Update README.md

top_p, metrics, model performance comments

README.md

@@ -7,7 +7,7 @@ inference:
     use_cache: False
     max_length: 200
     top_k: 5
-    
+    top_p: 0.9
     
 widget:
 - text: "Sony TV"
@@ -18,13 +18,17 @@ widget:
   example_title: "Ads for shopping deals"
 - text: "Labor Day discounts for"
   example_title: "Ads for Holiday deals"
+  
+metrics:
+- bleu
+
 ---
 
 Generates Ad text copy, for ads for Amazon shopping (fine tuned for electronics and wearables).   
 
 The model is fine tuned on the EleutherAI/gpt-neo-125M model using the Amazon Ads dataset. 
 
-**Examples:**   
+**Usage Examples:**   
 Select from among the examples in the dropdown or enter your own prompts.   
 
 You can try entering brand and product names like Samsung Galaxy to see the ad text generator in action.   
@@ -33,10 +37,15 @@ Feel free to play around with native Amazon ads which are product descriptions l
 **Sony** BDPS3700 Streaming Blu-Ray Disc Player with Wi-Fi (Black).  
 **AmazonBasics** TV Trolley for 24-43" TVs with Swivel Feature.  
 
-Or try other ad formats like on other shopping sites for holiday deals like:   
+Or try additional ad formats, similar to other shopping sites for holiday deals, like:   
 **Big savings on the new** Roku Streaming Device  
 **Mothers Day discounts for** Apple Watch Wireless Charger USB Charging Cable  
 **Last minute shopping for Samsung headphones for** 
 
+**Model Performance:**   
+The model does quite well on the Electronics and Wearables categories on which it has been fine-tuned. There are, however, occasional hallucinations, though the ad copy is mostly coherent.   
 
+In other domains, it doesn't do quite as well...  
+Tesla for Christmas today,   
+Honda on sale