Fix errors

Author: EricLBuehler

mistralrs-core/src/models/deepseek2.rs

@@ -93,8 +93,6 @@ pub struct DeepSeekV2Config {
     pub(crate) kv_lora_rank: usize,
     pub(crate) v_head_dim: usize,
     pub(crate) qk_nope_head_dim: usize,
-    #[serde(default = "use_flash_attn_default")]
-    pub(crate) use_flash_attn: bool,
     pub(crate) quantization_config: Option<QuantizedConfig>,
     pub(crate) n_group: usize,
     pub(crate) topk_group: usize,

mistralrs-core/src/models/deepseek3.rs

@@ -93,8 +93,6 @@ pub struct DeepSeekV3Config {
     pub(crate) kv_lora_rank: usize,
     pub(crate) v_head_dim: usize,
     pub(crate) qk_nope_head_dim: usize,
-    #[serde(default = "use_flash_attn_default")]
-    pub(crate) use_flash_attn: bool,
     pub(crate) quantization_config: Option<QuantizedConfig>,
     pub(crate) n_group: usize,
     pub(crate) topk_group: usize,

mistralrs-core/src/models/gemma.rs

@@ -50,7 +50,6 @@ pub struct Config {
 
     #[serde(default = "default_max_position_embeddings")]
     pub max_position_embeddings: usize,
-    pub use_flash_attn: bool,
     pub quantization_config: Option<QuantizedConfig>,
     #[serde(default = "word_emb_default")]
     #[allow(dead_code)]

mistralrs-core/src/models/gemma2.rs

@@ -24,7 +24,7 @@ use crate::{
     utils::{progress::NiceProgressBar, unvarbuilder::UnVarBuilder},
 };
 
-#[derive(Debug, Clone, Default, serde::Serialize)]
+#[derive(Debug, Clone, Default, serde::Deserialize)]
 pub struct Config {
     pub attention_bias: bool,
     pub head_dim: usize,
@@ -45,7 +45,6 @@ pub struct Config {
     pub query_pre_attn_scalar: usize,
     pub max_position_embeddings: usize,
     pub quantization_config: Option<QuantizedConfig>,
-    pub use_flash_attn: bool,
     #[allow(dead_code)]
     pub tie_word_embeddings: bool,
 }

mistralrs-core/src/models/llama.rs

@@ -41,8 +41,6 @@ pub struct Config {
     pub num_hidden_layers: usize,
     pub num_attention_heads: usize,
     pub num_key_value_heads: usize,
-    #[serde(default = "use_flash_attn_default")]
-    pub use_flash_attn: bool,
     pub rms_norm_eps: f64,
     pub rope_theta: f32,
     pub max_position_embeddings: usize,

mistralrs-core/src/models/mistral.rs

@@ -42,8 +42,6 @@ pub struct Config {
     pub(crate) rms_norm_eps: f64,
     pub(crate) rope_theta: f64,
     pub(crate) sliding_window: Option<usize>,
-    #[serde(default = "use_flash_attn")]
-    pub(crate) use_flash_attn: bool,
     pub(crate) head_dim: Option<usize>,
     pub(crate) quantization_config: Option<QuantizedConfig>,
     #[serde(default = "tie_word_embeddings")]

mistralrs-core/src/models/mixtral.rs

@@ -45,7 +45,6 @@ pub struct Config {
     pub(crate) sliding_window: Option<usize>,
     pub(crate) num_experts_per_tok: usize,
     pub(crate) num_local_experts: usize,
-    pub(crate) use_flash_attn: bool,
     pub(crate) quantization_config: Option<QuantizedConfig>,
     #[serde(default = "word_emb_default")]
     pub(crate) tie_word_embeddings: bool,

mistralrs-core/src/models/phi2.rs

@@ -51,7 +51,6 @@ pub struct Config {
     pub(crate) rope_theta: f32,
     pub(crate) partial_rotary_factor: f64,
     pub(crate) qk_layernorm: bool,
-    pub(crate) use_flash_attn: bool,
     pub(crate) quantization_config: Option<QuantizedConfig>,
     #[serde(default = "word_emb_default")]
     pub(crate) tie_word_embeddings: bool,

mistralrs-core/src/models/phi3.rs

@@ -47,7 +47,6 @@ pub struct Config {
     pub eos_token_id: Option<u32>,
     pub rope_scaling: Option<PhiRopeScalingConfig>,
     pub max_position_embeddings: usize,
-    pub use_flash_attn: bool,
     pub sliding_window: Option<usize>,
     pub original_max_position_embeddings: usize,
     pub quantization_config: Option<QuantizedConfig>,

mistralrs-core/src/models/phi3_5_moe.rs

@@ -43,7 +43,6 @@ pub struct Config {
     pub(crate) rope_theta: f64,
     pub(crate) rope_scaling: Option<PhiRopeScalingConfig>,
     pub(crate) max_position_embeddings: usize,
-    pub(crate) use_flash_attn: bool,
     pub(crate) sliding_window: Option<usize>,
     pub(crate) original_max_position_embeddings: usize,
     pub(crate) quantization_config: Option<QuantizedConfig>,