Spaces:
Sleeping
Sleeping
cosmos loading
Browse files- app.py +32 -15
- app_cosmos.py → app_gemma.py +15 -32
app.py
CHANGED
@@ -1,26 +1,26 @@
|
|
1 |
from fastapi import FastAPI, Request
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import torch
|
4 |
-
from huggingface_hub import login
|
5 |
-
import os
|
6 |
|
7 |
-
print("
|
8 |
|
9 |
-
|
10 |
-
access_token = os.getenv('HF_TOKEN')
|
11 |
-
login(access_token)
|
12 |
-
|
13 |
-
model_id = "google/gemma-2-9b-it"
|
14 |
|
15 |
print("Model loading started")
|
16 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
17 |
model = AutoModelForCausalLM.from_pretrained(
|
18 |
model_id,
|
19 |
-
device_map="auto",
|
20 |
torch_dtype=torch.bfloat16,
|
|
|
21 |
)
|
22 |
print("Model loading completed")
|
23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
25 |
print("Selected device:", device)
|
26 |
|
@@ -40,18 +40,35 @@ async def ask(request: Request):
|
|
40 |
return {"error": "Prompt is missing"}
|
41 |
|
42 |
print("Device of the model:", model.device)
|
43 |
-
messages =
|
44 |
-
|
45 |
-
|
46 |
print("Messages:", messages)
|
47 |
print("Tokenizer process started")
|
48 |
-
input_ids = tokenizer.apply_chat_template(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
print("Tokenizer process completed")
|
50 |
|
51 |
print("Model process started")
|
52 |
-
outputs = model.generate(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
print("Tokenizer decode process started")
|
55 |
-
answer = tokenizer.decode(
|
56 |
|
57 |
return {"answer": answer}
|
|
|
1 |
from fastapi import FastAPI, Request
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import torch
|
|
|
|
|
4 |
|
5 |
+
print("COSMOS Llama Chatbot is starting...")
|
6 |
|
7 |
+
model_id = "ytu-ce-cosmos/Turkish-Llama-8b-DPO-v0.1"
|
|
|
|
|
|
|
|
|
8 |
|
9 |
print("Model loading started")
|
10 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
11 |
model = AutoModelForCausalLM.from_pretrained(
|
12 |
model_id,
|
|
|
13 |
torch_dtype=torch.bfloat16,
|
14 |
+
device_map="auto",
|
15 |
)
|
16 |
print("Model loading completed")
|
17 |
|
18 |
+
# bu mesaj değiştirilebilir ve chatbotun başlangıç mesajı olarak kullanılabilir
|
19 |
+
initial_message = [
|
20 |
+
{"role": "system", "content": "Sen bir yapay zeka asistanısın. Kullanıcı sana bir görev verecek. Amacın görevi olabildiğince sadık bir şekilde tamamlamak."}
|
21 |
+
# Görevi yerine getirirken adım adım düşün ve adımlarını gerekçelendir.
|
22 |
+
]
|
23 |
+
|
24 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
25 |
print("Selected device:", device)
|
26 |
|
|
|
40 |
return {"error": "Prompt is missing"}
|
41 |
|
42 |
print("Device of the model:", model.device)
|
43 |
+
messages = initial_message.copy()
|
44 |
+
messages.append({"role": "user", "content": f"{prompt}"})
|
45 |
+
|
46 |
print("Messages:", messages)
|
47 |
print("Tokenizer process started")
|
48 |
+
input_ids = tokenizer.apply_chat_template(
|
49 |
+
messages,
|
50 |
+
add_generation_prompt=True,
|
51 |
+
return_tensors="pt"
|
52 |
+
).to(model.device)
|
53 |
+
|
54 |
+
terminators = [
|
55 |
+
tokenizer.eos_token_id,
|
56 |
+
tokenizer.convert_tokens_to_ids("<|eot_id|>")
|
57 |
+
]
|
58 |
print("Tokenizer process completed")
|
59 |
|
60 |
print("Model process started")
|
61 |
+
outputs = model.generate(
|
62 |
+
input_ids,
|
63 |
+
max_new_tokens=256,
|
64 |
+
eos_token_id=terminators,
|
65 |
+
do_sample=True,
|
66 |
+
temperature=0.6,
|
67 |
+
top_p=0.9,
|
68 |
+
)
|
69 |
+
response = outputs[0][input_ids.shape[-1]:]
|
70 |
|
71 |
print("Tokenizer decode process started")
|
72 |
+
answer = tokenizer.decode(response, skip_special_tokens=True)
|
73 |
|
74 |
return {"answer": answer}
|
app_cosmos.py → app_gemma.py
RENAMED
@@ -1,26 +1,26 @@
|
|
1 |
from fastapi import FastAPI, Request
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import torch
|
|
|
|
|
4 |
|
5 |
-
print("
|
6 |
|
7 |
-
|
|
|
|
|
|
|
|
|
8 |
|
9 |
print("Model loading started")
|
10 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
11 |
model = AutoModelForCausalLM.from_pretrained(
|
12 |
model_id,
|
13 |
-
torch_dtype=torch.bfloat16,
|
14 |
device_map="auto",
|
|
|
15 |
)
|
16 |
print("Model loading completed")
|
17 |
|
18 |
-
# bu mesaj değiştirilebilir ve chatbotun başlangıç mesajı olarak kullanılabilir
|
19 |
-
initial_message = [
|
20 |
-
{"role": "system", "content": "Sen bir yapay zeka asistanısın. Kullanıcı sana bir görev verecek. Amacın görevi olabildiğince sadık bir şekilde tamamlamak."}
|
21 |
-
# Görevi yerine getirirken adım adım düşün ve adımlarını gerekçelendir.
|
22 |
-
]
|
23 |
-
|
24 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
25 |
print("Selected device:", device)
|
26 |
|
@@ -40,35 +40,18 @@ async def ask(request: Request):
|
|
40 |
return {"error": "Prompt is missing"}
|
41 |
|
42 |
print("Device of the model:", model.device)
|
43 |
-
messages =
|
44 |
-
|
45 |
-
|
46 |
print("Messages:", messages)
|
47 |
print("Tokenizer process started")
|
48 |
-
input_ids = tokenizer.apply_chat_template(
|
49 |
-
messages,
|
50 |
-
add_generation_prompt=True,
|
51 |
-
return_tensors="pt"
|
52 |
-
).to(model.device)
|
53 |
-
|
54 |
-
terminators = [
|
55 |
-
tokenizer.eos_token_id,
|
56 |
-
tokenizer.convert_tokens_to_ids("<|eot_id|>")
|
57 |
-
]
|
58 |
print("Tokenizer process completed")
|
59 |
|
60 |
print("Model process started")
|
61 |
-
outputs = model.generate(
|
62 |
-
input_ids,
|
63 |
-
max_new_tokens=256,
|
64 |
-
eos_token_id=terminators,
|
65 |
-
do_sample=True,
|
66 |
-
temperature=0.6,
|
67 |
-
top_p=0.9,
|
68 |
-
)
|
69 |
-
response = outputs[0][input_ids.shape[-1]:]
|
70 |
|
71 |
print("Tokenizer decode process started")
|
72 |
-
answer = tokenizer.decode(
|
73 |
|
74 |
return {"answer": answer}
|
|
|
1 |
from fastapi import FastAPI, Request
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import torch
|
4 |
+
from huggingface_hub import login
|
5 |
+
import os
|
6 |
|
7 |
+
print("Google Gemma 2 Chatbot is starting...")
|
8 |
|
9 |
+
# read access token from environment variable
|
10 |
+
access_token = os.getenv('HF_TOKEN')
|
11 |
+
login(access_token)
|
12 |
+
|
13 |
+
model_id = "google/gemma-2-9b-it"
|
14 |
|
15 |
print("Model loading started")
|
16 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
17 |
model = AutoModelForCausalLM.from_pretrained(
|
18 |
model_id,
|
|
|
19 |
device_map="auto",
|
20 |
+
torch_dtype=torch.bfloat16,
|
21 |
)
|
22 |
print("Model loading completed")
|
23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
25 |
print("Selected device:", device)
|
26 |
|
|
|
40 |
return {"error": "Prompt is missing"}
|
41 |
|
42 |
print("Device of the model:", model.device)
|
43 |
+
messages = [
|
44 |
+
{"role": "user", "content": f"{prompt}"},
|
45 |
+
]
|
46 |
print("Messages:", messages)
|
47 |
print("Tokenizer process started")
|
48 |
+
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt", return_dict=True).to("cuda")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
print("Tokenizer process completed")
|
50 |
|
51 |
print("Model process started")
|
52 |
+
outputs = model.generate(**input_ids, max_new_tokens=256)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
print("Tokenizer decode process started")
|
55 |
+
answer = tokenizer.decode(outputs[0]).split("<end_of_turn>")[1].strip()
|
56 |
|
57 |
return {"answer": answer}
|