Check shapes on transformer cache

Author: graemenail

src/models/transformer.h

@@ -28,7 +28,7 @@ class Transformer : public EncoderOrDecoderBase {
 
 protected:
   using Base::options_; using Base::inference_; using Base::batchIndex_; using Base::graph_;
-  std::unordered_map<std::string, Expr> cache_;    // caching transformation of the encoder that should not be created again
+  std::unordered_map<std::string, std::pair<Shape, Expr>> cache_;    // caching transformation of the encoder that should not be created again
   mutable/*lazy*/ std::vector<float> sinusoidalEmbeddingsFreq_, sinusoidalEmbeddingsOffs_;  // cached contributions to sinusoidal embeddings
 
   bool depthScaling_{false}; // As recommended in the GPT-2 paper, down-scale layer weights by a factor of 1 / sqrt(depth);
@@ -288,10 +288,10 @@ class Transformer : public EncoderOrDecoderBase {
     // Caching transformation of the encoder that should not be created again.
     // @TODO: set this automatically by memoizing encoder context and
     // memoization propagation (short-term)
-    if (cache                                                                          // if caching
-        && cache_.count(prefix + "_keys") > 0                                          // and the keys expression has been seen
-        && cache_[prefix + "_keys"]->shape().elements() == keys->shape().elements()) { // and the underlying element size did not change
-      kh = cache_[prefix + "_keys"];                                                   // then return cached tensor
+    if (cache                                                    // if caching
+        && cache_.count(prefix + "_keys") > 0                    // and the keys expression has been seen
+        && cache_[prefix + "_keys"].first == keys->shape()) {    // and the underlying element size did not change
+      kh = cache_[prefix + "_keys"].second;                      // then return cached tensor
     }
     else {
       int dimKeys =  keys->shape()[-1]; // different than dimModel when using lemma and factors combined with concatenation
@@ -300,22 +300,22 @@ class Transformer : public EncoderOrDecoderBase {
 
       kh = affine(keys, Wk, bk);     // [-4: beam depth, -3: batch size, -2: max length, -1: vector dim]
       kh = SplitHeads(kh, dimHeads); // [-4: batch size, -3: num heads, -2: max length, -1: split vector dim]
-      cache_[prefix + "_keys"] = kh;
+      cache_[prefix + "_keys"] = std::make_pair(keys->shape(), kh);
     }
 
     Expr vh;
     if (cache 
         && cache_.count(prefix + "_values") > 0 
-        && cache_[prefix + "_values"]->shape().elements() == values->shape().elements()) {
-      vh = cache_[prefix + "_values"];
+        && cache_[prefix + "_values"].first == values->shape()) {
+      vh = cache_[prefix + "_values"].second;
     } else {
       int dimValues = values->shape()[-1]; // different than dimModel when using lemma and factors combined with concatenation
       auto Wv = graph_->param(prefix + "_Wv", {dimValues, dimModel}, inits::glorotUniform(true, true, depthScaling_ ? 1.f / sqrtf((float)depth_) : 1.f));
       auto bv = graph_->param(prefix + "_bv", {1,        dimModel}, inits::zeros());
 
       vh = affine(values, Wv, bv); // [-4: batch size, -3: num heads, -2: max length, -1: split vector dim]
       vh = SplitHeads(vh, dimHeads);
-      cache_[prefix + "_values"] = vh;
+      cache_[prefix + "_values"] = std::make_pair(values->shape(), vh);
     }
 
     int dimBeam = q->shape()[-4];