huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 18:48:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Expose the t5 config fields + allow t5-large. (#1987)

Browse Source
This commit is contained in:
Laurent Mazare
2024-04-01 20:58:34 +02:00
committed by GitHub
parent ea0d8d3753
commit be9c200cbb
2 changed files with 18 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
candle-examples/examples/t5/main.rs
View File

@ -22,6 +22,7 @@ const DTYPE: DType = DType::F32;
enum Which { enum Which {
T5Base, T5Base,
T5Small, T5Small,
T5Large,
T5_3B, T5_3B,
Mt5Base, Mt5Base,
Mt5Small, Mt5Small,
@ -108,6 +109,7 @@ impl T5ModelBuilder {
let (default_model, default_revision) = match args.which { let (default_model, default_revision) = match args.which {
Which::T5Base => ("t5-base", "main"), Which::T5Base => ("t5-base", "main"),
Which::T5Small => ("t5-small", "refs/pr/15"), Which::T5Small => ("t5-small", "refs/pr/15"),
Which::T5Large => ("t5-large", "main"),
Which::T5_3B => ("t5-3b", "main"), Which::T5_3B => ("t5-3b", "main"),
Which::Mt5Base => ("google/mt5-base", "refs/pr/5"), Which::Mt5Base => ("google/mt5-base", "refs/pr/5"),
Which::Mt5Small => ("google/mt5-small", "refs/pr/6"), Which::Mt5Small => ("google/mt5-small", "refs/pr/6"),

32
candle-transformers/src/models/t5.rs
View File

@ -70,26 +70,26 @@ where
#[derive(Debug, Clone, PartialEq, Deserialize)] #[derive(Debug, Clone, PartialEq, Deserialize)]
pub struct Config { pub struct Config {
vocab_size: usize, pub vocab_size: usize,
d_model: usize, pub d_model: usize,
d_kv: usize, pub d_kv: usize,
d_ff: usize, pub d_ff: usize,
num_layers: usize, pub num_layers: usize,
num_decoder_layers: Option<usize>, pub num_decoder_layers: Option<usize>,
num_heads: usize, pub num_heads: usize,
relative_attention_num_buckets: usize, pub relative_attention_num_buckets: usize,
#[serde(default = "default_relative_attention_max_distance")] #[serde(default = "default_relative_attention_max_distance")]
relative_attention_max_distance: usize, pub relative_attention_max_distance: usize,
dropout_rate: f64, pub dropout_rate: f64,
layer_norm_epsilon: f64, pub layer_norm_epsilon: f64,
initializer_factor: f64, pub initializer_factor: f64,
#[serde(default, deserialize_with = "deserialize_feed_forward_proj_activation")] #[serde(default, deserialize_with = "deserialize_feed_forward_proj_activation")]
feed_forward_proj: ActivationWithOptionalGating, pub feed_forward_proj: ActivationWithOptionalGating,
#[serde(default = "default_tie_word_embeddings")] #[serde(default = "default_tie_word_embeddings")]
tie_word_embeddings: bool, pub tie_word_embeddings: bool,
#[serde(default = "default_is_decoder")] #[serde(default = "default_is_decoder")]
is_decoder: bool, pub is_decoder: bool,
is_encoder_decoder: bool, pub is_encoder_decoder: bool,
#[serde(default = "default_use_cache")] #[serde(default = "default_use_cache")]
pub use_cache: bool, pub use_cache: bool,
pub pad_token_id: usize, pub pad_token_id: usize,