Update app.py
Browse files
app.py
CHANGED
|
@@ -9,13 +9,10 @@ from typing import Tuple, Dict
|
|
| 9 |
# Конфигурация
|
| 10 |
# ------------------------------------------------------------
|
| 11 |
MODEL_NAMES = [
|
| 12 |
-
"tinkoff-ai/ruDialoGPT-small",
|
| 13 |
"thelamapi/next2-0.8b",
|
| 14 |
"Lamapi/next-1b",
|
| 15 |
"Vikhrmodels/Vikhr-Llama-3.2-1B-instruct",
|
| 16 |
-
"Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning"
|
| 17 |
-
"bigscience/mt0-base",
|
| 18 |
-
"BSC-LT/salamandra-2b-instruct"
|
| 19 |
]
|
| 20 |
DEFAULT_MODEL = MODEL_NAMES[0]
|
| 21 |
|
|
@@ -105,6 +102,7 @@ def generate_response(
|
|
| 105 |
repetition_penalty=repetition_penalty, # штраф за повторяющиеся токены
|
| 106 |
no_repeat_ngram_size=2, # запрет повторения триграмм
|
| 107 |
early_stopping=True, # остановка при достижении eos_token
|
|
|
|
| 108 |
pad_token_id=tokenizer.eos_token_id
|
| 109 |
)
|
| 110 |
latency = time.time() - start_time
|
|
|
|
| 9 |
# Конфигурация
|
| 10 |
# ------------------------------------------------------------
|
| 11 |
MODEL_NAMES = [
|
|
|
|
| 12 |
"thelamapi/next2-0.8b",
|
| 13 |
"Lamapi/next-1b",
|
| 14 |
"Vikhrmodels/Vikhr-Llama-3.2-1B-instruct",
|
| 15 |
+
"Vikhrmodels/QVikhr-3-1.7B-Instruction-noreasoning"
|
|
|
|
|
|
|
| 16 |
]
|
| 17 |
DEFAULT_MODEL = MODEL_NAMES[0]
|
| 18 |
|
|
|
|
| 102 |
repetition_penalty=repetition_penalty, # штраф за повторяющиеся токены
|
| 103 |
no_repeat_ngram_size=2, # запрет повторения триграмм
|
| 104 |
early_stopping=True, # остановка при достижении eos_token
|
| 105 |
+
top_k=50,
|
| 106 |
pad_token_id=tokenizer.eos_token_id
|
| 107 |
)
|
| 108 |
latency = time.time() - start_time
|