Add visionModel support #866
Conversation
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
| -sm, --small-model <string> small model for the run | ||
| -m, --model <string> 'large' model alias (default) | ||
| -sm, --small-model <string> 'small' alias model | ||
| -vm, --vision-model <string> 'vision' alias model |
There was a problem hiding this comment.
The use of 'large', 'small', and 'vision' as model aliases could benefit from additional clarification or context to ensure users understand what these terms specifically refer to in the context of the tool.
generated by pr-docs-review-commit
alias_clarity
| ) | ||
| return res | ||
| } | ||
| ) |
There was a problem hiding this comment.
The description for the system.vision_ask_image tool could be expanded to provide more context or examples of usage, which would enhance user understanding of its functionality.
generated by pr-docs-review-commit
tool_description
|
The changes in the Key Changes:
The implementation looks consistent, extending existing patterns to support the vision model. There are no apparent functional issues with the changes; they seem well-integrated and follow the existing architecture. LGTM 🚀
|
Investigator report, 65d81cf AnalysisThe log indicates that the TypeScript compilation process failed due to type incompatibility errors specifically related to the Errors Identified
Suggested FixTo address these type errors, add the Diff for Suggested Fix--- a/src/promptcontext.ts
+++ b/src/promptcontext.ts
@@ -300,6 +300,7 @@
function updateEnvVars(env: Environment, newVars: Record<string, any>) {
// Assuming proxifyVars ensures correct structure
if (newVars) {
+ newVars["copilot.terminalSelection?"] = newVars["copilot.terminalSelection?"] || ""; // Default value
env.vars = proxifyVars(env.vars)
}
}--- a/src/promptrunner.ts
+++ b/src/promptrunner.ts
@@ -89,6 +89,7 @@
function runPrompt(attrs: PromptParameters) {
const promptRunner = new PromptRunner(attrs)
+ attrs["copilot.terminalSelection?"] = attrs["copilot.terminalSelection?"] || ""; // Default value
return {
vars: attrs,
runner: promptRunner,
@@ -286,6 +287,7 @@
function executePrompt(vars: PromptParameters) {
const expanded = expandVariables(vars)
+ expanded.vars["copilot.terminalSelection?"] = expanded.vars["copilot.terminalSelection?"] || ""; // Default value
return {
dir: "/some/dir",
files: [],This fix ensures that the
|
|
It seems we're encountering persistent issues when trying to retrieve commit history and run details for the "vision-model" branch. This might be due to access restrictions, incorrect identifiers, or other repository-specific configurations. Here are a few steps we can take to address this:
If you have direct access to the repository or specific insights, please verify these points. Let me know if there is any additional information you can provide or if you would like to try another approach.
|
Detailed Analysis of the FailureFailed Run Details:
Errors Identified:
Root Cause:
Git Diff Information:Unfortunately, the specific diff for the source code could not be retrieved. However, these errors suggest a need to review recent changes in type definitions or interface contracts related to Recommended Fix:
Suggested Patch:Modify the type definitions to align with expected structures. The missing property // Update in src/types/prompt_template.d.ts
// Add the missing property in the PromptParameters type definition
interface PromptParameters {
// Existing properties
...
"copilot.terminalSelection?": string; // Add this line
}Report Summary:
HTML URLs to Relevant Information:
Implement the suggested patch and verify if the TypeScript compilation errors are resolved. Further investigation into recent changes in type definitions might be necessary.
|
| encoder | ||
| ) + "... (truncated)" | ||
| toolContent = | ||
| truncateTextToTokens( |
There was a problem hiding this comment.
The assignment of toolContent to itself is unnecessary and can be removed. It seems like a formatting issue.
generated by pr-review-commit
unnecessary_assignment
Introduce support for a vision model, allowing queries on images and enhancing model options to include vision capabilities. Update relevant interfaces and default settings accordingly.
🚀 New Vision Tool Added: Introduced the
vision_ask_imagetool to use a vision model to query images. This allows users to run image queries and get detailed responses about image content.📄 Documentation Updates:
vision_ask_imagetool.🛠️ CLI Enhancements:
-vm, --vision-modelis added for specifying a vision model.🐛 Configuration and Defaults:
DEFAULT_VISION_MODELto manage vision model defaults, similar to existing small and large model default configurations.🧪 Testing and API Updates:
visionModelconfiguration.🛡️ User-facing API Changes:
packages/core/src/types/prompt_template.d.ts, new types and interfaces for the vision model are added, making this a notable change in the public API.