-
-
Notifications
You must be signed in to change notification settings - Fork 220
/
example_lora.py
79 lines (55 loc) · 2.36 KB
/
example_lora.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
from model import ExLlama, ExLlamaCache, ExLlamaConfig
from tokenizer import ExLlamaTokenizer
from generator import ExLlamaGenerator
from lora import ExLlamaLora
import os, glob
import torch
# Directory containt model, tokenizer, generator
model_directory = "/mnt/str/models/_test_models/Neko-Institute-of-Science_LLaMA-7B-4bit-128g/"
# Directory containing LoRA config and weights
lora_directory = "/mnt/str/models/_test_loras/tloen_alpaca-lora-7b/"
# Locate files we need within those directories
tokenizer_path = os.path.join(model_directory, "tokenizer.model")
model_config_path = os.path.join(model_directory, "config.json")
st_pattern = os.path.join(model_directory, "*.safetensors")
model_path = glob.glob(st_pattern)
lora_config_path = os.path.join(lora_directory, "adapter_config.json")
lora_path = os.path.join(lora_directory, "adapter_model.bin")
# Create config, model, tokenizer and generator
config = ExLlamaConfig(model_config_path) # create config from config.json
config.model_path = model_path # supply path to model weights file
model = ExLlama(config) # create ExLlama instance and load the weights
tokenizer = ExLlamaTokenizer(tokenizer_path) # create tokenizer from tokenizer model file
cache = ExLlamaCache(model) # create cache for inference
generator = ExLlamaGenerator(model, tokenizer, cache) # create generator
# Load LoRA
lora = ExLlamaLora(model, lora_config_path, lora_path)
# Configure generator
generator.settings.token_repetition_penalty_max = 1.2
generator.settings.temperature = 0.65
generator.settings.top_p = 0.4
generator.settings.top_k = 0
generator.settings.typical = 0.0
# Alpaca prompt
prompt = \
"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n" \
"\n" \
"### Instruction:\n" \
"List five colors in alphabetical order.\n" \
"\n" \
"### Response:"
# Generate with LoRA
print(" --- LoRA ----------------- ")
print("")
generator.lora = lora
torch.manual_seed(1337)
output = generator.generate_simple(prompt, max_new_tokens = 200)
print(output)
# Generate without LoRA
print("")
print(" --- No LoRA -------------- ")
print("")
generator.lora = None
torch.manual_seed(1337)
output = generator.generate_simple(prompt, max_new_tokens = 200)
print(output)