update sampling parameters

so it turns out using LLaMA defaults was a horrible idea

generation_config.json

@@ -6,9 +6,10 @@
    "use_cache":false,
    "early_stopping":true,
    "do_sample":true,
-   "temperature":0.1,
-   "top_p":0.75,
+   "temperature":1.0,
+   "top_p":0.9,
    "top_k":40,
    "num_beams":4,
+   "no_repeat_ngram_size":6,
    "max_new_tokens":128
 }