Struct Model 

pub struct Model {

    pub id: i64,
    pub name: String,
    pub model_key: String,
    pub file_path: PathBuf,
    pub param_count_b: f64,
    pub architecture: Option<String>,
    pub quantization: Option<String>,
    pub context_length: Option<u64>,
    pub expert_count: Option<u32>,
    pub expert_used_count: Option<u32>,
    pub expert_shared_count: Option<u32>,
    pub metadata: HashMap<String, String>,
    pub added_at: DateTime<Utc>,
    pub hf_repo_id: Option<String>,
    pub hf_commit_sha: Option<String>,
    pub hf_filename: Option<String>,
    pub download_date: Option<DateTime<Utc>>,
    pub last_update_check: Option<DateTime<Utc>>,
    pub tags: Vec<String>,
    pub capabilities: ModelCapabilities,
    pub inference_defaults: Option<InferenceConfig>,
    pub defaults_origin: Option<DefaultsOrigin>,
    pub server_defaults: Option<ServerConfig>,
    pub dialect_spec: Option<DialectSpec>,
    pub template_caps: Option<TemplateCaps>,
    pub benchmark_summary: Option<ModelBenchmarkSummary>,
}

A model that exists in the system with a database ID.

This represents a persisted model with all its metadata. Use NewModel for models that haven’t been persisted yet.

Fields

id: i64

Database ID of the model (always present for persisted models).

name: String

Human-readable name for the model.

model_key: String

Canonical deduplication key (e.g., hf:repo@sha#file).

file_path: PathBuf

Absolute path to the GGUF file on the filesystem.

param_count_b: f64

Number of parameters in the model (in billions).

architecture: Option<String>

Model architecture (e.g., “llama”, “mistral”, “falcon”).

quantization: Option<String>

Quantization type (e.g., “Q4_0”, “Q8_0”, “F16”, “F32”).

context_length: Option<u64>

Maximum context length the model supports.

expert_count: Option<u32>

Number of experts (for MoE models).

expert_used_count: Option<u32>

Number of experts used during inference (for MoE models).

expert_shared_count: Option<u32>

Number of shared experts (for MoE models).

metadata: HashMap<String, String>

Additional metadata key-value pairs from the GGUF file.

added_at: DateTime<Utc>

UTC timestamp of when the model was added to the database.

hf_repo_id: Option<String>

HuggingFace repository ID (e.g., “TheBloke/Llama-2-7B-GGUF”).

hf_commit_sha: Option<String>

Git commit SHA from HuggingFace Hub.

hf_filename: Option<String>

Original filename on HuggingFace Hub.

download_date: Option<DateTime<Utc>>

Timestamp of when this model was downloaded from HuggingFace.

last_update_check: Option<DateTime<Utc>>

Last time we checked for updates on HuggingFace.

tags: Vec<String>

User-defined tags for organizing models.

capabilities: ModelCapabilities

Model capabilities inferred from chat template analysis.

inference_defaults: Option<InferenceConfig>

Per-model inference parameter defaults.

These are preferred over global settings when making inference requests. If not set, falls back to global settings or hardcoded defaults.

defaults_origin: Option<DefaultsOrigin>

Whether Self::inference_defaults was set by the user or auto-detected at import time from the reasoning tag.

Always None when Self::inference_defaults is None — there is nothing to have an origin. See DefaultsOrigin for why this changes how resolution ranks the field.

server_defaults: Option<ServerConfig>

Per-model server-level defaults (context_length, etc.).

Stored as JSON in the database. Overrides global settings but can be overridden at request time. Part of the 5-level fallback chain.

dialect_spec: Option<DialectSpec>

Tool-call dialect spec detected at import/retag time.

Stored as JSON in the database. None for rows imported before specs existed and for models whose dialect could not be derived — consumers fall back to the format:* tag mapping.

template_caps: Option<TemplateCaps>

llama-server’s template-capability self-report (chat_template_caps from GET /props), recorded once a launch has observed it.

Stored as JSON in the database. None means never observed — the third state of ADR 0007’s tri-state, never to be collapsed into “not supported”. Unlike Self::dialect_spec this is not derived at import time: it is a fact about the binary–model pair, so only a launch can learn it.

benchmark_summary: Option<ModelBenchmarkSummary>

Denormalised benchmark summary joined from model_benchmark_summaries.

None when no benchmark has been run for this model yet, or when the model is fetched without the summary join (e.g. lightweight lookups).

Trait Implementations

impl Clone for Model

fn clone(&self) -> Model

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Model

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Model

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl From<&Model> for ModelSummary

Borrow a stored model as the lightweight summary events carry.

Every emit site wants the same five fields off a Model it already has, so the mapping lives here rather than being spelled out per call site.

fn from(model: &Model) -> Self

Converts to this type from the input type.

impl Serialize for Model

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

where __S: Serializer,

Serialize this value into the given Serde serializer.