invalid keys | OpenRouter | Page 1

crisp saddle Aug 17, 2023, 4:51 AM

#

I think 403 is about moderation.

Error Codes (copied from https://openrouter.ai/docs#errors)

400: Bad Request (invalid or missing params, CORS)
401: Invalid credentials (OAuth session expired, disabled/invalid API key)
402: Out of credits
403: Your chosen model requires moderation and your input was flagged
408: Your request time out
429: You are being rate limited
502: Your chosen model is down or we received an invalid response from it

(I'm a user working on implementing code in my app to catch these errors, so I have no idea if this is relevant to you.)

sage arrow Aug 17, 2023, 5:06 AM

#

If your client is using fetch, I would serialize the json() output we returned with the error response - it should give you more details on why your input was flagged.

#

If you are calling Claude 2, 403 is indeed moderation, enforced by upstream API unfortunately.

#

For uncensored models, I'd recommend Llama2

fleet rover Aug 17, 2023, 8:44 AM

#

sage arrow If you are calling Claude 2, 403 is indeed moderation, enforced by upstream API ...

return is "Request failed with status code 403 API key disabled or exhausted"

sage arrow Aug 17, 2023, 9:19 AM

#

fleet rover return is "Request failed with status code 403 API key disabled or exhausted"

Does the key have any usage limit on it?

fleet rover Aug 17, 2023, 9:19 AM

#

sage arrow Does the key have any usage limit on it?

tried with both default of 5 and no limit

sage arrow Aug 17, 2023, 9:22 AM

#

fleet rover tried with both default of 5 and no limit

If you navigate to https://openrouter.ai/account, is your credits count positive?

fleet rover Aug 17, 2023, 9:23 AM

#

sage arrow If you navigate to https://openrouter.ai/account, is your credits count positive...

yes

#

or is that a rhetorical question

sage arrow Aug 17, 2023, 9:24 AM

#

Just making sure since I think you can use API key with the initial usage allowance

#

But not if you go under (it can go to negative)

fleet rover Aug 17, 2023, 9:26 AM

#

any ideas?

sage arrow Aug 17, 2023, 9:27 AM

#

Which model are you using btw?

fleet rover Aug 17, 2023, 9:27 AM

#

claude v1

sage arrow Aug 17, 2023, 9:28 AM

#

Does it work if you try over here: https://openrouter.ai/playground

OpenRouter

Authenticate and use your AI models in one place

#

(disable the other character, just pick 1 and swap to claude v1)

#

(also the default cut off is 142 max tokens so that shouldn't drain your credits. just a quick test to see if there's something wrong with your acc specifically or if there's something wrong with the whole system. I just tested it on there and seems to work for me :d)

fleet rover Aug 17, 2023, 9:30 AM

#

it works

sage arrow Aug 17, 2023, 9:30 AM

#

weird...

fleet rover Aug 17, 2023, 9:31 AM

#

also, I have managed to receive a response from gpt3.5 in the tavern

sage arrow Aug 17, 2023, 9:31 AM

#

???

fleet rover Aug 17, 2023, 9:31 AM

#

but not one from claude

sage arrow Aug 17, 2023, 9:31 AM

#

via openrouter??

fleet rover Aug 17, 2023, 9:32 AM

#

yeah

#

it seems gpt works, claude v1 doesn't

#

request appears properly in activity, key limit is adjusted correctly

sage arrow Aug 17, 2023, 9:34 AM

#

can you screenshot how you configured your sillytavern endpoint? Make sure you pick OpenRouter and not Claude itself

#

(also make sure to hide ur keys doh)

#

I suspect you might have picked the native claude v1 in the dropdown instead of OpenRouter. 403 in claude means wrong API key, to which that's indeed the case if you sent Claude an OpenRouter API key: https://docs.anthropic.com/claude/reference/errors-and-rate-limits#http-errors

Claude

Errors and rate limits

HTTP errorsOur API follows a predictable HTTP error code format: 400 - Invalid request: there was an issue with the format or content of your request.401 - Unauthorized: there's an issue with your API key.403 - Forbidden: your API key does not have permission to use the specified resource.404 - Not ...

fleet rover Aug 17, 2023, 9:37 AM

#

did I not mention that the requests worked until yesterday?

sage arrow Aug 17, 2023, 9:38 AM

#

yeah you did...

#

but why wouldn't it work now when gpt3.5 works lol

#

key disabled or exhausted <- also our API don't send this specific message back.

fleet rover Aug 17, 2023, 9:42 AM

#

do I need to intercept the packet to see what's wrong?

#

this is what tavern is showing me

#

obviously this is different from 403 Unauthorized when using a made up key

sage arrow Aug 17, 2023, 9:43 AM

#

This client right? https://github.com/SillyTavern/SillyTavern

fleet rover Aug 17, 2023, 9:43 AM

#

yes

sage arrow Aug 17, 2023, 9:44 AM

#

I tried looking up their repo for that string... nothing lol

#

The other explanation I can think of is that

#

claude v1 100k is pretty expensive per tokens: $0.01102 (in) $0.03268 (out)

fleet rover Aug 17, 2023, 9:46 AM

#

this is what your API returns
{"error":{"message":"API key disabled or exhausted"},"quota_error":false}

#

i intercepted the request

sage arrow Aug 17, 2023, 9:47 AM

#

yeah, and it's enclosed in a string right?

#

that means that error was proxied from upstream (i.e, claude is sending back that error)

#

if it's an error on our side, it would be a plain string message, not a json enclosed inside a string

fleet rover Aug 17, 2023, 9:48 AM

#

sage arrow claude v1 100k is pretty expensive per tokens: $0.01102 (in) $0.03268 (out)

the requests i'm making are about 0.1$ so it's not a funds issue

sage arrow Aug 17, 2023, 9:48 AM

#

yeah and it also worked on playground too

#

hmm.......

fleet rover Aug 17, 2023, 9:50 AM

#

all anthropic models send the same thing back

sage arrow Aug 17, 2023, 9:51 AM

#

I'm a bit baffled bc... if they were sending back err

#

we'd be seeing the exact issue on playground too

sage arrow Aug 17, 2023, 9:52 AM

#

fleet rover all anthropic models send the same thing back

including their latest v2 model right?

fleet rover Aug 17, 2023, 9:53 AM

#

yes

sage arrow Aug 17, 2023, 9:56 AM

#

The only potential diff between playground and tavern when it comes to API call is that we use streaming and tavern does not

fleet rover Aug 17, 2023, 9:59 AM

#

streaming works well in tavern

sage arrow Aug 17, 2023, 10:04 AM

#

fleet rover streaming works well in tavern

If it does... it's basically identical call as the playground :d.....

#

you're also on the latest branch of tavern right?

fleet rover Aug 17, 2023, 10:08 AM

#

sage arrow you're also on the latest branch of tavern right?

no

#

I'm not changing my setup though
again, as I've said, it worked before and I'm obviously not making any elementary mistakes

fleet rover Aug 17, 2023, 10:33 AM

#

@sage arrow i figured it out, it really was moderation, the antiquated tavern message error and the fact that you said only Claude 2 has moderation sent me sideways

#invalid keys