Restore hygate parity for Hunyuan3 and fix thinking separation
This commit is contained in:
@@ -3215,16 +3215,19 @@ func InspectSpace(client *http.Client, rawURL, userAgent string) (*SpaceDiscover
|
||||
|
||||
// Select protocol
|
||||
isCallV2 := false
|
||||
if strings.HasPrefix(discovery.GradioVersion, "6.") {
|
||||
isCallV2 = true
|
||||
}
|
||||
hasSnippet := false
|
||||
if bestEndpointInfo != nil && bestEndpointInfo.CodeSnippets != nil {
|
||||
if bashSnippet, ok := bestEndpointInfo.CodeSnippets["bash"].(string); ok {
|
||||
if bashSnippet, ok := bestEndpointInfo.CodeSnippets["bash"].(string); ok && bashSnippet != "" {
|
||||
hasSnippet = true
|
||||
if strings.Contains(bashSnippet, "/call/v2/") {
|
||||
isCallV2 = true
|
||||
}
|
||||
}
|
||||
}
|
||||
if !hasSnippet && strings.HasPrefix(discovery.GradioVersion, "6.") {
|
||||
isCallV2 = true
|
||||
}
|
||||
|
||||
|
||||
if strings.HasPrefix(discovery.GradioVersion, "3.") {
|
||||
discovery.Protocol = "predict"
|
||||
@@ -3299,11 +3302,12 @@ func InspectSpace(client *http.Client, rawURL, userAgent string) (*SpaceDiscover
|
||||
if strings.Contains(pName, "function") || strings.Contains(pName, "tool") || strings.Contains(pLabel, "tool") || strings.Contains(pLabel, "function") || strings.Contains(cLabel, "tool") || strings.Contains(cLabel, "function") {
|
||||
mapping.ParamType = "tools"
|
||||
discovery.FunctionsJSONIndex = idx
|
||||
} else if strings.Contains(pName, "preserved") || strings.Contains(cLabel, "preserved") {
|
||||
mapping.ParamType = "preserved_thinking"
|
||||
discovery.PreservedThinkingIndex = idx
|
||||
} else if strings.Contains(pName, "think_level") || strings.Contains(pName, "thinking") || strings.Contains(pLabel, "think") || strings.Contains(cLabel, "think") {
|
||||
mapping.ParamType = "think_level"
|
||||
discovery.ThinkLevelIndex = idx
|
||||
} else if strings.Contains(pName, "preserved") || strings.Contains(cLabel, "preserved") {
|
||||
discovery.PreservedThinkingIndex = idx
|
||||
} else if strings.Contains(pName, "system") || strings.Contains(pLabel, "system") || strings.Contains(cLabel, "system") || strings.Contains(cLabel, "instruction") {
|
||||
mapping.ParamType = "system_prompt"
|
||||
discovery.SystemIndex = idx
|
||||
@@ -3532,6 +3536,9 @@ func InspectSpace(client *http.Client, rawURL, userAgent string) (*SpaceDiscover
|
||||
if discovery.PrimaryModel == "gradio-chat" || discovery.PrimaryModel == "" {
|
||||
discovery.PrimaryModel = "hy3"
|
||||
}
|
||||
if discovery.Protocol == "call_v2" {
|
||||
discovery.Protocol = "call"
|
||||
}
|
||||
}
|
||||
|
||||
// 9. Tool Calling Support Picture Resolution
|
||||
@@ -3991,6 +3998,10 @@ func (g *GradioGateway) BuildGradioPayload(disc *SpaceDiscovery, req ChatComplet
|
||||
data[disc.ThinkLevelIndex] = thinkLevel
|
||||
}
|
||||
|
||||
if disc.PreservedThinkingIndex >= 0 && disc.PreservedThinkingIndex != disc.MessageIndex && disc.PreservedThinkingIndex != disc.HistoryIndex && disc.PreservedThinkingIndex != disc.SystemIndex && disc.PreservedThinkingIndex < len(data) {
|
||||
data[disc.PreservedThinkingIndex] = nil
|
||||
}
|
||||
|
||||
if disc.TempIndex >= 0 && disc.TempIndex != disc.MessageIndex && disc.TempIndex != disc.HistoryIndex && disc.TempIndex < len(data) {
|
||||
if req.Temperature != nil {
|
||||
data[disc.TempIndex] = *req.Temperature
|
||||
@@ -4046,39 +4057,136 @@ type GradioOutputFrame struct {
|
||||
OK bool
|
||||
}
|
||||
|
||||
func toInt(v interface{}) (int, bool) {
|
||||
switch n := v.(type) {
|
||||
case int:
|
||||
return n, true
|
||||
case int64:
|
||||
return int(n), true
|
||||
case float64:
|
||||
return int(n), true
|
||||
case json.Number:
|
||||
i, err := n.Int64()
|
||||
return int(i), err == nil
|
||||
default:
|
||||
return 0, false
|
||||
}
|
||||
}
|
||||
|
||||
func isReasoningPath(path interface{}) bool {
|
||||
switch p := path.(type) {
|
||||
case []interface{}:
|
||||
for i, elem := range p {
|
||||
if s, ok := elem.(string); ok {
|
||||
sLow := strings.ToLower(s)
|
||||
if strings.Contains(sLow, "reason") || strings.Contains(sLow, "thought") || strings.Contains(sLow, "think") {
|
||||
return true
|
||||
}
|
||||
}
|
||||
if num, ok := toInt(elem); ok {
|
||||
if len(p) == 1 && num == 1 {
|
||||
return true
|
||||
}
|
||||
if len(p) == 2 && i == 1 && num == 1 {
|
||||
if p0, ok0 := toInt(p[0]); ok0 && p0 == 0 {
|
||||
return true
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
case string:
|
||||
sLow := strings.ToLower(p)
|
||||
if strings.Contains(sLow, "reason") || strings.Contains(sLow, "thought") || strings.Contains(sLow, "think") {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func isContentPath(path interface{}) bool {
|
||||
switch p := path.(type) {
|
||||
case []interface{}:
|
||||
if len(p) == 0 {
|
||||
return true
|
||||
}
|
||||
for _, elem := range p {
|
||||
if s, ok := elem.(string); ok {
|
||||
sLow := strings.ToLower(s)
|
||||
if strings.Contains(sLow, "content") || strings.Contains(sLow, "text") || strings.Contains(sLow, "value") {
|
||||
return true
|
||||
}
|
||||
}
|
||||
}
|
||||
if num, ok := toInt(p[0]); ok && num == 0 {
|
||||
if len(p) == 1 {
|
||||
return true
|
||||
}
|
||||
if len(p) >= 2 {
|
||||
if p1, ok1 := toInt(p[1]); ok1 && p1 == 0 {
|
||||
return true
|
||||
}
|
||||
if s1, ok1 := p[1].(string); ok1 && (strings.Contains(s1, "content") || strings.Contains(s1, "text")) {
|
||||
return true
|
||||
}
|
||||
}
|
||||
}
|
||||
case string:
|
||||
sLow := strings.ToLower(p)
|
||||
if strings.Contains(sLow, "content") || strings.Contains(sLow, "text") || strings.Contains(sLow, "value") {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
// extractGradioDiffDelta extracts delta string from Gradio 6 streaming diff operations
|
||||
func extractGradioDiffDelta(v []interface{}) (string, bool) {
|
||||
if len(v) == 0 {
|
||||
return "", false
|
||||
}
|
||||
if len(v) >= 3 {
|
||||
if op, ok := v[0].(string); ok && (op == "append" || op == "add") {
|
||||
if delta, ok := v[2].(string); ok {
|
||||
return delta, true
|
||||
func extractGradioDiffDelta(v []interface{}) (contentDelta string, reasoningDelta string, ok bool) {
|
||||
var diffItems [][]interface{}
|
||||
var findOps func(items []interface{})
|
||||
findOps = func(items []interface{}) {
|
||||
if len(items) >= 3 {
|
||||
if op, ok := items[0].(string); ok && (op == "append" || op == "add") {
|
||||
diffItems = append(diffItems, items)
|
||||
return
|
||||
}
|
||||
}
|
||||
for _, it := range items {
|
||||
if sub, ok := it.([]interface{}); ok {
|
||||
findOps(sub)
|
||||
}
|
||||
}
|
||||
}
|
||||
for _, item := range v {
|
||||
if innerList, ok := item.([]interface{}); ok && len(innerList) > 0 {
|
||||
for _, opItem := range innerList {
|
||||
if opArr, ok := opItem.([]interface{}); ok && len(opArr) >= 3 {
|
||||
if op, ok := opArr[0].(string); ok && (op == "append" || op == "add") {
|
||||
if delta, ok := opArr[2].(string); ok {
|
||||
return delta, true
|
||||
}
|
||||
}
|
||||
}
|
||||
findOps(v)
|
||||
|
||||
if len(diffItems) == 0 {
|
||||
return "", "", false
|
||||
}
|
||||
|
||||
hasMatch := false
|
||||
for _, item := range diffItems {
|
||||
delta, isStr := item[2].(string)
|
||||
if !isStr || delta == "" {
|
||||
continue
|
||||
}
|
||||
path := item[1]
|
||||
if isReasoningPath(path) {
|
||||
if reasoningDelta == "" {
|
||||
reasoningDelta = delta
|
||||
hasMatch = true
|
||||
}
|
||||
if len(innerList) >= 3 {
|
||||
if op, ok := innerList[0].(string); ok && (op == "append" || op == "add") {
|
||||
if delta, ok := innerList[2].(string); ok {
|
||||
return delta, true
|
||||
}
|
||||
}
|
||||
} else if isContentPath(path) {
|
||||
if contentDelta == "" {
|
||||
contentDelta = delta
|
||||
hasMatch = true
|
||||
}
|
||||
} else {
|
||||
if contentDelta == "" {
|
||||
contentDelta = delta
|
||||
hasMatch = true
|
||||
}
|
||||
}
|
||||
}
|
||||
return "", false
|
||||
return contentDelta, reasoningDelta, hasMatch
|
||||
}
|
||||
|
||||
// ParseGradioStreamOutput extracts structured content, reasoning, and tool calls from Gradio output
|
||||
@@ -4109,30 +4217,33 @@ func ParseGradioStreamOutput(rawJSON string) (frame GradioOutputFrame) {
|
||||
return frame
|
||||
}
|
||||
|
||||
if delta, ok := extractGradioDiffDelta(v); ok {
|
||||
frame.Content = delta
|
||||
if cDelta, rDelta, ok := extractGradioDiffDelta(v); ok {
|
||||
frame.Content = cDelta
|
||||
frame.Reasoning = rDelta
|
||||
frame.IsDelta = true
|
||||
frame.OK = true
|
||||
return frame
|
||||
}
|
||||
|
||||
// 1. Check if v[0] is an inner slice with len >= 3 (e.g. Hy3: [[content, reasoning, tool_calls, history]])
|
||||
if inner, ok := v[0].([]interface{}); ok && len(inner) >= 3 {
|
||||
s0, _ := inner[0].(string)
|
||||
s1, _ := inner[1].(string)
|
||||
frame.Content = s0
|
||||
frame.Reasoning = s1
|
||||
if inner[2] != nil {
|
||||
b, err := json.Marshal(inner[2])
|
||||
if err == nil {
|
||||
var tcs []ToolCall
|
||||
if json.Unmarshal(b, &tcs) == nil && len(tcs) > 0 {
|
||||
frame.ToolCalls = tcs
|
||||
// 1. Check if v[0] is an inner slice with len >= 2 (e.g. Hy3: [[content, reasoning, tool_calls, history]])
|
||||
if inner, ok := v[0].([]interface{}); ok && len(inner) >= 2 {
|
||||
s0, ok0 := inner[0].(string)
|
||||
s1, ok1 := inner[1].(string)
|
||||
if ok0 || ok1 {
|
||||
frame.Content = s0
|
||||
frame.Reasoning = s1
|
||||
if len(inner) >= 3 && inner[2] != nil {
|
||||
b, err := json.Marshal(inner[2])
|
||||
if err == nil {
|
||||
var tcs []ToolCall
|
||||
if json.Unmarshal(b, &tcs) == nil && len(tcs) > 0 {
|
||||
frame.ToolCalls = tcs
|
||||
}
|
||||
}
|
||||
}
|
||||
frame.OK = true
|
||||
return frame
|
||||
}
|
||||
frame.OK = true
|
||||
return frame
|
||||
}
|
||||
|
||||
// 2. Check if any element of v is a Chatbot message list or Chatbot pair list
|
||||
@@ -4270,22 +4381,6 @@ func ParseGradioStreamOutput(rawJSON string) (frame GradioOutputFrame) {
|
||||
return frame
|
||||
}
|
||||
|
||||
// 4. Check if v[0] is an inner pair [content, reasoning]
|
||||
if inner, ok := v[0].([]interface{}); ok && len(inner) == 2 {
|
||||
s0, ok0 := inner[0].(string)
|
||||
s1, ok1 := inner[1].(string)
|
||||
if ok0 && ok1 {
|
||||
frame.Content = s0
|
||||
frame.Reasoning = s1
|
||||
frame.OK = true
|
||||
return frame
|
||||
}
|
||||
if ok1 {
|
||||
frame.Content = s1
|
||||
frame.OK = true
|
||||
return frame
|
||||
}
|
||||
}
|
||||
|
||||
// 5. Check if v[0] is a non-empty string or single output
|
||||
if s, ok := v[0].(string); ok && (s != "" || len(v) == 1) {
|
||||
@@ -4618,15 +4713,16 @@ func (g *GradioGateway) executeQueueCompletion(w http.ResponseWriter, r *http.Re
|
||||
if frame.OK {
|
||||
if frame.IsDelta {
|
||||
latestFrame.Content += frame.Content
|
||||
latestFrame.Reasoning += frame.Reasoning
|
||||
} else {
|
||||
if len(frame.ToolCalls) > 0 {
|
||||
latestFrame.Content = frame.Content
|
||||
} else if frame.Content != "" || latestFrame.Content == "" {
|
||||
latestFrame.Content = frame.Content
|
||||
}
|
||||
}
|
||||
if frame.Reasoning != "" || latestFrame.Reasoning == "" {
|
||||
latestFrame.Reasoning = frame.Reasoning
|
||||
if frame.Reasoning != "" || latestFrame.Reasoning == "" {
|
||||
latestFrame.Reasoning = frame.Reasoning
|
||||
}
|
||||
}
|
||||
if len(frame.ToolCalls) > 0 {
|
||||
latestFrame.ToolCalls = frame.ToolCalls
|
||||
@@ -4668,7 +4764,7 @@ func (g *GradioGateway) executeQueueCompletion(w http.ResponseWriter, r *http.Re
|
||||
var prevContent string
|
||||
var prevReasoning string
|
||||
prevToolArgs := make(map[int]string)
|
||||
nativeReasoningSeen := false
|
||||
nativeReasoningSeen := disc.IsHunyuan3
|
||||
nativeToolCallsSeen := false
|
||||
var streamErr error
|
||||
|
||||
@@ -4710,14 +4806,23 @@ func (g *GradioGateway) executeQueueCompletion(w http.ResponseWriter, r *http.Re
|
||||
// 1. Native reasoning handling
|
||||
if frame.Reasoning != "" || nativeReasoningSeen {
|
||||
nativeReasoningSeen = true
|
||||
if len(frame.Reasoning) > len(prevReasoning) && strings.HasPrefix(frame.Reasoning, prevReasoning) {
|
||||
delta := frame.Reasoning[len(prevReasoning):]
|
||||
streamer.Reasoning(delta)
|
||||
prevReasoning = frame.Reasoning
|
||||
} else if prevReasoning == "" && frame.Reasoning != "" {
|
||||
streamer.Reasoning(frame.Reasoning)
|
||||
var deltaReasoning string
|
||||
if frame.IsDelta {
|
||||
deltaReasoning = frame.Reasoning
|
||||
prevReasoning += deltaReasoning
|
||||
} else {
|
||||
if strings.HasPrefix(frame.Reasoning, prevReasoning) {
|
||||
deltaReasoning = frame.Reasoning[len(prevReasoning):]
|
||||
} else if prevReasoning == "" {
|
||||
deltaReasoning = frame.Reasoning
|
||||
} else {
|
||||
deltaReasoning = frame.Reasoning
|
||||
}
|
||||
prevReasoning = frame.Reasoning
|
||||
}
|
||||
if deltaReasoning != "" {
|
||||
streamer.Reasoning(deltaReasoning)
|
||||
}
|
||||
}
|
||||
|
||||
// 2. Native tool call handling
|
||||
@@ -4929,8 +5034,10 @@ func (g *GradioGateway) ExecuteChatCompletion(w http.ResponseWriter, r *http.Req
|
||||
return json.Marshal(v2Payload)
|
||||
}
|
||||
payloadMap := map[string]interface{}{
|
||||
"data": gradioData,
|
||||
"session_hash": GenerateUUID(),
|
||||
"data": gradioData,
|
||||
}
|
||||
if !disc.IsHunyuan3 {
|
||||
payloadMap["session_hash"] = GenerateUUID()
|
||||
}
|
||||
return json.Marshal(payloadMap)
|
||||
}
|
||||
@@ -5115,13 +5222,18 @@ func (g *GradioGateway) ExecuteChatCompletion(w http.ResponseWriter, r *http.Req
|
||||
return fmt.Errorf("upstream Gradio error: %s", errMsg)
|
||||
}
|
||||
if frame := ParseGradioStreamOutput(dataStr); frame.OK {
|
||||
if len(frame.ToolCalls) > 0 {
|
||||
latestFrame.Content = frame.Content
|
||||
} else if frame.Content != "" || latestFrame.Content == "" {
|
||||
latestFrame.Content = frame.Content
|
||||
}
|
||||
if frame.Reasoning != "" || latestFrame.Reasoning == "" {
|
||||
latestFrame.Reasoning = frame.Reasoning
|
||||
if frame.IsDelta {
|
||||
latestFrame.Content += frame.Content
|
||||
latestFrame.Reasoning += frame.Reasoning
|
||||
} else {
|
||||
if len(frame.ToolCalls) > 0 {
|
||||
latestFrame.Content = frame.Content
|
||||
} else if frame.Content != "" || latestFrame.Content == "" {
|
||||
latestFrame.Content = frame.Content
|
||||
}
|
||||
if frame.Reasoning != "" || latestFrame.Reasoning == "" {
|
||||
latestFrame.Reasoning = frame.Reasoning
|
||||
}
|
||||
}
|
||||
if len(frame.ToolCalls) > 0 {
|
||||
latestFrame.ToolCalls = frame.ToolCalls
|
||||
@@ -5161,7 +5273,7 @@ func (g *GradioGateway) ExecuteChatCompletion(w http.ResponseWriter, r *http.Req
|
||||
var prevContent string
|
||||
var prevReasoning string
|
||||
prevToolArgs := make(map[int]string)
|
||||
nativeReasoningSeen := false
|
||||
nativeReasoningSeen := disc.IsHunyuan3
|
||||
nativeToolCallsSeen := false
|
||||
currentEvent := ""
|
||||
var streamErr error
|
||||
@@ -5213,14 +5325,19 @@ func (g *GradioGateway) ExecuteChatCompletion(w http.ResponseWriter, r *http.Req
|
||||
if frame.Reasoning != "" || nativeReasoningSeen {
|
||||
nativeReasoningSeen = true
|
||||
var deltaReasoning string
|
||||
if strings.HasPrefix(frame.Reasoning, prevReasoning) {
|
||||
deltaReasoning = frame.Reasoning[len(prevReasoning):]
|
||||
} else if prevReasoning == "" {
|
||||
if frame.IsDelta {
|
||||
deltaReasoning = frame.Reasoning
|
||||
prevReasoning += deltaReasoning
|
||||
} else {
|
||||
deltaReasoning = frame.Reasoning
|
||||
if strings.HasPrefix(frame.Reasoning, prevReasoning) {
|
||||
deltaReasoning = frame.Reasoning[len(prevReasoning):]
|
||||
} else if prevReasoning == "" {
|
||||
deltaReasoning = frame.Reasoning
|
||||
} else {
|
||||
deltaReasoning = frame.Reasoning
|
||||
}
|
||||
prevReasoning = frame.Reasoning
|
||||
}
|
||||
prevReasoning = frame.Reasoning
|
||||
if deltaReasoning != "" {
|
||||
streamer.Reasoning(deltaReasoning)
|
||||
}
|
||||
@@ -5269,14 +5386,19 @@ func (g *GradioGateway) ExecuteChatCompletion(w http.ResponseWriter, r *http.Req
|
||||
// 3. Content handling
|
||||
currentText := frame.Content
|
||||
var delta string
|
||||
if strings.HasPrefix(currentText, prevContent) {
|
||||
delta = currentText[len(prevContent):]
|
||||
} else if prevContent == "" {
|
||||
if frame.IsDelta {
|
||||
delta = currentText
|
||||
prevContent += delta
|
||||
} else {
|
||||
delta = currentText
|
||||
if strings.HasPrefix(currentText, prevContent) {
|
||||
delta = currentText[len(prevContent):]
|
||||
} else if prevContent == "" {
|
||||
delta = currentText
|
||||
} else {
|
||||
delta = currentText
|
||||
}
|
||||
prevContent = currentText
|
||||
}
|
||||
prevContent = currentText
|
||||
|
||||
if delta != "" {
|
||||
if nativeReasoningSeen || nativeToolCallsSeen {
|
||||
|
||||
Reference in New Issue
Block a user