Update gpt2 README

examples/gpt-2/README.md

@@ -100,7 +100,12 @@ Texar's `TransformerDecoder` (and other RNN-based decoders) easily supports comm
 
 **For example**, after creating the language model
 ```python
-decoder = TransformerDecoder(embedder, hparams=gpt2_hparams)
+def _embedding_fn(ids, times):
+    return word_embedder(ids) + pos_embedder(times)
+
+decoder = TransformerDecoder(
+    output_layer=tf.transpose(word_embedder.embedding), 
+    hparams=gpt2_hparams)
 ```
 We can do
 
@@ -111,17 +116,18 @@ output, output_length = decoder(
     context=ctx,
     context_sequence_length=ctx_len,
     decoding_strategy='infer_greedy',
-    end_token=end_token)
+    end_token=end_token
+    embedding=_embedding_fn)
 
 sample_id = output.sample_id
 logits = output.logits
 ```
 
 **Ex. Use 2): Top-k sample decoding**
 
-```python
+```python    
 topk_helper = tx.modules.TopKSampleEmbeddingHelper(
-    embedding=embedder,
+    embedding=_embedding_fn,
     start_tokens=ctx[:,0],
     end_token=end_token,
     top_k=20,
@@ -136,10 +142,12 @@ output, output_length = decoder(
 **Ex. Use 3): Fine-tuning for conditional generation**
 
 ```python
+tgt_embed = word_embedder(truth_target[:, :-1]) + pos_embedder(sequence_length=tgt_len-1)
+
 output = decoder(
     memory=source_hidden_states, 
     memory_sequence_length=src_len,
-    inputs=embedder(truth_target[:, :-1]),
+    inputs=tgt_embed,
     decoding_strategy='train_greedy') # teacher-forcing decoding
 
 loss = tx.losses.sequence_sparse_softmax_cross_entropy(