actions/ai-inference
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge with main

Browse Source
This commit is contained in:
Maarten van Diemen
2025-11-24 13:06:01 +01:00
parent 48758ceaff 5022b33bc1
commit 95443f8d18
7 changed files with 62 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
__tests__/helpers-inference.test.ts
View File

@@ -106,6 +106,8 @@ describe('helpers.ts - inference request building', () => {
undefined,
undefined,
'gpt-4',
undefined,
undefined,
100,
'https://api.test.com',
'test-token',
@@ -117,6 +119,8 @@ describe('helpers.ts - inference request building', () => {
{role: 'user', content: 'User message'},
],
modelName: 'gpt-4',
temperature: undefined,
topP: undefined,
maxTokens: 100,
endpoint: 'https://api.test.com',
token: 'test-token',
@@ -136,6 +140,8 @@ describe('helpers.ts - inference request building', () => {
'System prompt',
'User prompt',
'gpt-4',
undefined,
undefined,
100,
'https://api.test.com',
'test-token',
@@ -147,6 +153,8 @@ describe('helpers.ts - inference request building', () => {
{role: 'user', content: 'User prompt'},
],
modelName: 'gpt-4',
temperature: undefined,
topP: undefined,
maxTokens: 100,
endpoint: 'https://api.test.com',
token: 'test-token',

26
dist/index.js generated vendored
View File

@@ -49505,6 +49505,8 @@ async function simpleInference(request) {
messages: request.messages,
max_tokens: request.maxTokens,
model: request.modelName,
temperature: request.temperature,
top_p: request.topP,
};
// Add response format if specified
if (request.responseFormat) {
@@ -49539,6 +49541,8 @@ async function mcpInference(request, githubMcpClient) {
messages: messages,
max_tokens: request.maxTokens,
model: request.modelName,
temperature: request.temperature,
top_p: request.topP,
};
// Add response format if specified (only on final iteration to avoid conflicts with tool calls)
if (finalMessage && request.responseFormat) {
@@ -49694,12 +49698,14 @@ function buildResponseFormat(promptConfig) {
/**
* Build complete InferenceRequest from prompt config and inputs
*/
function buildInferenceRequest(promptConfig, systemPrompt, prompt, modelName, maxTokens, endpoint, token) {
function buildInferenceRequest(promptConfig, systemPrompt, prompt, modelName, temperature, topP, maxTokens, endpoint, token) {
const messages = buildMessages(promptConfig, systemPrompt, prompt);
const responseFormat = buildResponseFormat(promptConfig);
return {
messages,
modelName,
temperature,
topP,
maxTokens,
endpoint,
token,
@@ -52581,10 +52587,8 @@ function loadPromptFile(filePath, templateVariables = {}) {
throw new Error(`Prompt file not found: ${filePath}`);
}
const fileContent = fs.readFileSync(filePath, 'utf-8');
// Apply template variable substitution
const processedContent = replaceTemplateVariables(fileContent, templateVariables);
try {
const config = load(processedContent);
const config = load(fileContent);
if (!config.messages || !Array.isArray(config.messages)) {
throw new Error('Prompt file must contain a "messages" array');
}
@@ -52597,6 +52601,13 @@ function loadPromptFile(filePath, templateVariables = {}) {
throw new Error(`Invalid message role: ${message.role}`);
}
}
// Prepare messages by replacing template variables with actual content
config.messages = config.messages.map(msg => {
return {
...msg,
content: replaceTemplateVariables(msg.content, templateVariables),
};
});
return config;
}
catch (error) {
@@ -52644,7 +52655,10 @@ async function run() {
}
// Get common parameters
const modelName = promptConfig?.model || coreExports.getInput('model');
const maxTokens = parseInt(coreExports.getInput('max-tokens'), 10);
let maxTokens = promptConfig?.modelParameters?.maxTokens ?? coreExports.getInput('max-tokens');
if (typeof maxTokens === 'string') {
maxTokens = parseInt(maxTokens, 10);
}
const token = process.env['GITHUB_TOKEN'] || coreExports.getInput('token');
if (token === undefined) {
throw new Error('GITHUB_TOKEN is not set');
@@ -52654,7 +52668,7 @@ async function run() {
const githubMcpToolsets = coreExports.getInput('github-mcp-toolsets');
const endpoint = coreExports.getInput('endpoint');
// Build the inference request with pre-processed messages and response format
const inferenceRequest = buildInferenceRequest(promptConfig, systemPrompt, prompt, modelName, maxTokens, endpoint, token);
const inferenceRequest = buildInferenceRequest(promptConfig, systemPrompt, prompt, modelName, promptConfig?.modelParameters?.temperature, promptConfig?.modelParameters?.topP, maxTokens, endpoint, token);
const enableMcp = coreExports.getBooleanInput('enable-github-mcp') || false;
let modelResponse = null;
if (enableMcp) {

2
dist/index.js.map generated vendored
View File

File diff suppressed because one or more lines are too long

4
src/helpers.ts
View File

@@ -82,6 +82,8 @@ export function buildInferenceRequest(
systemPrompt: string | undefined,
prompt: string | undefined,
modelName: string,
temperature: number | undefined,
topP: number | undefined,
maxTokens: number,
endpoint: string,
token: string,
@@ -92,6 +94,8 @@ export function buildInferenceRequest(
return {
messages,
modelName,
temperature,
topP,
maxTokens,
endpoint,
token,

6
src/inference.ts
View File

@@ -15,6 +15,8 @@ export interface InferenceRequest {
maxTokens: number
endpoint: string
token: string
temperature?: number
topP?: number
responseFormat?: {type: 'json_schema'; json_schema: unknown} // Processed response format for the API
}
@@ -45,6 +47,8 @@ export async function simpleInference(request: InferenceRequest): Promise<string
messages: request.messages as OpenAI.Chat.Completions.ChatCompletionMessageParam[],
max_tokens: request.maxTokens,
model: request.modelName,
temperature: request.temperature,
top_p: request.topP,
}
// Add response format if specified
@@ -90,6 +94,8 @@ export async function mcpInference(
messages: messages as OpenAI.Chat.Completions.ChatCompletionMessageParam[],
max_tokens: request.maxTokens,
model: request.modelName,
temperature: request.temperature,
top_p: request.topP,
}
// Add response format if specified (only on final iteration to avoid conflicts with tool calls)

8
src/main.ts
View File

@@ -53,7 +53,11 @@ export async function run(): Promise<void> {
// Get common parameters
const modelName = promptConfig?.model || core.getInput('model')
const maxTokens = parseInt(core.getInput('max-tokens'), 10)
let maxTokens = promptConfig?.modelParameters?.maxTokens ?? core.getInput('max-tokens')
if (typeof maxTokens === 'string') {
maxTokens = parseInt(maxTokens, 10)
}
const token = process.env['GITHUB_TOKEN'] || core.getInput('token')
if (token === undefined) {
@@ -72,6 +76,8 @@ export async function run(): Promise<void> {
systemPrompt,
prompt,
modelName,
promptConfig?.modelParameters?.temperature,
promptConfig?.modelParameters?.topP,
maxTokens,
endpoint,
token,

20
src/prompt.ts
View File

@@ -7,9 +7,16 @@ export interface PromptMessage {
content: string
}
export interface ModelParameters {
maxTokens?: number
temperature?: number
topP?: number
}
export interface PromptConfig {
messages: PromptMessage[]
model?: string
modelParameters?: ModelParameters
responseFormat?: 'text' | 'json_schema'
jsonSchema?: string
}
@@ -101,11 +108,8 @@ export function loadPromptFile(filePath: string, templateVariables: TemplateVari
const fileContent = fs.readFileSync(filePath, 'utf-8')
// Apply template variable substitution
const processedContent = replaceTemplateVariables(fileContent, templateVariables)
try {
const config = yaml.load(processedContent) as PromptConfig
const config = yaml.load(fileContent) as PromptConfig
if (!config.messages || !Array.isArray(config.messages)) {
throw new Error('Prompt file must contain a "messages" array')
@@ -121,6 +125,14 @@ export function loadPromptFile(filePath: string, templateVariables: TemplateVari
}
}
// Prepare messages by replacing template variables with actual content
config.messages = config.messages.map(msg => {
return {
...msg,
content: replaceTemplateVariables(msg.content, templateVariables),
}
})
return config
} catch (error) {
throw new Error(`Failed to parse prompt file: ${error instanceof Error ? error.message : 'Unknown error'}`)