"tests/vscode:/vscode.git/clone" did not exist on "8dd19e3bac53dd34d68eb38e876879b37b2180a5"
types.go 27 KB
Newer Older
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1
2
package api

3
import (
Michael Yang's avatar
Michael Yang committed
4
	"encoding/json"
5
	"fmt"
6
	"log/slog"
Michael Yang's avatar
Michael Yang committed
7
	"math"
8
	"os"
9
	"reflect"
10
	"strconv"
11
	"strings"
12
	"time"
13
14

	"github.com/ollama/ollama/envconfig"
15
	"github.com/ollama/ollama/types/model"
16
)
Michael Yang's avatar
Michael Yang committed
17

18
// StatusError is an error with an HTTP status code and message.
Patrick Devine's avatar
Patrick Devine committed
19
type StatusError struct {
20
21
22
	StatusCode   int
	Status       string
	ErrorMessage string `json:"error"`
Patrick Devine's avatar
Patrick Devine committed
23
24
25
}

func (e StatusError) Error() string {
26
27
28
29
30
31
32
33
34
35
	switch {
	case e.Status != "" && e.ErrorMessage != "":
		return fmt.Sprintf("%s: %s", e.Status, e.ErrorMessage)
	case e.Status != "":
		return e.Status
	case e.ErrorMessage != "":
		return e.ErrorMessage
	default:
		// this should not happen
		return "something went wrong, please see the ollama server logs for details"
Patrick Devine's avatar
Patrick Devine committed
36
37
38
	}
}

39
// ImageData represents the raw binary data of an image file.
Patrick Devine's avatar
Patrick Devine committed
40
41
type ImageData []byte

42
43
44
// GenerateRequest describes a request sent by [Client.Generate]. While you
// have to specify the Model and Prompt fields, all the other fields have
// reasonable defaults for basic uses.
45
type GenerateRequest struct {
46
47
48
	// Model is the model name; it should be a name familiar to Ollama from
	// the library at https://ollama.com/library
	Model string `json:"model"`
49

50
51
52
	// Prompt is the textual prompt to send to the model.
	Prompt string `json:"prompt"`

53
54
55
	// Suffix is the text that comes after the inserted text.
	Suffix string `json:"suffix"`

56
57
58
59
60
61
62
	// System overrides the model's default system message/prompt.
	System string `json:"system"`

	// Template overrides the model's default prompt template.
	Template string `json:"template"`

	// Context is the context parameter returned from a previous call to
63
	// [Client.Generate]. It can be used to keep a short conversational memory.
64
65
66
67
68
69
70
71
72
	Context []int `json:"context,omitempty"`

	// Stream specifies whether the response is streaming; it is true by default.
	Stream *bool `json:"stream,omitempty"`

	// Raw set to true means that no formatting will be applied to the prompt.
	Raw bool `json:"raw,omitempty"`

	// Format specifies the format to return a response in.
73
	Format json.RawMessage `json:"format,omitempty"`
74
75
76
77
78

	// KeepAlive controls how long the model will stay loaded in memory following
	// this request.
	KeepAlive *Duration `json:"keep_alive,omitempty"`

79
	// Images is an optional list of raw image bytes accompanying this
80
81
82
83
84
	// request, for multimodal models.
	Images []ImageData `json:"images,omitempty"`

	// Options lists model-specific options. For example, temperature can be
	// set through this field, if the model supports it.
85
	Options map[string]any `json:"options"`
86
87

	// Think controls whether thinking/reasoning models will think before
Michael Yang's avatar
Michael Yang committed
88
89
	// responding. Can be a boolean (true/false) or a string ("high", "medium", "low")
	// for supported models. Needs to be a pointer so we can distinguish between false
90
91
	// (request that thinking _not_ be used) and unset (use the old behavior
	// before this option was introduced)
Michael Yang's avatar
Michael Yang committed
92
	Think *ThinkValue `json:"think,omitempty"`
93
94
}

95
// ChatRequest describes a request sent by [Client.Chat].
Bruce MacDonald's avatar
Bruce MacDonald committed
96
type ChatRequest struct {
97
98
99
100
101
102
	// Model is the model name, as in [GenerateRequest].
	Model string `json:"model"`

	// Messages is the messages of the chat - can be used to keep a chat memory.
	Messages []Message `json:"messages"`

103
	// Stream enables streaming of returned responses; true by default.
104
105
106
	Stream *bool `json:"stream,omitempty"`

	// Format is the format to return the response in (e.g. "json").
107
	Format json.RawMessage `json:"format,omitempty"`
108
109

	// KeepAlive controls how long the model will stay loaded into memory
110
	// following the request.
111
	KeepAlive *Duration `json:"keep_alive,omitempty"`
Bruce MacDonald's avatar
Bruce MacDonald committed
112

Michael Yang's avatar
tools  
Michael Yang committed
113
	// Tools is an optional list of tools the model has access to.
114
	Tools `json:"tools,omitempty"`
Michael Yang's avatar
tools  
Michael Yang committed
115

116
	// Options lists model-specific options.
117
	Options map[string]any `json:"options"`
118
119

	// Think controls whether thinking/reasoning models will think before
Michael Yang's avatar
Michael Yang committed
120
121
122
	// responding. Can be a boolean (true/false) or a string ("high", "medium", "low")
	// for supported models.
	Think *ThinkValue `json:"think,omitempty"`
Bruce MacDonald's avatar
Bruce MacDonald committed
123
124
}

125
126
127
128
129
130
131
type Tools []Tool

func (t Tools) String() string {
	bts, _ := json.Marshal(t)
	return string(bts)
}

132
133
134
135
136
func (t Tool) String() string {
	bts, _ := json.Marshal(t)
	return string(bts)
}

137
138
139
// Message is a single message in a chat sequence. The message contains the
// role ("system", "user", or "assistant"), the content and an optional list
// of images.
Bruce MacDonald's avatar
Bruce MacDonald committed
140
type Message struct {
141
142
143
144
145
	Role    string `json:"role"`
	Content string `json:"content"`
	// Thinking contains the text that was inside thinking tags in the
	// original model output when ChatRequest.Think is enabled.
	Thinking  string      `json:"thinking,omitempty"`
Michael Yang's avatar
tools  
Michael Yang committed
146
147
	Images    []ImageData `json:"images,omitempty"`
	ToolCalls []ToolCall  `json:"tool_calls,omitempty"`
148
	ToolName  string      `json:"tool_name,omitempty"`
Michael Yang's avatar
tools  
Michael Yang committed
149
150
}

151
152
153
154
155
156
157
158
159
160
161
162
func (m *Message) UnmarshalJSON(b []byte) error {
	type Alias Message
	var a Alias
	if err := json.Unmarshal(b, &a); err != nil {
		return err
	}

	*m = Message(a)
	m.Role = strings.ToLower(m.Role)
	return nil
}

163
164
165
166
167
type ToolCall struct {
	Function ToolCallFunction `json:"function"`
}

type ToolCallFunction struct {
168
	Index     int                       `json:"index,omitempty"`
169
170
171
172
173
174
175
176
177
178
179
180
181
	Name      string                    `json:"name"`
	Arguments ToolCallFunctionArguments `json:"arguments"`
}

type ToolCallFunctionArguments map[string]any

func (t *ToolCallFunctionArguments) String() string {
	bts, _ := json.Marshal(t)
	return string(bts)
}

type Tool struct {
	Type     string       `json:"type"`
182
	Items    any          `json:"items,omitempty"`
183
184
185
	Function ToolFunction `json:"function"`
}

186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
// PropertyType can be either a string or an array of strings
type PropertyType []string

// UnmarshalJSON implements the json.Unmarshaler interface
func (pt *PropertyType) UnmarshalJSON(data []byte) error {
	// Try to unmarshal as a string first
	var s string
	if err := json.Unmarshal(data, &s); err == nil {
		*pt = []string{s}
		return nil
	}

	// If that fails, try to unmarshal as an array of strings
	var a []string
	if err := json.Unmarshal(data, &a); err != nil {
		return err
	}
	*pt = a
	return nil
}

// MarshalJSON implements the json.Marshaler interface
func (pt PropertyType) MarshalJSON() ([]byte, error) {
	if len(pt) == 1 {
		// If there's only one type, marshal as a string
		return json.Marshal(pt[0])
	}
	// Otherwise marshal as an array
	return json.Marshal([]string(pt))
}

// String returns a string representation of the PropertyType
func (pt PropertyType) String() string {
	if len(pt) == 0 {
		return ""
	}
	if len(pt) == 1 {
		return pt[0]
	}
	return fmt.Sprintf("%v", []string(pt))
}

Devon Rifkin's avatar
Devon Rifkin committed
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
type ToolProperty struct {
	AnyOf       []ToolProperty `json:"anyOf,omitempty"`
	Type        PropertyType   `json:"type"`
	Items       any            `json:"items,omitempty"`
	Description string         `json:"description"`
	Enum        []any          `json:"enum,omitempty"`
}

// ToTypeScriptType converts a ToolProperty to a TypeScript type string
func (tp ToolProperty) ToTypeScriptType() string {
	if len(tp.AnyOf) > 0 {
		var types []string
		for _, anyOf := range tp.AnyOf {
			types = append(types, anyOf.ToTypeScriptType())
		}
		return strings.Join(types, " | ")
	}

	if len(tp.Type) == 0 {
		return "any"
	}

	if len(tp.Type) == 1 {
		return mapToTypeScriptType(tp.Type[0])
	}

	var types []string
	for _, t := range tp.Type {
		types = append(types, mapToTypeScriptType(t))
	}
	return strings.Join(types, " | ")
}

// mapToTypeScriptType maps JSON Schema types to TypeScript types
func mapToTypeScriptType(jsonType string) string {
	switch jsonType {
	case "string":
		return "string"
	case "number", "integer":
		return "number"
	case "boolean":
		return "boolean"
	case "array":
		return "any[]"
	case "object":
		return "Record<string, any>"
	case "null":
		return "null"
	default:
		return "any"
	}
}

281
282
283
284
type ToolFunction struct {
	Name        string `json:"name"`
	Description string `json:"description"`
	Parameters  struct {
Devon Rifkin's avatar
Devon Rifkin committed
285
286
287
288
289
		Type       string                  `json:"type"`
		Defs       any                     `json:"$defs,omitempty"`
		Items      any                     `json:"items,omitempty"`
		Required   []string                `json:"required"`
		Properties map[string]ToolProperty `json:"properties"`
290
291
292
293
294
295
296
297
	} `json:"parameters"`
}

func (t *ToolFunction) String() string {
	bts, _ := json.Marshal(t)
	return string(bts)
}

298
299
// ChatResponse is the response returned by [Client.Chat]. Its fields are
// similar to [GenerateResponse].
Bruce MacDonald's avatar
Bruce MacDonald committed
300
type ChatResponse struct {
301
302
303
	Model      string    `json:"model"`
	CreatedAt  time.Time `json:"created_at"`
	Message    Message   `json:"message"`
Bruce MacDonald's avatar
Bruce MacDonald committed
304
	DoneReason string    `json:"done_reason,omitempty"`
Bruce MacDonald's avatar
Bruce MacDonald committed
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319

	Done bool `json:"done"`

	Metrics
}

type Metrics struct {
	TotalDuration      time.Duration `json:"total_duration,omitempty"`
	LoadDuration       time.Duration `json:"load_duration,omitempty"`
	PromptEvalCount    int           `json:"prompt_eval_count,omitempty"`
	PromptEvalDuration time.Duration `json:"prompt_eval_duration,omitempty"`
	EvalCount          int           `json:"eval_count,omitempty"`
	EvalDuration       time.Duration `json:"eval_duration,omitempty"`
}

320
321
// Options specified in [GenerateRequest].  If you add a new option here, also
// add it to the API docs.
322
323
324
325
326
327
328
329
330
type Options struct {
	Runner

	// Predict options used at runtime
	NumKeep          int      `json:"num_keep,omitempty"`
	Seed             int      `json:"seed,omitempty"`
	NumPredict       int      `json:"num_predict,omitempty"`
	TopK             int      `json:"top_k,omitempty"`
	TopP             float32  `json:"top_p,omitempty"`
331
	MinP             float32  `json:"min_p,omitempty"`
332
333
334
335
336
337
338
339
340
341
342
	TypicalP         float32  `json:"typical_p,omitempty"`
	RepeatLastN      int      `json:"repeat_last_n,omitempty"`
	Temperature      float32  `json:"temperature,omitempty"`
	RepeatPenalty    float32  `json:"repeat_penalty,omitempty"`
	PresencePenalty  float32  `json:"presence_penalty,omitempty"`
	FrequencyPenalty float32  `json:"frequency_penalty,omitempty"`
	Stop             []string `json:"stop,omitempty"`
}

// Runner options which must be set when the model is loaded into memory
type Runner struct {
343
344
345
346
347
348
	NumCtx    int   `json:"num_ctx,omitempty"`
	NumBatch  int   `json:"num_batch,omitempty"`
	NumGPU    int   `json:"num_gpu,omitempty"`
	MainGPU   int   `json:"main_gpu,omitempty"`
	UseMMap   *bool `json:"use_mmap,omitempty"`
	NumThread int   `json:"num_thread,omitempty"`
349
350
}

351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
// EmbedRequest is the request passed to [Client.Embed].
type EmbedRequest struct {
	// Model is the model name.
	Model string `json:"model"`

	// Input is the input to embed.
	Input any `json:"input"`

	// KeepAlive controls how long the model will stay loaded in memory following
	// this request.
	KeepAlive *Duration `json:"keep_alive,omitempty"`

	Truncate *bool `json:"truncate,omitempty"`

	// Options lists model-specific options.
366
	Options map[string]any `json:"options"`
367
368
369
370
371
}

// EmbedResponse is the response from [Client.Embed].
type EmbedResponse struct {
	Model      string      `json:"model"`
372
	Embeddings [][]float32 `json:"embeddings"`
373
374
375
376

	TotalDuration   time.Duration `json:"total_duration,omitempty"`
	LoadDuration    time.Duration `json:"load_duration,omitempty"`
	PromptEvalCount int           `json:"prompt_eval_count,omitempty"`
377
378
}

379
// EmbeddingRequest is the request passed to [Client.Embeddings].
Bruce MacDonald's avatar
Bruce MacDonald committed
380
type EmbeddingRequest struct {
381
382
383
384
385
386
387
388
	// Model is the model name.
	Model string `json:"model"`

	// Prompt is the textual prompt to embed.
	Prompt string `json:"prompt"`

	// KeepAlive controls how long the model will stay loaded in memory following
	// this request.
389
	KeepAlive *Duration `json:"keep_alive,omitempty"`
Bruce MacDonald's avatar
Bruce MacDonald committed
390

391
	// Options lists model-specific options.
392
	Options map[string]any `json:"options"`
Bruce MacDonald's avatar
Bruce MacDonald committed
393
394
}

395
// EmbeddingResponse is the response from [Client.Embeddings].
Bruce MacDonald's avatar
Bruce MacDonald committed
396
397
398
399
type EmbeddingResponse struct {
	Embedding []float64 `json:"embedding"`
}

400
// CreateRequest is the request passed to [Client.Create].
401
type CreateRequest struct {
402
403
404
405
406
407
408
409
410
411
412
413
414
	Model    string `json:"model"`
	Stream   *bool  `json:"stream,omitempty"`
	Quantize string `json:"quantize,omitempty"`

	From       string            `json:"from,omitempty"`
	Files      map[string]string `json:"files,omitempty"`
	Adapters   map[string]string `json:"adapters,omitempty"`
	Template   string            `json:"template,omitempty"`
	License    any               `json:"license,omitempty"`
	System     string            `json:"system,omitempty"`
	Parameters map[string]any    `json:"parameters,omitempty"`
	Messages   []Message         `json:"messages,omitempty"`

Michael Yang's avatar
Michael Yang committed
415
	// Deprecated: set the model name with Model instead
Michael Yang's avatar
Michael Yang committed
416
	Name string `json:"name"`
Michael Yang's avatar
Michael Yang committed
417
	// Deprecated: use Quantize instead
418
	Quantization string `json:"quantization,omitempty"`
419
420
}

421
// DeleteRequest is the request passed to [Client.Delete].
422
type DeleteRequest struct {
Michael Yang's avatar
Michael Yang committed
423
424
	Model string `json:"model"`

Michael Yang's avatar
Michael Yang committed
425
	// Deprecated: set the model name with Model instead
426
427
428
	Name string `json:"name"`
}

429
// ShowRequest is the request passed to [Client.Show].
Patrick Devine's avatar
Patrick Devine committed
430
type ShowRequest struct {
431
432
	Model  string `json:"model"`
	System string `json:"system"`
Patrick Devine's avatar
Patrick Devine committed
433
434

	// Template is deprecated
435
	Template string `json:"template"`
436
	Verbose  bool   `json:"verbose"`
437

438
	Options map[string]any `json:"options"`
Michael Yang's avatar
Michael Yang committed
439

Michael Yang's avatar
Michael Yang committed
440
	// Deprecated: set the model name with Model instead
Michael Yang's avatar
Michael Yang committed
441
	Name string `json:"name"`
Patrick Devine's avatar
Patrick Devine committed
442
443
}

444
// ShowResponse is the response returned from [Client.Show].
Patrick Devine's avatar
Patrick Devine committed
445
type ShowResponse struct {
446
447
448
449
450
451
452
453
454
455
456
457
	License       string             `json:"license,omitempty"`
	Modelfile     string             `json:"modelfile,omitempty"`
	Parameters    string             `json:"parameters,omitempty"`
	Template      string             `json:"template,omitempty"`
	System        string             `json:"system,omitempty"`
	Details       ModelDetails       `json:"details,omitempty"`
	Messages      []Message          `json:"messages,omitempty"`
	ModelInfo     map[string]any     `json:"model_info,omitempty"`
	ProjectorInfo map[string]any     `json:"projector_info,omitempty"`
	Tensors       []Tensor           `json:"tensors,omitempty"`
	Capabilities  []model.Capability `json:"capabilities,omitempty"`
	ModifiedAt    time.Time          `json:"modified_at,omitempty"`
Patrick Devine's avatar
Patrick Devine committed
458
459
}

460
// CopyRequest is the request passed to [Client.Copy].
Patrick Devine's avatar
Patrick Devine committed
461
462
463
464
465
type CopyRequest struct {
	Source      string `json:"source"`
	Destination string `json:"destination"`
}

466
// PullRequest is the request passed to [Client.Pull].
Bruce MacDonald's avatar
Bruce MacDonald committed
467
type PullRequest struct {
Michael Yang's avatar
Michael Yang committed
468
	Model    string `json:"model"`
469
470
471
	Insecure bool   `json:"insecure,omitempty"` // Deprecated: ignored
	Username string `json:"username"`           // Deprecated: ignored
	Password string `json:"password"`           // Deprecated: ignored
472
	Stream   *bool  `json:"stream,omitempty"`
Michael Yang's avatar
Michael Yang committed
473

Michael Yang's avatar
Michael Yang committed
474
	// Deprecated: set the model name with Model instead
Michael Yang's avatar
Michael Yang committed
475
	Name string `json:"name"`
Bruce MacDonald's avatar
Bruce MacDonald committed
476
477
}

478
479
// ProgressResponse is the response passed to progress functions like
// [PullProgressFunc] and [PushProgressFunc].
480
type ProgressResponse struct {
481
482
	Status    string `json:"status"`
	Digest    string `json:"digest,omitempty"`
Michael Yang's avatar
Michael Yang committed
483
484
	Total     int64  `json:"total,omitempty"`
	Completed int64  `json:"completed,omitempty"`
Bruce MacDonald's avatar
Bruce MacDonald committed
485
486
}

487
// PushRequest is the request passed to [Client.Push].
488
type PushRequest struct {
Michael Yang's avatar
Michael Yang committed
489
	Model    string `json:"model"`
490
	Insecure bool   `json:"insecure,omitempty"`
491
492
493
	Username string `json:"username"`
	Password string `json:"password"`
	Stream   *bool  `json:"stream,omitempty"`
Michael Yang's avatar
Michael Yang committed
494

Michael Yang's avatar
Michael Yang committed
495
	// Deprecated: set the model name with Model instead
Michael Yang's avatar
Michael Yang committed
496
	Name string `json:"name"`
497
}
498

499
// ListResponse is the response from [Client.List].
Patrick Devine's avatar
Patrick Devine committed
500
type ListResponse struct {
501
	Models []ListModelResponse `json:"models"`
Patrick Devine's avatar
Patrick Devine committed
502
503
}

504
505
506
507
508
509
510
// ProcessResponse is the response from [Client.Process].
type ProcessResponse struct {
	Models []ProcessModelResponse `json:"models"`
}

// ListModelResponse is a single model description in [ListResponse].
type ListModelResponse struct {
511
512
513
514
515
516
	Name       string       `json:"name"`
	Model      string       `json:"model"`
	ModifiedAt time.Time    `json:"modified_at"`
	Size       int64        `json:"size"`
	Digest     string       `json:"digest"`
	Details    ModelDetails `json:"details,omitempty"`
517
518
519
520
}

// ProcessModelResponse is a single model description in [ProcessResponse].
type ProcessModelResponse struct {
521
522
523
524
525
526
527
528
	Name          string       `json:"name"`
	Model         string       `json:"model"`
	Size          int64        `json:"size"`
	Digest        string       `json:"digest"`
	Details       ModelDetails `json:"details,omitempty"`
	ExpiresAt     time.Time    `json:"expires_at"`
	SizeVRAM      int64        `json:"size_vram"`
	ContextLength int          `json:"context_length"`
Patrick Devine's avatar
Patrick Devine committed
529
530
}

Patrick Devine's avatar
Patrick Devine committed
531
532
533
534
type TokenResponse struct {
	Token string `json:"token"`
}

535
// GenerateResponse is the response passed into [GenerateResponseFunc].
Michael Yang's avatar
Michael Yang committed
536
type GenerateResponse struct {
537
538
539
	// Model is the model name that generated the response.
	Model string `json:"model"`

Michael Yang's avatar
Michael Yang committed
540
	// CreatedAt is the timestamp of the response.
541
542
	CreatedAt time.Time `json:"created_at"`

543
544
545
	// Response is the textual response itself.
	Response string `json:"response"`

546
547
548
549
	// Thinking contains the text that was inside thinking tags in the
	// original model output when ChatRequest.Think is enabled.
	Thinking string `json:"thinking,omitempty"`

550
551
552
	// Done specifies if the response is complete.
	Done bool `json:"done"`

553
	// DoneReason is the reason the model stopped generating text.
Bruce MacDonald's avatar
Bruce MacDonald committed
554
	DoneReason string `json:"done_reason,omitempty"`
555

556
557
	// Context is an encoding of the conversation used in this response; this
	// can be sent in the next request to keep a conversational memory.
Michael Yang's avatar
Michael Yang committed
558
	Context []int `json:"context,omitempty"`
559

Bruce MacDonald's avatar
Bruce MacDonald committed
560
	Metrics
Michael Yang's avatar
Michael Yang committed
561
562

	ToolCalls []ToolCall `json:"tool_calls,omitempty"`
563
564
}

565
// ModelDetails provides details about a model.
Patrick Devine's avatar
Patrick Devine committed
566
type ModelDetails struct {
567
	ParentModel       string   `json:"parent_model"`
Patrick Devine's avatar
Patrick Devine committed
568
569
570
571
572
573
574
	Format            string   `json:"format"`
	Family            string   `json:"family"`
	Families          []string `json:"families"`
	ParameterSize     string   `json:"parameter_size"`
	QuantizationLevel string   `json:"quantization_level"`
}

575
576
577
578
579
580
581
// Tensor describes the metadata for a given tensor.
type Tensor struct {
	Name  string   `json:"name"`
	Type  string   `json:"type"`
	Shape []uint64 `json:"shape"`
}

Bruce MacDonald's avatar
Bruce MacDonald committed
582
583
584
func (m *Metrics) Summary() {
	if m.TotalDuration > 0 {
		fmt.Fprintf(os.Stderr, "total duration:       %v\n", m.TotalDuration)
585
586
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
587
588
	if m.LoadDuration > 0 {
		fmt.Fprintf(os.Stderr, "load duration:        %v\n", m.LoadDuration)
Michael Yang's avatar
Michael Yang committed
589
590
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
591
592
	if m.PromptEvalCount > 0 {
		fmt.Fprintf(os.Stderr, "prompt eval count:    %d token(s)\n", m.PromptEvalCount)
593
594
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
595
596
597
	if m.PromptEvalDuration > 0 {
		fmt.Fprintf(os.Stderr, "prompt eval duration: %s\n", m.PromptEvalDuration)
		fmt.Fprintf(os.Stderr, "prompt eval rate:     %.2f tokens/s\n", float64(m.PromptEvalCount)/m.PromptEvalDuration.Seconds())
598
599
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
600
601
	if m.EvalCount > 0 {
		fmt.Fprintf(os.Stderr, "eval count:           %d token(s)\n", m.EvalCount)
602
603
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
604
605
606
	if m.EvalDuration > 0 {
		fmt.Fprintf(os.Stderr, "eval duration:        %s\n", m.EvalDuration)
		fmt.Fprintf(os.Stderr, "eval rate:            %.2f tokens/s\n", float64(m.EvalCount)/m.EvalDuration.Seconds())
607
	}
608
609
}

610
func (opts *Options) FromMap(m map[string]any) error {
611
612
613
614
615
616
617
618
619
620
621
622
623
	valueOpts := reflect.ValueOf(opts).Elem() // names of the fields in the options struct
	typeOpts := reflect.TypeOf(opts).Elem()   // types of the fields in the options struct

	// build map of json struct tags to their types
	jsonOpts := make(map[string]reflect.StructField)
	for _, field := range reflect.VisibleFields(typeOpts) {
		jsonTag := strings.Split(field.Tag.Get("json"), ",")[0]
		if jsonTag != "" {
			jsonOpts[jsonTag] = field
		}
	}

	for key, val := range m {
624
625
		opt, ok := jsonOpts[key]
		if !ok {
626
			slog.Warn("invalid option provided", "option", key)
627
628
			continue
		}
Michael Yang's avatar
Michael Yang committed
629

630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
		field := valueOpts.FieldByName(opt.Name)
		if field.IsValid() && field.CanSet() {
			if val == nil {
				continue
			}

			switch field.Kind() {
			case reflect.Int:
				switch t := val.(type) {
				case int64:
					field.SetInt(t)
				case float64:
					// when JSON unmarshals numbers, it uses float64, not int
					field.SetInt(int64(t))
				default:
					return fmt.Errorf("option %q must be of type integer", key)
				}
			case reflect.Bool:
				val, ok := val.(bool)
				if !ok {
					return fmt.Errorf("option %q must be of type boolean", key)
				}
				field.SetBool(val)
			case reflect.Float32:
				// JSON unmarshals to float64
				val, ok := val.(float64)
				if !ok {
					return fmt.Errorf("option %q must be of type float32", key)
				}
				field.SetFloat(val)
			case reflect.String:
				val, ok := val.(string)
				if !ok {
					return fmt.Errorf("option %q must be of type string", key)
				}
				field.SetString(val)
			case reflect.Slice:
667
668
				// JSON unmarshals to []any, not []string
				val, ok := val.([]any)
669
670
671
				if !ok {
					return fmt.Errorf("option %q must be of type array", key)
				}
672
				// convert []any to []string
673
674
675
				slice := make([]string, len(val))
				for i, item := range val {
					str, ok := item.(string)
676
					if !ok {
677
						return fmt.Errorf("option %q must be of an array of strings", key)
678
					}
679
					slice[i] = str
680
				}
681
				field.Set(reflect.ValueOf(slice))
682
683
684
685
686
687
688
689
690
691
692
			case reflect.Pointer:
				var b bool
				if field.Type() == reflect.TypeOf(&b) {
					val, ok := val.(bool)
					if !ok {
						return fmt.Errorf("option %q must be of type boolean", key)
					}
					field.Set(reflect.ValueOf(&val))
				} else {
					return fmt.Errorf("unknown type loading config params: %v %v", field.Kind(), field.Type())
				}
693
694
			default:
				return fmt.Errorf("unknown type loading config params: %v", field.Kind())
695
696
697
			}
		}
	}
698

699
700
701
	return nil
}

702
703
// DefaultOptions is the default set of options for [GenerateRequest]; these
// values are used unless the user specifies other values explicitly.
Michael Yang's avatar
Michael Yang committed
704
705
func DefaultOptions() Options {
	return Options{
706
		// options set on request to runner
707
708
709
710
		NumPredict: -1,

		// set a minimal num_keep to avoid issues on context shifts
		NumKeep:          4,
Michael Yang's avatar
Michael Yang committed
711
712
713
714
		Temperature:      0.8,
		TopK:             40,
		TopP:             0.9,
		TypicalP:         1.0,
715
716
717
718
719
		RepeatLastN:      64,
		RepeatPenalty:    1.1,
		PresencePenalty:  0.0,
		FrequencyPenalty: 0.0,
		Seed:             -1,
Michael Yang's avatar
Michael Yang committed
720

721
722
		Runner: Runner{
			// options set when the model is loaded
723
			NumCtx:    int(envconfig.ContextLength()),
Michael Yang's avatar
Michael Yang committed
724
725
			NumBatch:  512,
			NumGPU:    -1, // -1 here indicates that NumGPU should be set dynamically
726
			NumThread: 0,  // let the runtime decide
727
			UseMMap:   nil,
728
		},
Michael Yang's avatar
Michael Yang committed
729
	}
Jeffrey Morgan's avatar
Jeffrey Morgan committed
730
}
Michael Yang's avatar
Michael Yang committed
731

Michael Yang's avatar
Michael Yang committed
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
// ThinkValue represents a value that can be a boolean or a string ("high", "medium", "low")
type ThinkValue struct {
	// Value can be a bool or string
	Value interface{}
}

// IsValid checks if the ThinkValue is valid
func (t *ThinkValue) IsValid() bool {
	if t == nil || t.Value == nil {
		return true // nil is valid (means not set)
	}

	switch v := t.Value.(type) {
	case bool:
		return true
	case string:
		return v == "high" || v == "medium" || v == "low"
	default:
		return false
	}
}

// IsBool returns true if the value is a boolean
func (t *ThinkValue) IsBool() bool {
	if t == nil || t.Value == nil {
		return false
	}
	_, ok := t.Value.(bool)
	return ok
}

// IsString returns true if the value is a string
func (t *ThinkValue) IsString() bool {
	if t == nil || t.Value == nil {
		return false
	}
	_, ok := t.Value.(string)
	return ok
}

772
773
// Bool returns the value as a bool (true if enabled in any way)
func (t *ThinkValue) Bool() bool {
Michael Yang's avatar
Michael Yang committed
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
	if t == nil || t.Value == nil {
		return false
	}

	switch v := t.Value.(type) {
	case bool:
		return v
	case string:
		// Any string value ("high", "medium", "low") means thinking is enabled
		return v == "high" || v == "medium" || v == "low"
	default:
		return false
	}
}

789
790
// String returns the value as a string
func (t *ThinkValue) String() string {
Michael Yang's avatar
Michael Yang committed
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
	if t == nil || t.Value == nil {
		return ""
	}

	switch v := t.Value.(type) {
	case string:
		return v
	case bool:
		if v {
			return "medium" // Default level when just true
		}
		return ""
	default:
		return ""
	}
}

// UnmarshalJSON implements json.Unmarshaler
func (t *ThinkValue) UnmarshalJSON(data []byte) error {
	// Try to unmarshal as bool first
	var b bool
	if err := json.Unmarshal(data, &b); err == nil {
		t.Value = b
		return nil
	}

	// Try to unmarshal as string
	var s string
	if err := json.Unmarshal(data, &s); err == nil {
		// Validate string values
		if s != "high" && s != "medium" && s != "low" {
			return fmt.Errorf("invalid think value: %q (must be \"high\", \"medium\", \"low\", true, or false)", s)
		}
		t.Value = s
		return nil
	}

	return fmt.Errorf("think must be a boolean or string (\"high\", \"medium\", \"low\")")
}

// MarshalJSON implements json.Marshaler
func (t *ThinkValue) MarshalJSON() ([]byte, error) {
	if t == nil || t.Value == nil {
		return []byte("null"), nil
	}
	return json.Marshal(t.Value)
}

Michael Yang's avatar
Michael Yang committed
839
840
841
842
type Duration struct {
	time.Duration
}

843
844
845
846
847
848
849
func (d Duration) MarshalJSON() ([]byte, error) {
	if d.Duration < 0 {
		return []byte("-1"), nil
	}
	return []byte("\"" + d.Duration.String() + "\""), nil
}

Michael Yang's avatar
Michael Yang committed
850
851
852
853
854
855
856
857
858
859
860
func (d *Duration) UnmarshalJSON(b []byte) (err error) {
	var v any
	if err := json.Unmarshal(b, &v); err != nil {
		return err
	}

	d.Duration = 5 * time.Minute

	switch t := v.(type) {
	case float64:
		if t < 0 {
bnorick's avatar
bnorick committed
861
			d.Duration = time.Duration(math.MaxInt64)
862
		} else {
863
			d.Duration = time.Duration(int(t) * int(time.Second))
Michael Yang's avatar
Michael Yang committed
864
865
866
867
868
869
		}
	case string:
		d.Duration, err = time.ParseDuration(t)
		if err != nil {
			return err
		}
870
		if d.Duration < 0 {
bnorick's avatar
bnorick committed
871
			d.Duration = time.Duration(math.MaxInt64)
872
		}
873
874
	default:
		return fmt.Errorf("Unsupported type: '%s'", reflect.TypeOf(v))
Michael Yang's avatar
Michael Yang committed
875
876
877
878
	}

	return nil
}
879
880

// FormatParams converts specified parameter options to their correct types
881
func FormatParams(params map[string][]string) (map[string]any, error) {
882
883
884
885
886
887
888
889
890
891
892
893
894
	opts := Options{}
	valueOpts := reflect.ValueOf(&opts).Elem() // names of the fields in the options struct
	typeOpts := reflect.TypeOf(opts)           // types of the fields in the options struct

	// build map of json struct tags to their types
	jsonOpts := make(map[string]reflect.StructField)
	for _, field := range reflect.VisibleFields(typeOpts) {
		jsonTag := strings.Split(field.Tag.Get("json"), ",")[0]
		if jsonTag != "" {
			jsonOpts[jsonTag] = field
		}
	}

895
	out := make(map[string]any)
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
	// iterate params and set values based on json struct tags
	for key, vals := range params {
		if opt, ok := jsonOpts[key]; !ok {
			return nil, fmt.Errorf("unknown parameter '%s'", key)
		} else {
			field := valueOpts.FieldByName(opt.Name)
			if field.IsValid() && field.CanSet() {
				switch field.Kind() {
				case reflect.Float32:
					floatVal, err := strconv.ParseFloat(vals[0], 32)
					if err != nil {
						return nil, fmt.Errorf("invalid float value %s", vals)
					}

					out[key] = float32(floatVal)
				case reflect.Int:
					intVal, err := strconv.ParseInt(vals[0], 10, 64)
					if err != nil {
						return nil, fmt.Errorf("invalid int value %s", vals)
					}

					out[key] = intVal
				case reflect.Bool:
					boolVal, err := strconv.ParseBool(vals[0])
					if err != nil {
						return nil, fmt.Errorf("invalid bool value %s", vals)
					}

					out[key] = boolVal
				case reflect.String:
					out[key] = vals[0]
				case reflect.Slice:
					// TODO: only string slices are supported right now
					out[key] = vals
930
931
932
933
934
935
936
937
938
939
940
				case reflect.Pointer:
					var b bool
					if field.Type() == reflect.TypeOf(&b) {
						boolVal, err := strconv.ParseBool(vals[0])
						if err != nil {
							return nil, fmt.Errorf("invalid bool value %s", vals)
						}
						out[key] = &boolVal
					} else {
						return nil, fmt.Errorf("unknown type %s for %s", field.Kind(), key)
					}
941
942
943
944
945
946
947
948
949
				default:
					return nil, fmt.Errorf("unknown type %s for %s", field.Kind(), key)
				}
			}
		}
	}

	return out, nil
}