Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix clippy shadowing #446

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
+28 −25
Open

Fix clippy shadowing #446

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
998813f
Deny clippy shadowing
EricLBuehler Jun 18, 2024
d5bc486
Fix shadowing for quantized llama
EricLBuehler Jun 18, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 28 additions & 25 deletions mistralrs-core/src/models/quantized_llama.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -126,6 +126,7 @@ struct LayerWeights {
}

impl LayerWeights {
#[deny(clippy::shadow_same, clippy::shadow_reuse, clippy::shadow_unrelated)]
fn forward_attn(
&mut self,
x: &Tensor,
Expand All @@ -136,13 +137,13 @@ impl LayerWeights {
) -> Result<Tensor> {
let (b_sz, seq_len, n_embd) = x.dims3()?;

let q = MatMul.qmatmul(x, &self.attention_wq)?;
let k = MatMul.qmatmul(x, &self.attention_wk)?;
let v = MatMul.qmatmul(x, &self.attention_wv)?;
let mut q = MatMul.qmatmul(x, &self.attention_wq)?;
let mut k = MatMul.qmatmul(x, &self.attention_wk)?;
let mut v = MatMul.qmatmul(x, &self.attention_wv)?;

let mut q = q.reshape((b_sz * seq_len, self.n_head, self.head_dim))?;
let mut k = k.reshape((b_sz * seq_len, self.n_kv_head, self.head_dim))?;
let v = v
q = q.reshape((b_sz * seq_len, self.n_head, self.head_dim))?;
k = k.reshape((b_sz * seq_len, self.n_kv_head, self.head_dim))?;
v = v
.reshape((b_sz, seq_len, self.n_kv_head, self.head_dim))?
.transpose(1, 2)?;

Expand All @@ -159,12 +160,14 @@ impl LayerWeights {
.transpose(1, 2)?;
}

let (k, v) = Cache::update_kv_cache(kv_cache, k, v, false)?;
let (k_, v_) = Cache::update_kv_cache(kv_cache, k, v, false)?;
k = k_;
v = v_;

let k = repeat_kv(k, self.n_head / self.n_kv_head)?;
let v = repeat_kv(v, self.n_head / self.n_kv_head)?;
k = repeat_kv(k, self.n_head / self.n_kv_head)?;
v = repeat_kv(v, self.n_head / self.n_kv_head)?;

let y = ScaledDotProductAttention.run_attention(
let mut y = ScaledDotProductAttention.run_attention(
&q,
&k,
&v,
Expand All @@ -176,9 +179,8 @@ impl LayerWeights {
seq_len,
)?;

let y = y.transpose(1, 2)?.reshape(&[b_sz, seq_len, n_embd])?;
let y = MatMul.qmatmul(&y, &self.attention_wo)?;
Ok(y)
y = y.transpose(1, 2)?.reshape(&[b_sz, seq_len, n_embd])?;
MatMul.qmatmul(&y, &self.attention_wo)
}
}

Expand Down Expand Up @@ -481,17 +483,18 @@ impl ModelConfig::FromGGUF for ModelWeights {
}

impl ModelWeights {
#[deny(clippy::shadow_same, clippy::shadow_reuse, clippy::shadow_unrelated)]
pub fn forward(
&mut self,
x: &Tensor,
input_ids: &Tensor,
start_offsets: &[usize],
start_offsets_kernel: Tensor,
context_lens: Vec<(usize, usize)>,
) -> Result<Tensor> {
let mut layer_in = self.tok_embeddings.forward(x)?;
let mut layer_in = self.tok_embeddings.forward(input_ids)?;
let mut cache = self.cache.lock();
let mask = CausalMasker.make_causal_mask_as_attn_bias(
x,
input_ids,
&cache,
DType::F32,
self.layers[0].n_head,
Expand All @@ -500,9 +503,9 @@ impl ModelWeights {
if let Some(ref mapper) = self.mapper {
layer_in = mapper.map(layer_in, i)?;
}
let x = layer_in;
let residual = &x;
let x = layer.attention_norm.forward(&x)?;
let mut x = layer_in;
let mut residual = x.clone();
x = layer.attention_norm.forward(&x)?;
let attn = layer.forward_attn(
&x,
mask.as_ref()
Expand All @@ -512,16 +515,16 @@ impl ModelWeights {
start_offsets_kernel.clone(),
&mut cache[i],
)?;
let x = (attn + residual)?;
x = (attn + residual)?;

// MLP
let residual = &x;
let x = layer.ffn_norm.forward(&x)?;
let x = layer.mlp_or_moe.forward(&x)?;
let x = (x + residual)?;
residual = x.clone();
x = layer.ffn_norm.forward(&x)?;
x = layer.mlp_or_moe.forward(&x)?;
x = (x + residual)?;
layer_in = x;
}
let layer_in = layer_in.to_device(&self.device)?;
layer_in = layer_in.to_device(&self.device)?;
let x = self.norm.forward(&layer_in)?;
extract_logits(
&MatMul.qmatmul(&x.contiguous()?, &self.output)?,
Expand Down
Loading