Upload LanceAI

generation_config.json

@@ -1,7 +1,7 @@
 {
   "_from_model_config": true,
   "do_sample": true,
-  "max_new_tokens": 200,
+  "max_new_tokens": 250,
   "no_repeat_ngram_size": 3,
   "repetition_penalty": 1.3,
   "temperature": 0.8,

lance_ai_model.py

@@ -35,7 +35,7 @@ class LanceAI(PreTrainedModel, GenerationMixin):
         self.loss_fct = nn.CrossEntropyLoss(ignore_index=-100)
 
         # Set generation config defaults for more natural responses
-        self.generation_config.max_new_tokens = 200
+        self.generation_config.max_new_tokens = 250
         self.generation_config.temperature = 0.8
         self.generation_config.top_k = 40
         self.generation_config.top_p = 0.9