InvalidOperationException: Tensor data cannot be read from, use .ReadbackAndClone() to allow reading from tensor.

[HeyThisWasRandomlyMade]

Describe the bug
When setting the inference device to "Compute Shader" in Unity ML-Agents, this error occurs:

InvalidOperationException: Tensor data cannot be read from, use .ReadbackAndClone() to allow reading from tensor.

Burst works fine, this is probably related to me using unity 6 preview.

To Reproduce
Steps to reproduce the behavior:

1. Go to Behavior Parameters in any agent and set inference device to Compute Shader.
2. Press play and you'll get this error.

Console logs / stack traces

InvalidOperationException: Tensor data cannot be read from, use .ReadbackAndClone() to allow reading from tensor.
Unity.Sentis.Tensor.GetItem[T] (System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/Tensor.cs:257)
Unity.Sentis.Tensor`1[T].get_Item (System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/TensorGeneric.cs:256)
Unity.Sentis.Tensor`1[T].get_Item (System.Int32 d1, System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/TensorGeneric.cs:246)
Unity.MLAgents.Inference.ContinuousActionOutputApplier.Apply (Unity.MLAgents.Inference.TensorProxy tensorProxy, System.Collections.Generic.IList`1[T] actionIds, System.Collections.Generic.Dictionary`2[TKey,TValue] lastActions) (at ./Packages/com.unity.ml-agents/Runtime/Inference/ApplierImpl.cs:47)
Unity.MLAgents.Inference.TensorApplier.ApplyTensors (System.Collections.Generic.IReadOnlyList`1[T] tensors, System.Collections.Generic.IList`1[T] actionIds, System.Collections.Generic.Dictionary`2[TKey,TValue] lastActions) (at ./Packages/com.unity.ml-agents/Runtime/Inference/TensorApplier.cs:108)
Unity.MLAgents.Inference.ModelRunner.DecideBatch () (at ./Packages/com.unity.ml-agents/Runtime/Inference/ModelRunner.cs:242)
Unity.MLAgents.Policies.SentisPolicy.DecideAction () (at ./Packages/com.unity.ml-agents/Runtime/Policies/SentisPolicy.cs:134)
Unity.MLAgents.Agent.DecideAction () (at ./Packages/com.unity.ml-agents/Runtime/Agent.cs:1411)
Unity.MLAgents.Academy.EnvironmentStep () (at ./Packages/com.unity.ml-agents/Runtime/Academy.cs:590)
Unity.MLAgents.AcademyFixedUpdateStepper.FixedUpdate () (at ./Packages/com.unity.ml-agents/Runtime/Academy.cs:43)

Environment (please complete the following information):

• Unity Version: Unity 6 Preview, 6000.0.22f1
• OS + version: Kubuntu 24.04
• ML-Agents version: Latest development
• Torch version: 2.4.1+rocm6.1

[paulo101977]

Describe the bug When setting the inference device to "Compute Shader" in Unity ML-Agents, this error occurs:

InvalidOperationException: Tensor data cannot be read from, use .ReadbackAndClone() to allow reading from tensor.

Burst works fine, this is probably related to me using unity 6 preview.

To Reproduce Steps to reproduce the behavior:

1. Go to Behavior Parameters in any agent and set inference device to Compute Shader.
2. Press play and you'll get this error.

Console logs / stack traces

InvalidOperationException: Tensor data cannot be read from, use .ReadbackAndClone() to allow reading from tensor.
Unity.Sentis.Tensor.GetItem[T] (System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/Tensor.cs:257)
Unity.Sentis.Tensor`1[T].get_Item (System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/TensorGeneric.cs:256)
Unity.Sentis.Tensor`1[T].get_Item (System.Int32 d1, System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/TensorGeneric.cs:246)
Unity.MLAgents.Inference.ContinuousActionOutputApplier.Apply (Unity.MLAgents.Inference.TensorProxy tensorProxy, System.Collections.Generic.IList`1[T] actionIds, System.Collections.Generic.Dictionary`2[TKey,TValue] lastActions) (at ./Packages/com.unity.ml-agents/Runtime/Inference/ApplierImpl.cs:47)
Unity.MLAgents.Inference.TensorApplier.ApplyTensors (System.Collections.Generic.IReadOnlyList`1[T] tensors, System.Collections.Generic.IList`1[T] actionIds, System.Collections.Generic.Dictionary`2[TKey,TValue] lastActions) (at ./Packages/com.unity.ml-agents/Runtime/Inference/TensorApplier.cs:108)
Unity.MLAgents.Inference.ModelRunner.DecideBatch () (at ./Packages/com.unity.ml-agents/Runtime/Inference/ModelRunner.cs:242)
Unity.MLAgents.Policies.SentisPolicy.DecideAction () (at ./Packages/com.unity.ml-agents/Runtime/Policies/SentisPolicy.cs:134)
Unity.MLAgents.Agent.DecideAction () (at ./Packages/com.unity.ml-agents/Runtime/Agent.cs:1411)
Unity.MLAgents.Academy.EnvironmentStep () (at ./Packages/com.unity.ml-agents/Runtime/Academy.cs:590)
Unity.MLAgents.AcademyFixedUpdateStepper.FixedUpdate () (at ./Packages/com.unity.ml-agents/Runtime/Academy.cs:43)

Environment (please complete the following information):

• Unity Version: Unity 6 Preview, 6000.0.22f1
• OS + version: Kubuntu 24.04
• ML-Agents version: Latest development
• Torch version: 2.4.1+rocm6.1

Probably Tensorflow version incompatibility. I was opening an old project here and had the same error.

[paulo101977]

Describe the bug When setting the inference device to "Compute Shader" in Unity ML-Agents, this error occurs:

InvalidOperationException: Tensor data cannot be read from, use .ReadbackAndClone() to allow reading from tensor.

Burst works fine, this is probably related to me using unity 6 preview.

To Reproduce Steps to reproduce the behavior:

1. Go to Behavior Parameters in any agent and set inference device to Compute Shader.
2. Press play and you'll get this error.

Console logs / stack traces

InvalidOperationException: Tensor data cannot be read from, use .ReadbackAndClone() to allow reading from tensor.
Unity.Sentis.Tensor.GetItem[T] (System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/Tensor.cs:257)
Unity.Sentis.Tensor`1[T].get_Item (System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/TensorGeneric.cs:256)
Unity.Sentis.Tensor`1[T].get_Item (System.Int32 d1, System.Int32 d0) (at ./Library/PackageCache/com.unity.sentis@a9353f8ec25d/Runtime/Core/TensorGeneric.cs:246)
Unity.MLAgents.Inference.ContinuousActionOutputApplier.Apply (Unity.MLAgents.Inference.TensorProxy tensorProxy, System.Collections.Generic.IList`1[T] actionIds, System.Collections.Generic.Dictionary`2[TKey,TValue] lastActions) (at ./Packages/com.unity.ml-agents/Runtime/Inference/ApplierImpl.cs:47)
Unity.MLAgents.Inference.TensorApplier.ApplyTensors (System.Collections.Generic.IReadOnlyList`1[T] tensors, System.Collections.Generic.IList`1[T] actionIds, System.Collections.Generic.Dictionary`2[TKey,TValue] lastActions) (at ./Packages/com.unity.ml-agents/Runtime/Inference/TensorApplier.cs:108)
Unity.MLAgents.Inference.ModelRunner.DecideBatch () (at ./Packages/com.unity.ml-agents/Runtime/Inference/ModelRunner.cs:242)
Unity.MLAgents.Policies.SentisPolicy.DecideAction () (at ./Packages/com.unity.ml-agents/Runtime/Policies/SentisPolicy.cs:134)
Unity.MLAgents.Agent.DecideAction () (at ./Packages/com.unity.ml-agents/Runtime/Agent.cs:1411)
Unity.MLAgents.Academy.EnvironmentStep () (at ./Packages/com.unity.ml-agents/Runtime/Academy.cs:590)
Unity.MLAgents.AcademyFixedUpdateStepper.FixedUpdate () (at ./Packages/com.unity.ml-agents/Runtime/Academy.cs:43)

Environment (please complete the following information):

• Unity Version: Unity 6 Preview, 6000.0.22f1
• OS + version: Kubuntu 24.04
• ML-Agents version: Latest development
• Torch version: 2.4.1+rocm6.1

I think I discovered where the problem is: for me, the error only occurs when I choose the 'Pixel Shader' as the inference device. Both Computer Shader and Pixel Shader use the GPU. Switching to CPU the problem disappears.