Hugging Face Model Not Supported

Question

I have a question about the ARC "Hugging Face" Robot Skill.

I set up the API as instructed. I'm using a USB webcam (connected directly to the PC running ARC) and enabled "face" and "motion" tracking. I tried the first example from the Hugging Face Skill documentation, but all I get is this error:

enabled.","type":"invalid_request_error","param":"model","code":"model_not_supported"}}
20:55:27 Error: HTTP 400 from Hugging Face. {"error":{"message":"The requested model 'Qwen/Qwen2.5-VL-7B-Instruct' is not supported by any provider you have enabled.","type":"invalid_request_error","param":"model","code":"model_not_supported"}}

What am I missing?

Athena · Answer

That error is coming from Hugging Face’s provider routing, not from ARC itself. ## What it means Qwen/Qwen2.5-VL-7B-Instruct is a vision-language model, and Hugging Face is saying: > "You don’t currently have any enabled provider that can serve this model." So ARC is successfully sending the request, but Hugging Face is rejecting the model because your account/token/provider setup doesn’t have access to a backend that can run it. ## What to check ### 1) Verify the model supports the task you chose In the Hugging Face skill configuration, make sure you’re using the correct task: - Vision Chat for asking the model to see and respond - Object Detection for finding faces/objects in the image - Image Classification for labeling the whole image - Image Captioning for describing the scene If you enabled face and motion tracking, that sounds more like Object Detection, not Vision Chat. ### 2) Try a known-good default model for the selected task In the skill’s kebab/gear menu configuration, switch to Default for that task and test again. For example: - Object Detection: facebook/detr-resnet-50 - Image Captioning: Salesforce/blip-image-captioning-large These are generally easier to get working than a large vision-chat model. ### 3) Check Hugging Face model access / provider availability That specific error usually means one of these: - Your token is valid, but the model is not available on the enabled provider - The model may require a different provider than the one currently enabled in your Hugging Face account - You may need to select a different model that is supported by serverless inference Hugging Face has a support article for model-based robot use in ARC here: - https://synthiam.com/Support/Skills/Artificial-Intelligence/Hugging-Face?id=23410 ## Important note about your webcam Your USB webcam on the PC is fine. ARC can use that with the Camera Device skill. The webcam is not the issue here. Make sure you have: 1. Camera Device skill added 2. Webcam selected in that skill 3. Camera skill started 4. Hugging Face skill configured to use that camera feed If the camera is not running, Hugging Face would usually complain about no active camera, not model support. ## Best next step Try this: 1. Open the Hugging Face skill settings 2. Switch to Object Detection 3. Use the default detection model 4. Press Save 5. Run Query Now If that works, then the issue is specifically the Qwen vision-chat model/provider combo. ## If you want Vision Chat If your goal is the "robot sees and talks" behavior, you may need to: - choose a different vision model that your enabled provider supports, or - use a Hugging Face Inference Endpoint / compatible provider that supports that model ## Quick question Are you trying to do: - face detection / tracking, or - vision chat where the model describes the scene and answers questions? If you tell me which task mode you selected in the skill, I can point you to the exact model/settings to use in ARC.

RoboHappy · Answer

The image above shows my settings, This is the default setting according to the provided info about this skill. As shown I see a HTTP 400 error showing up, which I am guessing is an issue with connecting to the Huggyface site?

Athena · Answer

Thanks for the screenshot - that helps.

What I see is that you’re on Vision Chat with the default Qwen/Qwen2.5-VL-7B-Instruct model, and ARC is successfully reaching Hugging Face. So this is not a webcam problem and not a general "can’t connect to Hugging Face" problem.

What the error means

The key part is:

model_not_supported
The requested model 'Qwen/Qwen2.5-VL-7B-Instruct' is not supported by any provider you have enabled.

That means:

ARC is sending the request correctly
Hugging Face received it
Hugging Face does not have an enabled provider on your account/session that can run that model

So the issue is model/provider availability, not ARC connectivity.

Why the default model can fail

Even if the skill documentation lists a default model, Hugging Face serverless inference may not currently serve that model for your account/provider setup. Model availability can change, and some models require:

a different provider
an inference endpoint
accepted model terms
or a smaller/more common vision model

What to try next

1) Test with a simpler model

In the Hugging Face skill:

open the kebab/gear menu
go to the Hugging Face tab
change the model to something more commonly supported

If you want to test Vision Chat, try another vision model that is known to work with your provider setup.

If you want to test face/motion tracking, that should usually be done with Object Detection, not Vision Chat.

2) Make sure you selected the right task

From your description, you mentioned face and motion tracking. That sounds like:

Object Detection mode for detecting a face/person
then Camera Tracking to move servos

For that, the default object detection model is a much better test than Qwen vision chat.

3) If you want the exact default docs behavior

The docs may assume a provider setup that isn’t available in every Hugging Face account. If the default model doesn’t run, the fix is usually to pick a different supported model, not to change ARC networking.

In short

This is not a connection failure to Hugging Face.

It is Hugging Face saying:

"I received your request, but I can’t run that model with any provider currently enabled for your account."

Recommended next step

Try this in ARC:

Switch the Hugging Face skill to Object Detection
Leave the model at the default detection model
Make sure your Camera Device skill is started
Press Query Now

If that works, then your ARC setup and webcam are fine, and the issue is only the Vision Chat model choice.

Helpful reference

Hugging Face skill overview: https://synthiam.com/Support/Skills/Artificial-Intelligence/Hugging-Face?id=23410

If you want, I can help you pick a working Vision Chat model or walk you through setting up object/face tracking correctly in ARC step by step.

I am a customer support AI and still learning. Please verify important answers in our support documentation. For advanced AI features, use the AI Script Agent built into ARC.

RoboHappy · Answer

Yes, I would like some help on picking a working vision chat model since I have no idea on what to look for or setup. This is my first time doing this.

Athena · Answer

Absolutely - for a first-time setup, the safest path is to use a smaller, commonly supported vision model rather than the big default one.

What to look for

In Hugging Face Vision Chat mode, you want a model that:

supports vision-language / image-to-text / image-text-to-text
is available through the provider enabled on your Hugging Face account
is not a huge model that requires special access or an inference endpoint

The default Qwen/Qwen2.5-VL-7B-Instruct is a good example of a modern model, but it may not be available through your current provider setup, which is why you got the model_not_supported error.

Good first models to try

Try one of these in ARC one at a time:

HuggingFaceTB/SmolVLM-2B-Instruct
HuggingFaceTB/SmolVLM-Instruct
microsoft/Phi-3.5-vision-instruct
llava-hf/llava-1.5-7b-hf

If one fails with provider/model support, try the next one.

My recommendation for a beginner

Start with:

HuggingFaceTB/SmolVLM-2B-Instruct

It’s smaller and often easier to get working than larger vision models.

How to change it in ARC

Open the Hugging Face robot skill.
Press the kebab/gear menu for the skill settings.
Go to the Hugging Face tab.
Leave Task set to Vision Chat.
Replace the model name with one of the models above.
Press Save.
Press Query Now.

If it works, you should see a response in the log and $HfSpeak / $HfResponse should populate.

If it still errors

If you get another model_not_supported error, that means the selected model is still not available from your enabled Hugging Face provider.

In that case, the quickest way forward is:

try the next model in the list above
or use Object Detection mode first, which is much more likely to work out of the box for tracking a face/person

Important note about your goal

You mentioned face and motion tracking. For that, you may actually want:

Object Detection mode
Track label: person
Pan/Tilt servos assigned on the Camera Tracking tab

That is the correct setup if you want the camera to follow a person’s face/body. Vision Chat is better for "look at the camera and describe what you see."

Helpful ARC docs

Hugging Face skill:
https://synthiam.com/Support/Skills/Artificial-Intelligence/Hugging-Face?id=23410
Create a custom robot skill if you want to extend behavior later:
https://synthiam.com/Support/Create-Robot-Skill/Overview

Quick question

Do you want your robot to:

talk about what it sees with Vision Chat, or
track a person’s face with the camera?

If you tell me which one, I can give you the exact ARC settings to use next.

I am a customer support AI and still learning. Please verify important answers in our support documentation. For advanced AI features, use the AI Script Agent built into ARC.

RoboHappy · Answer

Not successful. I have tried all four of the models you recommended, none of them worked. I have tried different tracking settings in the camera,face or motion, or object. Again nothing is working.

Each time., I still get this message....

 20:38:08 Error: Attempting to read GetCurrentBitmapManaged while locked.
20:39:02 Error: Attempting to read GetCurrentBitmapManaged while locked.
20:39:42 Error: HTTP 400 from Hugging Face. {"error":{"message":"The requested model 'microsoft/Phi-3.5-vision-instruct' is not supported by any provider you have enabled.","type":"invalid\_request\_error","param":"model","code":"model\_not\_supported"}}
20:40:17 Error: Attempting to read GetCurrentBitmapManaged while locked.
20:41:00 Error: Attempting to read GetCurrentBitmapManaged while locked.

While I created the API thru the HuggyFace website, are there other settings on the website I need to set/choose, thats not mentioned in the original information of this skill?

Athena · Answer

Yes - there are two separate issues in your log, and the Hugging Face website setup is likely part of one of them. ## 1) model_not_supported = Hugging Face provider issue This one is still the same root problem: > The requested model ... is not supported by any provider you have enabled. That means your token is valid enough for ARC to reach Hugging Face, but your Hugging Face account is not enabled for a provider that can serve that model. ### What to check on Hugging Face Please log into Hugging Face and check: - Settings Access Tokens - Confirm the token is correct and not expired - Any billing / provider / inference settings - Some models require a provider that isn’t enabled by default - Model page access - Some models are gated and need you to accept terms on the model’s page The exact wording in the error is the giveaway: this is not ARC failing to connect; it’s Hugging Face rejecting the model because no enabled backend can run it. ## 2) Attempting to read GetCurrentBitmapManaged while locked This is a separate ARC camera-image contention problem. It usually means more than one skill is trying to read the camera frame at the same time, or the camera feed is being accessed while another process is holding the bitmap. ### Common causes - Camera Device is open in one skill, while Hugging Face is also trying to pull frames - Another vision skill is active at the same time - The camera stream is paused/starting/stopping repeatedly - Tracking is enabled in multiple places at once ### What to do Try this clean test setup: 1. New empty project 2. Add Camera Device 3. Select your USB webcam 4. Press Start on Camera Device 5. Add Hugging Face 6. In Hugging Face, use only one mode at a time: - either Vision Chat - or Object Detection - not multiple tracking modes in the Camera skill while testing 7. Press Query Now Also make sure no other robot skills are trying to use the camera. ## Important: for face tracking, don’t use Vision Chat If your goal is to track a face/person, use: - Camera Device for the webcam - Hugging Face Object Detection - target label: person - then use Camera Tracking tab for pan/tilt servos if you have them That is much more reliable than trying to use Vision Chat for tracking. ## Most likely answer to your question No, there usually isn’t a special hidden setting on Hugging Face that ARC forgot to mention. The big issue is: - the model/provider combination you’re trying is not enabled on Hugging Face, and - ARC is also seeing a camera frame lock conflict ## Next best step Please try this exact diagnostic: ### Clean camera test - Create a brand-new ARC project - Add only Camera Device - Start the webcam ### Then add Hugging Face - Use Object Detection - Try the default model - Do not enable other camera/vision skills If that works, then we know the camera lock issue was caused by another skill or setup. ## If you want, I can help you with either of these next: 1. a known-good ARC camera + Hugging Face test setup 2. a face-tracking setup using the Camera Device skill with pan/tilt 3. checking your Hugging Face token / provider settings step-by-step If you’d like, send a screenshot of: - the Hugging Face skill settings - the Camera Device skill - and your Hugging Face token/settings page with sensitive info hidden and I’ll help you narrow it down.

RoboHappy · Answer

a known-good ARC camera + Hugging Face test setup 2. a face-tracking setup using the Camera Device skill with pan/tilt 3. checking your Hugging Face token / provider settings step-by-step Yes, I would like a step by step of these three items......

Athena · Answer

Absolutely - let’s break this into three clean steps and keep each test isolated so we can find the exact failure point. --- # 1) Known-good ARC camera + Hugging Face test setup This is the best first test because it removes extra variables. ## Goal Confirm that: - your USB webcam works in ARC - the Camera Device skill can stream video - the Hugging Face skill can read one camera frame ## Steps ### A. Create a new project 1. Open ARC 2. Create a new blank project ### B. Add only the Camera Device skill 1. Go to Project Add Robot Skill 2. Add Camera Camera Device ### C. Configure the webcam 1. Open the Camera Device skill 2. Select your USB webcam from the device list 3. Choose a supported resolution like: - 320x240 or - 640x480 if the camera supports it 4. Press Start You should see the live camera image. ### D. Add the Hugging Face skill 1. Go to Project Add Robot Skill 2. Add Artificial Intelligence Hugging Face ### E. Configure Hugging Face 1. Open the kebab menu / gear menu on the Hugging Face skill 2. Go to the Hugging Face tab 3. Enter your Hugging Face access token 4. For now, test with Object Detection first, not Vision Chat 5. Use the default model for Object Detection 6. Press Save ### F. Query once 1. In the Hugging Face skill, press Query Now ## What should happen - The skill should process a frame from the camera - You should see results in the log - If a person is in view, it should detect them ## Important If you still get: - Attempting to read GetCurrentBitmapManaged while locked that usually means: - the camera is being accessed by something else, or - another vision skill is also trying to read it So for this test, make sure only Camera Device + Hugging Face are in the project. --- # 2) Face-tracking setup using Camera Device skill with pan/tilt This is the correct setup if you want the camera to physically follow a person. ## Goal Use the camera to detect a person and move pan/tilt servos. --- ## A. Add Camera Device If not already added: 1. Add Camera Device 2. Start the webcam --- ## B. Add pan/tilt servos You need two servos connected to the EZ-B: - pan servo - tilt servo Typical ports: - pan on D0 - tilt on D1 (You can use different ports if needed.) --- ## C. Add a servo control skill If you want manual servo testing first: 1. Add Auto Position or Servo skill 2. Test that both servos move correctly This is important before using tracking. --- ## D. Configure Camera Tracking 1. Open the Camera Device skill 2. Go to the Tracking tab 3. Enable Object Tracking 4. On the Camera Tracking or Servo Tracking section: - assign the pan servo - assign the tilt servo 5. Set tracking for person if available in the object detection mode 6. Adjust: - invert if movement is backward - deadband if it jitters - movement speed if it moves too fast --- ## E. Start the camera and test 1. Press Start on Camera Device 2. Let a person appear in the camera view 3. Watch whether the camera follows the person --- ## Best practice For face-following, Object Detection is usually better than Vision Chat. If you want it to track faces specifically, note: - Face Tracking in Camera Device can work for simpler detection - Object Detection with the label person is often more practical for follow behavior --- # 3) Checking your Hugging Face token / provider settings step-by-step This is the most likely reason for your model_not_supported error. ## What ARC needs ARC only needs: - a valid Hugging Face token - a model that is available through an enabled provider If Hugging Face says the model is not supported, it usually means the account/provider side needs adjustment. --- ## A. Verify the token 1. Sign in to Hugging Face 2. Go to your token page: - https://huggingface.co/settings/tokens 3. Confirm your token: - exists - is not expired - is copied exactly into ARC 4. In ARC, paste it into the Hugging Face skill and save ### Tip Use a read token unless the skill documentation says otherwise. --- ## B. Check if the model is gated Some models require you to accept terms before use. 1. Go to the model page on Hugging Face 2. Look for an Agree and access / Request access / Accept terms button 3. If present, accept the model terms while logged in This is especially important for vision models. --- ## C. Check provider availability Your error says: > not supported by any provider you have enabled That means your account may not have an enabled inference provider for that model. ### What to look for On Hugging Face, check: - billing / inference / provider-related settings - whether you’re using the free serverless option - whether the chosen model is actually hosted by that service ### Practical advice If a model gives model_not_supported, try a model that is: - smaller - more common - explicitly supported by serverless inference But even then, provider support can vary over time. --- ## D. Try the model page directly To confirm whether Hugging Face can serve it at all: 1. Open the model page in Hugging Face 2. Look for an Inference / Playground / Try it out area 3. If the model can’t run there either, the issue is definitely on the Hugging Face side --- # Recommended order to test Do these in this order: ### Test 1 - Camera only - Camera Device skill - webcam starts successfully ### Test 2 - Camera + Object Detection - add Hugging Face - use Object Detection - press Query Now ### Test 3 - Vision Chat - only after Test 2 works - then try Vision Chat with a supported vision model --- # Why you’re seeing two errors ## model_not_supported This is Hugging Face saying the model is not available through your enabled provider. ## GetCurrentBitmapManaged while locked This is ARC’s camera image being accessed while another process is already using it. That usually means the project needs to be simplified before testing. --- # Useful links - Hugging Face skill manual: https://synthiam.com/Support/Skills/Artificial-Intelligence/Hugging-Face?id=23410 - Camera Device skill: https://synthiam.com/Support/Skills/Camera/Camera-Device?id=16120 - ARC downloads: https://synthiam.com/Products/ARC --- # Quick question Would you like me to give you: 1. a very exact click-by-click setup for Camera Device + Object Detection, or 2. a click-by-click setup for pan/tilt face tracking? If you want, I can also help you inspect the Hugging Face model page and token settings one screenshot at a time.

RoboHappy · Answer

So I started with the 3 clean steps as suggested. Happy to report that everything worked. looked like opening a whole new project was best idea, along with latest huggingface update. I tried the object detection and face detection, and HF was able to report back what it was seeing. I would like the click-by-click setup for the pan/tilt face tracking

Athena · Answer

Great - that narrows it down nicely. Since the clean project worked, the earlier issue was almost certainly a project/camera-lock conflict, and the latest Hugging Face update helped too. Let’s set up pan/tilt face tracking in ARC step by step. ## What you need - Camera Device skill - 2 servos for pan and tilt - Hugging Face running in Object Detection mode - A camera view with a person in it --- # Click-by-click pan/tilt face tracking setup ## 1) Add and test the servos first Before camera tracking, make sure the servos move manually. ### Add a servo control 1. In ARC, go to Project 2. Select Add Robot Skill 3. Add a servo skill such as: - Auto Position, or - Servo 4. Assign: - Pan servo to one port, for example d0 - Tilt servo to another port, for example d1 ### Test movement 1. Move the pan servo a little left/right 2. Move the tilt servo up/down 3. If a servo moves backward, note that - we’ll fix it with Invert later If you already have a servo skill set up, you can skip this part. --- ## 2) Open the Camera Device skill 1. Open your Camera Device skill 2. Make sure the webcam is selected 3. Press Start 4. Confirm you can see live video --- ## 3) Enable Object Detection For face/person following, use Object Detection. 1. In the Camera Device skill, go to the Tracking tab 2. Enable Object Tracking 3. Make sure other tracking types are not competing while testing 4. If the skill gives you a list of labels, use person as the target If you want the robot to follow a face specifically, the detection usually still works best as person. --- ## 4) Assign pan and tilt servos 1. In the Camera Device skill, go to the Servo Tracking section 2. Enable Servo Tracking 3. Assign the pan servo to horizontal tracking 4. Assign the tilt servo to vertical tracking Typical setup: - Pan = left/right motion - Tilt = up/down motion If there’s an option for relative tracking versus absolute tracking: - Start with relative tracking if you want smoother centering behavior - Use absolute tracking only if you have a specific reason --- ## 5) Set movement direction correctly This is very important. If the camera moves the wrong way: - check Invert on the pan servo - check Invert on the tilt servo Examples: - If the face moves left but the camera turns right, invert pan - If the face moves up but the camera tilts down, invert tilt --- ## 6) Set tracking behavior Now tune the motion so it doesn’t jitter. Look for these settings in the camera tracking section: - Movement speed: start low or medium - Deadband: increase this a little if the servos hunt back and forth - Min/Max limits: keep servo motion safely within the physical range A good starting point: - speed: small to medium - deadband: moderate - min/max: the safe range for your mechanism --- ## 7) Start tracking 1. Leave the Camera Device skill running 2. Make sure a person is visible in the camera 3. Watch the servos move to center the person 4. If it overshoots, reduce speed 5. If it jitters, increase deadband --- # Suggested simple starting settings Use these as a baseline: - Camera Device: running - Tracking mode: Object Detection - Target label: person - Servo Tracking: enabled - Pan servo: assigned - Tilt servo: assigned - Invert: only if needed - Speed: low to medium - Deadband: moderate --- # Common fixes ## If the servos do nothing - confirm the servo skill is added - confirm the correct ports are selected - confirm the servos have power - confirm the Camera Device is actually detecting a person ## If the servos move the wrong way - invert pan or tilt ## If the servos jitter - increase deadband - lower movement speed - improve lighting ## If tracking is slow - use a lower camera resolution like 320x240 - make sure only one camera-processing skill is using the webcam --- # Helpful ARC references - Camera Device skill: https://synthiam.com/Support/Skills/Camera/Camera-Device?id=16120 - Hugging Face skill: https://synthiam.com/Support/Skills/Artificial-Intelligence/Hugging-Face?id=23410 - How to make a robot: https://synthiam.com/Support/Get-Started/how-to-make-a-robot/make-a-diy-robot --- If you want, I can do the next step as a literal click-by-click guide for your exact layout: 1. where to click in Camera Device, or 2. how to wire and configure the pan/tilt servos in ARC If you share a screenshot of your Camera Device skill tracking tab, I can mark exactly what to set.

RoboHappy

Upgrade to ARC Pro

What it means

What to check

1) Verify the model supports the task you chose

2) Try a known-good default model for the selected task

3) Check Hugging Face model access / provider availability

Important note about your webcam

Best next step

If you want Vision Chat

Quick question

What the error means

Why the default model can fail

What to try next

1) Test with a simpler model

2) Make sure you selected the right task

3) If you want the exact default docs behavior

In short

Recommended next step

Helpful reference

What to look for

Good first models to try

My recommendation for a beginner

How to change it in ARC

If it still errors

Important note about your goal

Helpful ARC docs

Quick question

1) model_not_supported = Hugging Face provider issue

What to check on Hugging Face

2) Attempting to read GetCurrentBitmapManaged while locked

Common causes

What to do

Important: for face tracking, don’t use Vision Chat

Most likely answer to your question

Next best step

Clean camera test

Then add Hugging Face

If you want, I can help you with either of these next:

1) `model_not_supported` = Hugging Face provider issue

2) `Attempting to read GetCurrentBitmapManaged while locked`