Suggested flag values on some benchmark data sets?

E.g. to run language_model on ptb data, which set of flag values is a reasonable starting point?
I'm using 

```
./language_model -train ptb/train.txt -min_occurence 1 -validation ptb/valid.txt -device -1 -j 8 -solver adagrad -show_wps -save lm.model                                                 
```

But so far the reconstructions look wired.
