Removing whitespace between Thinking and Content in Qwen3VL (#12838)

Eats extra whitespace at the end/beginning of content

model/parsers/qwen3vl.go

@@ -16,6 +16,8 @@ const (
 	CollectingThinkingContent qwenParserState = iota
 	CollectingContent
 	CollectingToolContent
+	ThinkingDoneEatingWhitespace
+	ToolCallDoneEatingWhitespace
 )
 
 const (
@@ -111,17 +113,28 @@ func (p *Qwen3VLParser) parseEvents() []qwenEvent {
 	return all
 }
 
-func emitContentBeforeTag(p *Qwen3VLParser, events []qwenEvent, tag string) []qwenEvent {
+func splitAtTag(p *Qwen3VLParser, tag string, trimAfter bool) (string, string) {
 	split := strings.SplitN(p.buffer.String(), tag, 2)
 	before := split[0]
 	before = strings.TrimRightFunc(before, unicode.IsSpace)
-	if len(before) > 0 {
-		events = append(events, qwenEventContent{content: before})
-	}
 	after := split[1]
+	if trimAfter {
+		after = strings.TrimLeftFunc(after, unicode.IsSpace)
+	}
 	p.buffer.Reset()
 	p.buffer.WriteString(after)
-	return events
+	return before, after // return events
+}
+
+func (p *Qwen3VLParser) eatLeadingWhitespaceAndTransitionTo(nextState qwenParserState) ([]qwenEvent, bool) {
+	trimmed := strings.TrimLeftFunc(p.buffer.String(), unicode.IsSpace)
+	p.buffer.Reset()
+	if trimmed == "" {
+		return nil, false
+	}
+	p.state = nextState
+	p.buffer.WriteString(trimmed)
+	return nil, true
 }
 
 func (p *Qwen3VLParser) eat() ([]qwenEvent, bool) {
@@ -130,7 +143,11 @@ func (p *Qwen3VLParser) eat() ([]qwenEvent, bool) {
 	switch p.state {
 	case CollectingContent:
 		if strings.Contains(p.buffer.String(), toolOpenTag) {
-			events = emitContentBeforeTag(p, events, toolOpenTag)
+			// events = emitContentBeforeTag(p, events, toolOpenTag)
+			before, _ := splitAtTag(p, toolOpenTag, false)
+			if len(before) > 0 {
+				events = append(events, qwenEventContent{content: before})
+			}
 			p.state = CollectingToolContent
 			return events, true
 		} else if overlapLen := overlap(p.buffer.String(), toolOpenTag); overlapLen > 0 {
@@ -167,27 +184,26 @@ func (p *Qwen3VLParser) eat() ([]qwenEvent, bool) {
 				slog.Warn("qwen tool call closing tag found but no content before it")
 			}
 
-			after := strings.TrimLeftFunc(split[1], unicode.IsSpace)
+			after := split[1]
 			events = append(events, qwenEventRawToolCall{raw: before})
 			p.buffer.Reset()
 			p.buffer.WriteString(after)
-			p.state = CollectingContent
+			p.state = ToolCallDoneEatingWhitespace
 			return events, true
 		} else {
 			return events, false
 		}
 	case CollectingThinkingContent:
 		if strings.Contains(p.buffer.String(), thinkingCloseTag) {
-			split := strings.SplitN(p.buffer.String(), thinkingCloseTag, 2)
-			// before := split[0]
-			before := strings.TrimRightFunc(split[0], unicode.IsSpace)
-			after := strings.TrimLeftFunc(split[1], unicode.IsSpace)
-			if len(before) > 0 {
-				events = append(events, qwenEventThinkingContent{content: before})
+			thinking, remaining := splitAtTag(p, thinkingCloseTag, true)
+			if len(thinking) > 0 {
+				events = append(events, qwenEventThinkingContent{content: thinking})
+			}
+			if remaining == "" {
+				p.state = ThinkingDoneEatingWhitespace
+			} else {
+				p.state = CollectingContent
 			}
-			p.buffer.Reset()
-			p.buffer.WriteString(after)
-			p.state = CollectingContent
 			return events, true
 		} else if overlapLen := overlap(p.buffer.String(), thinkingCloseTag); overlapLen > 0 {
 			beforePartialTag := p.buffer.String()[:len(p.buffer.String())-overlapLen]
@@ -215,6 +231,10 @@ func (p *Qwen3VLParser) eat() ([]qwenEvent, bool) {
 			}
 			return events, false
 		}
+	case ThinkingDoneEatingWhitespace:
+		return p.eatLeadingWhitespaceAndTransitionTo(CollectingContent)
+	case ToolCallDoneEatingWhitespace:
+		return p.eatLeadingWhitespaceAndTransitionTo(CollectingContent)
 	default:
 		panic("unreachable")
 	}

model/parsers/qwen3vl_nonthinking_test.go

@@ -653,3 +653,189 @@ func TestQwen3VLNonThinkingToolParser(t *testing.T) {
 		}
 	}
 }
+
+func TestQwen3VLNonThinkingToolCallWhitespaceHandling(t *testing.T) {
+	type step struct {
+		input      string
+		wantEvents []qwenEvent
+	}
+
+	cases := []struct {
+		desc  string
+		steps []step
+		only  bool
+	}{
+		{
+			desc: "whitespace inside tool call preserves trailing space",
+			steps: []step{
+				{
+					input: "before<tool_call>   tool content   </tool_call>after",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "before"},
+						qwenEventRawToolCall{raw: "   tool content   "},
+						qwenEventContent{content: "after"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace inside tool call preserves trailing space",
+			steps: []step{
+				{
+					input: "\n \n \n \n \n \n blahhhhhhhhhh blahhhh blahhhh \n\n\n\t\t     <tool_call>   tool content   </tool_call> \n\n\n\n\n\n\n after",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "\n \n \n \n \n \n blahhhhhhhhhh blahhhh blahhhh"},
+						qwenEventRawToolCall{raw: "   tool content   "},
+						qwenEventContent{content: "after"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace inside tool call preserves trailing space",
+			steps: []step{
+				{
+					input: "<tool_call>   tool content   </tool_call>            ",
+					wantEvents: []qwenEvent{
+						qwenEventRawToolCall{raw: "   tool content   "},
+					},
+				},
+				{
+					input: "\n \n \n \n \n \n blahhhhhhhhhh blahhhh blahhhh \n\n\n\t\t     <tool_call>   anotha one   </tool_call> \n\n\n\n\n\n\n after \n\n\n\n\n\n blep",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "blahhhhhhhhhh blahhhh blahhhh"},
+						qwenEventRawToolCall{raw: "   anotha one   "},
+						qwenEventContent{content: "after \n\n\n\n\n\n blep"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace between content and tool call",
+			steps: []step{
+				{
+					input: "content   \n  <tool_call>tool</tool_call>  \n  more content",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "content"},
+						qwenEventRawToolCall{raw: "tool"},
+						qwenEventContent{content: "more content"},
+					},
+				},
+			},
+		},
+		{
+			desc: "consecutive tool calls with whitespace",
+			steps: []step{
+				{
+					input: "<tool_call>first</tool_call>  \n  <tool_call>second</tool_call>  \n  <tool_call>third</tool_call>",
+					wantEvents: []qwenEvent{
+						qwenEventRawToolCall{raw: "first"},
+						qwenEventRawToolCall{raw: "second"},
+						qwenEventRawToolCall{raw: "third"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace before and after tool open tag",
+			steps: []step{
+				{
+					input: "text   \n   <tool_call>content</tool_call>",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "text"},
+						qwenEventRawToolCall{raw: "content"},
+					},
+				},
+			},
+		},
+		{
+			desc: "unicode whitespace around tool calls",
+			steps: []step{
+				{
+					input: "text\u00a0\u3000<tool_call>content</tool_call>\u00a0\u3000text",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "text"},
+						qwenEventRawToolCall{raw: "content"},
+						qwenEventContent{content: "text"},
+					},
+				},
+			},
+		},
+		{
+			desc: "empty tool call with surrounding whitespace",
+			steps: []step{
+				{
+					input: "before  <tool_call></tool_call>  after",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "before"},
+						qwenEventRawToolCall{raw: ""},
+						qwenEventContent{content: "after"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace in tool call split across chunks",
+			steps: []step{
+				{
+					input:      "before<tool_call>  ",
+					wantEvents: []qwenEvent{qwenEventContent{content: "before"}},
+				},
+				{
+					input:      "tool",
+					wantEvents: []qwenEvent{},
+				},
+				{
+					input: "  </tool_call>after",
+					wantEvents: []qwenEvent{
+						qwenEventRawToolCall{raw: "  tool  "},
+						qwenEventContent{content: "after"},
+					},
+				},
+			},
+		},
+		{
+			desc: "mixed whitespace types between tool calls",
+			steps: []step{
+				{
+					input: "<tool_call>first</tool_call> \t\n\r <tool_call>second</tool_call>",
+					wantEvents: []qwenEvent{
+						qwenEventRawToolCall{raw: "first"},
+						qwenEventRawToolCall{raw: "second"},
+					},
+				},
+			},
+		},
+	}
+
+	anyOnlies := false
+	for _, tc := range cases {
+		if tc.only {
+			anyOnlies = true
+		}
+	}
+
+	for _, tc := range cases {
+		if anyOnlies && !tc.only {
+			continue
+		}
+
+		t.Run(tc.desc, func(t *testing.T) {
+			parser := Qwen3VLParser{hasThinkingSupport: false}
+			parser.Init([]api.Tool{}, nil)
+
+			for i, step := range tc.steps {
+				parser.buffer.WriteString(step.input)
+				gotEvents := parser.parseEvents()
+
+				if len(gotEvents) == 0 && len(step.wantEvents) == 0 {
+					continue
+				}
+
+				if !reflect.DeepEqual(gotEvents, step.wantEvents) {
+					t.Errorf("step %d: input %q: got events %#v, want %#v", i, step.input, gotEvents, step.wantEvents)
+				}
+			}
+		})
+	}
+}

model/parsers/qwen3vl_thinking_test.go

@@ -546,3 +546,333 @@ func TestQwen3VLThinkingParserStreamingAssistantPrefillContent(t *testing.T) {
 		}
 	}
 }
+
+func TestQwen3VLThinkingWhitespaceHandling(t *testing.T) {
+	type step struct {
+		input      string
+		wantEvents []qwenEvent
+	}
+
+	cases := []struct {
+		desc  string
+		steps []step
+		only  bool
+	}{
+		{
+			desc: "whitespace after thinking tag is trimmed",
+			steps: []step{
+				{
+					input: "thinking content</think>   \n\t  content starts here",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "thinking content"},
+						qwenEventContent{content: "content starts here"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace after thinking tag split across chunks",
+			steps: []step{
+				{
+					input:      "thinking content</think>   ",
+					wantEvents: []qwenEvent{qwenEventThinkingContent{content: "thinking content"}},
+				},
+				{
+					input:      "  \n\t",
+					wantEvents: []qwenEvent{},
+				},
+				{
+					input: "content",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "content"},
+					},
+				},
+			},
+		},
+		{
+			desc: "only whitespace after thinking tag",
+			steps: []step{
+				{
+					input:      "thinking content</think>   \n\t  ",
+					wantEvents: []qwenEvent{qwenEventThinkingContent{content: "thinking content"}},
+				},
+			},
+		},
+		{
+			desc: "multiple spaces and tabs after thinking",
+			steps: []step{
+				{
+					input: "think</think>     \t\t\n\n   text",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "think"},
+						qwenEventContent{content: "text"},
+					},
+				},
+			},
+		},
+		{
+			desc: "trailing whitespace before thinking tag is preserved in content",
+			steps: []step{
+				{
+					input: "thinking with spaces   </think>text",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "thinking with spaces"},
+						qwenEventContent{content: "text"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace between thinking and tool call",
+			steps: []step{
+				{
+					input: "thinking</think>  \n  <tool_call>{\"name\":\"test\"}</tool_call>",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "thinking"},
+						qwenEventRawToolCall{raw: "{\"name\":\"test\"}"},
+					},
+				},
+			},
+		},
+		{
+			desc: "no whitespace after thinking tag",
+			steps: []step{
+				{
+					input: "thinking</think>content",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "thinking"},
+						qwenEventContent{content: "content"},
+					},
+				},
+			},
+		},
+		{
+			desc: "unicode whitespace after thinking tag",
+			steps: []step{
+				{
+					input: "thinking</think>\u00a0\u3000content",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "thinking"},
+						qwenEventContent{content: "content"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace split with partial thinking tag",
+			steps: []step{
+				{
+					input:      "thinking</th",
+					wantEvents: []qwenEvent{qwenEventThinkingContent{content: "thinking"}},
+				},
+				{
+					input:      "ink>  \n",
+					wantEvents: []qwenEvent{},
+				},
+				{
+					input: "  content",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "content"},
+					},
+				},
+			},
+		},
+		{
+			desc: "empty thinking tag with whitespace after",
+			steps: []step{
+				{
+					input: "</think>   \ncontent",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "content"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace inside tool call preserves trailing space",
+			steps: []step{
+				{
+					input: "bruh</think> \n \n \n \n \n \n blahhhhhhhhhh blahhhh blahhhh \n\n\n\t\t     <tool_call>   tool content   </tool_call> \n\n\n\n\n\n\n after",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "bruh"},
+						qwenEventContent{content: "blahhhhhhhhhh blahhhh blahhhh"},
+						qwenEventRawToolCall{raw: "   tool content   "},
+						qwenEventContent{content: "after"},
+					},
+				},
+			},
+		},
+		{
+			desc: "whitespace inside tool call preserves trailing space",
+			steps: []step{
+				{
+					input: "bruh</think>          shdjfhksdhfj  ",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "bruh"},
+						qwenEventContent{content: "shdjfhksdhfj"},
+					},
+				},
+				{
+					input: "another word  ",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "  another word"},
+					},
+				},
+				{
+					input: "<tool_call>   tool content   </tool_call>            ",
+					wantEvents: []qwenEvent{
+						qwenEventRawToolCall{raw: "   tool content   "},
+					},
+				},
+				{
+					input: "\n \n \n \n \n \n blahhhhhhhhhh blahhhh blahhhh \n\n\n\t\t     <tool_call>   anotha one   </tool_call> \n\n\n\n\n\n\n after \n\n\n\n\n\n blep",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "blahhhhhhhhhh blahhhh blahhhh"},
+						qwenEventRawToolCall{raw: "   anotha one   "},
+						qwenEventContent{content: "after \n\n\n\n\n\n blep"},
+					},
+				},
+			},
+		},
+	}
+
+	anyOnlies := false
+	for _, tc := range cases {
+		if tc.only {
+			anyOnlies = true
+		}
+	}
+
+	for _, tc := range cases {
+		if anyOnlies && !tc.only {
+			continue
+		}
+
+		t.Run(tc.desc, func(t *testing.T) {
+			parser := Qwen3VLParser{hasThinkingSupport: true}
+			parser.Init([]api.Tool{}, nil)
+
+			for i, step := range tc.steps {
+				parser.buffer.WriteString(step.input)
+				gotEvents := parser.parseEvents()
+
+				if len(gotEvents) == 0 && len(step.wantEvents) == 0 {
+					continue
+				}
+
+				if !reflect.DeepEqual(gotEvents, step.wantEvents) {
+					t.Errorf("step %d: input %q: got events %#v, want %#v", i, step.input, gotEvents, step.wantEvents)
+				}
+			}
+		})
+	}
+}
+
+func TestQwen3VLToolCallWhitespaceHandling(t *testing.T) {
+	type step struct {
+		input      string
+		wantEvents []qwenEvent
+	}
+
+	cases := []struct {
+		desc       string
+		steps      []step
+		only       bool
+		prefillMsg *api.Message // allows starting in content mode instead of thinking mode
+	}{
+		{
+			desc:       "whitespace inside tool call is fully preserved (with content prefill)",
+			prefillMsg: &api.Message{Role: "assistant", Content: "prefill"},
+			steps: []step{
+				{
+					input: "before<tool_call>   tool content   </tool_call>  \n  after",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "before"},
+						qwenEventRawToolCall{raw: "   tool content   "},
+						qwenEventContent{content: "after"},
+					},
+				},
+			},
+		},
+		{
+			desc:       "whitespace after tool call trimmed across chunks (with content prefill)",
+			prefillMsg: &api.Message{Role: "assistant", Content: "prefill"},
+			steps: []step{
+				{
+					input: "before<tool_call>tool</tool_call>   ",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "before"},
+						qwenEventRawToolCall{raw: "tool"},
+					},
+				},
+				{
+					input:      "\n\t",
+					wantEvents: []qwenEvent{},
+				},
+				{
+					input: "after \n this is a song",
+					wantEvents: []qwenEvent{
+						qwenEventContent{content: "after \n this is a song"},
+					},
+				},
+			},
+		},
+		{
+			desc:       "multiple tool calls with whitespace between (with content prefill)",
+			prefillMsg: &api.Message{Role: "assistant", Content: "prefill"},
+			steps: []step{
+				{
+					input: "<tool_call>first</tool_call>  \n  <tool_call>second</tool_call>",
+					wantEvents: []qwenEvent{
+						qwenEventRawToolCall{raw: "first"},
+						qwenEventRawToolCall{raw: "second"},
+					},
+				},
+			},
+		},
+		{
+			desc: "thinking with whitespace then tool call",
+			steps: []step{
+				{
+					input: "thinking</think>   \n   <tool_call>tool</tool_call>   \n   content",
+					wantEvents: []qwenEvent{
+						qwenEventThinkingContent{content: "thinking"},
+						qwenEventRawToolCall{raw: "tool"},
+						qwenEventContent{content: "content"},
+					},
+				},
+			},
+		},
+	}
+
+	anyOnlies := false
+	for _, tc := range cases {
+		if tc.only {
+			anyOnlies = true
+		}
+	}
+
+	for _, tc := range cases {
+		if anyOnlies && !tc.only {
+			continue
+		}
+
+		t.Run(tc.desc, func(t *testing.T) {
+			parser := Qwen3VLParser{hasThinkingSupport: true}
+			parser.Init([]api.Tool{}, tc.prefillMsg)
+
+			for i, step := range tc.steps {
+				parser.buffer.WriteString(step.input)
+				gotEvents := parser.parseEvents()
+
+				if len(gotEvents) == 0 && len(step.wantEvents) == 0 {
+					continue
+				}
+
+				if !reflect.DeepEqual(gotEvents, step.wantEvents) {
+					t.Errorf("step %d: input %q: got events %#v, want %#v", i, step.input, gotEvents, step.wantEvents)
+				}
+			}
+		})
+	}
+}