⚡️ add min temp and epsilon_cutoff

Signed-off-by: peter szemraj <peterszemraj@gmail.com>

app.py

@@ -4,7 +4,7 @@ from gradio.themes.utils import sizes
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 import utils
-from constants import END_OF_TEXT
+from constants import END_OF_TEXT, MIN_TEMPERATURE
 
 # Load the tokenizer and model
 tokenizer = AutoTokenizer.from_pretrained(
@@ -55,13 +55,13 @@ def run_inference(
     outputs = model.generate(
         **inputs,
         do_sample=True,
-        # early_stopping=True,
+        epsilon_cutoff=1e-3,
         max_new_tokens=max_new_tokens,
         min_new_tokens=2,
         no_repeat_ngram_size=6,
         renormalize_logits=True,
         repetition_penalty=repetition_penalty,
-        temperature=temperature,
+        temperature=max(temperature, MIN_TEMPERATURE)
         top_p=top_p,
     )
     text = tokenizer.batch_decode(