Make some model cloneable. (#1125)

candle-transformers/src/models/mistral.rs

@@ -39,7 +39,7 @@ impl Config {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct RmsNorm {
     inner: candle_nn::RmsNorm,
     span: tracing::Span,
@@ -60,7 +60,7 @@ impl Module for RmsNorm {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct RotaryEmbedding {
     sin: Tensor,
     cos: Tensor,
@@ -111,7 +111,7 @@ impl RotaryEmbedding {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 #[allow(clippy::upper_case_acronyms)]
 struct MLP {
     gate_proj: Linear,
@@ -160,7 +160,7 @@ fn flash_attn(_: &Tensor, _: &Tensor, _: &Tensor, _: f32, _: bool) -> Result<Ten
     unimplemented!("compile with '--features flash-attn'")
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct Attention {
     q_proj: Linear,
     k_proj: Linear,
@@ -279,7 +279,7 @@ impl Attention {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct DecoderLayer {
     self_attn: Attention,
     mlp: MLP,
@@ -322,7 +322,7 @@ impl DecoderLayer {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct Model {
     embed_tokens: candle_nn::Embedding,
     layers: Vec<DecoderLayer>,

candle-transformers/src/models/mixformer.rs

@@ -74,7 +74,7 @@ impl Config {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct Embedding {
     wte: E,
 }
@@ -106,7 +106,7 @@ fn masked_fill(on_false: &Tensor, mask: &Tensor, on_true: f32) -> Result<Tensor>
     Ok(m)
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct RotaryEmbedding {
     sin: Tensor,
     cos: Tensor,
@@ -172,7 +172,7 @@ impl RotaryEmbedding {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 #[allow(clippy::upper_case_acronyms)]
 struct MLP {
     fc1: Linear,
@@ -199,7 +199,7 @@ impl Module for MLP {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct CausalLMHead {
     ln: candle_nn::LayerNorm,
     linear: Linear,
@@ -221,7 +221,7 @@ impl Module for CausalLMHead {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 #[allow(clippy::upper_case_acronyms)]
 struct MHA {
     wqkv: Linear,
@@ -310,7 +310,7 @@ impl MHA {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct ParallelBlock {
     ln: candle_nn::LayerNorm,
     mixer: MHA,
@@ -345,7 +345,7 @@ impl ParallelBlock {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct MixFormerSequentialForCausalLM {
     embedding: Embedding,
     blocks: Vec<ParallelBlock>,

candle-transformers/src/models/mpt.rs

@@ -40,7 +40,7 @@ impl Config {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct GroupedQueryAttention {
     wqkv: Linear,
     out_proj: Linear,
@@ -148,7 +148,7 @@ pub(crate) fn repeat_kv(xs: Tensor, n_rep: usize) -> Result<Tensor> {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct Ffn {
     up_proj: Linear,
     down_proj: Linear,
@@ -169,7 +169,7 @@ impl Module for Ffn {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 struct MPTBlock {
     norm1: LayerNorm, // Do we need the low-precision variant?
     attn: GroupedQueryAttention,
@@ -240,7 +240,7 @@ pub(crate) fn build_alibi_bias(cfg: &Config) -> Result<Tensor> {
     alibi_bias.to_dtype(DType::F32)?.broadcast_mul(&slopes)
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct Model {
     wte: Embedding,
     blocks: Vec<MPTBlock>,

candle-transformers/src/models/quantized_llama.rs

@@ -7,6 +7,7 @@ use candle_nn::{Embedding, Module};
 
 pub const MAX_SEQ_LEN: usize = 4096;
 
+#[derive(Debug, Clone)]
 struct RmsNorm {
     inner: candle_nn::LayerNorm,
     span: tracing::Span,
@@ -27,6 +28,7 @@ impl RmsNorm {
 }
 
 // QMatMul wrapper adding some tracing.
+#[derive(Debug, Clone)]
 struct QMatMul {
     inner: candle::quantized::QMatMul,
     span: tracing::Span,
@@ -45,6 +47,7 @@ impl QMatMul {
     }
 }
 
+#[derive(Debug, Clone)]
 struct LayerWeights {
     attention_wq: QMatMul,
     attention_wk: QMatMul,
@@ -167,6 +170,7 @@ impl LayerWeights {
     }
 }
 
+#[derive(Debug, Clone)]
 pub struct ModelWeights {
     tok_embeddings: Embedding,
     layers: Vec<LayerWeights>,

candle-transformers/src/models/with_tracing.rs

@@ -1,7 +1,7 @@
 use candle::{Module, Result, Tensor};
 use candle_nn::VarBuilder;
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct Embedding {
     inner: candle_nn::Embedding,
     span: tracing::Span,
@@ -26,7 +26,7 @@ impl Module for Embedding {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct Linear {
     inner: candle_nn::Linear,
     span: tracing::Span,
@@ -52,7 +52,7 @@ impl Module for Linear {
 }
 
 // Wrap the conv2d op to provide some tracing.
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub struct Conv2d {
     inner: candle_nn::Conv2d,
     span: tracing::Span,
@@ -78,6 +78,7 @@ pub fn conv2d(
 }
 
 // QMatMul wrapper adding some tracing.
+#[derive(Clone)]
 pub struct QMatMul {
     inner: candle::quantized::QMatMul,
     span: tracing::Span,