Nvidia NIM provider error(via custom proxy in github actions agent) #1799

Closed
opened 2026-02-16 17:32:40 -05:00 by yindo · 6 comments
Owner

Originally created by @Mirrowel on GitHub (Sep 22, 2025).

Originally assigned to: @fwang on GitHub.

Creating comment...
opencode session ses_68bdb071fffetigw0iHuswRYqv
Checking out new branch...
Switched to a new branch 'opencode/issue5-20250923011644'
Fetching prompt data for issue...
Sending message to opencode...

Let me check the relevant files to understand how logging is currently implemented. I'll look at the error handling and logging components.

59469 |       constructor(data3, options2) {
59470 |         super(name2, options2);
59471 |         this.data = data3;
59472 |         this.name = name2;
59473 |       }
59474 |         return "name" in input && input.name === name2;
                                 ^
TypeError: "Failed to complete the streaming request. Last error: litellm.InternalServerError: InternalServerError: Nvidia_nimException - Internal server error: can only concatenate str (not "dict") to str" is not an Object. (evaluating '"name" in input')
      at isInstance (/$bunfs/root/opencode:59474:26)
      at process (/$bunfs/root/opencode:113636:56)

This happens when using models hosted by Nvidia_Nim.

Context: I am using opencode github app via actions, and have it set up on a separate repo(copy of existing) for testing. I use my own hosted proxy(not litellm proxy) to route requests to many possible providers via same openai-compatible endpoint. This was never an issue, until this happened. This exact error was on deepseek-r1-0528 via nvidia nim.
Gemini via same proxy works fine, not counting how bad gemini is at following instructions and formatting tool calls.

Originally created by @Mirrowel on GitHub (Sep 22, 2025). Originally assigned to: @fwang on GitHub. ``` Creating comment... opencode session ses_68bdb071fffetigw0iHuswRYqv Checking out new branch... Switched to a new branch 'opencode/issue5-20250923011644' Fetching prompt data for issue... Sending message to opencode... Let me check the relevant files to understand how logging is currently implemented. I'll look at the error handling and logging components. 59469 | constructor(data3, options2) { 59470 | super(name2, options2); 59471 | this.data = data3; 59472 | this.name = name2; 59473 | } 59474 | return "name" in input && input.name === name2; ^ TypeError: "Failed to complete the streaming request. Last error: litellm.InternalServerError: InternalServerError: Nvidia_nimException - Internal server error: can only concatenate str (not "dict") to str" is not an Object. (evaluating '"name" in input') at isInstance (/$bunfs/root/opencode:59474:26) at process (/$bunfs/root/opencode:113636:56) ``` This happens when using models hosted by Nvidia_Nim. Context: I am using opencode github app via actions, and have it set up on a separate repo(copy of existing) for testing. I use my own hosted proxy(not litellm proxy) to route requests to many possible providers via same openai-compatible endpoint. This was never an issue, until this happened. This exact error was on deepseek-r1-0528 via nvidia nim. Gemini via same proxy works fine, not counting how bad gemini is at following instructions and formatting tool calls.
yindo closed this issue 2026-02-16 17:32:40 -05:00
Author
Owner

@Mirrowel commented on GitHub (Sep 22, 2025):

Addendum:
After getting pinged by the action and inspecting it - this repo(official) uses a different way of running opencode agent on github, it seems. And opencode version that runs with the app is 0.5.5, which is very old vs the current one that this uses. I will check out if this would potentially solve the problem tomorrow.

@Mirrowel commented on GitHub (Sep 22, 2025): Addendum: After getting pinged by the action and inspecting it - this repo(official) uses a different way of running opencode agent on github, it seems. And opencode version that runs with the app is 0.5.5, which is very old vs the current one that this uses. I will check out if this would potentially solve the problem tomorrow.
Author
Owner

@rekram1-node commented on GitHub (Sep 22, 2025):

@Mirrowel this is normal opencode github action usage:

https://github.com/sst/opencode/blob/dev/.github/workflows/opencode.yml

this is using opencode for a discrete task:

https://github.com/sst/opencode/blob/dev/.github/workflows/duplicate-issues.yml

I don't think the issue is the github action though it appears that your proxy is erroring not opencode:

 can only concatenate str (not "dict") to str

that is a python error

@rekram1-node commented on GitHub (Sep 22, 2025): @Mirrowel this is normal opencode github action usage: https://github.com/sst/opencode/blob/dev/.github/workflows/opencode.yml this is using opencode for a discrete task: https://github.com/sst/opencode/blob/dev/.github/workflows/duplicate-issues.yml I don't think the issue is the github action though it appears that your proxy is erroring not opencode: ``` can only concatenate str (not "dict") to str ``` that is a python error
Author
Owner

@Mirrowel commented on GitHub (Sep 22, 2025):

@Mirrowel this is normal opencode github action usage:

https://github.com/sst/opencode/blob/dev/.github/workflows/opencode.yml

this is using opencode for a discrete task:

https://github.com/sst/opencode/blob/dev/.github/workflows/duplicate-issues.yml

I don't think the issue is the github action though it appears that your proxy is erroring not opencode:

 can only concatenate str (not "dict") to str

that is a python error

I noticed severe version mismatch between those workflows, as it installs 0.5.5 as action. If that is not a problem - that can be concluded.

I will look at the proxy, but i am not sure why this would happen with nvidia provider and only it, consistently. And only in opencode.
Will update later

@Mirrowel commented on GitHub (Sep 22, 2025): > [@Mirrowel](https://github.com/Mirrowel) this is normal opencode github action usage: > > https://github.com/sst/opencode/blob/dev/.github/workflows/opencode.yml > > this is using opencode for a discrete task: > > https://github.com/sst/opencode/blob/dev/.github/workflows/duplicate-issues.yml > > I don't think the issue is the github action though it appears that your proxy is erroring not opencode: > > ``` > can only concatenate str (not "dict") to str > ``` > > that is a python error I noticed severe version mismatch between those workflows, as it installs 0.5.5 as action. If that is not a problem - that can be concluded. I will look at the proxy, but i am not sure why this would happen with nvidia provider and only it, consistently. And only in opencode. Will update later
Author
Owner

@rekram1-node commented on GitHub (Sep 22, 2025):

I noticed severe version mismatch between those workflows, as it installs 0.5.5 as action. If that is not a problem - that can be concluded.

Yeah the github action is due for some updates, @fwang is more aware of the state of it than I am

@rekram1-node commented on GitHub (Sep 22, 2025): > I noticed severe version mismatch between those workflows, as it installs 0.5.5 as action. If that is not a problem - that can be concluded. Yeah the github action is due for some updates, @fwang is more aware of the state of it than I am
Author
Owner

@Mirrowel commented on GitHub (Sep 24, 2025):

After fixing the type error(was in the debug logging) i could not get litellm error to happen again, even though i logged 2 separate errors(litellm one cause the type error). In either case - the issue was indeed not with opencode, i think.

It is still inconsistent on github: got chats ending before model actually finishes doing what it is doing (https://opencode.ai/s/5KZJT0KU, https://opencode.ai/s/EjPDMlJm, https://opencode.ai/s/VMBt0QOO). Though that could be model-specific or generally related to being a project in active development - don't have that much time to extensively test.

Can be closed

@Mirrowel commented on GitHub (Sep 24, 2025): After fixing the type error(was in the debug logging) i could not get litellm error to happen again, even though i logged 2 separate errors(litellm one cause the type error). In either case - the issue was indeed not with opencode, i think. It is still inconsistent on github: got chats ending before model actually finishes doing what it is doing (https://opencode.ai/s/5KZJT0KU, https://opencode.ai/s/EjPDMlJm, https://opencode.ai/s/VMBt0QOO). Though that could be model-specific or generally related to being a project in active development - don't have that much time to extensively test. Can be closed
Author
Owner

@rekram1-node commented on GitHub (Sep 24, 2025):

Yeah that looks like model issues

@rekram1-node commented on GitHub (Sep 24, 2025): Yeah that looks like model issues
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: anomalyco/opencode#1799