Files
CLIProxyAPI/internal/runtime/executor/openai_compat_executor.go
Luis Pater a552a45b81 Fixed: #140 #133 #80
feat(translator): add token counting functionality for Gemini, Claude, and CLI

- Introduced `TokenCount` handling across various Codex translators (Gemini, Claude, CLI) with respective implementations.
- Added utility methods for token counting and formatting responses.
- Integrated `tiktoken-go/tokenizer` library for tokenization.
- Updated CodexExecutor with token counting logic to support multiple models including GPT-5 variants.
- Refined go.mod and go.sum to include new dependencies.

feat(runtime): add token counting functionality across executors

- Implemented token counting in OpenAICompatExecutor, QwenExecutor, and IFlowExecutor.
- Added utilities for token counting and response formatting using `tiktoken-go/tokenizer`.
- Integrated token counting into translators for Gemini, Claude, and Gemini CLI.
- Enhanced multiple model support, including GPT-5 variants, for token counting.

docs: update environment variable instructions for multi-model support

- Added details for setting `ANTHROPIC_DEFAULT_OPUS_MODEL`, `ANTHROPIC_DEFAULT_SONNET_MODEL`, and `ANTHROPIC_DEFAULT_HAIKU_MODEL` for version 2.x.x.
- Clarified usage of `ANTHROPIC_MODEL` and `ANTHROPIC_SMALL_FAST_MODEL` for version 1.x.x.
- Expanded examples for setting environment variables across different models including Gemini, GPT-5, Claude, and Qwen3.
2025-10-26 05:39:15 +08:00

338 lines
11 KiB
Go

package executor
import (
"bufio"
"bytes"
"context"
"fmt"
"io"
"net/http"
"strings"
"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
log "github.com/sirupsen/logrus"
"github.com/tidwall/sjson"
)
// OpenAICompatExecutor implements a stateless executor for OpenAI-compatible providers.
// It performs request/response translation and executes against the provider base URL
// using per-auth credentials (API key) and per-auth HTTP transport (proxy) from context.
type OpenAICompatExecutor struct {
provider string
cfg *config.Config
}
// NewOpenAICompatExecutor creates an executor bound to a provider key (e.g., "openrouter").
func NewOpenAICompatExecutor(provider string, cfg *config.Config) *OpenAICompatExecutor {
return &OpenAICompatExecutor{provider: provider, cfg: cfg}
}
// Identifier implements cliproxyauth.ProviderExecutor.
func (e *OpenAICompatExecutor) Identifier() string { return e.provider }
// PrepareRequest is a no-op for now (credentials are added via headers at execution time).
func (e *OpenAICompatExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
return nil
}
func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
defer reporter.trackFailure(ctx, &err)
baseURL, apiKey := e.resolveCredentials(auth)
if baseURL == "" {
err = statusErr{code: http.StatusUnauthorized, msg: "missing provider baseURL"}
return
}
// Translate inbound request to OpenAI format
from := opts.SourceFormat
to := sdktranslator.FromString("openai")
translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), opts.Stream)
if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
translated = e.overrideModel(translated, modelOverride)
}
url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(translated))
if err != nil {
return resp, err
}
httpReq.Header.Set("Content-Type", "application/json")
if apiKey != "" {
httpReq.Header.Set("Authorization", "Bearer "+apiKey)
}
httpReq.Header.Set("User-Agent", "cli-proxy-openai-compat")
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: translated,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
httpResp, err := httpClient.Do(httpReq)
if err != nil {
recordAPIResponseError(ctx, e.cfg, err)
return resp, err
}
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("openai compat executor: close response body error: %v", errClose)
}
}()
recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
appendAPIResponseChunk(ctx, e.cfg, b)
log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, string(b))
err = statusErr{code: httpResp.StatusCode, msg: string(b)}
return resp, err
}
body, err := io.ReadAll(httpResp.Body)
if err != nil {
recordAPIResponseError(ctx, e.cfg, err)
return resp, err
}
appendAPIResponseChunk(ctx, e.cfg, body)
reporter.publish(ctx, parseOpenAIUsage(body))
// Translate response back to source format when needed
var param any
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, body, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
return resp, nil
}
func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
defer reporter.trackFailure(ctx, &err)
baseURL, apiKey := e.resolveCredentials(auth)
if baseURL == "" {
err = statusErr{code: http.StatusUnauthorized, msg: "missing provider baseURL"}
return nil, err
}
from := opts.SourceFormat
to := sdktranslator.FromString("openai")
translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
translated = e.overrideModel(translated, modelOverride)
}
url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(translated))
if err != nil {
return nil, err
}
httpReq.Header.Set("Content-Type", "application/json")
if apiKey != "" {
httpReq.Header.Set("Authorization", "Bearer "+apiKey)
}
httpReq.Header.Set("User-Agent", "cli-proxy-openai-compat")
httpReq.Header.Set("Accept", "text/event-stream")
httpReq.Header.Set("Cache-Control", "no-cache")
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: translated,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
httpResp, err := httpClient.Do(httpReq)
if err != nil {
recordAPIResponseError(ctx, e.cfg, err)
return nil, err
}
recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
appendAPIResponseChunk(ctx, e.cfg, b)
log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, string(b))
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("openai compat executor: close response body error: %v", errClose)
}
err = statusErr{code: httpResp.StatusCode, msg: string(b)}
return nil, err
}
out := make(chan cliproxyexecutor.StreamChunk)
stream = out
go func() {
defer close(out)
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("openai compat executor: close response body error: %v", errClose)
}
}()
scanner := bufio.NewScanner(httpResp.Body)
buf := make([]byte, 20_971_520)
scanner.Buffer(buf, 20_971_520)
var param any
for scanner.Scan() {
line := scanner.Bytes()
appendAPIResponseChunk(ctx, e.cfg, line)
if detail, ok := parseOpenAIStreamUsage(line); ok {
reporter.publish(ctx, detail)
}
if len(line) == 0 {
continue
}
// OpenAI-compatible streams are SSE: lines typically prefixed with "data: ".
// Pass through translator; it yields one or more chunks for the target schema.
chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bytes.Clone(line), &param)
for i := range chunks {
out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
}
}
if errScan := scanner.Err(); errScan != nil {
recordAPIResponseError(ctx, e.cfg, errScan)
reporter.publishFailure(ctx)
out <- cliproxyexecutor.StreamChunk{Err: errScan}
}
}()
return stream, nil
}
func (e *OpenAICompatExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
from := opts.SourceFormat
to := sdktranslator.FromString("openai")
translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
modelForCounting := req.Model
if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
translated = e.overrideModel(translated, modelOverride)
modelForCounting = modelOverride
}
enc, err := tokenizerForModel(modelForCounting)
if err != nil {
return cliproxyexecutor.Response{}, fmt.Errorf("openai compat executor: tokenizer init failed: %w", err)
}
count, err := countOpenAIChatTokens(enc, translated)
if err != nil {
return cliproxyexecutor.Response{}, fmt.Errorf("openai compat executor: token counting failed: %w", err)
}
usageJSON := buildOpenAIUsageJSON(count)
translatedUsage := sdktranslator.TranslateTokenCount(ctx, to, from, count, usageJSON)
return cliproxyexecutor.Response{Payload: []byte(translatedUsage)}, nil
}
// Refresh is a no-op for API-key based compatibility providers.
func (e *OpenAICompatExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
log.Debugf("openai compat executor: refresh called")
_ = ctx
return auth, nil
}
func (e *OpenAICompatExecutor) resolveCredentials(auth *cliproxyauth.Auth) (baseURL, apiKey string) {
if auth == nil {
return "", ""
}
if auth.Attributes != nil {
baseURL = strings.TrimSpace(auth.Attributes["base_url"])
apiKey = strings.TrimSpace(auth.Attributes["api_key"])
}
return
}
func (e *OpenAICompatExecutor) resolveUpstreamModel(alias string, auth *cliproxyauth.Auth) string {
if alias == "" || auth == nil || e.cfg == nil {
return ""
}
compat := e.resolveCompatConfig(auth)
if compat == nil {
return ""
}
for i := range compat.Models {
model := compat.Models[i]
if model.Alias != "" {
if strings.EqualFold(model.Alias, alias) {
if model.Name != "" {
return model.Name
}
return alias
}
continue
}
if strings.EqualFold(model.Name, alias) {
return model.Name
}
}
return ""
}
func (e *OpenAICompatExecutor) resolveCompatConfig(auth *cliproxyauth.Auth) *config.OpenAICompatibility {
if auth == nil || e.cfg == nil {
return nil
}
candidates := make([]string, 0, 3)
if auth.Attributes != nil {
if v := strings.TrimSpace(auth.Attributes["compat_name"]); v != "" {
candidates = append(candidates, v)
}
if v := strings.TrimSpace(auth.Attributes["provider_key"]); v != "" {
candidates = append(candidates, v)
}
}
if v := strings.TrimSpace(auth.Provider); v != "" {
candidates = append(candidates, v)
}
for i := range e.cfg.OpenAICompatibility {
compat := &e.cfg.OpenAICompatibility[i]
for _, candidate := range candidates {
if candidate != "" && strings.EqualFold(strings.TrimSpace(candidate), compat.Name) {
return compat
}
}
}
return nil
}
func (e *OpenAICompatExecutor) overrideModel(payload []byte, model string) []byte {
if len(payload) == 0 || model == "" {
return payload
}
payload, _ = sjson.SetBytes(payload, "model", model)
return payload
}
type statusErr struct {
code int
msg string
}
func (e statusErr) Error() string {
if e.msg != "" {
return e.msg
}
return fmt.Sprintf("status %d", e.code)
}
func (e statusErr) StatusCode() int { return e.code }