[M2M100] fix positional embeddings (#10590)

* fix tests

* emb should be a parameter

* fix positional embeddings

* fix make_weights

* don't save pos embeds

* add comment to describe the clamping

src/transformers/models/m2m_100/modeling_m2m_100.py

@@ -121,8 +121,17 @@ class M2M100SinusoidalPositionalEmbedding(nn.Module):
         self.offset = 2
         self.embedding_dim = embedding_dim
         self.padding_idx = padding_idx
-        self.weights = self.get_embedding(num_positions + self.offset, embedding_dim, padding_idx)
-        self.register_buffer("_float_tensor", torch.FloatTensor(1))
+        self.make_weights(num_positions + self.offset, embedding_dim, padding_idx)
+
+    def make_weights(self, num_embeddings: int, embedding_dim: int, padding_idx: Optional[int] = None):
+        emb_weights = self.get_embedding(num_embeddings, embedding_dim, padding_idx)
+        if hasattr(self, "weights"):
+            # in forward, put the weights on correct device
+            emb_weights = emb_weights.to(self.weights.device)
+
+        self.weights = nn.Parameter(emb_weights)
+        self.weights.requires_grad = False
+        self.weights.detach_()
 
     @staticmethod
     def get_embedding(num_embeddings: int, embedding_dim: int, padding_idx: Optional[int] = None):
@@ -142,6 +151,7 @@ class M2M100SinusoidalPositionalEmbedding(nn.Module):
             emb = torch.cat([emb, torch.zeros(num_embeddings, 1)], dim=1)
         if padding_idx is not None:
             emb[padding_idx, :] = 0
+
         return emb
 
     @torch.no_grad()
@@ -161,9 +171,7 @@ class M2M100SinusoidalPositionalEmbedding(nn.Module):
         # expand embeddings if needed
         max_pos = self.padding_idx + 1 + seq_len
         if max_pos > self.weights.size(0):
-            self.weights = self.get_embedding(max_pos, self.embedding_dim, self.padding_idx)
-
-        self.weights = self.weights.to(self._float_tensor)
+            self.make_weights(max_pos + self.offset, self.embedding_dim, self.padding_idx)
 
         return self.weights.index_select(0, position_ids.view(-1)).view(bsz, seq_len, -1).detach()
 
@@ -1149,6 +1157,12 @@ class M2M100ForConditionalGeneration(M2M100PreTrainedModel):
         r"encoder\.version",
         r"decoder\.version",
         r"lm_head\.weight",
+        r"model.encoder.embed_positions.weights",
+        r"model.decoder.embed_positions.weights",
+    ]
+    _keys_to_ignore_on_save = [
+        r"model.encoder.embed_positions.weights",
+        r"model.decoder.embed_positions.weights",
     ]
 
     def __init__(self, config: M2M100Config):

tests/test_modeling_m2m_100.py

@@ -96,13 +96,19 @@ class M2M100ModelTester:
 
     def prepare_config_and_inputs(self):
         input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
-        input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size).clamp(
-            3,
-        )
         input_ids[:, -1] = self.eos_token_id  # Eos Token
-
         decoder_input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
 
+        # we need to clamp the input ids here to avoid having pad token in between
+        # this is because for M2M100 the position_ids are prepared such that
+        # all pad tokens have pos id = 2 and rest are between 2..seq_length
+        # and the seq_length here is seq_length - num_pad_tokens
+        # but when using past, there is no way of knowing if the past input ids had
+        # pad tokens in them, which results in incorrect seq_lenth and which in turn results in
+        # position_ids being off by num_pad_tokens in past input
+        input_ids = input_ids.clamp(self.pad_token_id + 1)
+        decoder_input_ids = decoder_input_ids.clamp(self.pad_token_id + 1)
+
         config = M2M100Config(
             vocab_size=self.vocab_size,
             d_model=self.hidden_size,