The limit is "fuzzy". Around 2.5-3MB depending on the content. Seems to apply to all models. (`JSON.

Rrichw Hi all. I've got my Worker running AI models, which is great. I can see on https...

K

kian•10/30/24, 4:16 PM

Your error mentions

8b

8b

8b

8b, but there is no

8b

8b

8b

8b Llama 3.2 model

K

kian•10/30/24, 4:17 PM

Sounds like you typo'd the model name, as you mentioned it's

@cf/meta/llama-3.2-3b-instruct

@cf/meta/llama-3.2-3b-instruct

@cf/meta/llama-3.2-3b-instruct

@cf/meta/llama-3.2-3b-instruct

RRaylight The limit is "fuzzy". Around 2.5-3MB depending on the content. Seems to apply to...

M

Martijn•10/30/24, 4:36 PM

The image is around 3,5mb so you recommend compressing them?

MMartijn The image is around 3,5mb so you recommend compressing them?

R

RaylightOP•10/30/24, 5:11 PM

Yeah, you could try that. The limit for compressed/random data is roughly 2.6MB. In my experience, sending smaller images (file size) has resulted in a significant improvement in latency.

RRaylight Yeah, you could try that. The limit for compressed/random data is roughly 2.6MB....

F

fry69•10/30/24, 5:14 PM

These are still gigantic image sizes for LLMs, it should be possible to downscale them to ~100-300kb size and get similar results.

F

fry69•10/30/24, 5:15 PM

Most likely the AI API will scale those images down anyway before processing.

M

Martijn•10/30/24, 5:37 PM

What format do you recommend to compress them to jpg? With what setting

MMartijn What format do you recommend to compress them to jpg? With what setting

I

Isaac McFadyen•10/30/24, 5:38 PM

JPEG would work well. Keep in mind that LLMs actually use very small sizes internally anyway - a size of 640 pixels is good enough, no need for 4k images.

M

Martijn•10/30/24, 5:38 PM

With the object classification from fb i feel like if you resize the box dimensions dont work anymore

R

RaylightOP•10/30/24, 5:39 PM

Either way will work. Compressing or downscaling. The main thing is to get the file size down. When I tested I could shave of half a second of latency simply by sending the (1024x1024) image as a ~200kb jpg instead of a 2MB png.

MMartijn With the object classification from fb i feel like if you resize the box dimensi...

I

Isaac McFadyen•10/30/24, 5:39 PM

The box dimensions should be based on the pixels you give, so it should be consistent?

M

Martijn•10/30/24, 5:39 PM

Yeah but if I want to then visualize the boxes on the original image it wouldnt match anymore right

RRaylight Either way will work. Compressing or downscaling. The main thing is to get the f...

I

Isaac McFadyen•10/30/24, 5:39 PM

I'd recommend resizing over compressing. A compressed 4k image is still going to be resized internally by the model anyway, and by aggressively compressing big images you're going to lose some of the details.

MMartijn Yeah but if I want to then visualize the boxes on the original image it wouldnt ...

I

Isaac McFadyen•10/30/24, 5:40 PM

Just make sure to calculate the resize ratio.

M

Martijn•10/30/24, 5:40 PM

I guess yeah

I

Isaac McFadyen•10/30/24, 5:40 PM

If you have an image that is 3840 and resize to 640 then just multiple the returned coordinates by 6 (3840 divided by 640)

I

Isaac McFadyen•10/30/24, 5:41 PM

Just an FYI that I'm not picking 640 pixels arbitrarily - that's the actual size used by lots of models (ex. YOLO object detection)

M

Martijn•10/30/24, 5:42 PM

Kind of difficult since the images I want to classify are user input and could differ but I’m sure ill figure it out

MMartijn Kind of difficult since the images I want to classify are user input and could d...

I

Isaac McFadyen•10/30/24, 5:43 PM

Hmmm yeah. So depending on what your input pipeline looks like, I'd take the longest dimension (height or width) and then divide that by 640, and then resize to 640, send the image, and multiple all results by that ratio

I

Isaac McFadyen•10/30/24, 5:43 PM

Might be tricky if your pipeline is browser based but there's still options even then

M

Martijn•10/30/24, 5:44 PM

Users upload through an app, it gets stored in R2 and then i’ll likely run some automation on it for it to be classified

I

Isaac McFadyen•10/30/24, 5:44 PM

If you're using a Worker for that automation you can try

sharp

sharp

, I believe they have a WASM backend now that might work in Workers

M

Martijn•10/30/24, 5:45 PM

Probably be a job on our api thatll send it to the worker and update the db with the response

IIsaac McFadyen I'd recommend resizing over compressing. A compressed 4k image is still going to...

R

RaylightOP•10/30/24, 6:04 PM

Jpeg blocks are 8x8 pixels, so you're unlikely to lose details if the image is resized by a large factor. I believe resnet expects 224x224 as input, which would amount to ~150kb of uncompressed image data. If you're going to handle the resizing yourself you still have to decide on which kernel to use and in which color space to do the operation - e.g. lancoz, linear gamma. Worst case scenario is if you happen to resize to image dimensions that doesn't quite match the model, since the low res image will be resized once more. So idk, I'd probably go with lossy compression. Although if the image dimensions are huge, it could make sense to bring them down a bit.

H

hNico21•10/30/24, 9:21 PM

Hey, when I send

{
  "messages": [
    {
      "role": "system",
      "content": "You are a bot that parses user input into recipe ingredients..."
    },
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "[\"2 tbsp minced cilantro, leaves and stems\"]"
        }
      ]
    }
  ],
  "return_citations": true,
  "return_images": true
}

{
  "messages": [
    {
      "role": "system",
      "content": "You are a bot that parses user input into recipe ingredients..."
    },
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "[\"2 tbsp minced cilantro, leaves and stems\"]"
        }
      ]
    }
  ],
  "return_citations": true,
  "return_images": true
}

{
  "messages": [
    {
      "role": "system",
      "content": "You are a bot that parses user input into recipe ingredients..."
    },
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "[\"2 tbsp minced cilantro, leaves and stems\"]"
        }
      ]
    }
  ],
  "return_citations": true,
  "return_images": true
}

{
  "messages": [
    {
      "role": "system",
      "content": "You are a bot that parses user input into recipe ingredients..."
    },
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "[\"2 tbsp minced cilantro, leaves and stems\"]"
        }
      ]
    }
  ],
  "return_citations": true,
  "return_images": true
}

I'll get: Answer HTTP 400:

{
  "errors": [
    {
      "message": "AiError: Bad input: must have required property 'prompt', 'messages1content' must be string, must match exactly one schema in oneOf",
      "code": 5006
    }
  ],
  "success": false,
  "result": {},
  "messages": []
}

{
  "errors": [
    {
      "message": "AiError: Bad input: must have required property 'prompt', 'messages1content' must be string, must match exactly one schema in oneOf",
      "code": 5006
    }
  ],
  "success": false,
  "result": {},
  "messages": []
}

{
  "errors": [
    {
      "message": "AiError: Bad input: must have required property 'prompt', 'messages1content' must be string, must match exactly one schema in oneOf",
      "code": 5006
    }
  ],
  "success": false,
  "result": {},
  "messages": []
}

{
  "errors": [
    {
      "message": "AiError: Bad input: must have required property 'prompt', 'messages1content' must be string, must match exactly one schema in oneOf",
      "code": 5006
    }
  ],
  "success": false,
  "result": {},
  "messages": []
}

H

hNico21•10/30/24, 9:24 PM

so it says content must be a string, but Openai-API says string or Array https://platform.openai.com/docs/api-reference/messages/createMessage#messages-createmessage-content

A

ajgeiss0702•10/31/24, 3:31 AM

any idea how I can get more useful error messages? No matter what, I only get

AiError: AiError: Unknown internal error

AiError: AiError: Unknown internal error

A

ajgeiss0702•10/31/24, 3:31 AM

im trying to use a lora adapter

A

ajgeiss0702•10/31/24, 3:37 AM

please ping me if you reply

T

Tanis Wolfbane•10/31/24, 10:42 AM

Any one manage to get flux to work? When I run the exactly example in the documentation it errors with NSFW:

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run/@cf/black-forest-labs/flux-1-schnell    -X POST    -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN"    -d '{ "prompt": "cyberpunk cat" }'
{"errors":[{"message":"AiError: AiError: Input prompt contains NSFW content.","code":3030}],"success":false,"result":{},"messages":[]}

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run/@cf/black-forest-labs/flux-1-schnell    -X POST    -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN"    -d '{ "prompt": "cyberpunk cat" }'
{"errors":[{"message":"AiError: AiError: Input prompt contains NSFW content.","code":3030}],"success":false,"result":{},"messages":[]}

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run/@cf/black-forest-labs/flux-1-schnell    -X POST    -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN"    -d '{ "prompt": "cyberpunk cat" }'
{"errors":[{"message":"AiError: AiError: Input prompt contains NSFW content.","code":3030}],"success":false,"result":{},"messages":[]}

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run/@cf/black-forest-labs/flux-1-schnell    -X POST    -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN"    -d '{ "prompt": "cyberpunk cat" }'
{"errors":[{"message":"AiError: AiError: Input prompt contains NSFW content.","code":3030}],"success":false,"result":{},"messages":[]}

F

fry69•10/31/24, 2:32 PM

FYI -> https://discord.com/channels/595317990191398933/895794943182909470/1301548773620715563

F

Fanitullen•10/31/24, 6:59 PM

Am I the only one that can't get

@cf/black-forest-labs/flux-1-schnell

@cf/black-forest-labs/flux-1-schnell

to work? Even the short example at https://developers.cloudflare.com/workers-ai/models/flux-1-schnell/ is broken for me.

TTanis Wolfbane Any one manage to get flux to work? When I run the exactly example in the docume...

D

DaniFoldi•10/31/24, 7:02 PM

Hey, is anyone else still seeing this with flux?

I

Isaac McFadyen•10/31/24, 7:08 PM

We're going to try escalating it

D

DaniFoldi•10/31/24, 7:10 PM

Isaac has escalated this, we'll let you know if we have any updates

C

Chaika•10/31/24, 7:13 PM

Looks like the example is categorized as nsfw but most other inputs aren't

FFanitullen Am I the only one that can't get `@cf/black-forest-labs/flux-1-schnell` to work?...

R

RaylightOP•10/31/24, 7:14 PM

The example code doesn't work. Flux returns a json object with the image encoded as base64. So you need to decode it.

RRaylight The example code doesn't work. Flux returns a json object with the image encoded...

F

Fanitullen•10/31/24, 7:17 PM

Thank you, got it working now.

R

RaylightOP•10/31/24, 7:31 PM

A lot of prompts triggers the NSFW error. "cyber", "bar", "toe", "arm", "nose", "drill", "boot", "spy", "cow", "chicken" just from trying a few random one word prompts. "a cyberpunk cat" is ok, "cyberpunk cat" isn't.

RRaylight A *lot* of prompts triggers the NSFW error. "cyber", "bar", "toe", "arm", "nose"...

D

DaniFoldi•10/31/24, 9:45 PM

The team is looking into the filter, for the time being they recommend using longer prompts as they are less likely to be caught - they recommend at least 4-5 words

Aajgeiss0702 any idea how I can get more useful error messages? No matter what, I only get `A...

A

ajgeiss0702•10/31/24, 10:08 PM

any ideas?

Original message was deleted

R

RaylightOP•11/1/24, 9:06 AM

This. Although I wish https://developers.cloudflare.com/workers-ai/privacy/ were more clear. It feels well-intended and as if it is written to provide peace of mind, but "Cloudflare does not use .. to" can be read as "does use, except for these two things" in the same way that "may be stored by Cloudflare if you specifically use a storage service" can be read as "does not store, except if".

R

RaylightOP•11/1/24, 10:30 AM

Right now, however, I'm more concerned about Cloudflare throwing NSFW errors and wonder if this filter is Cloudflare's own doing, because if so then IMO Cloudflare is on a slippery slope of engaging in content moderation, and I can't find anything about NFSW in terms and conditions. So I would very much like to have some clarity regarding this. Might be raising an "unpopular option" thing here, but well.. I feel it is important.

Original message was deleted

F

fry69•11/1/24, 11:39 AM

?pings

F

Flare•11/1/24, 11:39 AM

Please do not ping community members for non-moderation reasons. Doing so will not solve your issue faster and will make people less likely to want to help you.

RRaylight This. Although I wish <https://developers.cloudflare.com/workers-ai/privacy/> we...

F

Fanitullen•11/1/24, 5:21 PM

The link you posted defines all input/output as Customer Content. This is a term specified in their Self-Serve Subscription Agreement. Specifically 2.5.1 states Subject to the terms of this Agreement, you hereby grant us a non-exclusive, fully sublicensable, worldwide, royalty-free right to collect, use, copy, store, transmit, modify and create derivative works of Customer Content, in each case to the extent necessary to provide the Services. With the current trend of excessive data retention among AI providers, I would be very hesitant to assume any sort of privacy regarding the use of these services. At least I find it hard to read it any other way.

A

ac•11/2/24, 3:39 PM

I am noticing a large number of AiError: 3001: Unknown internal error s when doing some AI/Vectorize queries (I believe when creating embeddings). Is there any way to get more insight into what these are? I have no clue if it's not able to get the model, if I'm getting rate limited, or if it's something I can control

LLogan Grasby We definitely don't intend to moderate content and we'll hopefully add documenta...

R

RaylightOP•11/2/24, 8:27 PM

That's very helpful and reassuring. Thanks for replying!

The limit is "fuzzy". Around 2.5-3MB depending on the content. Seems to apply to all models. (`JSON.

Similar Threads

Similar Threads

Similar Threads