pszemraj commited on
Commit
c2e8e0a
1 Parent(s): 5a475bb

⚡️ add min temp and epsilon_cutoff

Browse files

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

Files changed (1) hide show
  1. app.py +3 -3
app.py CHANGED
@@ -4,7 +4,7 @@ from gradio.themes.utils import sizes
4
  from transformers import AutoModelForCausalLM, AutoTokenizer
5
 
6
  import utils
7
- from constants import END_OF_TEXT
8
 
9
  # Load the tokenizer and model
10
  tokenizer = AutoTokenizer.from_pretrained(
@@ -55,13 +55,13 @@ def run_inference(
55
  outputs = model.generate(
56
  **inputs,
57
  do_sample=True,
58
- # early_stopping=True,
59
  max_new_tokens=max_new_tokens,
60
  min_new_tokens=2,
61
  no_repeat_ngram_size=6,
62
  renormalize_logits=True,
63
  repetition_penalty=repetition_penalty,
64
- temperature=temperature,
65
  top_p=top_p,
66
  )
67
  text = tokenizer.batch_decode(
 
4
  from transformers import AutoModelForCausalLM, AutoTokenizer
5
 
6
  import utils
7
+ from constants import END_OF_TEXT, MIN_TEMPERATURE
8
 
9
  # Load the tokenizer and model
10
  tokenizer = AutoTokenizer.from_pretrained(
 
55
  outputs = model.generate(
56
  **inputs,
57
  do_sample=True,
58
+ epsilon_cutoff=1e-3,
59
  max_new_tokens=max_new_tokens,
60
  min_new_tokens=2,
61
  no_repeat_ngram_size=6,
62
  renormalize_logits=True,
63
  repetition_penalty=repetition_penalty,
64
+ temperature=max(temperature, MIN_TEMPERATURE)
65
  top_p=top_p,
66
  )
67
  text = tokenizer.batch_decode(