use std::collections::HashMap;
use std::path::PathBuf;
use std::sync::Arc;
use ndarray::{Array2, ArrayD, Axis, IxDyn};
use ort::environment::Environment;
use ort::tensor::{FromArray, InputTensor};
use ort::{GraphOptimizationLevel, SessionBuilder};
use crate::common::Device;
use crate::common::{apply_device, match_to_inputs};
use crate::error::Result;
use crate::{try_extract_to_f32, ORTSession};
pub struct ClassificationModel<'a> {
model_session: ORTSession<'a>,
token_type_support: bool,
num_labels: usize,
is_tok_classification: bool,
}
impl<'a> ClassificationModel<'a> {
pub fn new_from_memory(
env: Arc<Environment>,
model_bytes: &'a [u8],
device: Device,
optimization_level: GraphOptimizationLevel,
) -> Result<Self> {
let mut session_builder = SessionBuilder::new(&env)?;
session_builder = apply_device(session_builder, device)?;
let session = session_builder
.with_optimization_level(optimization_level)?
.with_model_from_memory(model_bytes)?;
let token_type_support = session.inputs.len() == 3
&& session
.inputs
.iter()
.filter(|i| i.name == "token_type_ids")
.count()
> 0;
let num_dims = session.outputs[0].dimensions.len();
let num_labels = session.outputs[0].dimensions[num_dims - 1].unwrap() as usize;
let is_tok_classification = num_dims == 3;
Ok(Self {
model_session: ORTSession::InMemory(session),
token_type_support,
num_labels,
is_tok_classification,
})
}
pub fn new_from_file<'path>(
env: Arc<Environment>,
model_path: PathBuf,
device: Device,
optimization_level: GraphOptimizationLevel,
) -> Result<Self> {
let mut session_builder = SessionBuilder::new(&env)?;
session_builder = apply_device(session_builder, device)?;
let session = session_builder
.with_optimization_level(optimization_level)?
.with_model_from_file(model_path)?;
let token_type_support = session.inputs.len() == 3
&& session
.inputs
.iter()
.filter(|i| i.name == "token_type_ids")
.count()
> 0;
let num_dims = session.outputs[0].dimensions.len();
let num_labels = session.outputs[0].dimensions[num_dims - 1].unwrap() as usize;
let is_tok_classification = num_dims == 3;
Ok(Self {
model_session: ORTSession::Owned(session),
token_type_support,
num_labels,
is_tok_classification,
})
}
pub fn forward(
&self,
input_ids: Array2<u32>,
attention_mask: Array2<u32>,
token_type_ids: Option<Array2<u32>>,
) -> Result<ArrayD<f32>> {
let input_map = self.prepare_input_map(input_ids, attention_mask, token_type_ids)?;
let model = match &self.model_session {
ORTSession::Owned(s) => s,
ORTSession::InMemory(s) => s,
};
let input_tensor = match_to_inputs(&model.inputs, input_map)?;
let output_names = model
.outputs
.iter()
.map(|o| o.name.clone())
.collect::<Vec<_>>();
let outputs_tensors = model.run(input_tensor)?;
let mut output_map = HashMap::new();
for (name, tensor) in output_names.iter().zip(outputs_tensors) {
let extracted = try_extract_to_f32(tensor)?;
let view = extracted.view();
let owned = view.to_owned();
let dimensionality = owned.into_dimensionality::<IxDyn>()?;
output_map.insert(name.to_string(), dimensionality);
}
let logits = output_map.remove("logits").unwrap();
let exps = logits.mapv(|x: f32| x.exp());
let sum = exps
.sum_axis(Axis(logits.ndim() - 1))
.insert_axis(Axis(logits.ndim() - 1));
let softmax = exps / sum;
Ok(softmax.into_dimensionality()?)
}
fn prepare_input_map(
&self,
input_ids: Array2<u32>,
attention_mask: Array2<u32>,
token_type_ids: Option<Array2<u32>>,
) -> Result<HashMap<String, InputTensor>> {
let mut input_map = HashMap::<String, InputTensor>::new();
if self.token_type_support {
if let Some(token_types_array) = token_type_ids {
input_map.insert(
"token_type_ids".to_string(),
InputTensor::from_array(token_types_array.into_dimensionality()?),
);
} else {
input_map.insert(
"token_type_ids".to_string(),
InputTensor::from_array(
Array2::<u32>::zeros((input_ids.nrows(), input_ids.ncols()))
.into_dimensionality()?,
),
);
}
}
input_map.insert(
"input_ids".to_string(),
InputTensor::from_array(input_ids.into_dimensionality()?),
);
input_map.insert(
"attention_mask".to_string(),
InputTensor::from_array(attention_mask.into_dimensionality()?),
);
Ok(input_map)
}
pub fn get_token_type_support(&self) -> bool {
self.token_type_support
}
pub fn get_num_labels(&self) -> usize {
self.num_labels
}
pub fn is_token_classification(&self) -> bool {
self.is_tok_classification
}
}
#[cfg(test)]
mod tests {
use crate::hf_hub::hf_hub_download;
use super::*;
#[test]
fn test_seq_classify() {
let env = Environment::builder().build().unwrap();
let bert = ClassificationModel::new_from_file(
env.into_arc(),
hf_hub_download(
"npc-engine/deberta-v3-small-finetuned-hate_speech18",
"model.onnx",
None,
None,
)
.unwrap(),
Device::CPU,
GraphOptimizationLevel::Disable,
)
.unwrap();
let input_ids1 =
Array2::from_shape_vec((1, 8), vec![101, 2000, 1037, 1037, 1037, 1037, 1037, 102])
.unwrap();
let attention_mask = Array2::from_shape_vec((1, 8), vec![1, 1, 1, 1, 1, 1, 1, 1]).unwrap();
let scores = bert.forward(input_ids1, attention_mask, None).unwrap();
let allzeros = (scores.sum_axis(Axis(1)) - 1.0).mapv(|x| x.abs());
assert_eq!(scores.len(), 4);
assert!(allzeros.fold(true, |all_true, x| all_true && (*x < 1e-6)));
}
#[test]
fn test_tok_classify() {
let env = Environment::builder().build().unwrap();
let bert = ClassificationModel::new_from_file(
env.into_arc(),
hf_hub_download("optimum/bert-base-NER", "model.onnx", None, None).unwrap(),
Device::CPU,
GraphOptimizationLevel::Disable,
)
.unwrap();
let input_ids1 =
Array2::from_shape_vec((1, 8), vec![101, 2000, 1037, 1037, 1037, 1037, 1037, 102])
.unwrap();
let attention_mask = Array2::from_shape_vec((1, 8), vec![1, 1, 1, 1, 1, 1, 1, 1]).unwrap();
let scores = bert.forward(input_ids1, attention_mask, None).unwrap();
let allzeros = (scores.sum_axis(Axis(2)) - 1.0).mapv(|x| x.abs());
assert!(allzeros.fold(true, |all_true, x| all_true && (*x < 1e-6)));
}
}