use realizar::gguf::{MappedGGUFModel, OwnedQKVWeights, OwnedQuantizedModel};
use realizar::quantize::{fused_q4k_parallel_matvec, fused_q6k_parallel_matvec};
use realizar::rms_norm;
const GGUF_TYPE_Q4_K: u32 = 12;
const GGUF_TYPE_Q6_K: u32 = 14;
fn l2_norm(v: &[f32]) -> f32 {
(v.iter().map(|x| x * x).sum::<f32>()).sqrt()
}
fn silu(x: &mut [f32]) {
for v in x.iter_mut() {
*v = *v / (1.0 + (-*v).exp());
}
}
fn fused_matmul(input: &[f32], data: &[u8], qtype: u32, in_dim: usize, out_dim: usize) -> Vec<f32> {
match qtype {
GGUF_TYPE_Q4_K => fused_q4k_parallel_matvec(data, input, in_dim, out_dim).expect("test"),
GGUF_TYPE_Q6_K => fused_q6k_parallel_matvec(data, input, in_dim, out_dim).expect("test"),
_ => panic!("Unsupported qtype"),
}
}
fn main() {
let path = "/tmp/parity-bench/tinyllama-1.1b-q4_k_m.gguf";
let mapped = MappedGGUFModel::from_path(path).expect("Failed");
let model = OwnedQuantizedModel::from_mapped(&mapped).expect("test");
let hidden_dim = model.config().hidden_dim;
let intermediate_dim = model.config().intermediate_dim;
let eps = model.config().eps;
let start = 450 * hidden_dim;
let mut hidden: Vec<f32> = model.token_embedding()[start..start + hidden_dim].to_vec();
for layer_idx in 0..21 {
let layer = &model.layers()[layer_idx];
let normed = rms_norm(&hidden, &layer.attn_norm_weight, eps);
let OwnedQKVWeights::Separate {
q: q_weight,
k: k_weight,
v: v_weight,
} = &layer.qkv_weight
else {
panic!("Expected separate")
};
let _q = fused_matmul(
&normed,
&q_weight.data,
q_weight.qtype,
q_weight.in_dim,
q_weight.out_dim,
);
let _k = fused_matmul(
&normed,
&k_weight.data,
k_weight.qtype,
k_weight.in_dim,
k_weight.out_dim,
);
let v = fused_matmul(
&normed,
&v_weight.data,
v_weight.qtype,
v_weight.in_dim,
v_weight.out_dim,
);
let head_dim = hidden_dim / model.config().num_heads;
let group_size = model.config().num_heads / model.config().num_kv_heads;
let mut attn_out = Vec::with_capacity(hidden_dim);
for h in 0..model.config().num_heads {
let kv_head = h / group_size;
let start = kv_head * head_dim;
attn_out.extend_from_slice(&v[start..start + head_dim]);
}
let attn_proj = fused_matmul(
&attn_out,
&layer.attn_output_weight.data,
layer.attn_output_weight.qtype,
layer.attn_output_weight.in_dim,
layer.attn_output_weight.out_dim,
);
for i in 0..hidden_dim {
hidden[i] += attn_proj[i];
}
let ffn_input = if let Some(ref norm) = layer.ffn_norm_weight {
rms_norm(&hidden, norm, eps)
} else {
hidden.clone()
};
if let Some(ref gate_weight) = layer.ffn_gate_weight {
let ffn_up = fused_matmul(
&ffn_input,
&layer.ffn_up_weight.data,
layer.ffn_up_weight.qtype,
layer.ffn_up_weight.in_dim,
layer.ffn_up_weight.out_dim,
);
let mut ffn_gate = fused_matmul(
&ffn_input,
&gate_weight.data,
gate_weight.qtype,
gate_weight.in_dim,
gate_weight.out_dim,
);
silu(&mut ffn_gate);
let mut ffn_hidden = vec![0.0f32; intermediate_dim];
for i in 0..intermediate_dim {
ffn_hidden[i] = ffn_gate[i] * ffn_up[i];
}
let ffn_out = fused_matmul(
&ffn_hidden,
&layer.ffn_down_weight.data,
layer.ffn_down_weight.qtype,
layer.ffn_down_weight.in_dim,
layer.ffn_down_weight.out_dim,
);
for i in 0..hidden_dim {
hidden[i] += ffn_out[i];
}
}
}
println!("=== Layer 21 Detailed Trace ===\n");
println!("Input (after layer 20): L2={:.4}", l2_norm(&hidden));
println!(" First 5: {:?}", &hidden[..5]);
let layer = &model.layers()[21];
let normed = rms_norm(&hidden, &layer.attn_norm_weight, eps);
println!("\nAfter attn RMSNorm: L2={:.4}", l2_norm(&normed));
let OwnedQKVWeights::Separate {
q: q_weight,
k: k_weight,
v: v_weight,
} = &layer.qkv_weight
else {
panic!("Expected separate")
};
let q = fused_matmul(
&normed,
&q_weight.data,
q_weight.qtype,
q_weight.in_dim,
q_weight.out_dim,
);
let _k = fused_matmul(
&normed,
&k_weight.data,
k_weight.qtype,
k_weight.in_dim,
k_weight.out_dim,
);
let v = fused_matmul(
&normed,
&v_weight.data,
v_weight.qtype,
v_weight.in_dim,
v_weight.out_dim,
);
println!("Q L2: {:.4}", l2_norm(&q));
println!("V L2: {:.4}", l2_norm(&v));
let head_dim = hidden_dim / model.config().num_heads;
let group_size = model.config().num_heads / model.config().num_kv_heads;
let mut attn_out = Vec::with_capacity(hidden_dim);
for h in 0..model.config().num_heads {
let kv_head = h / group_size;
let start = kv_head * head_dim;
attn_out.extend_from_slice(&v[start..start + head_dim]);
}
println!("Attn out (GQA expanded V): L2={:.4}", l2_norm(&attn_out));
let attn_proj = fused_matmul(
&attn_out,
&layer.attn_output_weight.data,
layer.attn_output_weight.qtype,
layer.attn_output_weight.in_dim,
layer.attn_output_weight.out_dim,
);
println!("After attn output proj: L2={:.4}", l2_norm(&attn_proj));
let pre_attn = l2_norm(&hidden);
for i in 0..hidden_dim {
hidden[i] += attn_proj[i];
}
println!(
"After attn residual: L2={:.4} (was {:.4})",
l2_norm(&hidden),
pre_attn
);
let ffn_input = if let Some(ref norm) = layer.ffn_norm_weight {
rms_norm(&hidden, norm, eps)
} else {
hidden.clone()
};
println!("\nAfter FFN RMSNorm: L2={:.4}", l2_norm(&ffn_input));
if let Some(ref gate_weight) = layer.ffn_gate_weight {
let ffn_up = fused_matmul(
&ffn_input,
&layer.ffn_up_weight.data,
layer.ffn_up_weight.qtype,
layer.ffn_up_weight.in_dim,
layer.ffn_up_weight.out_dim,
);
println!("FFN up: L2={:.4}", l2_norm(&ffn_up));
let mut ffn_gate = fused_matmul(
&ffn_input,
&gate_weight.data,
gate_weight.qtype,
gate_weight.in_dim,
gate_weight.out_dim,
);
println!("FFN gate (pre-silu): L2={:.4}", l2_norm(&ffn_gate));
silu(&mut ffn_gate);
println!("FFN gate (post-silu): L2={:.4}", l2_norm(&ffn_gate));
let mut ffn_hidden = vec![0.0f32; intermediate_dim];
for i in 0..intermediate_dim {
ffn_hidden[i] = ffn_gate[i] * ffn_up[i];
}
println!("FFN hidden (gate*up): L2={:.4}", l2_norm(&ffn_hidden));
let ffn_out = fused_matmul(
&ffn_hidden,
&layer.ffn_down_weight.data,
layer.ffn_down_weight.qtype,
layer.ffn_down_weight.in_dim,
layer.ffn_down_weight.out_dim,
);
println!("FFN down: L2={:.4}", l2_norm(&ffn_out));
let pre_ffn = l2_norm(&hidden);
for i in 0..hidden_dim {
hidden[i] += ffn_out[i];
}
println!(
"\nAfter FFN residual: L2={:.4} (was {:.4})",
l2_norm(&hidden),
pre_ffn
);
}
println!("\n=== Final output: L2={:.4} ===", l2_norm(&hidden));
println!("\nHuggingFace reference:");
println!(" Layer 21 output L2: 8.0829");
}