vetrag/llm.go

109 lines
3.4 KiB
Go

package main
import (
"bytes"
"context"
"encoding/json"
"net/http"
"strings"
"github.com/sirupsen/logrus"
)
// LLMClient abstracts LLM API calls
type LLMClient struct {
APIKey string
BaseURL string
}
// ExtractKeywords calls LLM to extract keywords from user message
func (llm *LLMClient) ExtractKeywords(ctx context.Context, message string) ([]string, error) {
prompt, err := renderPrompt(appConfig.LLM.ExtractKeywordsPrompt, map[string]string{"Message": message})
if err != nil {
logrus.WithError(err).Error("[CONFIG] Failed to render ExtractKeywords prompt")
return nil, err
}
logrus.WithField("prompt", prompt).Info("[LLM] ExtractKeywords prompt")
resp, err := llm.openAICompletion(ctx, prompt)
logrus.WithFields(logrus.Fields{"response": resp, "err": err}).Info("[LLM] ExtractKeywords response")
if err != nil {
return nil, err
}
var keywords []string
if err := json.Unmarshal([]byte(resp), &keywords); err == nil {
return keywords, nil
}
// fallback: try splitting by comma
for _, k := range bytes.Split([]byte(resp), []byte{','}) {
kw := strings.TrimSpace(string(k))
if kw != "" {
keywords = append(keywords, kw)
}
}
return keywords, nil
}
// DisambiguateBestMatch calls LLM to pick best match from candidates
func (llm *LLMClient) DisambiguateBestMatch(ctx context.Context, message string, candidates []Reason) (string, error) {
entries, _ := json.Marshal(candidates)
prompt, err := renderPrompt(appConfig.LLM.DisambiguatePrompt, map[string]string{"Entries": string(entries), "Message": message})
if err != nil {
logrus.WithError(err).Error("[CONFIG] Failed to render Disambiguate prompt")
return "", err
}
logrus.WithField("prompt", prompt).Info("[LLM] DisambiguateBestMatch prompt")
resp, err := llm.openAICompletion(ctx, prompt)
logrus.WithFields(logrus.Fields{"response": resp, "err": err}).Info("[LLM] DisambiguateBestMatch response")
if err != nil {
return "", err
}
id := strings.TrimSpace(resp)
if id == "none" || id == "null" {
return "", nil
}
return id, nil
}
// openAICompletion is a minimal OpenAI API call (text-davinci-003 or gpt-3.5-turbo-instruct)
func (llm *LLMClient) openAICompletion(ctx context.Context, prompt string) (string, error) {
apiURL := llm.BaseURL
if apiURL == "" {
apiURL = "https://api.openai.com/v1/completions"
}
logrus.WithFields(logrus.Fields{"api_url": apiURL, "prompt": prompt}).Info("[LLM] openAICompletion POST")
body := map[string]interface{}{
"model": "text-davinci-003",
"prompt": prompt,
"max_tokens": 64,
"temperature": 0,
}
jsonBody, _ := json.Marshal(body)
req, _ := http.NewRequestWithContext(ctx, "POST", apiURL, bytes.NewBuffer(jsonBody))
if llm.APIKey != "" {
req.Header.Set("Authorization", "Bearer "+llm.APIKey)
}
req.Header.Set("Content-Type", "application/json")
client := &http.Client{}
resp, err := client.Do(req)
if err != nil {
logrus.WithError(err).Error("[LLM] openAICompletion error")
return "", err
}
defer resp.Body.Close()
var result struct {
Choices []struct {
Text string `json:"text"`
} `json:"choices"`
}
if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
logrus.WithError(err).Error("[LLM] openAICompletion decode error")
return "", err
}
if len(result.Choices) == 0 {
logrus.Warn("[LLM] openAICompletion: no choices returned")
return "", nil
}
logrus.WithField("text", result.Choices[0].Text).Info("[LLM] openAICompletion: got text")
return result.Choices[0].Text, nil
}