checkpoint

server/routes.go

@@ -1485,7 +1485,6 @@ func (s *Server) ChatHandler(c *gin.Context) {
 	ch := make(chan any)
 	go func() {
 		defer close(ch)
-		// var sb strings.Builder
 		var toolParser *ToolParser
 		if len(req.Tools) > 0 {
 			toolParser = NewToolParser(m)
@@ -1518,6 +1517,9 @@ func (s *Server) ChatHandler(c *gin.Context) {
 
 			if len(req.Tools) > 0 && !toolParser.Done {
 				toolCalls, leftover := toolParser.ParseToolCalls(r.Content)
+				// * This can be abstracted again to a .handleState(tp.state)
+				// * However, we'd need a flag to indicate whether to send the response or not
+				// * happy to take whatever is more idiomatic
 				switch toolParser.ParserState {
 				case ToolCallAccumulate:
 					// tokens are accumulated in the tool parser
@@ -1526,7 +1528,10 @@ func (s *Server) ChatHandler(c *gin.Context) {
 					// tokens are sent back in the response
 				case ToolCallSendPartial:
 					// tokens not needed for parsing are sent back in the response
-					res.Message.Content = leftover
+					if len(leftover) > 0 {
+						res.Message.Content = leftover
+					}
+					// ! state is needed as we need to not match on the other states
 				case ToolCallFound:
 					res.Message.ToolCalls = toolCalls
 					res.Message.Content = ""
@@ -1534,6 +1539,7 @@ func (s *Server) ChatHandler(c *gin.Context) {
 			}
 
 			fmt.Println("sending response", res.Message.Content)
+			// * this is where we'd need the flag if we have a .handleState(tp.state)
 			ch <- res
 		}); err != nil {
 			ch <- gin.H{"error": err.Error()}

server/tools.go

@@ -5,7 +5,6 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"log/slog"
 	"strings"
 	gotmpl "text/template"
 
@@ -24,7 +23,9 @@ const (
 	GreedyToolNoPrefix
 	ForceTools
 	ToolSuffix
-	ContainsPartialPrefix
+	ContainsPrefix
+	PartialPrefix
+	NotPartialPrefix
 	Done
 )
 
@@ -64,9 +65,11 @@ func (s State) String() string {
 		return "ForceTools"
 	case ToolSuffix:
 		return "ToolSuffix"
+	case PartialPrefix:
+		return "PossiblePrefix"
 	case Done:
 		return "Done"
-	case ContainsPartialPrefix:
+	case ContainsPrefix:
 		return "PartialPrefix"
 	default:
 		return fmt.Sprintf("Unknown State (%d)", s)
@@ -88,6 +91,8 @@ type ToolParser struct {
 // parseJSONToolCalls attempts to parse a JSON string into a slice of ToolCalls.
 // Returns parsed tool calls, a boolean indicating if the JSON is incomplete, and a boolean indicating if the tool calls were found
 func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
+	fmt.Printf("attempting to parse JSON tool calls: input=%s\n", s)
+
 	var b bytes.Buffer
 	if err := p.tmpl.Execute(&b, map[string][]api.ToolCall{
 		"ToolCalls": {
@@ -101,6 +106,7 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 			},
 		},
 	}); err != nil {
+		fmt.Printf("failed to execute template: error=%v\n", err)
 		return nil, false, false
 	}
 
@@ -108,6 +114,7 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 	var temp any
 	err := jsonv2.Unmarshal(b.Bytes(), &temp)
 	if err != nil {
+		fmt.Printf("failed to unmarshal template: error=%v\n", err)
 		return nil, false, false
 	}
 
@@ -125,6 +132,7 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 			}
 		default:
 			// TODO: err or fallback
+			fmt.Printf("collect encountered unknown type: type=%T\n", obj)
 			return nil
 		}
 
@@ -142,6 +150,7 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 		templateObjects = collect(t)
 	}
 	if len(templateObjects) == 0 {
+		fmt.Println("no template objects found")
 		return nil, false, false
 	}
 
@@ -151,12 +160,15 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 		switch v.(type) {
 		case string:
 			name = k
+			fmt.Printf("found name field: key=%s\n", k)
 		case map[string]any:
 			arguments = k
+			fmt.Printf("found arguments field: key=%s\n", k)
 		}
 	}
 
 	if name == "" || arguments == "" {
+		fmt.Printf("missing required fields: name_found=%v arguments_found=%v\n", name != "", arguments != "")
 		return nil, false, false
 	}
 
@@ -165,18 +177,17 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 	dec := jsontext.NewDecoder(strings.NewReader(s))
 	if got, err := dec.ReadValue(); err == nil {
 		s = got.String()
+		fmt.Printf("decoded JSON value: value=%s\n", s)
 	}
 
 	var responseObjects any
 	err = jsonv2.Unmarshal([]byte(s), &responseObjects)
 	if err != nil {
 		if errors.Is(err, io.ErrUnexpectedEOF) || err.Error() == "unexpected end of JSON input" {
-			fmt.Println("Detected partial or incomplete JSON.")
+			fmt.Println("incomplete JSON detected")
-			fmt.Println("state", p.state)
 			return nil, true, false
 		} else {
-			fmt.Printf("Other error: %v\n", err)
+			fmt.Printf("failed to unmarshal response: error=%v\n", err)
-			fmt.Println("exiting from JSON parsing", p.state)
 			return nil, false, false
 		}
 	}
@@ -187,14 +198,14 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 		return nil, false, false
 	}
 
-	slog.Debug("collected objects", "count", len(objs))
+	fmt.Printf("collected objects: count=%d\n", len(objs))
 
 	var toolCalls []api.ToolCall
 	for _, kv := range objs {
 		n, nok := kv[name].(string)
 		a, aok := kv[arguments].(map[string]any)
 		if nok && aok {
-			slog.Debug("found valid tool call", "name", n)
+			fmt.Printf("found valid tool call: name=%s\n", n)
 			toolCalls = append(toolCalls, api.ToolCall{
 				Function: api.ToolCallFunction{
 					Name:      n,
@@ -204,84 +215,89 @@ func (p *ToolParser) parseJSONToolCalls(s string) ([]api.ToolCall, bool, bool) {
 		}
 	}
 
-	slog.Debug("parsed tool calls", "count", len(toolCalls))
+	fmt.Printf("parsed tool calls: count=%d\n", len(toolCalls))
 	return toolCalls, false, true
 }
 
-func (p *ToolParser) updateOutputState(ok bool, partial bool, tcs []api.ToolCall) {
+// TODO: clean up the boundary of internal and external state transitions
+func (p *ToolParser) updateStateAfterJSONParse(ok bool, partial bool, tcs []api.ToolCall) {
+	fmt.Printf("updating output state: ok=%v partial=%v tool_calls=%d current_state=%s\n", ok, partial, len(tcs), p.state)
+
+	// state transition logic
 	switch {
 	case !ok && !partial && p.state == ForceTools:
-		fmt.Println("Case: !ok && !partial && ForceTools - staying in force tools, resetting buffer")
 		// force partial tool if we have a prefix
 		// no op and stay in force tools
 		p.sb.Reset()
 	case !ok && !partial:
-		fmt.Println("Case: !ok && !partial")
-		fmt.Println("state", p.state)
 		if p.state == GreedyToolNoPrefix {
-			fmt.Println("  Subcase: GreedyToolNoPrefix - marking as done")
 			p.state = Done
-			// p.ParserState = DoneFR
+			// ? the output parser state is the same even though internal can we not leak the external state?
-			p.ParserState = ToolCallSendTokens
 			p.Done = true
 		}
 		if p.state == GreedyToolWithPrefix {
-			fmt.Println("  Subcase: GreedyToolWithPrefix - switching to SendTokens")
 			p.state = SendTokens
-			p.ParserState = ToolCallSendTokens
 		}
-		p.sb.Reset()
+		if p.state == PartialPrefix {
+			p.state = NotPartialPrefix
+		}
 	case !ok && partial:
-		fmt.Println("Case: !ok && partial - accumulating partial content")
+		// acucumulate
-		// ! acucumulate
 
 	case len(tcs) > 0:
-		fmt.Println("Case: tool calls found")
 		// do not parse again in the greedy JSON case as soon as we have a tool call
-		if p.state == GreedyToolWithPrefix {
-			p.state = SendTokens
-			p.ParserState = ToolCallFound
-			p.state = Done
-			p.Done = true
-		} else if p.state == GreedyToolNoPrefix {
-			fmt.Println("  Subcase: Greedy modes - marking done and switching to SendTokens")
-			p.state = Done
-			p.Done = true
-		}
 		p.sb.Reset()
 	}
 	p.updateExternalState(tcs)
+	fmt.Printf("state updated: new_state=%s parser_state=%s\n", p.state, p.ParserState)
 }
 
 func (p *ToolParser) updateExternalState(tcs []api.ToolCall) {
-	if (p.state == GreedyToolWithPrefix || p.state == GreedyToolNoPrefix || p.state == ToolSuffix) || (p.state == ForceTools && len(tcs) == 0) {
+	fmt.Printf("updating external state: current_state=%s tool_calls=%d\n", p.state, len(tcs))
-		p.ParserState = ToolCallAccumulate
+
-	} else if p.state == ContainsPartialPrefix {
+	switch {
-		p.ParserState = ToolCallSendPartial
+	case len(tcs) > 0:
-	} else if len(tcs) > 0 {
+		// do not parse again in the greedy JSON case as soon as we have a tool call
+		if p.state == GreedyToolWithPrefix {
+			p.state = SendTokens
+		} else if p.state == GreedyToolNoPrefix {
+			p.state = Done
+			p.Done = true
+		}
 		p.ParserState = ToolCallFound
-	} else if p.state == SendTokens {
+	case p.state == GreedyToolWithPrefix || p.state == GreedyToolNoPrefix ||
+		p.state == ToolSuffix || p.state == PartialPrefix ||
+		(p.state == ForceTools && len(tcs) == 0):
+		p.ParserState = ToolCallAccumulate
+	case p.state == ContainsPrefix:
+		p.ParserState = ToolCallSendPartial
+	case p.state == SendTokens || p.state == Done:
 		p.ParserState = ToolCallSendTokens
+	case p.state == NotPartialPrefix:
+		p.ParserState = ToolCallSendPartial
+	default:
+		p.ParserState = ToolCallSendTokens
+		p.sb.Reset()
+		p.state = SendTokens
 	}
 }
 
 // string, and if it has a prefix
 func (p *ToolParser) checkPrefix(s string) (string, bool) {
+	fmt.Printf("checking prefix: input=%s prefix=%s\n", s, p.toolPrefix)
+
 	if p.toolPrefix == "" {
 		return s, true
 	}
 	original := s
-	// s = strings.TrimSpace(s)
 	s, hasPrefix := strings.CutPrefix(s, p.toolPrefix)
 	if hasPrefix {
-		fmt.Println("has prefix", s)
 		p.state = ForceTools
-		// partial tool possibly
+		fmt.Printf("found exact prefix match: remaining=%s\n", s)
-	} else if strings.HasPrefix(p.toolPrefix, s) {
+		// partial tool possibly - accumulate
-		slog.Debug("tool prefix partially", "prefix", p.toolPrefix, "content", s)
+	} else if suffixOverlap(s, p.toolPrefix) > 0 {
-		// TODO: could possibly err maybe this should be greedy instead?
+		p.state = PartialPrefix
-		p.state = ForceTools
+		fmt.Printf("found partial prefix: remaining=%s\n", s)
-		// this would basically be a no op on rest of the input
 		return "", false
 		// the case where "token<tool_call>" - send "token" back
 		// accounts for spaces in prefix or suffix to avoid breaking cache
@@ -289,11 +305,13 @@ func (p *ToolParser) checkPrefix(s string) (string, bool) {
 		idx := strings.Index(original, p.toolPrefix)
 		if idx != -1 {
 			// still keeps the prefix
-			p.state = ContainsPartialPrefix
+			p.state = ContainsPrefix
 			p.sb.Reset()
 			// todo: see if there is a simpler way for this
 			idx2 := strings.Index(s, p.toolPrefix)
+			// buffer now only has the prefix
 			p.sb.WriteString(s[idx2:])
+			fmt.Printf("found prefix in middle: prefix_start=%d content_before=%s\n", idx, original[:idx])
 			return original[:idx], false
 		}
 	}
@@ -305,51 +323,71 @@ func (p *ToolParser) checkPrefix(s string) (string, bool) {
 // ParseToolCalls extracts tool calls from a string using a tool token prefix or direct JSON parsing.
 // Returns tool calls, whether parsing is incomplete, and any errors.
 func (p *ToolParser) ParseToolCalls(s string) ([]api.ToolCall, string) {
-	fmt.Println("checking tool calls", s)
+	fmt.Printf("parsing tool calls: input=%s current_state=%s\n", s, p.state)
-	fmt.Println("external state", p.ParserState)
+
-	fmt.Println("internal state", p.state)
 	p.sb.WriteString(s)
 	s = p.sb.String()
 
 	s = strings.TrimSpace(s)
-	fmt.Println("sb", s)
 
-	p.updateExternalState(nil)
 	if len(s) == 0 {
+		p.updateExternalState(nil)
 		return nil, ""
 	}
 
 	s, cont := p.checkPrefix(s)
 	if !cont {
 		p.updateExternalState(nil)
-		if p.state == ContainsPartialPrefix {
+		if p.state == ContainsPrefix {
+			fmt.Printf("returning partial prefix: remaining=%s\n", s)
 			return nil, s
 		}
+		// * we'd be returning here for just accumulating with possible prefix
+		// * ext state is accumulation
 		return nil, ""
 	}
+	// * lets say the check fails here and now we're still in external state accumulation here
 
 	// stay in SendTokens unless we have a prefix
 	if p.state == SendTokens {
-		fmt.Println("SendTokens - resetting buffer")
 		p.updateExternalState(nil)
 		p.sb.Reset()
-		return nil, ""
+		fmt.Printf("returning send tokens: remaining=%s\n", s)
+		return nil, s
 	}
 
+	// * we'd parse here as json to see if it's a tool call
 	tcs, partial, ok := p.parseJSONToolCalls(s)
-	p.updateOutputState(ok, partial, tcs)
+	// * it would not be a tool call here
-	fmt.Println("output state", p.ParserState, p.state)
+	p.updateStateAfterJSONParse(ok, partial, tcs)
 	if !ok {
-		fmt.Println("returning empty tool calls")
+		// * and so we should send the data here
+		// * we also need to move out of that internal state after sending the tokens
+		if p.state == NotPartialPrefix {
+			p.state = SendTokens
+			// the string would have acc until here
+			return nil, p.sb.String()
+		}
 		return nil, ""
 	}
 	for _, tc := range tcs {
 		tc.Function.Index = p.toolIndex
 		p.toolIndex++
 	}
+	fmt.Printf("finished parsing tool calls: tool_calls_found=%d\n", len(tcs))
 	return tcs, ""
 }
 
+func suffixOverlap(s, delim string) int {
+	max := min(len(delim), len(s))
+	for i := max; i > 0; i-- {
+		if strings.HasSuffix(s, delim[:i]) {
+			return i
+		}
+	}
+	return 0
+}
+
 func NewToolParser(model *Model) *ToolParser {
 	// TODO: use new template parsing to get all tokens for the prefix
 	templateToolPrefix, _ := ToolPrefix(model.Template.Template)
@@ -365,7 +403,7 @@ func NewToolParser(model *Model) *ToolParser {
 	} else {
 		state = GreedyToolWithPrefix
 	}
-	fmt.Println("setup state", state)
+	fmt.Printf("creating new tool parser: prefix=%s initial_state=%s\n", templateToolPrefix, state)
 	return &ToolParser{
 		tmpl:        tmpl,
 		sb:          &strings.Builder{},

server/tools_test.go

@@ -55,21 +55,21 @@ func TestParseToolCalls(t *testing.T) {
 		expectedTokens   string
 	}{
 		{
-			name:             "mistral invalid json",
+			name:             "mistral malformed json with tool calls prefix",
 			model:            "mistral",
 			output:           `[TOOL_CALLS]  [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_curren}]`,
 			expectedToolCall: []api.ToolCall{},
 			expectedTokens:   "",
 		},
 		{
-			name:             "mistral multiple tool calls - no prefix",
+			name:             "mistral multiple tool calls without prefix",
 			model:            "mistral",
 			output:           `[{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "",
 		},
 		{
-			name:  "mistral tool calls with text in between - no prefix",
+			name:  "mistral tool calls with text between no prefix",
 			model: "mistral",
 			output: `[{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}] 
 			model outputs more tokens here and then [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
@@ -77,15 +77,14 @@ func TestParseToolCalls(t *testing.T) {
 			expectedTokens:   `model outputs more tokens here and then [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
 		},
 		{
-			name:             "mistral valid json - with prefix",
+			name:             "mistral valid json with tool calls prefix",
 			model:            "mistral",
 			output:           `[TOOL_CALLS]  [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "",
 		},
 		{
-			// In this case we'd be ignoring the text in between and just returning the tool calls
+			name:  "mistral multiple tool calls with text between and prefix",
-			name:  "mistral valid json with text in between - with prefix",
 			model: "mistral",
 			output: `[TOOL_CALLS]  [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]
 			model outputs more tokens here and then [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
@@ -93,14 +92,14 @@ func TestParseToolCalls(t *testing.T) {
 			expectedTokens:   "",
 		},
 		{
-			name:             "mistral incomplete json",
+			name:             "mistral incomplete json with tool calls prefix",
 			model:            "mistral",
 			output:           `[TOOL_CALLS]  [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, `,
 			expectedToolCall: []api.ToolCall{},
 			expectedTokens:   "",
 		},
 		{
-			name:  "mistral without tool token",
+			name:  "mistral invalid tool call with explanatory text no prefix",
 			model: "mistral",
 			output: `I'm not aware of that information. However, I can suggest searching for the weather using the "get_current_weather" function:
 
@@ -109,14 +108,14 @@ func TestParseToolCalls(t *testing.T) {
 			expectedTokens:   `I'm not aware of that information. However, I can suggest searching for the weather using the "get_current_weather" function: [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
 		},
 		{
-			name:             "mistral without tool token - tool first",
+			name:             "mistral tool calls without prefix",
 			model:            "mistral",
 			output:           `[{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "",
 		},
 		{
-			name:  "command-r-plus with json block",
+			name:  "command r plus tool calls with json block format",
 			model: "command-r-plus",
 			output: "Action: ```json" + `
 		[
@@ -140,14 +139,14 @@ func TestParseToolCalls(t *testing.T) {
 			expectedTokens:   "",
 		},
 		{
-			name:             "firefunction with functools",
+			name:             "firefunction tool calls with functools prefix",
 			model:            "firefunction",
 			output:           ` functools[{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "",
 		},
 		{
-			name:  "llama3 with tool call tags",
+			name:  "llama3 groq single tool call with xml tags",
 			model: "llama3-groq-tool-use",
 			output: `<tool_call>
 		{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}
@@ -156,99 +155,126 @@ func TestParseToolCalls(t *testing.T) {
 			expectedTokens:   "",
 		},
 		{
-			name:             "xlam with tool_calls wrapper",
+			name:             "xlam tool calls with wrapper object",
 			model:            "xlam",
 			output:           `{"tool_calls": [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}},{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]}`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "",
 		},
 		{
-			name:             "qwen2.5 with single tool call",
+			name:             "qwen2.5-coder single tool call with prefix",
 			model:            "qwen2.5-coder",
 			output:           `<tool_call>{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}</tool_call>`,
 			expectedToolCall: []api.ToolCall{t1},
 			expectedTokens:   "",
 		},
 		{
-			name:             "qwen with no tool prefix",
+			name:             "qwen2.5-coder multiple tool calls with and without prefix",
+			model:            "qwen2.5-coder",
+			output:           `{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} <tool_call>{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}</tool_call> <tool_call>{"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}</tool_call>`,
+			expectedToolCall: []api.ToolCall{t1, t1, t2},
+			expectedTokens:   "",
+		},
+		{
+			name:             "qwen2.5-coder multiple tool calls without prefix",
 			model:            "qwen2.5-coder",
 			output:           `[{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}, {"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}]`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "",
 		},
 		{
-			name:             "qwen with no tool calls",
+			name:             "qwen2.5-coder plain text response no tool calls",
 			model:            "qwen2.5-coder",
 			output:           "The weather in San Francisco, CA is 70°F and in Toronto, Canada is 20°C.",
 			expectedToolCall: []api.ToolCall{},
 			expectedTokens:   "The weather in San Francisco, CA is 70°F and in Toronto, Canada is 20°C.",
 		},
 		{
-			name:             "qwen with no tool prefix",
+			name:             "qwen2.5-coder tool calls with trailing text",
 			model:            "qwen2.5-coder",
 			output:           `[{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}, {"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}] some tokens after call`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "some tokens after call",
 		},
 		{
-			name:             "qwen with prefix",
+			name:             "qwen2.5 tool calls with prefix and trailing text",
 			model:            "qwen2.5-coder",
 			output:           `<tool_call> [{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}, {"name": "get_current_weather", "arguments": {"format":"celsius","location":"Toronto, Canada"}}] </tool_call> some tokens after call`,
 			expectedToolCall: []api.ToolCall{t1, t2},
 			expectedTokens:   "",
 		},
 		{
-			// tests the leftover logic as well
+			name:             "qwen3 tool call with think prefix and tool prefix (sent as a single token)",
-			name:             "qwen3 with single tool call and thinking",
 			model:            "qwen3",
 			output:           `<think>Okay, let me think what tool we should use...</think><tool_call>{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}</tool_call>`,
 			expectedToolCall: []api.ToolCall{t1},
 			expectedTokens:   "<think>Okay, let me think what tool we should use...</think>",
 		},
 		{
-			name:             "qwen3 with single tool call and thinking spaces",
+			name:             "qwen3 tool call with think prefix, tool prefix, and whitespace (sent as separate tokens)",
 			model:            "qwen3",
 			output:           `<think>Okay, let me think what tool we should use...</think> <tool_call> {"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
 			expectedToolCall: []api.ToolCall{t1},
 			expectedTokens:   "<think>Okay, let me think what tool we should use...</think>",
 		},
 		{
-			name:             "qwen3 testing",
+			name:             "qwen3 empty think prefix without tool prefix and invalid tool call",
 			model:            "qwen3",
 			output:           `<think></think>{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
 			expectedToolCall: []api.ToolCall{},
 			expectedTokens:   `<think></think>{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
 		},
 		{
-			name:             "qwen3 testing 2",
+			name:             "qwen3 empty think prefix with tool prefix and valid tool call",
 			model:            "qwen3",
 			output:           `<think></think><tool_call>{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}}  </tool_call>`,
 			expectedToolCall: []api.ToolCall{t1},
 			expectedTokens:   `<think></think>`,
 		},
 		{
-			name:             "llama3.2 with tool call - no prefix",
+			name:             "qwen3 invalid tool call with fake tool prefix (single rune suffix match)",
+			model:            "qwen3",
+			output:           `<think></think>< fakeout{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
+			expectedToolCall: []api.ToolCall{},
+			expectedTokens:   `<think></think>< fakeout{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
+		},
+		{
+			name:             "qwen3 invalid tool call with partial tool prefix (multiple rune suffix match)",
+			model:            "qwen3",
+			output:           `<think></think><tool_c fakeout{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
+			expectedToolCall: []api.ToolCall{},
+			expectedTokens:   `<think></think><tool_c fakeout{"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
+		},
+		{
+			name:             "qwen3 invalid tool call with malformed tool prefix",
+			model:            "qwen3",
+			output:           `<think></think><tool_cfakeout {"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
+			expectedToolCall: []api.ToolCall{},
+			expectedTokens:   `<think></think><tool_cfakeout {"name": "get_current_weather", "arguments": {"format":"fahrenheit","location":"San Francisco, CA"}} </tool_call>`,
+		},
+		{
+			name:             "llama3.2 valid tool call without prefix",
 			model:            "llama3.2",
 			output:           `{"name": "get_current_weather", "parameters": {"format":"fahrenheit","location":"San Francisco, CA"}}`,
 			expectedToolCall: []api.ToolCall{t1},
 			expectedTokens:   "",
 		},
 		{
-			name:             "llama3.2 with incomplete tool call - no prefix",
+			name:             "llama3.2 incomplete tool call without prefix",
 			model:            "llama3.2",
 			output:           `{"name": "get_current_weather", "parameters": {"format":"fahrenheit","location":"San Francisco, `,
 			expectedToolCall: []api.ToolCall{},
 			expectedTokens:   "",
 		},
 		{
-			name:             "llama3.2 with tool call - in middle",
+			name:             "llama3.2 tool call with leading text",
 			model:            "llama3.2",
 			output:           `some non json text{"name": "get_current_weather", "parameters": {"format":"fahrenheit","location":"San Francisco, CA"}}`,
 			expectedToolCall: []api.ToolCall{},
 			expectedTokens:   `some non json text{"name": "get_current_weather", "parameters": {"format":"fahrenheit","location":"San Francisco, CA"}}`,
 		},
 		{
-			name:             "llama3.2 - fake tool prefix",
+			name:             "llama3.2 tool call with invalid tool prefix (no prefix in template)",
 			model:            "llama3.2",
 			output:           `<tool_call>{"name": "get_current_weather", "parameters": {"format":"fahrenheit","location":"San Francisco, CA"}}`,
 			expectedToolCall: []api.ToolCall{},
@@ -288,7 +314,7 @@ func TestParseToolCalls(t *testing.T) {
 				m := &Model{Template: tmpl}
 				tp := NewToolParser(m)
 				got := []api.ToolCall{}
-				var actualTokens strings.Builder
+				var gotTokens strings.Builder
 
 				tokens := strings.Fields(tt.output)
 				for _, tok := range tokens {
@@ -302,17 +328,18 @@ func TestParseToolCalls(t *testing.T) {
 							got = append(got, toolCalls...)
 							add = false
 						case ToolCallSendTokens:
-							actualTokens.WriteString(s)
+							gotTokens.WriteString(s)
 							add = false
 						case ToolCallAccumulate:
 							add = false
 						case ToolCallSendPartial:
-							actualTokens.WriteString(" " + leftover)
+							t.Log("send partial", "leftover", leftover)
+							gotTokens.WriteString(" " + leftover)
 							add = false
 						}
 					}
 					if add {
-						actualTokens.WriteString(s)
+						gotTokens.WriteString(s)
 					}
 				}
 
@@ -322,7 +349,7 @@ func TestParseToolCalls(t *testing.T) {
 				}
 
 				// Compare tokens if we expect any
-				stripped := strings.TrimSpace(actualTokens.String())
+				stripped := strings.TrimSpace(gotTokens.String())
 				if diff := cmp.Diff(stripped, tt.expectedTokens); diff != "" {
 					t.Log("actualTokens", stripped, "expectedTokens", tt.expectedTokens)
 					t.Errorf("tokens mismatch (-got +want):\n%s", diff)