toolapp

package
v1.21.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Mar 4, 2026 License: Apache-2.0 Imports: 19 Imported by: 0

Documentation

Overview

Package toolapp provides endpoints to handle tool management.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func Routes 

func Routes(app *web.App, cfg Config)

Routes adds specific routes for this group.

Types

type CatalogCapabilities added in v1.3.1 

type CatalogCapabilities struct {
	Endpoint  string `json:"endpoint"`
	Images    bool   `json:"images"`
	Audio     bool   `json:"audio"`
	Video     bool   `json:"video"`
	Streaming bool   `json:"streaming"`
	Reasoning bool   `json:"reasoning"`
	Tooling   bool   `json:"tooling"`
	Embedding bool   `json:"embedding"`
	Rerank    bool   `json:"rerank"`
}

CatalogCapabilities represents the capabilities of a model.

type CatalogFile added in v1.3.1 

type CatalogFile struct {
	URL  string `json:"url"`
	Size string `json:"size"`
}

CatalogFile represents the actual file url and size.

type CatalogFileInfoResponse added in v1.18.0 

type CatalogFileInfoResponse struct {
	Name       string `json:"name"`
	ModelCount int    `json:"model_count"`
}

CatalogFileInfoResponse represents a catalog file.

type CatalogFiles added in v1.3.1 

type CatalogFiles struct {
	Models []CatalogFile `json:"model"`
	Proj   CatalogFile   `json:"proj"`
}

CatalogFiles represents file information for a model.

type CatalogFilesListResponse added in v1.18.0 

type CatalogFilesListResponse []CatalogFileInfoResponse

CatalogFilesListResponse is a list of catalog files.

func (CatalogFilesListResponse) Encode added in v1.18.0 

func (app CatalogFilesListResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type CatalogMetadata added in v1.3.1 

type CatalogMetadata struct {
	Created     time.Time `json:"created"`
	Collections string    `json:"collections"`
	Description string    `json:"description"`
}

CatalogMetadata represents extra information about the model.

type CatalogModelResponse added in v1.3.1 

type CatalogModelResponse struct {
	ID             string              `json:"id"`
	Category       string              `json:"category"`
	OwnedBy        string              `json:"owned_by"`
	ModelFamily    string              `json:"model_family"`
	Architecture   string              `json:"architecture"`
	GGUFArch       string              `json:"gguf_arch"`
	Parameters     string              `json:"parameters"`
	ParameterCount int64               `json:"parameter_count"`
	WebPage        string              `json:"web_page"`
	GatedModel     bool                `json:"gated_model"`
	Template       string              `json:"template"`
	TotalSize      string              `json:"total_size"`
	TotalSizeBytes int64               `json:"total_size_bytes"`
	Files          CatalogFiles        `json:"files"`
	Capabilities   CatalogCapabilities `json:"capabilities"`
	Metadata       CatalogMetadata     `json:"metadata"`
	ModelConfig    *ModelConfig        `json:"model_config,omitempty"`
	BaseConfig     *ModelConfig        `json:"base_config,omitempty"`
	ModelMetadata  map[string]string   `json:"model_metadata,omitempty"`
	VRAM           *VRAM               `json:"vram,omitempty"`
	Downloaded     bool                `json:"downloaded"`
	Validated      bool                `json:"validated"`
	CatalogFile    string              `json:"catalog_file,omitempty"`
}

CatalogModelResponse represents information for a model.

func (CatalogModelResponse) Encode added in v1.3.1 

func (app CatalogModelResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type CatalogModelsResponse added in v1.3.1 

type CatalogModelsResponse []CatalogModelResponse

CatalogModelsResponse represents a list of catalog models.

func (CatalogModelsResponse) Encode added in v1.3.1 

func (app CatalogModelsResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type Config 

type Config struct {
	Log        *logger.Logger
	AuthClient *authclient.Client
	Cache      *cache.Cache
	Libs       *libs.Libs
	Models     *models.Models
	Catalog    *catalog.Catalog
}

Config contains all the mandatory systems required by handlers.

type GrammarContentResponse added in v1.18.9 

type GrammarContentResponse struct {
	Content string `json:"content"`
}

GrammarContentResponse returns the content of a grammar file.

func (GrammarContentResponse) Encode added in v1.18.9 

func (app GrammarContentResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type GrammarFilesResponse added in v1.18.9 

type GrammarFilesResponse struct {
	Files []string `json:"files"`
}

GrammarFilesResponse is a list of available grammar filenames.

func (GrammarFilesResponse) Encode added in v1.18.9 

func (app GrammarFilesResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type HFLookupRequest added in v1.18.0 

type HFLookupRequest struct {
	Input string `json:"input"`
}

HFLookupRequest represents the input for a HuggingFace model lookup.

func (*HFLookupRequest) Decode added in v1.18.0 

func (app *HFLookupRequest) Decode(data []byte) error

Decode implements the decoder interface.

type HFLookupResponse added in v1.18.0 

type HFLookupResponse struct {
	Model     CatalogModelResponse `json:"model"`
	RepoFiles []HFRepoFile         `json:"repo_files"`
}

HFLookupResponse contains the results from a HuggingFace lookup.

func (HFLookupResponse) Encode added in v1.18.0 

func (app HFLookupResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type HFRepoFile added in v1.18.0 

type HFRepoFile struct {
	Filename string `json:"filename"`
	Size     int64  `json:"size"`
	SizeStr  string `json:"size_str"`
}

HFRepoFile represents a GGUF file in a HuggingFace repository.

type KeyResponse added in v1.5.0 

type KeyResponse struct {
	ID      string `json:"id"`
	Created string `json:"created"`
}

KeyResponse represents a key in the system.

type KeysResponse added in v1.5.0 

type KeysResponse []KeyResponse

KeysResponse is a collection of keys.

func (KeysResponse) Encode added in v1.5.0 

func (app KeysResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type ListModelDetail 

type ListModelDetail struct {
	ID          string          `json:"id"`
	Object      string          `json:"object"`
	Created     int64           `json:"created"`
	OwnedBy     string          `json:"owned_by"`
	ModelFamily string          `json:"model_family"`
	Size        int64           `json:"size"`
	Modified    time.Time       `json:"modified"`
	Validated   bool            `json:"validated"`
	Sampling    *SamplingConfig `json:"sampling,omitempty"`
}

ListModelDetail provides information about a model.

type ListModelInfoResponse 

type ListModelInfoResponse struct {
	Object string            `json:"object"`
	Data   []ListModelDetail `json:"data"`
}

ListModelInfoResponse contains the list of models loaded in the system.

func (ListModelInfoResponse) Encode 

func (app ListModelInfoResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type ModelConfig added in v1.15.8 

type ModelConfig struct {
	Device               string                   `json:"device"`
	ContextWindow        int                      `json:"context-window"`
	NBatch               int                      `json:"nbatch"`
	NUBatch              int                      `json:"nubatch"`
	NThreads             int                      `json:"nthreads"`
	NThreadsBatch        int                      `json:"nthreads-batch"`
	CacheTypeK           model.GGMLType           `json:"cache-type-k"`
	CacheTypeV           model.GGMLType           `json:"cache-type-v"`
	UseDirectIO          bool                     `json:"use-direct-io"`
	FlashAttention       model.FlashAttentionType `json:"flash-attention"`
	IgnoreIntegrityCheck bool                     `json:"ignore-integrity-check"`
	NSeqMax              int                      `json:"nseq-max"`
	OffloadKQV           *bool                    `json:"offload-kqv"`
	OpOffload            *bool                    `json:"op-offload"`
	NGpuLayers           *int                     `json:"ngpu-layers"`
	SplitMode            model.SplitMode          `json:"split-mode"`
	SystemPromptCache    bool                     `json:"system-prompt-cache"`
	IncrementalCache     bool                     `json:"incremental-cache"`
	CacheMinTokens       int                      `json:"cache-min-tokens"`
	CacheSlotTimeout     int                      `json:"cache-slot-timeout"`
	Sampling             SamplingConfig           `json:"sampling-parameters"`
	RopeScaling          model.RopeScalingType    `json:"rope-scaling-type"`
	RopeFreqBase         *float32                 `json:"rope-freq-base"`
	RopeFreqScale        *float32                 `json:"rope-freq-scale"`
	YarnExtFactor        *float32                 `json:"yarn-ext-factor"`
	YarnAttnFactor       *float32                 `json:"yarn-attn-factor"`
	YarnBetaFast         *float32                 `json:"yarn-beta-fast"`
	YarnBetaSlow         *float32                 `json:"yarn-beta-slow"`
	YarnOrigCtx          *int                     `json:"yarn-orig-ctx"`
}

ModelConfig represents the model configuration the model will use by default.

type ModelDetail 

type ModelDetail struct {
	ID            string    `json:"id"`
	OwnedBy       string    `json:"owned_by"`
	ModelFamily   string    `json:"model_family"`
	Size          int64     `json:"size"`
	VRAMTotal     int64     `json:"vram_total"`
	SlotMemory    int64     `json:"slot_memory"`
	ExpiresAt     time.Time `json:"expires_at"`
	ActiveStreams int       `json:"active_streams"`
}

ModelDetail provides details for the models in the cache.

type ModelDetailsResponse added in v1.3.1 

type ModelDetailsResponse []ModelDetail

ModelDetailsResponse is a collection of model detail.

func (ModelDetailsResponse) Encode added in v1.3.1 

func (app ModelDetailsResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type ModelInfoResponse 

type ModelInfoResponse struct {
	ID            string            `json:"id"`
	Object        string            `json:"object"`
	Created       int64             `json:"created"`
	OwnedBy       string            `json:"owned_by"`
	Desc          string            `json:"desc"`
	Size          int64             `json:"size"`
	HasProjection bool              `json:"has_projection"`
	IsGPT         bool              `json:"is_gpt"`
	Metadata      map[string]string `json:"metadata"`
	VRAM          *VRAM             `json:"vram,omitempty"`
	ModelConfig   *ModelConfig      `json:"model_config,omitempty"`
}

ModelInfoResponse returns information about a model.

func (ModelInfoResponse) Encode 

func (app ModelInfoResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type PublishCatalogRequest added in v1.18.0 

type PublishCatalogRequest struct {
	CatalogFile string `json:"catalog_file"`
}

PublishCatalogRequest represents the input for publishing a catalog file to the cloned repository.

func (*PublishCatalogRequest) Decode added in v1.18.0 

func (app *PublishCatalogRequest) Decode(data []byte) error

Decode implements the decoder interface.

type PublishCatalogResponse added in v1.18.0 

type PublishCatalogResponse struct {
	Status string `json:"status"`
}

PublishCatalogResponse is returned after publishing a catalog file.

func (PublishCatalogResponse) Encode added in v1.18.0 

func (app PublishCatalogResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type PullCatalogRequest added in v1.20.1 

type PullCatalogRequest struct {
	DownloadServer string `json:"download_server"`
}

PullCatalogRequest represents the request to pull a catalog model.

func (*PullCatalogRequest) Decode added in v1.20.1 

func (app *PullCatalogRequest) Decode(data []byte) error

Decode implements the decoder interface.

type PullRequest 

type PullRequest struct {
	ModelURL  string   `json:"model_url"`
	ProjURL   string   `json:"proj_url"`
	SplitURLs []string `json:"-"`
}

PullRequest represents the input for the pull command.

func (*PullRequest) Decode 

func (app *PullRequest) Decode(data []byte) error

Decode implements the decoder interface.

type PullResponse 

type PullResponse struct {
	Status     string   `json:"status"`
	ModelFiles []string `json:"model_files,omitempty"`
	ProjFile   string   `json:"proj_file,omitempty"`
	Downloaded bool     `json:"downloaded,omitempty"`
}

PullResponse returns information about a model being downloaded.

func (PullResponse) Encode 

func (app PullResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type RateLimit added in v1.9.2 

type RateLimit struct {
	Limit  int    `json:"limit"`
	Window string `json:"window"`
}

RateLimit defines the rate limit configuration for an endpoint.

type RepoPathResponse added in v1.18.0 

type RepoPathResponse struct {
	RepoPath string `json:"repo_path"`
}

RepoPathResponse is returned for querying the repo path configuration.

func (RepoPathResponse) Encode added in v1.18.0 

func (app RepoPathResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type SamplingConfig added in v1.15.8 

type SamplingConfig struct {
	Temperature      float32 `json:"temperature"`
	TopK             int32   `json:"top_k"`
	TopP             float32 `json:"top_p"`
	MinP             float32 `json:"min_p"`
	MaxTokens        int     `json:"max_tokens"`
	RepeatPenalty    float32 `json:"repeat_penalty"`
	RepeatLastN      int32   `json:"repeat_last_n"`
	DryMultiplier    float32 `json:"dry_multiplier"`
	DryBase          float32 `json:"dry_base"`
	DryAllowedLen    int32   `json:"dry_allowed_length"`
	DryPenaltyLast   int32   `json:"dry_penalty_last_n"`
	XtcProbability   float32 `json:"xtc_probability"`
	XtcThreshold     float32 `json:"xtc_threshold"`
	XtcMinKeep       uint32  `json:"xtc_min_keep"`
	FrequencyPenalty float32 `json:"frequency_penalty"`
	PresencePenalty  float32 `json:"presence_penalty"`
	EnableThinking   string  `json:"enable_thinking"`
	ReasoningEffort  string  `json:"reasoning_effort"`
	Grammar          string  `json:"grammar"`
}

SamplingConfig represents sampling parameters for model inference.

type SaveCatalogRequest added in v1.18.0 

type SaveCatalogRequest struct {
	ID           string              `json:"id"`
	Category     string              `json:"category"`
	OwnedBy      string              `json:"owned_by"`
	ModelFamily  string              `json:"model_family"`
	Architecture string              `json:"architecture"`
	Parameters   string              `json:"parameters"`
	WebPage      string              `json:"web_page"`
	GatedModel   bool                `json:"gated_model"`
	Template     string              `json:"template"`
	Files        CatalogFiles        `json:"files"`
	Capabilities CatalogCapabilities `json:"capabilities"`
	Metadata     CatalogMetadata     `json:"metadata"`
	Config       *ModelConfig        `json:"config,omitempty"`
	CatalogFile  string              `json:"catalog_file"`
}

SaveCatalogRequest represents the input for saving a catalog entry.

func (*SaveCatalogRequest) Decode added in v1.18.0 

func (app *SaveCatalogRequest) Decode(data []byte) error

Decode implements the decoder interface.

type SaveCatalogResponse added in v1.18.0 

type SaveCatalogResponse struct {
	Status string `json:"status"`
	ID     string `json:"id"`
}

SaveCatalogResponse is returned after saving or deleting a catalog entry.

func (SaveCatalogResponse) Encode added in v1.18.0 

func (app SaveCatalogResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type TemplateFilesResponse added in v1.18.9 

type TemplateFilesResponse struct {
	Files []string `json:"files"`
}

TemplateFilesResponse is a list of available template filenames.

func (TemplateFilesResponse) Encode added in v1.18.9 

func (app TemplateFilesResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type TokenRequest added in v1.5.0 

type TokenRequest struct {
	Admin     bool                 `json:"admin"`
	Endpoints map[string]RateLimit `json:"endpoints"`
	Duration  time.Duration        `json:"duration"`
}

TokenRequest represents the input for the create token command.

func (*TokenRequest) Decode added in v1.5.0 

func (app *TokenRequest) Decode(data []byte) error

Decode implements the decoder interface.

type TokenResponse added in v1.5.0 

type TokenResponse struct {
	Token string `json:"token"`
}

TokenResponse represents the response for a successful token creation.

func (TokenResponse) Encode added in v1.5.0 

func (app TokenResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type UnloadRequest added in v1.19.5 

type UnloadRequest struct {
	ID string `json:"id"`
}

UnloadRequest represents the input for unloading a model from the cache.

func (*UnloadRequest) Decode added in v1.19.5 

func (app *UnloadRequest) Decode(data []byte) error

Decode implements the decoder interface.

func (*UnloadRequest) Validate added in v1.19.5 

func (app *UnloadRequest) Validate() error

Validate checks the request is valid.

type UnloadResponse added in v1.19.5 

type UnloadResponse struct {
	Status string `json:"status"`
	ID     string `json:"id"`
}

UnloadResponse represents the output for a model unload operation.

func (UnloadResponse) Encode added in v1.19.5 

func (app UnloadResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type VRAM added in v1.16.0 

type VRAM struct {
	Input              VRAMInput `json:"input"`
	KVPerTokenPerLayer int64     `json:"kv_per_token_per_layer"`
	KVPerSlot          int64     `json:"kv_per_slot"`
	SlotMemory         int64     `json:"slot_memory"`
	TotalVRAM          int64     `json:"total_vram"`
}

VRAM contains the calculated VRAM requirements.

type VRAMInput added in v1.16.4 

type VRAMInput struct {
	ModelSizeBytes  int64 `json:"model_size_bytes"`
	ContextWindow   int64 `json:"context_window"`
	BlockCount      int64 `json:"block_count"`
	HeadCountKV     int64 `json:"head_count_kv"`
	KeyLength       int64 `json:"key_length"`
	ValueLength     int64 `json:"value_length"`
	BytesPerElement int64 `json:"bytes_per_element"`
	Slots           int64 `json:"slots"`
}

VRAMInput contains the input parameters used for VRAM calculation.

type VRAMRequest added in v1.16.4 

type VRAMRequest struct {
	ModelURL        string `json:"model_url"`
	ContextWindow   int64  `json:"context_window"`
	BytesPerElement int64  `json:"bytes_per_element"`
	Slots           int64  `json:"slots"`
}

VRAMRequest represents the input for VRAM calculation.

func (*VRAMRequest) Decode added in v1.16.4 

func (app *VRAMRequest) Decode(data []byte) error

Decode implements the decoder interface.

type VRAMResponse added in v1.16.4 

type VRAMResponse struct {
	Input              VRAMInput `json:"input"`
	KVPerTokenPerLayer int64     `json:"kv_per_token_per_layer"`
	KVPerSlot          int64     `json:"kv_per_slot"`
	SlotMemory         int64     `json:"slot_memory"`
	TotalVRAM          int64     `json:"total_vram"`
}

VRAMResponse represents the VRAM calculation results.

func (VRAMResponse) Encode added in v1.16.4 

func (app VRAMResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

type VersionResponse 

type VersionResponse struct {
	Status       string `json:"status"`
	Arch         string `json:"arch,omitempty"`
	OS           string `json:"os,omitempty"`
	Processor    string `json:"processor,omitempty"`
	Latest       string `json:"latest,omitempty"`
	Current      string `json:"current,omitempty"`
	AllowUpgrade bool   `json:"allow_upgrade"`
}

VersionResponse returns information about the installed libraries.

func (VersionResponse) Encode 

func (app VersionResponse) Encode() ([]byte, string, error)

Encode implements the encoder interface.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.