huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-19 11:56:45 +00:00
Code Issues Packages Projects Releases Wiki Activity

Use candle_nn::embedding instead of local copies in a few models. (#1562)

Browse Source
This commit is contained in:
Jani Monoses
2024-01-10 22:36:27 +02:00
committed by GitHub
parent d3bdd788cf
commit 63944714f2
5 changed files with 6 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
candle-transformers/src/models/llama.rs
View File

@ -1,6 +1,6 @@
use super::with_tracing::{linear_no_bias as linear, Linear};
use candle::{DType, Device, IndexOp, Result, Tensor, D};
use candle_nn::{Embedding, Module, VarBuilder};
use candle_nn::{embedding, Embedding, Module, VarBuilder};
use serde::Deserialize;
use std::collections::HashMap;
use std::sync::{Arc, Mutex};
@ -136,11 +136,6 @@ impl Cache {
}
}
fn embedding(cfg: &Config, vb: VarBuilder) -> Result<Embedding> {
let embeddings = vb.get((cfg.vocab_size, cfg.hidden_size), "weight")?;
Ok(Embedding::new(embeddings, cfg.hidden_size))
}
struct RmsNorm {
inner: candle_nn::RmsNorm,
span: tracing::Span,
@ -409,7 +404,7 @@ impl Llama {
}
pub fn load(vb: VarBuilder, cache: &Cache, cfg: &Config) -> Result<Self> {
let wte = embedding(cfg, vb.pp("model.embed_tokens"))?;
let wte = embedding(cfg.vocab_size, cfg.hidden_size, vb.pp("model.embed_tokens"))?;
let lm_head = linear(cfg.hidden_size, cfg.vocab_size, vb.pp("lm_head"))?;
let ln_f = RmsNorm::load(cfg.hidden_size, cfg.rms_norm_eps, vb.pp("model.norm"))?;
let blocks: Vec<_> = (0..cfg.num_hidden_layers)