Internal change.

PiperOrigin-RevId: 762103223

Author: The gemma.cpp Authors

gemma/activations.h

@@ -18,6 +18,8 @@
 
 #include <stddef.h>
 
+#include <vector>
+
 #include "gemma/configs.h"   // ModelConfig
 #include "ops/matmul.h"      // MatMulEnv
 #include "ops/ops.h"         // CreateInvTimescale

gemma/gemma-inl.h

@@ -21,6 +21,7 @@
 #include <stdio.h>
 
 #include <algorithm>  // std::min
+#include <numeric>    // std::iota
 #include <vector>
 
 #include "gemma/activations.h"

gemma/weights.h

@@ -135,7 +135,8 @@ struct LayerWeightsPtrs {
         key_norm_scale(Concat("key_norm", suffix_), tensors),
         query_norm_scale(Concat("query_norm", suffix_), tensors),
 
-        layer_config(config) {}
+        layer_config(config) {
+  }
   ~LayerWeightsPtrs() = default;
 
   const std::string suffix_;
@@ -264,7 +265,7 @@ struct LayerWeightsPtrs {
       func(TENSOR_ARGS(gating_einsum_w, kMaybeRead));
       func(TENSOR_ARGS(gating_einsum_w1, kMaybeRead));
       func(TENSOR_ARGS(gating_einsum_w2, kMaybeRead));
-      func(TENSOR_ARGS(linear_w, kMustRead));
+      func(TENSOR_ARGS(linear_w, kMaybeRead));
       func(TENSOR_ARGS(pre_attention_norm_scale, kMustRead));
       func(TENSOR_ARGS(pre_ffw_norm_scale, kMustRead));
     }