2026-07-31 18:32:07 +08:00
|
|
|
//
|
|
|
|
|
// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
|
|
|
|
|
//
|
|
|
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
|
// you may not use this file except in compliance with the License.
|
|
|
|
|
// You may obtain a copy of the License at
|
|
|
|
|
//
|
|
|
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
|
//
|
|
|
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
|
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
|
// See the License for the specific language governing permissions and
|
|
|
|
|
// limitations under the License.
|
|
|
|
|
//
|
|
|
|
|
|
|
|
|
|
package models
|
|
|
|
|
|
|
|
|
|
import (
|
|
|
|
|
"encoding/json"
|
|
|
|
|
"fmt"
|
|
|
|
|
"io"
|
|
|
|
|
"ragflow/internal/common"
|
|
|
|
|
|
|
|
|
|
"go.uber.org/zap"
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
// HandleNonStreamingResponse processes a complete non-streaming chat
|
|
|
|
|
// response using the ParserConfig's ResponseParser to extract usage.
|
|
|
|
|
func HandleNonStreamingResponse(
|
|
|
|
|
body []byte,
|
|
|
|
|
modelUsage *common.ModelUsage,
|
|
|
|
|
chatConfig *ChatConfig,
|
|
|
|
|
cfg *ParserConfig,
|
|
|
|
|
) (*ChatResponse, error) {
|
|
|
|
|
var result map[string]any
|
|
|
|
|
if err := json.Unmarshal(body, &result); err != nil {
|
|
|
|
|
return nil, fmt.Errorf("failed to parse response: %w", err)
|
|
|
|
|
}
|
|
|
|
|
|
feat(go-models): migrate batch 4 model drivers to unified handlers (#17699)
## Summary
Relate to #17284
Migrate 10 OpenAI-compatible drivers (`orcarouter`, `perplexity`,
`ppio`, `qiniu`, `ragcon`, `stepfun`, `togetherai`, `tokenhub`,
`tokenpony`, `upstage`) to use the unified response handlers
(`HandleNonStreamingResponse` / `HandleStreamingResponse`), following
the same pattern established by `deepseek` in #17634.
- Cut ~150 lines per driver (1436 lines removed, 62 added across 10
files).
- No functional changes — pure deduplication of HTTP plumbing.
- Each driver now routes through `baseModel.doRequest()` and
`HandleNonStreamingResponse()`.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-08-03 17:48:45 +08:00
|
|
|
// Check for upstream error.
|
|
|
|
|
if apiErr, ok := result["error"]; ok && apiErr != nil {
|
|
|
|
|
return nil, fmt.Errorf("upstream error: %v", apiErr)
|
|
|
|
|
}
|
|
|
|
|
|
2026-07-31 18:32:07 +08:00
|
|
|
// Extract usage via the protocol-specific parser.
|
|
|
|
|
var usage *TokenUsage
|
|
|
|
|
if u, ok := cfg.ResponseParser(result); ok {
|
|
|
|
|
usage = u
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Extract content / reasoning_content / tool_calls.
|
|
|
|
|
content, reasonContent, toolCalls := extractContentAndChoices(result)
|
|
|
|
|
|
|
|
|
|
if content == nil && len(toolCalls) == 0 {
|
|
|
|
|
return nil, fmt.Errorf("no choices in response")
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if usage != nil {
|
|
|
|
|
recordResponseUsage(modelUsage, extractRequestID(result), usage, "chat")
|
|
|
|
|
if chatConfig != nil {
|
|
|
|
|
chatConfig.UsageResult = usage
|
|
|
|
|
model, _ := result["model"].(string)
|
|
|
|
|
common.Info("StreamUsage", zap.String("model", model), zap.Int("prompt", usage.PromptTokens), zap.Int("completion", usage.CompletionTokens), zap.Int("total", usage.TotalTokens))
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return &ChatResponse{
|
|
|
|
|
Answer: content,
|
|
|
|
|
ReasonContent: reasonContent,
|
|
|
|
|
ToolCalls: toolCalls,
|
|
|
|
|
Usage: usage,
|
|
|
|
|
}, nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// HandleStreamingResponse processes a streaming chat response using the
|
|
|
|
|
// ParserConfig's StreamParser to extract usage from each event.
|
|
|
|
|
func HandleStreamingResponse(
|
|
|
|
|
body io.Reader,
|
|
|
|
|
modelUsage *common.ModelUsage,
|
|
|
|
|
chatConfig *ChatConfig,
|
|
|
|
|
cfg *ParserConfig,
|
|
|
|
|
sender func(*string, *string) error,
|
|
|
|
|
) error {
|
|
|
|
|
if sender == nil {
|
|
|
|
|
return fmt.Errorf("sender is required")
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var streamUsage *TokenUsage
|
|
|
|
|
accumulatedToolCalls := make(map[int]map[string]any)
|
|
|
|
|
sawTerminal := false
|
|
|
|
|
|
|
|
|
|
var streamModel string
|
|
|
|
|
done, err := ParseSSEStream[map[string]any](body, func(event map[string]any) error {
|
|
|
|
|
if u, ok := cfg.StreamParser(event); ok {
|
|
|
|
|
streamUsage = u
|
|
|
|
|
}
|
|
|
|
|
if m, ok := event["model"].(string); ok {
|
|
|
|
|
streamModel = m
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if apiErr, ok := event["error"]; ok && apiErr != nil {
|
|
|
|
|
return fmt.Errorf("upstream stream error: %v", apiErr)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
choices, ok := event["choices"].([]any)
|
|
|
|
|
if !ok || len(choices) == 0 {
|
|
|
|
|
return nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
firstChoice, ok := choices[0].(map[string]any)
|
|
|
|
|
if !ok {
|
|
|
|
|
return nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
delta, ok := firstChoice["delta"].(map[string]any)
|
|
|
|
|
if !ok {
|
|
|
|
|
return nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
accumulateToolCallDeltas(delta, accumulatedToolCalls)
|
|
|
|
|
|
feat(go-models): migrate batch 3 model drivers to unified handlers (#17698)
## Summary
Relate to #17284
Migrate 10 OpenAI-compatible drivers (`minimax`, `mistral`,
`modelscope`, `moonshot`, `n1n`, `novita`, `ollama`, `openai`,
`openai_api_compatible`, `openrouter`) to use the unified response
handlers (`HandleNonStreamingResponse` / `HandleStreamingResponse`),
following the same pattern established by `deepseek` in #17634.
- Cut ~150 lines per driver (1692 lines removed, 172 added across 10
files).
- `openai_api_compatible` gains `ChatWithMessages` +
`ChatStreamlyWithSender` required by the unified handler infrastructure.
- `openai` driver preserves `reasoning_content` extraction for o-series
models.
- All drivers: pure deduplication of HTTP plumbing.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-08-03 15:08:55 +08:00
|
|
|
// Extract reasoning via the protocol hook so each provider can
|
|
|
|
|
// name its reasoning field differently (reasoning_content,
|
|
|
|
|
// reasoning, ...) without the shared handler knowing which.
|
|
|
|
|
extractReasoning := cfg.ExtractStreamReasoning
|
|
|
|
|
if extractReasoning == nil {
|
|
|
|
|
extractReasoning = extractDefaultStreamReasoning
|
|
|
|
|
}
|
|
|
|
|
if reasoning := extractReasoning(delta); reasoning != "" {
|
|
|
|
|
if err := sender(nil, &reasoning); err != nil {
|
2026-07-31 18:32:07 +08:00
|
|
|
return err
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if content, ok := delta["content"].(string); ok && content != "" {
|
|
|
|
|
if err := sender(&content, nil); err != nil {
|
|
|
|
|
return err
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if finishReason, ok := firstChoice["finish_reason"].(string); ok && finishReason != "" {
|
|
|
|
|
sawTerminal = true
|
feat(go-models): migrate batch 4 model drivers to unified handlers (#17699)
## Summary
Relate to #17284
Migrate 10 OpenAI-compatible drivers (`orcarouter`, `perplexity`,
`ppio`, `qiniu`, `ragcon`, `stepfun`, `togetherai`, `tokenhub`,
`tokenpony`, `upstage`) to use the unified response handlers
(`HandleNonStreamingResponse` / `HandleStreamingResponse`), following
the same pattern established by `deepseek` in #17634.
- Cut ~150 lines per driver (1436 lines removed, 62 added across 10
files).
- No functional changes — pure deduplication of HTTP plumbing.
- Each driver now routes through `baseModel.doRequest()` and
`HandleNonStreamingResponse()`.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-08-03 17:48:45 +08:00
|
|
|
}
|
|
|
|
|
if finishReason, ok := event["finish_reason"].(string); ok && finishReason != "" {
|
|
|
|
|
sawTerminal = true
|
2026-07-31 18:32:07 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return nil
|
|
|
|
|
})
|
|
|
|
|
if err != nil {
|
|
|
|
|
return fmt.Errorf("failed to scan response body: %w", err)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if chatConfig != nil {
|
|
|
|
|
setSortedToolCallsResult(chatConfig, accumulatedToolCalls)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if !done && !sawTerminal {
|
|
|
|
|
return fmt.Errorf("stream ended before [DONE] or finish_reason")
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if streamUsage != nil {
|
|
|
|
|
recordResponseUsage(modelUsage, "", streamUsage, "chat")
|
|
|
|
|
if chatConfig != nil {
|
|
|
|
|
chatConfig.UsageResult = streamUsage
|
|
|
|
|
common.Info("StreamUsage", zap.String("model", streamModel), zap.Int("prompt", streamUsage.PromptTokens), zap.Int("completion", streamUsage.CompletionTokens), zap.Int("total", streamUsage.TotalTokens))
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
endOfStream := "[DONE]"
|
|
|
|
|
return sender(&endOfStream, nil)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// extractContentAndChoices extracts content, reasoning_content, and tool_calls
|
|
|
|
|
// from a parsed non-streaming response.
|
|
|
|
|
func extractContentAndChoices(result map[string]any) (*string, *string, []map[string]any) {
|
|
|
|
|
choices, ok := result["choices"].([]any)
|
|
|
|
|
if !ok || len(choices) == 0 {
|
|
|
|
|
return nil, nil, nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
firstChoice, ok := choices[0].(map[string]any)
|
|
|
|
|
if !ok {
|
|
|
|
|
return nil, nil, nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
messageMap, ok := firstChoice["message"].(map[string]any)
|
|
|
|
|
if !ok {
|
|
|
|
|
return nil, nil, nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var content *string
|
|
|
|
|
if c, ok := messageMap["content"].(string); ok {
|
|
|
|
|
cc := c
|
|
|
|
|
content = &cc
|
|
|
|
|
} else if _, ok := messageMap["tool_calls"].([]any); ok {
|
|
|
|
|
// content may be nil when the response only carries tool calls.
|
|
|
|
|
// Return an empty-string pointer so callers can rely on a non-nil Answer.
|
|
|
|
|
empty := ""
|
|
|
|
|
content = &empty
|
|
|
|
|
} else if rc, ok := messageMap["reasoning_content"].(string); ok && rc != "" {
|
|
|
|
|
// content may be nil when the response only carries reasoning_content.
|
|
|
|
|
empty := ""
|
|
|
|
|
content = &empty
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var reasonContent *string
|
feat(go-models): migrate batch 1 model drivers to unified handlers (#17696)
## Summary
Relate to #17284
Migrate 10 OpenAI-compatible drivers (`302ai`, `aliyun`, `astraflow`,
`avian`, `azure_openai`, `baichuan`, `baidu`, `cometapi`, `deepinfra`,
`futurmix`) to use the unified response handlers
(`HandleNonStreamingResponse` / `HandleStreamingResponse`), following
the same pattern established by `deepseek` in #17634.
- Cut ~150 lines per driver (1507 lines removed, 144 added across 10
files).
- No functional changes — pure deduplication of HTTP plumbing.
- Each driver now routes through `baseModel.doRequest()` and
`HandleNonStreamingResponse()`.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-08-03 12:13:43 +08:00
|
|
|
if rc, ok := messageMap["reasoning_content"].(string); ok && rc != "" {
|
2026-07-31 18:32:07 +08:00
|
|
|
reason := rc
|
feat(go-models): migrate batch 1 model drivers to unified handlers (#17696)
## Summary
Relate to #17284
Migrate 10 OpenAI-compatible drivers (`302ai`, `aliyun`, `astraflow`,
`avian`, `azure_openai`, `baichuan`, `baidu`, `cometapi`, `deepinfra`,
`futurmix`) to use the unified response handlers
(`HandleNonStreamingResponse` / `HandleStreamingResponse`), following
the same pattern established by `deepseek` in #17634.
- Cut ~150 lines per driver (1507 lines removed, 144 added across 10
files).
- No functional changes — pure deduplication of HTTP plumbing.
- Each driver now routes through `baseModel.doRequest()` and
`HandleNonStreamingResponse()`.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-08-03 12:13:43 +08:00
|
|
|
if reason[0] == '\n' {
|
2026-07-31 18:32:07 +08:00
|
|
|
reason = reason[1:]
|
|
|
|
|
}
|
|
|
|
|
reasonContent = &reason
|
feat(go-models): migrate batch 1 model drivers to unified handlers (#17696)
## Summary
Relate to #17284
Migrate 10 OpenAI-compatible drivers (`302ai`, `aliyun`, `astraflow`,
`avian`, `azure_openai`, `baichuan`, `baidu`, `cometapi`, `deepinfra`,
`futurmix`) to use the unified response handlers
(`HandleNonStreamingResponse` / `HandleStreamingResponse`), following
the same pattern established by `deepseek` in #17634.
- Cut ~150 lines per driver (1507 lines removed, 144 added across 10
files).
- No functional changes — pure deduplication of HTTP plumbing.
- Each driver now routes through `baseModel.doRequest()` and
`HandleNonStreamingResponse()`.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-08-03 12:13:43 +08:00
|
|
|
} else if rc, ok := messageMap["reasoning"].(string); ok && rc != "" {
|
|
|
|
|
// Some providers (e.g. Avian) report reasoning under a top-level
|
|
|
|
|
// "reasoning" field instead of "reasoning_content".
|
|
|
|
|
reason := rc
|
|
|
|
|
reasonContent = &reason
|
2026-07-31 18:32:07 +08:00
|
|
|
} else {
|
|
|
|
|
// Always return a non-nil pointer so callers can rely on it.
|
|
|
|
|
empty := ""
|
|
|
|
|
reasonContent = &empty
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Some providers (e.g. SiliconFlow) embed thinking inline via <think> tags.
|
|
|
|
|
// Extract reasoning into ReasonContent and strip it from Answer.
|
|
|
|
|
if content != nil {
|
|
|
|
|
if reasoning, answer := extractThinkContent(content); reasoning != nil {
|
|
|
|
|
rc := *reasoning
|
|
|
|
|
reasonContent = &rc
|
|
|
|
|
a := *answer
|
|
|
|
|
content = &a
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var toolCalls []map[string]any
|
|
|
|
|
if tcs, ok := messageMap["tool_calls"].([]any); ok {
|
|
|
|
|
for _, tc := range tcs {
|
|
|
|
|
if tcMap, ok := tc.(map[string]any); ok {
|
|
|
|
|
toolCalls = append(toolCalls, tcMap)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return content, reasonContent, toolCalls
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// extractRequestID extracts the request ID from a parsed response.
|
|
|
|
|
func extractRequestID(result map[string]any) string {
|
|
|
|
|
if id, ok := result["id"].(string); ok {
|
|
|
|
|
return id
|
|
|
|
|
}
|
|
|
|
|
return ""
|
|
|
|
|
}
|