Version
- {`const Version = "1.32.9-rc"`}
+ {`const Version = "1.32.9"`}
Version contains the current version of the kronk package. Bump this in the same commit that prepares a `v<X.Y.Z>` release tag — CI guards the release pipeline (.github/workflows/release.yaml and docker.yml, via .github/scripts/check-version.sh) and refuses to publish a tag whose value does not match this constant.
diff --git a/cmd/server/api/services/kronk/static/assets/index.js b/cmd/server/api/services/kronk/static/assets/index.js
index 866d49062..be0ce1db6 100644
--- a/cmd/server/api/services/kronk/static/assets/index.js
+++ b/cmd/server/api/services/kronk/static/assets/index.js
@@ -231,7 +231,7 @@ kronk server start`})]})]})}function Ls(e){let t=/^([\d.]+)\s*(MB|GB|KB|B)?$/i.e
}`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ResponseTextFormat specifies the text format configuration.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`type-responseusage`,children:[(0,G.jsx)(`h4`,{children:`ResponseUsage`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:'type ResponseUsage struct {\n InputTokens int `json:"input_tokens"`\n InputTokensDetails InputTokensDetails `json:"input_tokens_details"`\n OutputTokens int `json:"output_tokens"`\n OutputTokenDetail OutputTokensDetails `json:"output_tokens_details"`\n TotalTokens int `json:"total_tokens"`\n}'})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ResponseUsage contains token usage information.`})]})]}),(0,G.jsxs)(`div`,{className:`card`,id:`methods`,children:[(0,G.jsx)(`h3`,{children:`Methods`}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-activestreams`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ActiveStreams`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ActiveStreams() int`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ActiveStreams returns the number of active streams.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-batchenginesnapshot`,children:[(0,G.jsx)(`h4`,{children:`Kronk.BatchEngineSnapshot`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) BatchEngineSnapshot() (model.BatchEngineSnapshot, bool)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`BatchEngineSnapshot returns the latest generation scheduler state. Models used only for embeddings or reranking do not expose a batch engine.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-chat`,children:[(0,G.jsx)(`h4`,{children:`Kronk.Chat`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) Chat(ctx context.Context, d model.D) (model.ChatResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Chat provides support to interact with an inference model. For text models, NSeqMax controls parallel sequence processing within a single model instance. For vision/audio models, NSeqMax creates multiple model instances in a pool for concurrent request handling.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-chatstreaming`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ChatStreaming`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ChatStreaming(ctx context.Context, d model.D) (<-chan model.ChatResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ChatStreaming provides support to interact with an inference model. For text models, NSeqMax controls parallel sequence processing within a single model instance. For vision/audio models, NSeqMax creates multiple model instances in a pool for concurrent request handling. When stream_options.include_usage is true, the terminal choice is followed by a usage response with an empty Choices slice.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-chatstreaminghttp`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ChatStreamingHTTP`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ChatStreamingHTTP(ctx context.Context, w http.ResponseWriter, d model.D) (model.ChatResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ChatStreamingHTTP provides http handler support for a chat/completions call. For text models, NSeqMax controls parallel sequence processing within a single model instance. For vision/audio models, NSeqMax creates multiple model instances in a pool for concurrent request handling.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-decision`,children:[(0,G.jsx)(`h4`,{children:`Kronk.Decision`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) Decision(ctx context.Context, req model.DecisionRequest) (model.DecisionResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Decision evaluates typed questions against shared state using a supported decision model.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-decisionhttp`,children:[(0,G.jsx)(`h4`,{children:`Kronk.DecisionHTTP`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) DecisionHTTP(ctx context.Context, log Logger, w http.ResponseWriter, req model.DecisionRequest) (model.DecisionResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`DecisionHTTP provides HTTP handler support for a decision call.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-embeddings`,children:[(0,G.jsx)(`h4`,{children:`Kronk.Embeddings`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) Embeddings(ctx context.Context, d model.D) (model.EmbedReponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Embeddings provides support to interact with an embedding model.`}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Supported options in d: - input (string): the text to embed (required) - truncate (bool): if true, truncate input to fit context window (default: false) - truncate_direction (string): "right" (default) or "left" - dimensions (int): reduce output to first N dimensions (for Matryoshka models)`}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Each model instance processes calls sequentially (llama.cpp only supports sequence 0 for embedding extraction). Use NSeqMax > 1 to create multiple model instances for concurrent request handling. Batch multiple texts in the input parameter for better performance within a single request.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-embeddingshttp`,children:[(0,G.jsx)(`h4`,{children:`Kronk.EmbeddingsHTTP`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) EmbeddingsHTTP(ctx context.Context, log Logger, w http.ResponseWriter, d model.D) (model.EmbedReponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`EmbeddingsHTTP provides http handler support for an embeddings call.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-imcsessions`,children:[(0,G.jsx)(`h4`,{children:`Kronk.IMCSessions`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) IMCSessions() []model.IMCSessionDetail`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`IMCSessions returns the current state of the model's allocated IMC cache entries. An unloaded model has no sessions.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-imcsystemcaches`,children:[(0,G.jsx)(`h4`,{children:`Kronk.IMCSystemCaches`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) IMCSystemCaches() []model.IMCSystemCacheDetail`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`IMCSystemCaches returns the model's immutable System cache pool entries.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-modelconfig`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ModelConfig`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ModelConfig() model.Config`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ModelConfig returns a copy of the configuration being used. This may be different from the configuration passed to New() if the model has overridden any of the settings.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-modelid`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ModelID`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ModelID() string`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ModelID returns the identifier emitted by public inference responses.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-modelinfo`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ModelInfo`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ModelInfo() model.ModelInfo`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ModelInfo returns the model information.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-rerank`,children:[(0,G.jsx)(`h4`,{children:`Kronk.Rerank`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) Rerank(ctx context.Context, d model.D) (model.RerankResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Rerank provides support to interact with a reranker model.`}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Supported options in d: - query (string): the query to rank documents against (required) - documents ([]string): the documents to rank (required) - top_n (int): return only the top N results (optional, default: all) - return_documents (bool): include document text in results (default: false)`}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Each model instance processes calls sequentially (llama.cpp only supports sequence 0 for rerank extraction). Use NSeqMax > 1 to create multiple model instances for concurrent request handling. Batch multiple texts in the input parameter for better performance within a single request.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-rerankhttp`,children:[(0,G.jsx)(`h4`,{children:`Kronk.RerankHTTP`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) RerankHTTP(ctx context.Context, log Logger, w http.ResponseWriter, d model.D) (model.RerankResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`RerankHTTP provides http handler support for a rerank call.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-response`,children:[(0,G.jsx)(`h4`,{children:`Kronk.Response`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) Response(ctx context.Context, d model.D) (ResponseResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Response provides support to interact with an inference model. For text models, NSeqMax controls parallel sequence processing within a single model instance. For vision/audio models, NSeqMax creates multiple model instances in a pool for concurrent request handling.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-responsestreaming`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ResponseStreaming`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ResponseStreaming(ctx context.Context, d model.D) (<-chan ResponseStreamEvent, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ResponseStreaming provides streaming support for the Responses API. For text models, NSeqMax controls parallel sequence processing within a single model instance. For vision/audio models, NSeqMax creates multiple model instances in a pool for concurrent request handling.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-responsestreaminghttp`,children:[(0,G.jsx)(`h4`,{children:`Kronk.ResponseStreamingHTTP`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) ResponseStreamingHTTP(ctx context.Context, w http.ResponseWriter, d model.D) (ResponseResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`ResponseStreamingHTTP provides http handler support for a responses call. For text models, NSeqMax controls parallel sequence processing within a single model instance. For vision/audio models, NSeqMax creates multiple model instances in a pool for concurrent request handling.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-systeminfo`,children:[(0,G.jsx)(`h4`,{children:`Kronk.SystemInfo`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) SystemInfo() map[string]string`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`SystemInfo returns system information.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-tokenize`,children:[(0,G.jsx)(`h4`,{children:`Kronk.Tokenize`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) Tokenize(ctx context.Context, d model.D) (model.TokenizeResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Tokenize returns the token count for a text input.`}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Supported options in d: - input (string): the text to tokenize (required) - apply_template (bool): if true, wrap input as a user message and apply the model's chat template before tokenizing (default: false) - add_generation_prompt (bool): when apply_template is true, controls whether the assistant role prefix is appended to the prompt (default: true)`}),(0,G.jsx)(`p`,{className:`doc-description`,children:`When apply_template is true, the returned count includes all template overhead (role markers, separators, generation prompt). This reflects the actual number of tokens that would be fed to the model.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-tokenizehttp`,children:[(0,G.jsx)(`h4`,{children:`Kronk.TokenizeHTTP`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) TokenizeHTTP(ctx context.Context, log Logger, w http.ResponseWriter, d model.D) (model.TokenizeResponse, error)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`TokenizeHTTP provides http handler support for a tokenize call.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`method-kronk-unload`,children:[(0,G.jsx)(`h4`,{children:`Kronk.Unload`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`func (krn *Kronk) Unload(ctx context.Context) error`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Unload will close down the loaded model. You should call this only when you are completely done using Kronk.`})]})]}),(0,G.jsxs)(`div`,{className:`card`,id:`constants`,children:[(0,G.jsx)(`h3`,{children:`Constants`}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`const-logsilent`,children:[(0,G.jsx)(`h4`,{children:`LogSilent`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`const (
LogSilent = applog.LogSilent
LogNormal = applog.LogNormal
-)`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:`Set of logging levels supported by llama.cpp.`})]}),(0,G.jsxs)(`div`,{className:`doc-section`,id:`const-version`,children:[(0,G.jsx)(`h4`,{children:`Version`}),(0,G.jsx)(`pre`,{className:`code-block`,children:(0,G.jsx)(`code`,{children:`const Version = "1.32.9-rc"`})}),(0,G.jsx)(`p`,{className:`doc-description`,children:"Version contains the current version of the kronk package. Bump this in the same commit that prepares a `v