Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -1,26 +1,89 @@
|
|
1 |
-
import
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
def
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
)
|
22 |
-
|
23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
|
26 |
|
|
|
1 |
+
import torch
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
|
4 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
5 |
+
|
6 |
+
|
7 |
+
def create_prompt_with_chat_format(messages, bos="<s>", eos="</s>", add_bos=True):
|
8 |
+
formatted_text = ""
|
9 |
+
for message in messages:
|
10 |
+
if message["role"] == "system":
|
11 |
+
formatted_text += "<|system|>\n" + message["content"] + "\n"
|
12 |
+
elif message["role"] == "user":
|
13 |
+
formatted_text += "<|user|>\n" + message["content"] + "\n"
|
14 |
+
elif message["role"] == "assistant":
|
15 |
+
formatted_text += "<|assistant|>\n" + message["content"].strip() + eos + "\n"
|
16 |
+
else:
|
17 |
+
raise ValueError(
|
18 |
+
"Tulu chat template only supports 'system', 'user' and 'assistant' roles. Invalid role: {}.".format(
|
19 |
+
message["role"]
|
20 |
+
)
|
21 |
+
)
|
22 |
+
formatted_text += "<|assistant|>\n"
|
23 |
+
formatted_text = bos + formatted_text if add_bos else formatted_text
|
24 |
+
return formatted_text
|
25 |
+
|
26 |
+
|
27 |
+
def inference(input_prompts, model, tokenizer):
|
28 |
+
input_prompts = [
|
29 |
+
create_prompt_with_chat_format([{"role": "user", "content": input_prompt}], add_bos=False)
|
30 |
+
for input_prompt in input_prompts
|
31 |
+
]
|
32 |
+
|
33 |
+
encodings = tokenizer(input_prompts, padding=True, return_tensors="pt")
|
34 |
+
encodings = encodings.to(device)
|
35 |
+
|
36 |
+
with torch.inference_mode():
|
37 |
+
outputs = model.generate(encodings.input_ids, do_sample=False, max_new_tokens=250)
|
38 |
+
|
39 |
+
output_texts = tokenizer.batch_decode(outputs.detach(), skip_special_tokens=True)
|
40 |
+
|
41 |
+
input_prompts = [
|
42 |
+
tokenizer.decode(tokenizer.encode(input_prompt), skip_special_tokens=True) for input_prompt in input_prompts
|
43 |
+
]
|
44 |
+
output_texts = [output_text[len(input_prompt) :] for input_prompt, output_text in zip(input_prompts, output_texts)]
|
45 |
+
return output_texts
|
46 |
+
|
47 |
+
|
48 |
+
model_name = "ai4bharat/Airavata"
|
49 |
+
|
50 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
|
51 |
+
tokenizer.pad_token = tokenizer.eos_token
|
52 |
+
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16).to(device)
|
53 |
+
|
54 |
+
input_prompts = [
|
55 |
+
"मैं अपने समय प्रबंधन कौशल को कैसे सुधार सकता हूँ? मुझे पांच बिंदु बताएं।",
|
56 |
+
"मैं अपने समय प्रबंधन कौशल को कैसे सुधार सकता हूँ? मुझे पांच बिंदु बताएं और उनका वर्णन करें।",
|
57 |
+
]
|
58 |
+
outputs = inference(input_prompts, model, tokenizer)
|
59 |
+
print(outputs)
|
60 |
+
|
61 |
+
|
62 |
+
|
63 |
+
|
64 |
+
# import gradio as gr
|
65 |
+
# from transformers import AutoTokenizer, AutoModelForCausalLM
|
66 |
+
|
67 |
+
# tokenizer = AutoTokenizer.from_pretrained("ai4bharat/Airavata")
|
68 |
+
# model = AutoModelForCausalLM.from_pretrained("ai4bharat/Airavata")
|
69 |
+
|
70 |
+
# def generate_response(prompt):
|
71 |
+
# input_ids = tokenizer.encode(prompt, return_tensors="pt", max_length=50)
|
72 |
+
# output_ids = model.generate(input_ids, max_length=100, num_beams=5, no_repeat_ngram_size=2)
|
73 |
+
# response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
|
74 |
+
# return response
|
75 |
+
|
76 |
+
# iface = gr.Interface(
|
77 |
+
# fn=generate_response,
|
78 |
+
# inputs="text",
|
79 |
+
# outputs="text",
|
80 |
+
# live=True,
|
81 |
+
# title="Airavata LLMs Chatbot",
|
82 |
+
# description="Ask me anything, and I'll generate a response!",
|
83 |
+
# theme="light",
|
84 |
+
# )
|
85 |
+
|
86 |
+
# iface.launch()
|
87 |
|
88 |
|
89 |
|