Copyright	(c) 2025 Tushar Adhatrao
License	MIT
Maintainer	Tushar Adhatrao <[email protected]>
Safe Haskell	None
Language	Haskell2010

Llama.Model

Description

Synopsis

defaultModelParams :: IO LlamaModelParams
loadModelFromFile :: FilePath -> LlamaModelParams -> IO (Either String Model)
initContextFromModel :: Model -> LlamaContextParams -> IO (Either String Context)
getModelVocab :: Model -> IO (Either String Vocab)
getContextModel :: Context -> IO Model
getVocabType :: Vocab -> IO (Maybe LlamaVocabType)
getModelRoPEFreqScale :: Model -> IO Float
getModelNumKVHeads :: Model -> IO Int
getModelNumHeads :: Model -> IO Int
getModelNumLayers :: Model -> IO Int
getModelEmbeddingDim :: Model -> IO Int
getModelTrainingContextSize :: Model -> IO Int
getModelSize :: Model -> IO Int64
getModelChatTemplate :: Model -> Maybe String -> IO (Either String String)
getModelHasEncoder :: Model -> IO Bool
getModelNumParams :: Model -> IO (Either String Int64)
getModelHasDecoder :: Model -> IO Bool
getModelDecoderStartToken :: Model -> IO (Either String LlamaToken)
getModelIsRecurrent :: Model -> IO Bool
quantizeModel :: FilePath -> FilePath -> LlamaModelQuantizeParams -> IO (Either String Word32)
quantizeModelDefault :: FilePath -> FilePath -> IO (Either String Word32)
defaultQuantizeParams :: IO LlamaModelQuantizeParams
getModelMetaCount :: Model -> IO Int
getModelMetaValue :: Model -> String -> IO (Either String String)
getModelMetaKeyByIndex :: Model -> Int -> IO (Either String String)
getModelMetaValueByIndex :: Model -> Int -> IO (Either String String)
getModelDescription :: Model -> IO String
loadModelFromSplits :: [FilePath] -> LlamaModelParams -> IO (Either String Model)
getModelRopeType :: Model -> IO (Maybe LlamaRopeTypeScaling)

Documentation

defaultModelParams :: IO LlamaModelParams Source #

Default model parameters

loadModelFromFile :: FilePath -> LlamaModelParams -> IO (Either String Model) Source #

Load a model from a file using the specified parameters

initContextFromModel :: Model -> LlamaContextParams -> IO (Either String Context) Source #

Create a context from a model using the specified parameters

getModelVocab :: Model -> IO (Either String Vocab) Source #

Get the vocabulary from a model

getContextModel :: Context -> IO Model Source #

Get the model associated with a context.

getVocabType :: Vocab -> IO (Maybe LlamaVocabType) Source #

Get the vocabulary type.

getModelRoPEFreqScale :: Model -> IO Float Source #

Get RoPE frequency scaling factor used during training.

getModelNumKVHeads :: Model -> IO Int Source #

Get the number of key/value heads in the model.

getModelNumHeads :: Model -> IO Int Source #

Get the number of attention heads in the model.

getModelNumLayers :: Model -> IO Int Source #

Get the number of transformer layers in the model.

getModelEmbeddingDim :: Model -> IO Int Source #

Get the embedding dimension of the model.

getModelTrainingContextSize :: Model -> IO Int Source #

Get the training context size of the model.

getModelSize :: Model -> IO Int64 Source #

Get the size of the model in bytes

getModelChatTemplate :: Model -> Maybe String -> IO (Either String String) Source #

Get the chat template from a model, optionally by name

getModelHasEncoder :: Model -> IO Bool Source #

Check if the model has an encoder

getModelNumParams :: Model -> IO (Either String Int64) Source #

Get the number of parameters in the model

getModelHasDecoder :: Model -> IO Bool Source #

Check if the model has a decoder

getModelDecoderStartToken :: Model -> IO (Either String LlamaToken) Source #

Get the decoder start token from the model

getModelIsRecurrent :: Model -> IO Bool Source #

Check if the model is recurrent

quantizeModel :: FilePath -> FilePath -> LlamaModelQuantizeParams -> IO (Either String Word32) Source #

Quantize a model from a file to another file using specified parameters

quantizeModelDefault :: FilePath -> FilePath -> IO (Either String Word32) Source #

Quantize a model from a file to another file using default parameters

defaultQuantizeParams :: IO LlamaModelQuantizeParams Source #

Get the default quantization parameters

getModelMetaCount :: Model -> IO Int Source #

Get the number of metadata entries in a model

getModelMetaValue :: Model -> String -> IO (Either String String) Source #

Get a metadata value as a string from a model

getModelMetaKeyByIndex :: Model -> Int -> IO (Either String String) Source #

Get a metadata key by index from a model

getModelMetaValueByIndex :: Model -> Int -> IO (Either String String) Source #

Get a metadata value by index from a model

getModelDescription :: Model -> IO String Source #

Get a model description

loadModelFromSplits :: [FilePath] -> LlamaModelParams -> IO (Either String Model) Source #

Load a model from multiple file paths using specified parameters

getModelRopeType :: Model -> IO (Maybe LlamaRopeTypeScaling) Source #

Get the RoPE type from a model