Use OpenAI SDK to avoid setting apiVersion manually
This commit is contained in:
Sean Goedecke
+69
-67
@@ -1,25 +1,16 @@
|
||||
import * as core from '@actions/core'
|
||||
import ModelClient, {isUnexpected} from '@azure-rest/ai-inference'
|
||||
import {AzureKeyCredential} from '@azure/core-auth'
|
||||
import {GitHubMCPClient, executeToolCalls, MCPTool, ToolCall} from './mcp.js'
|
||||
import {handleUnexpectedResponse} from './helpers.js'
|
||||
import OpenAI from 'openai'
|
||||
import {GitHubMCPClient, executeToolCalls, ToolCall} from './mcp.js'
|
||||
|
||||
interface ChatMessage {
|
||||
role: string
|
||||
role: 'system' | 'user' | 'assistant' | 'tool'
|
||||
content: string | null
|
||||
tool_calls?: ToolCall[]
|
||||
}
|
||||
|
||||
interface ChatCompletionsRequestBody {
|
||||
messages: ChatMessage[]
|
||||
max_tokens: number
|
||||
model: string
|
||||
response_format?: {type: 'json_schema'; json_schema: unknown}
|
||||
tools?: MCPTool[]
|
||||
tool_call_id?: string
|
||||
}
|
||||
|
||||
export interface InferenceRequest {
|
||||
messages: Array<{role: string; content: string}>
|
||||
messages: Array<{role: 'system' | 'user' | 'assistant' | 'tool'; content: string}>
|
||||
modelName: string
|
||||
maxTokens: number
|
||||
endpoint: string
|
||||
@@ -45,33 +36,38 @@ export interface InferenceResponse {
|
||||
export async function simpleInference(request: InferenceRequest): Promise<string | null> {
|
||||
core.info('Running simple inference without tools')
|
||||
|
||||
const client = ModelClient(request.endpoint, new AzureKeyCredential(request.token), {
|
||||
userAgentOptions: {userAgentPrefix: 'github-actions-ai-inference'},
|
||||
const client = new OpenAI({
|
||||
apiKey: request.token,
|
||||
baseURL: request.endpoint,
|
||||
})
|
||||
|
||||
const requestBody: ChatCompletionsRequestBody = {
|
||||
messages: request.messages,
|
||||
const chatCompletionRequest: OpenAI.Chat.Completions.ChatCompletionCreateParams = {
|
||||
messages: request.messages as OpenAI.Chat.Completions.ChatCompletionMessageParam[],
|
||||
max_tokens: request.maxTokens,
|
||||
model: request.modelName,
|
||||
}
|
||||
|
||||
// Add response format if specified
|
||||
if (request.responseFormat) {
|
||||
requestBody.response_format = request.responseFormat
|
||||
// eslint-disable-next-line @typescript-eslint/no-explicit-any
|
||||
chatCompletionRequest.response_format = request.responseFormat as any
|
||||
}
|
||||
|
||||
const response = await client.path('/chat/completions').post({
|
||||
body: requestBody,
|
||||
})
|
||||
try {
|
||||
const response = await client.chat.completions.create(chatCompletionRequest)
|
||||
|
||||
if (isUnexpected(response)) {
|
||||
handleUnexpectedResponse(response)
|
||||
if ('choices' in response) {
|
||||
const modelResponse = response.choices[0]?.message?.content
|
||||
core.info(`Model response: ${modelResponse || 'No response content'}`)
|
||||
return modelResponse || null
|
||||
} else {
|
||||
core.error('Unexpected response format from OpenAI API')
|
||||
return null
|
||||
}
|
||||
} catch (error) {
|
||||
core.error(`OpenAI API error: ${error}`)
|
||||
throw error
|
||||
}
|
||||
|
||||
const modelResponse = response.body.choices[0].message.content
|
||||
core.info(`Model response: ${modelResponse || 'No response content'}`)
|
||||
|
||||
return modelResponse
|
||||
}
|
||||
|
||||
/**
|
||||
@@ -83,8 +79,9 @@ export async function mcpInference(
|
||||
): Promise<string | null> {
|
||||
core.info('Running GitHub MCP inference with tools')
|
||||
|
||||
const client = ModelClient(request.endpoint, new AzureKeyCredential(request.token), {
|
||||
userAgentOptions: {userAgentPrefix: 'github-actions-ai-inference'},
|
||||
const client = new OpenAI({
|
||||
apiKey: request.token,
|
||||
baseURL: request.endpoint,
|
||||
})
|
||||
|
||||
// Start with the pre-processed messages
|
||||
@@ -97,52 +94,57 @@ export async function mcpInference(
|
||||
iterationCount++
|
||||
core.info(`MCP inference iteration ${iterationCount}`)
|
||||
|
||||
const requestBody: ChatCompletionsRequestBody = {
|
||||
messages: messages,
|
||||
const chatCompletionRequest: OpenAI.Chat.Completions.ChatCompletionCreateParams = {
|
||||
messages: messages as OpenAI.Chat.Completions.ChatCompletionMessageParam[],
|
||||
max_tokens: request.maxTokens,
|
||||
model: request.modelName,
|
||||
tools: githubMcpClient.tools,
|
||||
tools: githubMcpClient.tools as OpenAI.Chat.Completions.ChatCompletionTool[],
|
||||
}
|
||||
|
||||
// Add response format if specified (only on first iteration to avoid conflicts)
|
||||
if (iterationCount === 1 && request.responseFormat) {
|
||||
requestBody.response_format = request.responseFormat
|
||||
// eslint-disable-next-line @typescript-eslint/no-explicit-any
|
||||
chatCompletionRequest.response_format = request.responseFormat as any
|
||||
}
|
||||
|
||||
const response = await client.path('/chat/completions').post({
|
||||
body: requestBody,
|
||||
})
|
||||
try {
|
||||
const response = await client.chat.completions.create(chatCompletionRequest)
|
||||
|
||||
if (isUnexpected(response)) {
|
||||
handleUnexpectedResponse(response)
|
||||
if (!('choices' in response)) {
|
||||
core.error('Unexpected response format from OpenAI API')
|
||||
return null
|
||||
}
|
||||
|
||||
const assistantMessage = response.choices[0]?.message
|
||||
const modelResponse = assistantMessage?.content
|
||||
const toolCalls = assistantMessage?.tool_calls
|
||||
|
||||
core.info(`Model response: ${modelResponse || 'No response content'}`)
|
||||
|
||||
messages.push({
|
||||
role: 'assistant',
|
||||
content: modelResponse || '',
|
||||
...(toolCalls && {tool_calls: toolCalls as ToolCall[]}),
|
||||
})
|
||||
|
||||
if (!toolCalls || toolCalls.length === 0) {
|
||||
core.info('No tool calls requested, ending GitHub MCP inference loop')
|
||||
return modelResponse || null
|
||||
}
|
||||
|
||||
core.info(`Model requested ${toolCalls.length} tool calls`)
|
||||
|
||||
// Execute all tool calls via GitHub MCP
|
||||
const toolResults = await executeToolCalls(githubMcpClient.client, toolCalls as ToolCall[])
|
||||
|
||||
// Add tool results to the conversation
|
||||
messages.push(...toolResults)
|
||||
|
||||
core.info('Tool results added, continuing conversation...')
|
||||
} catch (error) {
|
||||
core.error(`OpenAI API error: ${error}`)
|
||||
throw error
|
||||
}
|
||||
|
||||
const assistantMessage = response.body.choices[0].message
|
||||
const modelResponse = assistantMessage.content
|
||||
const toolCalls = assistantMessage.tool_calls
|
||||
|
||||
core.info(`Model response: ${modelResponse || 'No response content'}`)
|
||||
|
||||
messages.push({
|
||||
role: 'assistant',
|
||||
content: modelResponse || '',
|
||||
...(toolCalls && {tool_calls: toolCalls}),
|
||||
})
|
||||
|
||||
if (!toolCalls || toolCalls.length === 0) {
|
||||
core.info('No tool calls requested, ending GitHub MCP inference loop')
|
||||
return modelResponse
|
||||
}
|
||||
|
||||
core.info(`Model requested ${toolCalls.length} tool calls`)
|
||||
|
||||
// Execute all tool calls via GitHub MCP
|
||||
const toolResults = await executeToolCalls(githubMcpClient.client, toolCalls)
|
||||
|
||||
// Add tool results to the conversation
|
||||
messages.push(...toolResults)
|
||||
|
||||
core.info('Tool results added, continuing conversation...')
|
||||
}
|
||||
|
||||
core.warning(`GitHub MCP inference loop exceeded maximum iterations (${maxIterations})`)
|
||||
|
||||
Reference in New Issue
Block a user