVYGYE6VHI2DSMVQWIX3LMV43ASLTON2WKOZTG42TCMRSHE3TSMY . You are receiving this because you are subscribed to this thread.Message ID: @.***>

purewjawkh-tech avatar Dec 22 '25 17:12 purewjawkh-tech

got it thanks for confirming the behavior

On Sun, Dec 21, 2025 at 3:06 PM wangaocheng @.***> wrote:

wangaocheng created an issue (exo-explore/exo#948) https://github.com/exo-explore/exo/issues/948

An error is thrown when using VS Code Client, while the normal chat mode works fine.

libc++abi: terminating due to uncaught exception of type std::runtime_error: [METAL] Command buffer execution failed: Caused GPU Timeout Error (00000002:kIOGPUCommandBufferCallbackErrorTimeout)

— Reply to this email directly, view it on GitHub https://github.com/exo-explore/exo/issues/948, or unsubscribe https://github.com/notifications/unsubscribe-auth/B3ZXMEAM6QZYRD66ERBPLLD4C2ZN3AVCNFSM6AAAAACPVYGYE6VHI2DSMVQWIX3LMV43ASLTON2WKOZTG42TCMRSHE3TSMY . You are receiving this because you are subscribed to this thread.Message ID: @.***>

joanwhitegt-art avatar Dec 22 '25 17:12 joanwhitegt-art

thanks for taking a look at this

On Sun, Dec 21, 2025 at 3:06 PM wangaocheng @.***> wrote:

wangaocheng created an issue (exo-explore/exo#948) https://github.com/exo-explore/exo/issues/948

An error is thrown when using VS Code Client, while the normal chat mode works fine.

libc++abi: terminating due to uncaught exception of type std::runtime_error: [METAL] Command buffer execution failed: Caused GPU Timeout Error (00000002:kIOGPUCommandBufferCallbackErrorTimeout)

— Reply to this email directly, view it on GitHub https://github.com/exo-explore/exo/issues/948, or unsubscribe https://github.com/notifications/unsubscribe-auth/B3ZY5KOGI5B3X7OG3PXHBET4C2ZN5AVCNFSM6AAAAACPVYGYE6VHI2DSMVQWIX3LMV43ASLTON2WKOZTG42TCMRSHE3TSMY . You are receiving this because you are subscribed to this thread.Message ID: @.***>

miguelfoster959-max avatar Dec 22 '25 17:12 miguelfoster959-max

got it thanks for clearing that up

On Sun, Dec 21, 2025 at 5:31 PM Evan Quiney @.***> wrote:

Evanev7 left a comment (exo-explore/exo#948) https://github.com/exo-explore/exo/issues/948#issuecomment-3679127667

Sounds like the other machine crashed - could be a particularly large prompt if its only happening during agentic workflows. Any more info?

— Reply to this email directly, view it on GitHub https://github.com/exo-explore/exo/issues/948#issuecomment-3679127667, or unsubscribe https://github.com/notifications/unsubscribe-auth/B3Z2FD3IUJT4W4SJBKG2YJD4C3KQXAVCNFSM6AAAAACPVYGYE6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNZZGEZDONRWG4 . You are receiving this because you are subscribed to this thread.Message ID: @.***>

hhztzf-a11y avatar Dec 22 '25 17:12 hhztzf-a11y