use crate::model_config;
use super::traits::*;
use anyhow::Result;
use serde::{Serialize, Deserialize};
model_config!(OPTConfig {
vocab_size: usize = 50272,
hidden_size: usize = 768,
intermediate_size: usize = 3072,
num_hidden_layers: usize = 12,
num_attention_heads: usize = 12,
num_key_value_heads: usize = 12,
hidden_act: String = "relu".to_string(),
max_position_embeddings: usize = 2048,
initializer_range: f32 = 0.02,
layer_norm_eps: f32 = 1e-5,
use_cache: bool = true,
pad_token_id: i64 = 1,
bos_token_id: i64 = 2,
eos_token_id: i64 = 2,
tie_word_embeddings: bool = false,
word_embed_proj_dim: usize = 768,
do_layer_norm_before: bool = true,
});
impl OPTConfig {
pub fn from_gguf_config(gguf: &crate::weight_loader_core::GGUFModelConfig) -> Self {
Self {
vocab_size: gguf.vocab_size,
hidden_size: gguf.hidden_size,
intermediate_size: gguf.intermediate_size,
num_hidden_layers: gguf.num_hidden_layers,
num_attention_heads: gguf.num_attention_heads,
num_key_value_heads: gguf.num_key_value_heads,
max_position_embeddings: gguf.max_position_embeddings,
..Default::default()
}
}
}
pub struct OPTModelV2 {
config: OPTConfig,
device: Device,
embed_tokens: Tensor,
embed_positions: Tensor,
project_in: Option<Tensor>,
project_out: Option<Tensor>,
layers: Vec<OPTLayer>,
final_layer_norm: Tensor,
lm_head: Tensor,
}
pub struct OPTLayer {
self_attn: OPTAttention,
fc1: Tensor,
fc2: Tensor,
self_attn_layer_norm: Tensor,
final_layer_norm: Tensor,
do_layer_norm_before: bool,
}
pub struct OPTAttention {
q_proj: Tensor,
k_proj: Tensor,
v_proj: Tensor,
out_proj: Tensor,
num_heads: usize,
head_dim: usize,
scale: f32,
}
impl Model for OPTModelV2 {
type Config = OPTConfig;
fn new(config: OPTConfig) -> Result<Self> {
let device = Device::CPU;
let embed_tokens = ops_fn::zeros(&[config.vocab_size, config.word_embed_proj_dim], DataType::Float32, &device)?;
let embed_positions = ops_fn::zeros(&[config.max_position_embeddings + 2, config.hidden_size], DataType::Float32, &device)?;
let project_in = if config.word_embed_proj_dim != config.hidden_size {
Some(ops_fn::zeros(&[config.word_embed_proj_dim, config.hidden_size], DataType::Float32, &device)?)
} else { None };
let project_out = if config.word_embed_proj_dim != config.hidden_size {
Some(ops_fn::zeros(&[config.hidden_size, config.word_embed_proj_dim], DataType::Float32, &device)?)
} else { None };
let final_layer_norm = ops_fn::zeros(&[config.hidden_size], DataType::Float32, &device)?;
let lm_head = ops_fn::zeros(&[config.word_embed_proj_dim, config.vocab_size], DataType::Float32, &device)?;
let mut layers = Vec::with_capacity(config.num_hidden_layers);
for _ in 0..config.num_hidden_layers {
layers.push(OPTLayer::new(&config, &device)?);
}
Ok(Self { config, device, embed_tokens, embed_positions, project_in, project_out, layers, final_layer_norm, lm_head })
}
fn from_weights(config: OPTConfig, weights: ModelWeights) -> Result<Self> {
let mut model = Self::new(config)?;
if let Some(w) = weights.get("model.decoder.embed_tokens.weight") { model.embed_tokens = w.clone(); }
if let Some(w) = weights.get("model.decoder.embed_positions.weight") { model.embed_positions = w.clone(); }
if let Some(w) = weights.get("model.decoder.project_in.weight") { model.project_in = Some(ops_fn::transpose(w)?); }
if let Some(w) = weights.get("model.decoder.project_out.weight") { model.project_out = Some(ops_fn::transpose(w)?); }
if let Some(w) = weights.get("model.decoder.final_layer_norm.weight") { model.final_layer_norm = w.clone(); }
if let Some(w) = weights.get("lm_head.weight") { model.lm_head = ops_fn::transpose(w)?; }
for (i, layer) in model.layers.iter_mut().enumerate() { layer.load_weights(&weights, i)?; }
Ok(model)
}
fn forward(&self, inputs: &ModelInputs) -> Result<ModelOutputs> {
match inputs {
ModelInputs::Text { input_ids, .. } => {
let shape = input_ids.shape();
let seq_len = shape[1];
let mut hidden = ops_fn::embedding(input_ids, &self.embed_tokens)?;
if let Some(ref proj) = self.project_in {
hidden = ops_fn::matmul(&hidden, proj)?;
}
let positions: Vec<i64> = (2..(seq_len + 2) as i64).collect();
let pos_tensor = Tensor::from_i64_slice(&positions, &[1, seq_len], &self.device)?;
let pos_embeds = ops_fn::embedding(&pos_tensor, &self.embed_positions)?;
hidden = ops_fn::add(&hidden, &pos_embeds)?;
for layer in &self.layers {
hidden = layer.forward(&hidden)?;
}
if self.config.do_layer_norm_before {
hidden = ops_fn::layer_norm(&hidden, &self.final_layer_norm, None, self.config.layer_norm_eps)?;
}
if let Some(ref proj) = self.project_out {
hidden = ops_fn::matmul(&hidden, proj)?;
}
let logits = ops_fn::matmul(&hidden, &self.lm_head)?;
Ok(ModelOutputs::Logits { logits, hidden_states: None })
}
_ => Err(anyhow::anyhow!("OPT only supports text inputs")),
}
}
fn generate(&self, prompt: &str, config: &GenerationConfig) -> Result<String> {
use crate::tokenizer::Tokenizer;
use rand::Rng;
let tokenizer = Tokenizer::new();
let mut tokens: Vec<u32> = tokenizer.encode(prompt);
for _ in 0..config.max_new_tokens {
let tokens_i64: Vec<i64> = tokens.iter().map(|&t| t as i64).collect();
let input = Tensor::from_i64_slice(&tokens_i64, &[1, tokens.len()], &self.device)?;
let outputs = self.forward(&ModelInputs::text(input))?;
let logits = match outputs { ModelOutputs::Logits { logits, .. } => logits, _ => return Err(anyhow::anyhow!("Expected logits")) };
let logits_candle = logits.to_candle()?;
let last = logits_candle.narrow(1, logits_candle.dims()[1] - 1, 1)?.squeeze(1)?.squeeze(0)?;
let logits_vec: Vec<f32> = last.to_vec1()?;
let next = if config.do_sample && config.temperature > 0.0 {
let scaled: Vec<f32> = logits_vec.iter().map(|&x| x / config.temperature).collect();
let max_v = scaled.iter().cloned().fold(f32::NEG_INFINITY, f32::max);
let exp_sum: f32 = scaled.iter().map(|&x| (x - max_v).exp()).sum();
let probs: Vec<f32> = scaled.iter().map(|&x| (x - max_v).exp() / exp_sum).collect();
let mut rng = rand::thread_rng();
let r: f32 = rng.gen();
let mut cum = 0.0;
let mut s = 0u32;
for (i, &p) in probs.iter().enumerate() { cum += p; if r <= cum { s = i as u32; break; } }
s
} else {
logits_vec.iter().enumerate().max_by(|a, b| a.1.partial_cmp(b.1).unwrap()).map(|(i, _)| i as u32).unwrap_or(0)
};
if next == config.eos_token_id { break; }
tokens.push(next);
}
Ok(tokenizer.decode(&tokens))
}
fn config(&self) -> &Self::Config { &self.config }
fn memory_requirements(&self) -> MemoryRequirements {
let p = self.config.vocab_size * self.config.hidden_size + self.config.num_hidden_layers * 8 * self.config.hidden_size.pow(2);
MemoryRequirements { gpu_memory: p * 4, cpu_memory: p, kv_cache_memory: 2 * self.config.num_hidden_layers * self.config.max_position_embeddings * self.config.hidden_size * 4, peak_memory: p * 5 }
}
fn to_device(&mut self, device: &Device) -> Result<()> {
self.embed_tokens = self.embed_tokens.to_device(device)?;
self.embed_positions = self.embed_positions.to_device(device)?;
if let Some(ref mut p) = self.project_in { *p = p.to_device(device)?; }
if let Some(ref mut p) = self.project_out { *p = p.to_device(device)?; }
self.final_layer_norm = self.final_layer_norm.to_device(device)?;
self.lm_head = self.lm_head.to_device(device)?;
for l in &mut self.layers { l.to_device(device)?; }
self.device = device.clone();
Ok(())
}
}
impl OPTLayer {
fn new(config: &OPTConfig, device: &Device) -> Result<Self> {
Ok(Self {
self_attn: OPTAttention::new(config, device)?,
fc1: ops_fn::zeros(&[config.hidden_size, config.intermediate_size], DataType::Float32, device)?,
fc2: ops_fn::zeros(&[config.intermediate_size, config.hidden_size], DataType::Float32, device)?,
self_attn_layer_norm: ops_fn::zeros(&[config.hidden_size], DataType::Float32, device)?,
final_layer_norm: ops_fn::zeros(&[config.hidden_size], DataType::Float32, device)?,
do_layer_norm_before: config.do_layer_norm_before,
})
}
fn forward(&self, hidden_states: &Tensor) -> Result<Tensor> {
let residual = hidden_states.clone();
let h = if self.do_layer_norm_before {
ops_fn::layer_norm(hidden_states, &self.self_attn_layer_norm, None, 1e-5)?
} else { hidden_states.clone() };
let attn_out = self.self_attn.forward(&h)?;
let h = ops_fn::add(&residual, &attn_out)?;
let h = if !self.do_layer_norm_before {
ops_fn::layer_norm(&h, &self.self_attn_layer_norm, None, 1e-5)?
} else { h };
let residual = h.clone();
let h = if self.do_layer_norm_before {
ops_fn::layer_norm(&h, &self.final_layer_norm, None, 1e-5)?
} else { h };
let fc1_out = ops_fn::matmul(&h, &self.fc1)?;
let activated = {
let x = fc1_out.to_candle()?;
Tensor::from_candle(x.relu()?)
};
let fc2_out = ops_fn::matmul(&activated, &self.fc2)?;
let h = ops_fn::add(&residual, &fc2_out)?;
if !self.do_layer_norm_before {
ops_fn::layer_norm(&h, &self.final_layer_norm, None, 1e-5)
} else { Ok(h) }
}
fn load_weights(&mut self, weights: &ModelWeights, idx: usize) -> Result<()> {
let p = format!("model.decoder.layers.{}", idx);
if let Some(w) = weights.get(&format!("{}.self_attn.q_proj.weight", p)) { self.self_attn.q_proj = ops_fn::transpose(w)?; }
if let Some(w) = weights.get(&format!("{}.self_attn.k_proj.weight", p)) { self.self_attn.k_proj = ops_fn::transpose(w)?; }
if let Some(w) = weights.get(&format!("{}.self_attn.v_proj.weight", p)) { self.self_attn.v_proj = ops_fn::transpose(w)?; }
if let Some(w) = weights.get(&format!("{}.self_attn.out_proj.weight", p)) { self.self_attn.out_proj = ops_fn::transpose(w)?; }
if let Some(w) = weights.get(&format!("{}.fc1.weight", p)) { self.fc1 = ops_fn::transpose(w)?; }
if let Some(w) = weights.get(&format!("{}.fc2.weight", p)) { self.fc2 = ops_fn::transpose(w)?; }
if let Some(w) = weights.get(&format!("{}.self_attn_layer_norm.weight", p)) { self.self_attn_layer_norm = w.clone(); }
if let Some(w) = weights.get(&format!("{}.final_layer_norm.weight", p)) { self.final_layer_norm = w.clone(); }
Ok(())
}
fn to_device(&mut self, device: &Device) -> Result<()> {
self.self_attn.to_device(device)?;
self.fc1 = self.fc1.to_device(device)?;
self.fc2 = self.fc2.to_device(device)?;
self.self_attn_layer_norm = self.self_attn_layer_norm.to_device(device)?;
self.final_layer_norm = self.final_layer_norm.to_device(device)?;
Ok(())
}
}
impl OPTAttention {
fn new(config: &OPTConfig, device: &Device) -> Result<Self> {
let head_dim = config.hidden_size / config.num_attention_heads;
Ok(Self {
q_proj: ops_fn::zeros(&[config.hidden_size, config.hidden_size], DataType::Float32, device)?,
k_proj: ops_fn::zeros(&[config.hidden_size, config.hidden_size], DataType::Float32, device)?,
v_proj: ops_fn::zeros(&[config.hidden_size, config.hidden_size], DataType::Float32, device)?,
out_proj: ops_fn::zeros(&[config.hidden_size, config.hidden_size], DataType::Float32, device)?,
num_heads: config.num_attention_heads,
head_dim,
scale: 1.0 / (head_dim as f32).sqrt(),
})
}
fn forward(&self, hidden_states: &Tensor) -> Result<Tensor> {
let shape = hidden_states.shape();
let (batch, seq_len, _) = (shape[0], shape[1], shape[2]);
let q = ops_fn::matmul(hidden_states, &self.q_proj)?.to_candle()?.reshape(&[batch, seq_len, self.num_heads, self.head_dim])?.transpose(1, 2)?;
let k = ops_fn::matmul(hidden_states, &self.k_proj)?.to_candle()?.reshape(&[batch, seq_len, self.num_heads, self.head_dim])?.transpose(1, 2)?;
let v = ops_fn::matmul(hidden_states, &self.v_proj)?.to_candle()?.reshape(&[batch, seq_len, self.num_heads, self.head_dim])?.transpose(1, 2)?;
let q = q.contiguous()?;
let k_t = k.transpose(2, 3)?.contiguous()?;
let scores = (q.matmul(&k_t)? * (self.scale as f64))?;
let device = scores.device();
let mask = {
let mut m = vec![0.0f32; seq_len * seq_len];
for i in 0..seq_len { for j in (i+1)..seq_len { m[i*seq_len+j] = f32::NEG_INFINITY; } }
candle_core::Tensor::from_vec(m, &[1, 1, seq_len, seq_len], device)?
};
let v = v.contiguous()?;
let attn = candle_nn::ops::softmax_last_dim(&scores.broadcast_add(&mask)?)?.matmul(&v)?;
let out = attn.transpose(1, 2)?.reshape(&[batch, seq_len, self.num_heads * self.head_dim])?;
ops_fn::matmul(&Tensor::from_candle(out), &self.out_proj)
}
fn to_device(&mut self, device: &Device) -> Result<()> {
self.q_proj = self.q_proj.to_device(device)?;
self.k_proj = self.k_proj.to_device(device)?;
self.v_proj = self.v_proj.to_device(device)?;
self.out_proj = self.out_proj.to_device(device)?;
Ok(())
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_opt_creation() {
let config = OPTConfig { vocab_size: 1000, hidden_size: 128, intermediate_size: 512, num_hidden_layers: 2, num_attention_heads: 4, num_key_value_heads: 4, ..Default::default() };
let model = OPTModelV2::new(config).unwrap();
assert_eq!(model.config().vocab_size(), 1000);
}
#[test]
fn test_opt_forward() {
let config = OPTConfig { vocab_size: 100, hidden_size: 64, intermediate_size: 256, num_hidden_layers: 1, num_attention_heads: 4, num_key_value_heads: 4, word_embed_proj_dim: 64, ..Default::default() };
let model = OPTModelV2::new(config).unwrap();
let inputs = ModelInputs::text(ops_fn::zeros(&[2, 8], DataType::Int64, &Device::CPU).unwrap());
match model.forward(&inputs).unwrap() { ModelOutputs::Logits { logits, .. } => assert_eq!(logits.shape(), &[2, 8, 100]), _ => panic!() }
}
}