Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,7 @@
|
|
1 |
---
|
2 |
library_name: peft
|
|
|
|
|
3 |
---
|
4 |
## Training procedure
|
5 |
|
@@ -14,7 +16,54 @@ The following `bitsandbytes` quantization config was used during training:
|
|
14 |
- bnb_4bit_quant_type: nf4
|
15 |
- bnb_4bit_use_double_quant: False
|
16 |
- bnb_4bit_compute_dtype: float16
|
17 |
-
### Framework versions
|
18 |
|
|
|
19 |
|
20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
library_name: peft
|
3 |
+
language:
|
4 |
+
- pt
|
5 |
---
|
6 |
## Training procedure
|
7 |
|
|
|
16 |
- bnb_4bit_quant_type: nf4
|
17 |
- bnb_4bit_use_double_quant: False
|
18 |
- bnb_4bit_compute_dtype: float16
|
|
|
19 |
|
20 |
+
## Algoritmo para utilização do modelo
|
21 |
|
22 |
+
import torch
|
23 |
+
from peft import PeftModel, PeftConfig
|
24 |
+
from transformers import AutoModelForCausalLM,BitsAndBytesConfig,AutoTokenizer
|
25 |
+
|
26 |
+
#Quantização
|
27 |
+
use_4bit = True
|
28 |
+
bnb_4bit_compute_dtype = "float16"
|
29 |
+
bnb_4bit_quant_type = "nf4"
|
30 |
+
use_nested_quant = False
|
31 |
+
|
32 |
+
|
33 |
+
# Carrega o tokenizer e modelo com configuração QLoRA
|
34 |
+
compute_dtype = getattr(torch, bnb_4bit_compute_dtype)
|
35 |
+
|
36 |
+
bnb_config = BitsAndBytesConfig(load_in_4bit = use_4bit,
|
37 |
+
bnb_4bit_quant_type = bnb_4bit_quant_type,
|
38 |
+
bnb_4bit_compute_dtype = compute_dtype,
|
39 |
+
bnb_4bit_use_double_quant = use_nested_quant)
|
40 |
+
|
41 |
+
|
42 |
+
#Import do modelo
|
43 |
+
config = PeftConfig.from_pretrained("MatNLP/Sectrum")
|
44 |
+
base_model = AutoModelForCausalLM.from_pretrained("NousResearch/Llama-2-7b-chat-hf",quantization_config = bnb_config)
|
45 |
+
model = PeftModel.from_pretrained(base_model, "MatNLP/Sectrum")
|
46 |
+
|
47 |
+
|
48 |
+
# Carrega o tokenizador
|
49 |
+
tokenizer = AutoTokenizer.from_pretrained("NousResearch/Llama-2-7b-chat-hf", trust_remote_code = True,skip_special_tokens=True)
|
50 |
+
tokenizer.pad_token = tokenizer.eos_token
|
51 |
+
tokenizer.padding_side = "right"
|
52 |
+
|
53 |
+
|
54 |
+
# Prepara o prompt
|
55 |
+
prompt = "Como proteger meu e-mail?"
|
56 |
+
|
57 |
+
# Cria o pipeline
|
58 |
+
pipe = pipeline(task = "text-generation",
|
59 |
+
model = model,
|
60 |
+
tokenizer = tokenizer,
|
61 |
+
max_length = 200)
|
62 |
+
#streamer=TextStreamer(tokenizer,skip_prompt=True)
|
63 |
+
|
64 |
+
|
65 |
+
# Executa o pipeline e gera o texto a partir do prompt inicial
|
66 |
+
resultado = pipe(f"<s>[INST] {prompt} [/INST]")
|
67 |
+
|
68 |
+
|
69 |
+
print(resultado[0]['generated_text'].split("[/INST]")[1].split('<\s>')[0])
|