aiha 0.0.4

//! Module for the OPT model
use serde::Deserialize;
use serde_json::Value;

use crate::models::{ModelConfigTrait, ModelError, ModelLibraries};

/// A struct representing the OPT architecture parameters
#[derive(Clone, Debug, Deserialize)]
pub struct OPTParams {
    /// OPT model hidden_size
    hidden_size: i32,
    /// OPT model intermediate_size
    ffn_dim: i32,
    /// OPT model max_position_embeddings
    max_position_embeddings: i32,
    /// OPT model num_attention_heads
    num_attention_heads: i32,
    /// OPT model num_hidden_layers
    num_hidden_layers: i32,
}

/// OPT model parameters implementation
impl OPTParams {
    /// Build a new `OPTParams` struct based on the provided parameters
    pub fn new(
        hidden_size: i32,
        ffn_dim: i32,
        max_position_embeddings: i32,
        num_attention_heads: i32,
        num_hidden_layers: i32,
    ) -> OPTParams {
        OPTParams {
            hidden_size,
            ffn_dim,
            max_position_embeddings,
            num_attention_heads,
            num_hidden_layers,
        }
    }
    /// Build from a JSON value
    pub fn from_json(value: Value) -> Result<OPTParams, ModelError> {
        let hidden_size = value["hidden_size"]
            .as_i64()
            .ok_or(ModelError::MissingField("hidden_size".to_string()))?
            as i32;

        let ffn_dim = value["ffn_dim"]
            .as_i64()
            .ok_or(ModelError::MissingField("ffn_dim".to_string()))? as i32;

        let max_position_embeddings =
            value["max_position_embeddings"]
                .as_i64()
                .ok_or(ModelError::MissingField(
                    "max_position_embeddings".to_string(),
                ))? as i32;

        let num_attention_heads = value["num_attention_heads"]
            .as_i64()
            .ok_or(ModelError::MissingField("num_attention_heads".to_string()))?
            as i32;

        let num_hidden_layers = value["num_hidden_layers"]
            .as_i64()
            .ok_or(ModelError::MissingField("num_hidden_layers".to_string()))?
            as i32;

        Ok(OPTParams::new(
            hidden_size,
            ffn_dim,
            max_position_embeddings,
            num_attention_heads,
            num_hidden_layers,
        ))
    }
}

/// A struct representing a OPT model configuration
#[derive(Clone, Debug, Deserialize)]
pub struct OPTModelConfig {
    /// OPT model parameters
    params: OPTParams,
    /// OPT model type
    model_type: String,
    /// OPT model available libraries
    available_libraries: Vec<ModelLibraries>,
}

/// OPT model implementation
impl OPTModelConfig {
    /// Build a new `OPTModelConfig` struct based on the provided parameters
    pub fn new(
        params: OPTParams,
        model_type: String,
        available_libraries: Vec<ModelLibraries>,
    ) -> OPTModelConfig {
        OPTModelConfig {
            params,
            model_type,
            available_libraries,
        }
    }
}

/// Implementation of the `ModelConfigTrait` trait for `OPTModelConfig`
impl ModelConfigTrait for OPTModelConfig {
    fn hidden_size(&self) -> i32 {
        self.params.hidden_size
    }

    fn intermediate_size(&self) -> i32 {
        self.params.ffn_dim
    }

    fn max_position_embeddings(&self) -> i32 {
        self.params.max_position_embeddings
    }

    fn num_attention_heads(&self) -> i32 {
        self.params.num_attention_heads
    }

    fn num_hidden_layers(&self) -> i32 {
        self.params.num_hidden_layers
    }

    fn model_type(&self) -> &str {
        &self.model_type
    }

    fn available_libraries(&self) -> &[ModelLibraries] {
        &self.available_libraries
    }

    fn from_json(value: Value) -> Result<Self, ModelError>
    where
        Self: Sized,
    {
        let params = OPTParams::from_json(value.clone())?;

        let model_type = match value["model_type"].as_str() {
            Some(model_type) => model_type.to_string(),
            None => return Err(ModelError::MissingField("model_type".to_string())),
        };

        // TODO: Implement this
        let available_libraries = vec![ModelLibraries::PyTorch];
        // let available_libraries = match value["available_libraries"].as_array() {
        //     Some(al) => al.iter().map(|v| ModelLibraries::from_str(v.as_str().unwrap()).unwrap()).collect(),
        //     None => return Err(ModelError::MissingField("available_libraries".to_string())),
        // };

        Ok(OPTModelConfig::new(params, model_type, available_libraries))
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_opt_model_params() {
        let opt_params = OPTParams::new(768, 3072, 512, 12, 12);

        assert_eq!(opt_params.hidden_size, 768);
        assert_eq!(opt_params.ffn_dim, 3072);
        assert_eq!(opt_params.max_position_embeddings, 512);
        assert_eq!(opt_params.num_attention_heads, 12);
        assert_eq!(opt_params.num_hidden_layers, 12);
    }

    #[test]
    fn test_opt_model_config() {
        let opt_params = OPTParams::new(768, 3072, 512, 12, 12);

        let opt_model_config = OPTModelConfig::new(
            opt_params,
            String::from("opt"),
            vec![ModelLibraries::TensorFlow, ModelLibraries::PyTorch],
        );

        assert_eq!(opt_model_config.params.hidden_size, 768);
        assert_eq!(opt_model_config.params.ffn_dim, 3072);
        assert_eq!(opt_model_config.params.max_position_embeddings, 512);
        assert_eq!(opt_model_config.params.num_attention_heads, 12);
        assert_eq!(opt_model_config.params.num_hidden_layers, 12);
        assert_eq!(opt_model_config.model_type, "opt");
        assert_eq!(
            opt_model_config.available_libraries,
            vec![ModelLibraries::TensorFlow, ModelLibraries::PyTorch]
        );
    }

    #[test]
    fn test_opt_model_trait_implementation() {
        let opt_params = OPTParams::new(768, 3072, 512, 12, 12);

        let opt_model_config = OPTModelConfig::new(
            opt_params,
            String::from("opt"),
            vec![ModelLibraries::TensorFlow, ModelLibraries::PyTorch],
        );

        assert_eq!(opt_model_config.hidden_size(), 768);
        assert_eq!(opt_model_config.intermediate_size(), 3072);
        assert_eq!(opt_model_config.max_position_embeddings(), 512);
        assert_eq!(opt_model_config.num_attention_heads(), 12);
        assert_eq!(opt_model_config.num_hidden_layers(), 12);
        assert_eq!(opt_model_config.model_type(), "opt");
        assert_eq!(
            opt_model_config.available_libraries(),
            vec![ModelLibraries::TensorFlow, ModelLibraries::PyTorch]
        );
    }
}