Trait ModelPaths

pub trait ModelPaths: AsAny + Debug {
    // Required methods
    fn get_weight_filenames(&self) -> &[PathBuf];
    fn get_config_filename(&self) -> &PathBuf;
    fn get_tokenizer_filename(&self) -> &PathBuf;
    fn get_template_filename(&self) -> &Option<PathBuf>;
    fn get_adapter_filenames(&self) -> &Option<Vec<(String, PathBuf)>>;
    fn get_adapter_configs(
        &self,
    ) -> &Option<Vec<((String, String), LoraConfig)>>;
    fn get_classifier_path(&self) -> &Option<PathBuf>;
    fn get_classifier_config(&self) -> &Option<XLoraConfig>;
    fn get_ordering(&self) -> &Option<Ordering>;
    fn get_gen_conf_filename(&self) -> Option<&PathBuf>;
    fn get_lora_preload_adapter_info(
        &self,
    ) -> &Option<HashMap<String, (PathBuf, LoraConfig)>>;
    fn get_preprocessor_config(&self) -> &Option<PathBuf>;
    fn get_processor_config(&self) -> &Option<PathBuf>;
    fn get_chat_template_json(&self) -> &Option<PathBuf>;
}

ModelPaths abstracts the mechanism to get all necessary files for running a model. For example LocalModelPaths implements ModelPaths when all files are in the local file system.

Required Methods

fn get_weight_filenames(&self) -> &[PathBuf]

Model weights files (multiple files supported).

fn get_config_filename(&self) -> &PathBuf

Retrieve the PretrainedConfig file.

fn get_tokenizer_filename(&self) -> &PathBuf

A serialised tokenizers.Tokenizer HuggingFace object.

fn get_template_filename(&self) -> &Option<PathBuf>

File where the content is expected to deserialize to ChatTemplate.

fn get_adapter_filenames(&self) -> &Option<Vec<(String, PathBuf)>>

Optional adapter files. (String, PathBuf) is of the form (id name, path).

fn get_adapter_configs(&self) -> &Option<Vec<((String, String), LoraConfig)>>

Configuration of optional adapters. (String, String) is of the form (id name, name).

fn get_classifier_path(&self) -> &Option<PathBuf>

Filepath for the XLORA classifier

fn get_classifier_config(&self) -> &Option<XLoraConfig>

XLoraConfig for the XLORA classifier

fn get_ordering(&self) -> &Option<Ordering>

Return the defined ordering of adapters and layers within the model.

fn get_gen_conf_filename(&self) -> Option<&PathBuf>

Filepath for general model configuration.

fn get_lora_preload_adapter_info( &self, ) -> &Option<HashMap<String, (PathBuf, LoraConfig)>>

Information for preloading LoRA adapters (adapter name, the weight file, and the config).

fn get_preprocessor_config(&self) -> &Option<PathBuf>

Get the preprocessor config (for the vision models). This is used to pre process images.

fn get_processor_config(&self) -> &Option<PathBuf>

Get the processor config (for the vision models). This is primarily used for the chat template.

fn get_chat_template_json(&self) -> &Option<PathBuf>

Get the explicit chat template. If specified, this overwrites anything in the tokenizer_config.json

Implementors

impl ModelPaths for LocalModelPaths<PathBuf>