use crate::model_config;
use super::traits::*;
use anyhow::Result;
use serde::{Serialize, Deserialize};
model_config!(ArcticConfig {
vocab_size: usize = 32000,
hidden_size: usize = 4096,
intermediate_size: usize = 11008,
moe_intermediate_size: usize = 2816,
num_hidden_layers: usize = 32,
num_attention_heads: usize = 32,
num_key_value_heads: usize = 8,
hidden_act: String = "silu".to_string(),
max_position_embeddings: usize = 4096,
rms_norm_eps: f32 = 1e-5,
use_cache: bool = true,
pad_token_id: i64 = 0,
bos_token_id: i64 = 1,
eos_token_id: i64 = 2,
tie_word_embeddings: bool = false,
rope_theta: f32 = 10000.0,
num_experts: usize = 128,
num_experts_per_tok: usize = 2,
});
impl ArcticConfig {
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,
rope_theta: gguf.rope_theta,
..Default::default()
}
}
}
pub struct ArcticModelV2 {
config: ArcticConfig,
device: Device,
embed_tokens: Tensor,
layers: Vec<ArcticLayer>,
norm: Tensor,
lm_head: Tensor,
}
pub struct ArcticLayer {
self_attn: ArcticAttention,
dense_mlp: ArcticMLP,
moe: ArcticMoE,
input_layernorm: Tensor,
post_attention_layernorm: Tensor,
}
pub struct ArcticAttention {
q_proj: Tensor,
k_proj: Tensor,
v_proj: Tensor,
o_proj: Tensor,
num_heads: usize,
num_kv_heads: usize,
head_dim: usize,
scale: f32,
}
pub struct ArcticMLP {
gate_proj: Tensor,
up_proj: Tensor,
down_proj: Tensor,
}
pub struct ArcticMoE {
router: Tensor,
experts: Vec<ArcticExpert>,
num_experts_per_tok: usize,
}
pub struct ArcticExpert {
gate_proj: Tensor,
up_proj: Tensor,
down_proj: Tensor,
}
fn apply_rope_arctic(
q: &candle_core::Tensor, k: &candle_core::Tensor, seq_len: usize, head_dim: usize, rope_theta: f32,
) -> Result<(candle_core::Tensor, candle_core::Tensor)> {
let device = q.device();
let half_dim = head_dim / 2;
let inv_freq: Vec<f32> = (0..half_dim).map(|i| 1.0 / rope_theta.powf((2 * i) as f32 / head_dim as f32)).collect();
let positions: Vec<f32> = (0..seq_len).map(|p| p as f32).collect();
let mut angles = Vec::with_capacity(seq_len * half_dim);
for pos in &positions { for freq in &inv_freq { angles.push(pos * freq); } }
let angles_tensor = candle_core::Tensor::from_vec(angles, &[seq_len, half_dim], device)?;
let cos = angles_tensor.cos()?.unsqueeze(0)?.unsqueeze(0)?;
let sin = angles_tensor.sin()?.unsqueeze(0)?.unsqueeze(0)?;
let (q_half1, q_half2) = (q.narrow(3, 0, half_dim)?, q.narrow(3, half_dim, half_dim)?);
let (k_half1, k_half2) = (k.narrow(3, 0, half_dim)?, k.narrow(3, half_dim, half_dim)?);
Ok((
candle_core::Tensor::cat(&[&(q_half1.broadcast_mul(&cos)? - q_half2.broadcast_mul(&sin)?)?, &(q_half1.broadcast_mul(&sin)? + q_half2.broadcast_mul(&cos)?)?], 3)?,
candle_core::Tensor::cat(&[&(k_half1.broadcast_mul(&cos)? - k_half2.broadcast_mul(&sin)?)?, &(k_half1.broadcast_mul(&sin)? + k_half2.broadcast_mul(&cos)?)?], 3)?
))
}
impl Model for ArcticModelV2 {
type Config = ArcticConfig;
fn new(config: ArcticConfig) -> Result<Self> {
let device = Device::CPU;
let embed_tokens = ops_fn::zeros(&[config.vocab_size, config.hidden_size], DataType::Float32, &device)?;
let norm = ops_fn::zeros(&[config.hidden_size], DataType::Float32, &device)?;
let lm_head = ops_fn::zeros(&[config.vocab_size, config.hidden_size], DataType::Float32, &device)?;
let mut layers = Vec::with_capacity(config.num_hidden_layers);
for _ in 0..config.num_hidden_layers { layers.push(ArcticLayer::new(&config, &device)?); }
Ok(Self { config, device, embed_tokens, layers, norm, lm_head })
}
fn from_weights(config: ArcticConfig, weights: ModelWeights) -> Result<Self> {
let mut model = Self::new(config)?;
if let Some(w) = weights.get("model.embed_tokens.weight") { model.embed_tokens = w.clone(); }
if let Some(w) = weights.get("model.norm.weight") { model.norm = w.clone(); }
if let Some(w) = weights.get("lm_head.weight") { model.lm_head = w.clone(); }
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 seq_len = input_ids.shape()[1];
let mut hidden = ops_fn::embedding(input_ids, &self.embed_tokens)?;
for layer in &self.layers { hidden = layer.forward(&hidden, seq_len, self.config.rope_theta)?; }
hidden = ops_fn::rms_norm(&hidden, &self.norm, self.config.rms_norm_eps)?;
let logits = ops_fn::matmul(&hidden, &ops_fn::transpose(&self.lm_head)?)?;
Ok(ModelOutputs::Logits { logits, hidden_states: None })
}
_ => Err(anyhow::anyhow!("Arctic 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.norm = self.norm.to_device(device)?;
self.lm_head = self.lm_head.to_device(device)?;
for layer in &mut self.layers { layer.to_device(device)?; }
self.device = device.clone();
Ok(())
}
}
impl ArcticLayer {
fn new(config: &ArcticConfig, device: &Device) -> Result<Self> {
Ok(Self {
self_attn: ArcticAttention::new(config, device)?,
dense_mlp: ArcticMLP::new(config.hidden_size, config.intermediate_size, device)?,
moe: ArcticMoE::new(config, device)?,
input_layernorm: ops_fn::zeros(&[config.hidden_size], DataType::Float32, device)?,
post_attention_layernorm: ops_fn::zeros(&[config.hidden_size], DataType::Float32, device)?,
})
}
fn forward(&self, hidden_states: &Tensor, seq_len: usize, rope_theta: f32) -> Result<Tensor> {
let residual = hidden_states.clone();
let h = ops_fn::rms_norm(hidden_states, &self.input_layernorm, 1e-5)?;
let attn_out = self.self_attn.forward(&h, seq_len, rope_theta)?;
let h = ops_fn::add(&residual, &attn_out)?;
let residual = h.clone();
let h = ops_fn::rms_norm(&h, &self.post_attention_layernorm, 1e-5)?;
let dense_out = self.dense_mlp.forward(&h)?;
let moe_out = self.moe.forward(&h)?;
let mlp_out = ops_fn::add(&dense_out, &moe_out)?;
ops_fn::add(&residual, &mlp_out)
}
fn load_weights(&mut self, weights: &ModelWeights, idx: usize) -> Result<()> {
let p = format!("model.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.o_proj.weight", p)) { self.self_attn.o_proj = ops_fn::transpose(w)?; }
if let Some(w) = weights.get(&format!("{}.input_layernorm.weight", p)) { self.input_layernorm = w.clone(); }
if let Some(w) = weights.get(&format!("{}.post_attention_layernorm.weight", p)) { self.post_attention_layernorm = w.clone(); }
Ok(())
}
fn to_device(&mut self, device: &Device) -> Result<()> {
self.self_attn.to_device(device)?;
self.dense_mlp.to_device(device)?;
self.moe.to_device(device)?;
self.input_layernorm = self.input_layernorm.to_device(device)?;
self.post_attention_layernorm = self.post_attention_layernorm.to_device(device)?;
Ok(())
}
}
impl ArcticAttention {
fn new(config: &ArcticConfig, 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.num_attention_heads * head_dim], DataType::Float32, device)?,
k_proj: ops_fn::zeros(&[config.hidden_size, config.num_key_value_heads * head_dim], DataType::Float32, device)?,
v_proj: ops_fn::zeros(&[config.hidden_size, config.num_key_value_heads * head_dim], DataType::Float32, device)?,
o_proj: ops_fn::zeros(&[config.num_attention_heads * head_dim, config.hidden_size], DataType::Float32, device)?,
num_heads: config.num_attention_heads, num_kv_heads: config.num_key_value_heads, head_dim,
scale: 1.0 / (head_dim as f32).sqrt(),
})
}
fn forward(&self, hidden_states: &Tensor, seq_len: usize, rope_theta: f32) -> Result<Tensor> {
let shape = hidden_states.shape();
let batch = shape[0];
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_kv_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_kv_heads, self.head_dim])?.transpose(1, 2)?;
let (q, k) = apply_rope_arctic(&q, &k, seq_len, self.head_dim, rope_theta)?;
let num_groups = self.num_heads / self.num_kv_heads;
let (k, v) = if num_groups > 1 {
(k.unsqueeze(2)?.broadcast_as(&[batch, self.num_kv_heads, num_groups, seq_len, self.head_dim])?.reshape(&[batch, self.num_heads, seq_len, self.head_dim])?,
v.unsqueeze(2)?.broadcast_as(&[batch, self.num_kv_heads, num_groups, seq_len, self.head_dim])?.reshape(&[batch, self.num_heads, seq_len, self.head_dim])?)
} else { (k, v) };
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 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; } }
let mask = candle_core::Tensor::from_vec(m, &[1, 1, seq_len, seq_len], device)?;
let scores = scores.broadcast_add(&mask)?;
let v = v.contiguous()?;
let attn = candle_nn::ops::softmax_last_dim(&scores)?.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.o_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.o_proj = self.o_proj.to_device(device)?;
Ok(())
}
}
impl ArcticMLP {
fn new(hidden_size: usize, intermediate_size: usize, device: &Device) -> Result<Self> {
Ok(Self {
gate_proj: ops_fn::zeros(&[hidden_size, intermediate_size], DataType::Float32, device)?,
up_proj: ops_fn::zeros(&[hidden_size, intermediate_size], DataType::Float32, device)?,
down_proj: ops_fn::zeros(&[intermediate_size, hidden_size], DataType::Float32, device)?,
})
}
fn forward(&self, x: &Tensor) -> Result<Tensor> {
let gate = ops_fn::matmul(x, &self.gate_proj)?;
let up = ops_fn::matmul(x, &self.up_proj)?;
let h = ops_fn::mul(&ops_fn::silu(&gate)?, &up)?;
ops_fn::matmul(&h, &self.down_proj)
}
fn to_device(&mut self, device: &Device) -> Result<()> {
self.gate_proj = self.gate_proj.to_device(device)?;
self.up_proj = self.up_proj.to_device(device)?;
self.down_proj = self.down_proj.to_device(device)?;
Ok(())
}
}
impl ArcticMoE {
fn new(config: &ArcticConfig, device: &Device) -> Result<Self> {
let router = ops_fn::zeros(&[config.hidden_size, config.num_experts], DataType::Float32, device)?;
let mut experts = Vec::with_capacity(config.num_experts);
for _ in 0..config.num_experts { experts.push(ArcticExpert::new(config.hidden_size, config.moe_intermediate_size, device)?); }
Ok(Self { router, experts, num_experts_per_tok: config.num_experts_per_tok })
}
fn forward(&self, hidden_states: &Tensor) -> Result<Tensor> {
let shape = hidden_states.shape();
let (batch_size, seq_len, hidden_size) = (shape[0], shape[1], shape[2]);
let num_tokens = batch_size * seq_len;
let k = self.num_experts_per_tok;
let flat_hidden = hidden_states.reshape(&[num_tokens, hidden_size])?;
let router_logits = ops_fn::matmul(&flat_hidden, &self.router)?;
let (topk_weights, topk_indices) = ops_fn::topk(&router_logits, k, -1)?;
let routing_weights = ops_fn::softmax(&topk_weights, -1)?;
let all_indices: Vec<i64> = topk_indices.to_candle()?.flatten_all()?.to_vec1()?;
let all_weights: Vec<f32> = routing_weights.to_candle()?.flatten_all()?.to_vec1()?;
let flat_hidden_candle = flat_hidden.to_candle()?;
let mut output_data = vec![0.0f32; num_tokens * hidden_size];
for tok_idx in 0..num_tokens {
let token_hidden = flat_hidden_candle.get(tok_idx)?;
let token_tensor = Tensor::from_candle(token_hidden.unsqueeze(0)?);
let start = tok_idx * k;
let indices = &all_indices[start..start + k];
let weights = &all_weights[start..start + k];
let mut token_output = ops_fn::zeros(&[1, hidden_size], hidden_states.dtype(), hidden_states.device())?;
for (i, &expert_idx) in indices.iter().enumerate() {
if (expert_idx as usize) < self.experts.len() {
let expert = &self.experts[expert_idx as usize];
let expert_output = expert.forward(&token_tensor)?;
let scaled_output = ops_fn::scale(&expert_output, weights[i])?;
token_output = ops_fn::add(&token_output, &scaled_output)?;
}
}
let token_data: Vec<f32> = token_output.to_candle()?.flatten_all()?.to_vec1()?;
for (i, &v) in token_data.iter().enumerate() { output_data[tok_idx * hidden_size + i] = v; }
}
let output = Tensor::from_f32_slice(&output_data, &[num_tokens, hidden_size], hidden_states.device())?;
output.reshape(&[batch_size, seq_len, hidden_size])
}
fn to_device(&mut self, device: &Device) -> Result<()> {
self.router = self.router.to_device(device)?;
for expert in &mut self.experts { expert.to_device(device)?; }
Ok(())
}
}
impl ArcticExpert {
fn new(hidden_size: usize, intermediate_size: usize, device: &Device) -> Result<Self> {
Ok(Self {
gate_proj: ops_fn::zeros(&[hidden_size, intermediate_size], DataType::Float32, device)?,
up_proj: ops_fn::zeros(&[hidden_size, intermediate_size], DataType::Float32, device)?,
down_proj: ops_fn::zeros(&[intermediate_size, hidden_size], DataType::Float32, device)?,
})
}
fn forward(&self, x: &Tensor) -> Result<Tensor> {
let gate = ops_fn::matmul(x, &self.gate_proj)?;
let up = ops_fn::matmul(x, &self.up_proj)?;
let h = ops_fn::mul(&ops_fn::silu(&gate)?, &up)?;
ops_fn::matmul(&h, &self.down_proj)
}
fn to_device(&mut self, device: &Device) -> Result<()> {
self.gate_proj = self.gate_proj.to_device(device)?;
self.up_proj = self.up_proj.to_device(device)?;
self.down_proj = self.down_proj.to_device(device)?;
Ok(())
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_arctic_creation() {
let config = ArcticConfig { vocab_size: 1000, hidden_size: 64, intermediate_size: 256, moe_intermediate_size: 64, num_hidden_layers: 2, num_attention_heads: 4, num_key_value_heads: 2, num_experts: 4, num_experts_per_tok: 2, ..Default::default() };
let model = ArcticModelV2::new(config).unwrap();
assert_eq!(model.config().vocab_size(), 1000);
}
#[test]
fn test_arctic_forward() {
let config = ArcticConfig { vocab_size: 100, hidden_size: 64, intermediate_size: 256, moe_intermediate_size: 64, num_hidden_layers: 1, num_attention_heads: 4, num_key_value_heads: 2, num_experts: 4, num_experts_per_tok: 2, ..Default::default() };
let model = ArcticModelV2::new(config).unwrap();
let inputs = ModelInputs::text(ops_fn::zeros(&[1, 4], DataType::Int64, &Device::CPU).unwrap());
match model.forward(&inputs).unwrap() { ModelOutputs::Logits { logits, .. } => assert_eq!(logits.shape(), &[1, 4, 100]), _ => panic!() }
}
}