Struct ModelSettings 

pub struct ModelSettings {

    pub context_length: u32,
    pub threads: u32,
    pub threads_batch: u32,
    pub batch_size: u32,
    pub ubatch_size: u32,
    pub parallel: u32,
    pub max_concurrent_predictions: Option<u32>,
    pub uniform_cache: bool,
    pub kv_cache_offload: bool,
    pub cache_type_k: Option<CacheTypeK>,
    pub cache_type_v: Option<CacheTypeV>,
    pub keep: i32,
    pub swa_full: bool,
    pub mlock: bool,
    pub mmap: bool,
    pub numa: NumMode,
    pub system_prompt: String,
    pub system_prompt_preset_name: String,
    pub gpu_layers_mode: GpuLayersMode,
    pub split_mode: SplitMode,
    pub tensor_split: String,
    pub main_gpu: i32,
    pub fit: bool,
    pub lora: Option<PathBuf>,
    pub lora_scaled: Option<(PathBuf, f32)>,
    pub rpc: String,
    pub embedding: bool,
    pub flash_attn: bool,
    pub expert_count: i32,
    pub jinja: bool,
    pub chat_template: Option<String>,
    pub chat_template_kwargs: Option<String>,
    pub seed: i32,
    pub temperature: f32,
    pub top_k: i32,
    pub top_p: f32,
    pub min_p: f32,
    pub typical_p: f32,
    pub mirostat: Mirostat,
    pub mirostat_lr: f32,
    pub mirostat_ent: f32,
    pub ignore_eos: bool,
    pub samplers: Samplers,
    pub repeat_penalty: f32,
    pub repeat_last_n: i32,
    pub presence_penalty: Option<f32>,
    pub frequency_penalty: Option<f32>,
    pub dry_multiplier: f32,
    pub dry_base: f32,
    pub dry_allowed_length: i32,
    pub dry_penalty_last_n: i32,
    pub rope_scaling: RopeScaling,
    pub rope_scale: f32,
    pub rope_freq_base: f32,
    pub rope_freq_scale: f32,
    pub rope_yarn_enabled: bool,
    pub host: String,
    pub port: u16,
    pub timeout: u32,
    pub cache_prompt: bool,
    pub cache_reuse: u32,
    pub webui: bool,
    pub max_tokens: Option<u32>,
    pub cache_type: CacheType,
    pub backend: Backend,
    pub llama_cpp_version_cpu: Option<String>,
    pub llama_cpp_version_vulkan: Option<String>,
    pub llama_cpp_version_rocm: Option<String>,
    pub llama_cpp_version_rocm_lemonade: Option<String>,
    pub llama_cpp_version_cuda: Option<String>,
    pub api_endpoint_enabled: bool,
    pub api_endpoint_port: u16,
    pub spec_type: String,
    pub draft_tokens: u32,
    pub tags: Vec<String>,
    pub ws_server_enabled: bool,
    pub ws_server_port: u16,
    pub ws_server_auth_key: Option<String>,
    pub ws_server_tls_enabled: bool,
    pub ws_server_tls_cert: Option<String>,
    pub ws_server_tls_key: Option<String>,
}

Settings for loading a model via llama.cpp server.

Fields

context_length: u32

Size of the prompt context.

threads: u32

Number of CPU threads for generation.

threads_batch: u32

Number of CPU threads for batch processing.

batch_size: u32

Logical maximum batch size.

ubatch_size: u32

Physical maximum batch size (micro-batch).

parallel: u32

Max concurrent predictions (sequences).

max_concurrent_predictions: Option<u32>

Max concurrent predictions (requests in flight). None means no –parallel argument.

uniform_cache: bool

Use uniform (unified) KV cache across all sequences.

kv_cache_offload: bool

Offload KV cache to system RAM.

cache_type_k: Option<CacheTypeK>

KV cache data type for K.

cache_type_v: Option<CacheTypeV>

KV cache data type for V.

keep: i32

Keep N tokens from the initial prompt.

swa_full: bool

Use full-size SWA cache.

mlock: bool

Force system to keep model in RAM.

mmap: bool

Memory-map the model.

numa: NumMode

NUMA optimization.

system_prompt: String

System prompt.

system_prompt_preset_name: String

Name of the system prompt preset currently selected.

gpu_layers_mode: GpuLayersMode

GPU layer offloading mode.

split_mode: SplitMode

Split mode across multiple GPUs.

tensor_split: String

Fraction of model offloaded to each GPU (comma-separated).

main_gpu: i32

Main GPU index.

fit: bool

Whether to adjust arguments to fit device memory.

lora: Option<PathBuf>

Path to LoRA adapter.

lora_scaled: Option<(PathBuf, f32)>

Path to LoRA adapter with scale.

rpc: String

RPC servers.

embedding: bool

Restrict to embedding use case.

flash_attn: bool

Enable Flash Attention.

expert_count: i32

Active experts per token (MoE models, -1 = model default).

jinja: bool

Use Jinja template engine for chat.

chat_template: Option<String>

Custom chat template string.

chat_template_kwargs: Option<String>

JSON string for –chat-template-kwargs (e.g. {“enable_thinking”: false}).

seed: i32

RNG seed (-1 = random).

temperature: f32

Temperature.

top_k: i32

Top-k sampling (0 = disabled).

top_p: f32

Top-p sampling (1.0 = disabled).

min_p: f32

Minimum probability for a token.

typical_p: f32

Locally typical sampling parameter p.

mirostat: Mirostat

Mirostat version (0=off, 1=Mirostat, 2=Mirostat2).

mirostat_lr: f32

Mirostat learning rate (eta).

mirostat_ent: f32

Mirostat target entropy (tau).

ignore_eos: bool

Ignore end-of-stream token.

samplers: Samplers

Sampler order string.

repeat_penalty: f32

Penalize repeat sequence of tokens.

repeat_last_n: i32

Last N tokens to consider for repeat penalty.

presence_penalty: Option<f32>

Repeat alpha presence penalty.

frequency_penalty: Option<f32>

Repeat alpha frequency penalty.

dry_multiplier: f32

DRY sampling multiplier.

dry_base: f32

DRY sampling base value.

dry_allowed_length: i32

DRY allowed length.

dry_penalty_last_n: i32

DRY penalty last N.

rope_scaling: RopeScaling

RoPE frequency scaling method.

rope_scale: f32

RoPE context scaling factor.

rope_freq_base: f32

RoPE base frequency.

rope_freq_scale: f32

RoPE frequency scaling factor.

rope_yarn_enabled: bool

Enable Yarn RoPE scaling mode.

host: String

Host address.

port: u16

Port.

timeout: u32

Server timeout in seconds.

cache_prompt: bool

Whether to enable prompt caching.

cache_reuse: u32

Min chunk size for cache reuse.

webui: bool

Whether to enable WebUI.

max_tokens: Option<u32>

Max tokens to predict.

cache_type: CacheType

Cache type (legacy, kept for compatibility).

backend: Backend

Backend (cpu/vulkan).

llama_cpp_version_cpu: Option<String>

llama.cpp release tag for CPU backend (e.g. “b1234” or None for latest).

llama_cpp_version_vulkan: Option<String>

llama.cpp release tag for Vulkan backend (e.g. “b1234” or None for latest).

llama_cpp_version_rocm: Option<String>

llama.cpp release tag for ROCm backend (e.g. “b1234” or None for latest).

llama_cpp_version_rocm_lemonade: Option<String>

Lemonade llama.cpp release tag for ROCm backend.

llama_cpp_version_cuda: Option<String>

llama.cpp release tag for CUDA backend.

api_endpoint_enabled: bool

Whether to enable the API proxy server.

api_endpoint_port: u16

Port for the API proxy server.

spec_type: String

Speculative decoding type (e.g., “draft-mtp”, “ngram-simple”, “” for off).

draft_tokens: u32

Number of draft tokens for MTP.

tags: Vec<String>

Tags for the model.

ws_server_enabled: bool

Whether to enable the WebSocket dashboard server.

ws_server_port: u16

ws_server_auth_key: Option<String>

ws_server_tls_enabled: bool

ws_server_tls_cert: Option<String>

ws_server_tls_key: Option<String>

Implementations

impl ModelSettings

pub fn get_active_backend_version(&self) -> Option<&String>

Get the version string for the currently active backend.

pub fn get_active_backend_version_display(&self) -> &str

Get the display version string for the currently active backend (defaults to “latest”).

pub fn set_active_backend_version(&mut self, tag: Option<String>)

Set the version string for the currently active backend.

impl ModelSettings

pub fn from_config(config: &Config) -> Self

Create ModelSettings from config defaults, applying model-specific overrides.

impl ModelSettings

pub fn is_dirty(&self, other: &Self) -> bool

Check if this settings differs from other in any field. Uses derived PartialEq which compares all fields — compiler-enforced.

Trait Implementations

impl Clone for ModelSettings

fn clone(&self) -> ModelSettings

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ModelSettings

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ModelSettings

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ModelSettings

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl From<DefaultParams> for ModelSettings

fn from(dp: DefaultParams) -> Self

Converts to this type from the input type.

impl PartialEq for ModelSettings

fn eq(&self, other: &ModelSettings) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for ModelSettings

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for ModelSettings