DeploymentOutputEvaluationFailed: Unable to evaluate template outputs: 'deployments'

[ahotrod]

Dev container running in VSCode gets this error message:

"DeploymentOutputEvaluationFailed: Unable to evaluate template outputs: 'deployments'. Please see error details and deployment operations. Please see https://aka.ms/arm-common-errors for usage details.
DeploymentOutputEvaluationFailed: The template output 'deployments' is not valid: The language expression property 'outputs' doesn't exist, available properties are 'name, platform, roles, model, regions, role'.."

Checked "https://aka.ms/arm-common-errors" as suggested for more info, but "DeploymentOutputEvaluationFailed" is not listed.

Guidance for resolving this issue is appreciated.

Complete output after running "configure_models.py" and then "azd_up:":

vscode ➜ /workspaces/raft-distillation-recipe (main) $ azd up
? Select an Azure Subscription to use: 1. XXXXXXXXXXXXXX

Packaging services (azd package)

Provisioning Azure resources (azd provision)
Provisioning Azure resources can take some time.

Subscription: XXXXXXXXXXXXXX
Location: West US 3

You can view detailed progress in the Azure Portal:
https://portal.azure.com/#view/HubsExtension/DeploymentDetailsBlade/~/overview/id/XXXXXXXXXXXXXX

(✓) Done: Resource group: rg-raft-recipe (1.392s)
(✓) Done: Storage account: stxfixqzwraftrecipe (21.094s)
(✓) Done: Key Vault: kv-xfixqzw-raft-recipe (19.815s)
(✓) Done: Azure AI Services: aoai-xfixqzw-raft-recipe (23.307s)
(✓) Done: Azure AI Services Model Deployment: aoai-xfixqzw-raft-recipe/openai-text-embedding-ada-002 (801ms)
(✓) Done: Azure AI Services Model Deployment: aoai-xfixqzw-raft-recipe/openai-gpt-4 (1.077s)
(✓) Done: Machine Learning Workspace: ai-hub-xfixqzw-raft-recipe (18.742s)
(✓) Done: Machine Learning Connection: ai-hub-xfixqzw-raft-recipe/aoai-connection (4.517s)
(✓) Done: Machine Learning Workspace: ai-project-xfixqzw-raft-recipe (18.699s)
|== | Creating/Updating resources

ERROR: error executing step command 'provision': deployment failed: error deploying infrastructure: deploying to subscription:

Deployment Error Details:
DeploymentOutputEvaluationFailed: Unable to evaluate template outputs: 'deployments'. Please see error details and deployment operations. Please see https://aka.ms/arm-common-errors for usage details.
DeploymentOutputEvaluationFailed: The template output 'deployments' is not valid: The language expression property 'outputs' doesn't exist, available properties are 'name, platform, roles, model, regions, role'

[ahotrod]

.env for reference:

AZURE_ENV_NAME="raft-recipe"
AZURE_LOCATION="westus3"
AZURE_SUBSCRIPTION_ID="XXXXXXXXXXXXXX"
BASELINE_DEPLOYMENT_NAME="meta-llama-2-7b-chat"
BASELINE_MODEL_API="chat"
BASELINE_MODEL_NAME="Meta-Llama-3.1-8B-Instruct"
EMBEDDING_DEPLOYMENT_NAME="openai-text-embedding-ada-002"
EMBEDDING_MODEL_API="embeddings"
EMBEDDING_MODEL_NAME="text-embedding-ada-002"
JUDGE_DEPLOYMENT_NAME="openai-gpt-4"
JUDGE_MODEL_API="chat"
JUDGE_MODEL_NAME="gpt-4"
STUDENT_DEPLOYMENT_NAME="meta-llama-2-7b-chat"
STUDENT_MODEL_API="chat"
STUDENT_MODEL_NAME="Meta-Llama-3.1-8B-Instruct"
TEACHER_DEPLOYMENT_NAME="meta-llama-3-1-405B-chat"
TEACHER_MODEL_API="chat"
TEACHER_MODEL_NAME="Meta-Llama-3.1-405B-Instruct"

[cedricvidal]

Hey @ahotrod ,

thanks for your feedback. Would you be able to provide the log generated by running azd up --debug?

Also, can you provide the result of git rev-parse --short HEAD ?

Cedric

[ahotrod]

@cedricvidal - thanks for your prompt response.

This one is easy:
vscode ➜ /workspaces/raft-distillation-recipe (main) $ git rev-parse --short HEAD
33ee9d1

"azd up --debug" results in a lengthy output which exceeds the capacity of the Jupyter terminal display. I tried piping the terminal output to a text file, "azd up --debug >> azd_up--debug.txt", but only the error message from above shows up in the file. I emailed that to you. How else to capture the output and/or is there a portion of it you need?

[ahotrod]

Hi Cedric, File attached. Best, Dennis — Piped file from "azd up --debug >> azd_up--debug.txt" Initialize bicep provider Packaging services (azd package) Provisioning Azure resources (azd provision) Provisioning Azure resources can take some time. Initialize bicep provider Reading subscription and location from environment... Subscription: XXXXXXXXXXXXXX Location: West US 3 Comparing deployment state Creating/Updating resources You can view detailed progress in the Azure Portal: https://portal.azure.com/#view/HubsExtension/DeploymentDetailsBlade/~/overview/id/XXXXXXXXXXXXXX (✓) Done: Resource group: rg-raft-recipe (1.673s) (✓) Done: Key Vault: kv-xfixqzw-raft-recipe (599ms) (✓) Done: Storage account: stxfixqzwraftrecipe (517ms) (✓) Done: Azure AI Services: aoai-xfixqzw-raft-recipe (16.754s) (✓) Done: Azure AI Services Model Deployment: aoai-xfixqzw-raft-recipe/openai-text-embedding-ada-002 (247ms) (✓) Done: Azure AI Services Model Deployment: aoai-xfixqzw-raft-recipe/openai-gpt-4 (517ms) (✓) Done: Machine Learning Workspace: ai-hub-xfixqzw-raft-recipe (17.431s) (✓) Done: Machine Learning Connection: ai-hub-xfixqzw-raft-recipe/aoai-connection (1.127s) (✓) Done: Machine Learning Workspace: ai-project-xfixqzw-raft-recipe (17.222s) ERROR: error executing step command 'provision': deployment failed: error deploying infrastructure: deploying to subscription: Deployment Error Details: DeploymentOutputEvaluationFailed: Unable to evaluate template outputs: 'deployments'. Please see error details and deployment operations. Please see https://aka.ms/arm-common-errors for usage details. DeploymentOutputEvaluationFailed: The template output 'deployments' is not valid: The language expression property 'outputs' doesn't exist, available properties are 'name, platform, roles, model, regions, role'.. TraceID: c0298fa200695b62201de2a4fa428994

[ahotrod]

@cedricvidal

Line #191 in file /infra/main.bicep references "ai.outputs.deployments" as mentioned in the Deployment Error Details shown above.

Researching bicep to determine why "The language expression property 'outputs' doesn't exist", and verifying the listed "ai.name", "ai.platform", "ai.roles", etc. are the only available properties.

[ahotrod]

This is an awesome Azure Sample, thanks for sharing!

Which do you use, recommend using, and have worked for you:

 1) Codespaces,
 2) Dev Containers,
 3) virtual environments?

I tried Dev Containers first, thinking the automatic deployment would be easiest. Stuck with issue above. Next, I tried virtual environments (.venv & .conda) with manual model deployments via Azure AI Studio, both Meta-Llama & OpenAI GPT-4 models. Several issues and sorting them out is on the agenda, if recommended. One example, the environment variables in neither the .env.sample and the raft .env files match the corresponding callouts in the notebooks.

Thanks again.

[cedricvidal]

Hey Dennis,

Sorry for the lack of reply, I've been super busy, and I'm traveling for AI Tour for the next two weeks so not a lot of time to look into this. Have you made any progress or are you still stuck?

Line #191 in file /infra/main.bicep references "ai.outputs.deployments" as mentioned in the Deployment Error Details shown above.

This is very weird. outputs should be there, it's standard for a azd module to have outputs. This suggests a bigger problem but I sadly don't have the time to look into it now.

1. Codespaces,
2. Dev Containers,
3. virtual environments?

Dev Container is my daily driver.

This repo is verified to work with both Dev Container and Codespaces.

I haven't tried this repo with venv for some time

One example, the environment variables in neither the .env.sample and the raft .env files match the corresponding callouts in the notebooks.

Hum, yeah, it's possible that the doc and sample .env files are not up to date. Feel free to submit a PR if you make it work :)