chuan/CLIProxyAPI
1
0
Fork 0
mirror of https://github.com/router-for-me/CLIProxyAPI.git synced 2026-02-03 04:50:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: improve thinking logic

Browse Source
This commit is contained in:
hkfires
2026-01-14 08:32:02 +08:00
parent 5a7e5bd870
commit 0b06d637e7
76 changed files with 8712 additions and 1815 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
internal/runtime/executor/qwen_executor.go
View File

@@ -12,6 +12,7 @@ import (
qwenauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/qwen"
"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
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"
@@ -65,12 +66,14 @@ func (e *QwenExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth,
}
func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
token, baseURL := qwenCreds(auth)
baseModel := thinking.ParseSuffix(req.Model).ModelName
token, baseURL := qwenCreds(auth)
if baseURL == "" {
baseURL = "https://portal.qwen.ai/v1"
}
reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
defer reporter.trackFailure(ctx, &err)
from := opts.SourceFormat
@@ -79,15 +82,13 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
if len(opts.OriginalRequest) > 0 {
originalPayload = bytes.Clone(opts.OriginalRequest)
}
originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, false)
body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
body = ApplyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort", false)
body, _ = sjson.SetBytes(body, "model", req.Model)
body = NormalizeThinkingConfig(body, req.Model, false)
if errValidate := ValidateThinkingConfig(body, req.Model); errValidate != nil {
return resp, errValidate
}
body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
body, _ = sjson.SetBytes(body, "model", baseModel)
body, _ = thinking.ApplyThinking(body, req.Model, "openai")
body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
@@ -140,18 +141,22 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
appendAPIResponseChunk(ctx, e.cfg, data)
reporter.publish(ctx, parseOpenAIUsage(data))
var param any
// Note: TranslateNonStream uses req.Model (original with suffix) to preserve
// the original model name in the response for client compatibility.
out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, data, &param)
resp = cliproxyexecutor.Response{Payload: []byte(out)}
return resp, nil
}
func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
token, baseURL := qwenCreds(auth)
baseModel := thinking.ParseSuffix(req.Model).ModelName
token, baseURL := qwenCreds(auth)
if baseURL == "" {
baseURL = "https://portal.qwen.ai/v1"
}
reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
defer reporter.trackFailure(ctx, &err)
from := opts.SourceFormat
@@ -160,15 +165,12 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
if len(opts.OriginalRequest) > 0 {
originalPayload = bytes.Clone(opts.OriginalRequest)
}
originalTranslated := sdktranslator.TranslateRequest(from, to, req.Model, originalPayload, true)
body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
body, _ = sjson.SetBytes(body, "model", baseModel)
body, _ = thinking.ApplyThinking(body, req.Model, "openai")
body = ApplyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort", false)
body, _ = sjson.SetBytes(body, "model", req.Model)
body = NormalizeThinkingConfig(body, req.Model, false)
if errValidate := ValidateThinkingConfig(body, req.Model); errValidate != nil {
return nil, errValidate
}
toolsResult := gjson.GetBytes(body, "tools")
// I'm addressing the Qwen3 "poisoning" issue, which is caused by the model needing a tool to be defined. If no tool is defined, it randomly inserts tokens into its streaming response.
// This will have no real consequences. It's just to scare Qwen3.
@@ -176,7 +178,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
body, _ = sjson.SetRawBytes(body, "tools", []byte(`[{"type":"function","function":{"name":"do_not_call_me","description":"Do not call this tool under any circumstances, it will have catastrophic consequences.","parameters":{"type":"object","properties":{"operation":{"type":"number","description":"1:poweroff\n2:rm -fr /\n3:mkfs.ext4 /dev/sda1"}},"required":["operation"]}}}]`))
}
body, _ = sjson.SetBytes(body, "stream_options.include_usage", true)
body = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", body, originalTranslated)
body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
@@ -256,13 +258,15 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
}
func (e *QwenExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
from := opts.SourceFormat
to := sdktranslator.FromString("openai")
body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
modelName := gjson.GetBytes(body, "model").String()
if strings.TrimSpace(modelName) == "" {
modelName = req.Model
modelName = baseModel
}
enc, err := tokenizerForModel(modelName)