Introduce metadata-driven AI model features and parameters - #122
Open
MikeAlhayek wants to merge 8 commits into
Open
Introduce metadata-driven AI model features and parameters#122MikeAlhayek wants to merge 8 commits into
MikeAlhayek wants to merge 8 commits into
Conversation
Replaces hardcoded, provider-specific model options with an extensible registry of model features (binary capabilities) and model parameters (configurable options carrying kind, allowed values, ranges, and defaults). AI deployments declare which registered definitions their model exposes and may narrow the allowed values, default, or numeric bounds. AI profiles, profile templates, and chat interactions store the selected values, and the runtime binds them into the outgoing request. Values for parameters a deployment does not expose are never rendered and never sent. Framework: - adds the capability abstractions, descriptors, and metadata models - adds AddCoreAIModelCapabilities, AddAIModelFeature, and AddAIModelParameter - registers eight built-in features and the reasoningEffort parameter - adds DefaultAIModelCapabilityService to merge registrations with deployment metadata without mutating the registered descriptors - adds ModelParametersAICompletionServiceHandler as the single enforcement point, with deployment-default fallback and binder dispatch - adds IAIModelParameterBinder and a reasoning effort binder that sets ChatOptions.Reasoning.Effort - maps the resolved effort onto ChatCompletionOptions.ReasoningEffortLevel in AzureOpenAICompletionClient so both request paths behave the same - adds CompletionServiceConfigureContext.Deployment - adds a ModelParameters front-matter key to the markdown template parser Sample hosts: - adds a deployment capability editor and a metadata-driven parameter editor to the AI deployment, profile, template, and chat interaction screens in CrestApps.Core.Mvc.Web - adds the equivalent ModelCapabilitiesEditor and ModelParametersEditor components to CrestApps.Core.Blazor.Web Docs: - adds the Model Capabilities guide and records the change in the 1.1.0 changelog Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
This pull request has merge conflicts. Please resolve those before requesting a review. |
- List only genuine trained model capabilities under "Trained features" (add image/audio/video input/output; drop provider-hosted web search) - Add AIModelFeatureDescriptor.EnabledByDefault; default toolCalling + streaming on - Add ModelFeaturesAICompletionServiceHandler to enforce features at runtime, stripping tools/ToolMode and JSON response format when not declared (opt-in on metadata) - MVC + Blazor: "Trained features" heading, default-on features, read-only capability badges in profile/template/chat editors - MVC: bootstrap-select searchable multi-select on deployment allowed-values pickers - Update docs and 1.1.0 changelog; add enforcement + registration tests Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Remove form-select class from the deployment allowed-values multiselect so the native control no longer shows nested inside the bootstrap-select picker - Upgrade @crestapps/bootstrap-select CDN assets from 1.2.0 to 1.2.1 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Add AIModelParameterDescriptor.RequiredFeature so a parameter can declare a dependency on a trained feature; wire the built-in reasoningEffort parameter to require the reasoning feature - MVC and Blazor deployment editors only show a dependent parameter while its feature is enabled, and clear/disable it when the feature is turned off so a contradictory config (e.g. reasoningEffort on a non-reasoning model) can't be saved - Update docs and 1.1.0 changelog; add RequiredFeature registration and Clone tests Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…n framework - Deployment editors now render a parameter that declares a RequiredFeature inline beneath its feature checkbox; the "Model parameters" heading only appears for parameters not linked to a feature - Extract the MVC parameter card into a _DeploymentModelParameterCard partial and the Blazor card into a shared RenderFragment to reuse in both locations - Enforce the parameter/feature relationship in the framework: DefaultAIModelCapabilityService.GetCapabilities now excludes a parameter whose RequiredFeature is not declared by the deployment, so the runtime handlers never apply it regardless of how the metadata was authored - Add framework tests for the new capability filtering; update docs and changelog Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add extensible model capability enforcement so unsupported request options never reach a provider, avoiding provider validation errors (e.g. HTTP 400). Framework: - Add shared ModelFeatureEnforcement helper that strips tools/ToolMode when toolCalling is not declared, removes a JSON response format when structuredOutputs is not declared, removes reasoning options when reasoning is not declared, removes the reasoning effort when the reasoningEffort parameter is not exposed, and coerces an unsupported effort to the deployment default (preserving other reasoning state). - Add CapabilityEnforcingChatClient and wire it into DefaultAIClientFactory as the terminal layer above the provider client so enforcement also runs for callers that resolve an IChatClient directly. Enforcement is opt-in to deployments that declare capability metadata; the capability service is optional so unconfigured hosts are unchanged. - Convert model parameter values written to AdditionalProperties to their typed primitive; parse integers through decimal for exact Int64 bounds, reject non-finite numbers, and skip values that cannot be represented. - Log the Azure ExtraHigh -> High reasoning-effort clamp instead of downgrading silently. - Replace the computerUse feature with videoOutput. UI: - Unify the Blazor and MVC deployment editors on bootstrap-select (1.2.1) and sort trained features alphabetically. Docs and tests updated; all tests pass with zero warnings. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Complete streaming requests as a single buffered (non-streaming) response when a deployment does not declare the streaming trained feature, replaying the result via ToChatResponseUpdates(). Enforced on both the client-factory path (CapabilityEnforcingChatClient) and the Azure SDK direct path (AzureOpenAICompletionClient). The Azure non-streaming completion no longer swallows OperationCanceledException, so a suppressed streaming fallback propagates cancellation like a normal streaming request. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
This pull request has merge conflicts. Please resolve those before requesting a review. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fix #120
Replaces hardcoded, provider-specific model options with an extensible registry of model features (binary capabilities) and model parameters (configurable options carrying kind, allowed values, ranges, and defaults).
AI deployments declare which registered definitions their model exposes and may narrow the allowed values, default, or numeric bounds. AI profiles, profile templates, and chat interactions store the selected values, and the runtime binds them into the outgoing request. Values for parameters a deployment does not expose are never rendered and never sent.
Framework:
Sample hosts:
Docs: