Add the quantized mixformer model. (#953)

* Add the quantized mixformer model.

* Add the quantized option in the phi example.

candle-transformers/src/models/mixformer.rs

@@ -10,17 +10,17 @@ const MAX_SEQ_LEN: usize = 4096;
 // https://huggingface.co/microsoft/phi-1_5/blob/main/configuration_mixformer_sequential.py
 #[derive(Debug, Clone, PartialEq)]
 pub struct Config {
-    vocab_size: usize,
-    n_positions: usize,
-    n_embd: usize,
-    n_layer: usize,
-    n_inner: Option<usize>,
-    n_head: usize,
-    rotary_dim: usize,
-    activation_function: Activation,
-    layer_norm_epsilon: f64,
-    tie_word_embeddings: bool,
-    pad_vocab_size_multiple: usize,
+    pub(crate) vocab_size: usize,
+    pub(crate) n_positions: usize,
+    pub(crate) n_embd: usize,
+    pub(crate) n_layer: usize,
+    pub(crate) n_inner: Option<usize>,
+    pub(crate) n_head: usize,
+    pub(crate) rotary_dim: usize,
+    pub(crate) activation_function: Activation,
+    pub(crate) layer_norm_epsilon: f64,
+    pub(crate) tie_word_embeddings: bool,
+    pub(crate) pad_vocab_size_multiple: usize,
 }
 
 impl Config {