Spaces:
Build error
Build error
File size: 6,283 Bytes
5002792 c8f289a 5f8e100 5002792 5f8e100 5002792 4203a9b 5f8e100 5002792 c8f289a 5002792 c8f289a 5f8e100 5002792 5f8e100 5002792 c8f289a 5002792 c8f289a 5002792 c8f289a 5f8e100 5002792 5f8e100 5002792 c8f289a 5002792 c8f289a |
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 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 |
import os
import re
import torch
from transformers import (
AutoModelForCausalLM,
AutoTokenizer,
BitsAndBytesConfig,
TextStreamer,
)
from tqdm import tqdm
def load_model(
model_name,
dtype=torch.bfloat16,
load_in_4bit=False,
adapter_name_or_path=None,
):
print(f"loading model: {model_name}")
if adapter_name_or_path:
from llamafactory.chat import ChatModel
template = "llama3" if "llama-3" in model_name.lower() else "chatml"
args = dict(
model_name_or_path=model_name,
adapter_name_or_path=adapter_name_or_path, # load the saved LoRA adapters
template=template, # same to the one in training
finetuning_type="lora", # same to the one in training
quantization_bit=4 if load_in_4bit else None, # load 4-bit quantized model
)
chat_model = ChatModel(args)
if os.getenv("RESIZE_TOKEN_EMBEDDINGS") == "true":
chat_model.engine.model.resize_token_embeddings(
len(chat_model.engine.tokenizer), pad_to_multiple_of=32
)
return chat_model.engine.model, chat_model.engine.tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
bnb_config = BitsAndBytesConfig(
load_in_4bit=load_in_4bit,
bnb_4bit_quant_type="nf4",
bnb_4bit_use_double_quant=False,
bnb_4bit_compute_dtype=dtype,
)
model = (
AutoModelForCausalLM.from_pretrained(
model_name,
quantization_config=bnb_config,
torch_dtype=dtype,
trust_remote_code=True,
device_map="auto",
)
if load_in_4bit
else AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=dtype,
trust_remote_code=True,
device_map="auto",
)
)
return model, tokenizer
def check_gpu():
# torch.cuda.is_available() checks and returns a Boolean True if a GPU is available, else it'll return False
is_cuda = torch.cuda.is_available()
# If we have a GPU available, we'll set our device to GPU. We'll use this device variable later in our code.
if is_cuda:
device = torch.device("cuda")
print("GPU is available")
elif torch.backends.mps.is_available():
device = torch.device("mps")
print("MPS is available")
else:
device = torch.device("cpu")
print("GPU/MPS not available, CPU used")
return device
def test_model(model, tokenizer, prompt, device="cuda"):
inputs = tokenizer(
[prompt],
return_tensors="pt",
).to(device)
text_streamer = TextStreamer(tokenizer)
_ = model.generate(
**inputs, max_new_tokens=2048, streamer=text_streamer, use_cache=True
)
def extract_answer(text, debug=False):
if text:
# Remove the begin and end tokens
text = re.sub(
r".*?(assistant|\[/INST\]).+?\b",
"",
text,
flags=re.DOTALL | re.MULTILINE,
)
if debug:
print("--------\nstep 1:", text)
text = re.sub(r"<.+?>.*", "", text, flags=re.DOTALL | re.MULTILINE)
if debug:
print("--------\nstep 2:", text)
text = re.sub(
r".*?end_header_id\|>\n\n", "", text, flags=re.DOTALL | re.MULTILINE
)
if debug:
print("--------\nstep 3:", text)
text = text.split(".")[0].strip()
if debug:
print("--------\nstep 4:", text)
text = re.sub(
r"^Response:.+?\b",
"",
text,
flags=re.DOTALL | re.MULTILINE,
)
if debug:
print("--------\nstep 5:", text)
return text
def eval_model(model, tokenizer, eval_dataset, device="cuda"):
total = len(eval_dataset)
predictions = []
for i in tqdm(range(total)):
inputs = tokenizer(
eval_dataset["prompt"][i : i + 1],
return_tensors="pt",
).to(device)
outputs = model.generate(**inputs, max_new_tokens=4096, use_cache=False)
decoded_output = tokenizer.batch_decode(outputs)
debug = i == 0
decoded_output = [
extract_answer(output, debug=debug) for output in decoded_output
]
predictions.extend(decoded_output)
return predictions
def save_model(
model,
tokenizer,
include_gguf=True,
include_merged=True,
publish=True,
):
try:
token = os.getenv("HF_TOKEN") or None
model_name = os.getenv("MODEL_NAME")
save_method = "lora"
quantization_method = "q5_k_m"
model_names = get_model_names(
model_name, save_method=save_method, quantization_method=quantization_method
)
model.save_pretrained(model_names["local"])
tokenizer.save_pretrained(model_names["local"])
if publish:
model.push_to_hub(
model_names["hub"],
token=token,
)
tokenizer.push_to_hub(
model_names["hub"],
token=token,
)
if include_merged:
model.save_pretrained_merged(
model_names["local"] + "-merged", tokenizer, save_method=save_method
)
if publish:
model.push_to_hub_merged(
model_names["hub"] + "-merged",
tokenizer,
save_method="lora",
token="",
)
if include_gguf:
model.save_pretrained_gguf(
model_names["local-gguf"],
tokenizer,
quantization_method=quantization_method,
)
if publish:
model.push_to_hub_gguf(
model_names["hub-gguf"],
tokenizer,
quantization_method=quantization_method,
token=token,
)
except Exception as e:
print(e)
def print_row_details(df, indices=[0]):
for index in indices:
for col in df.columns:
print("-" * 50)
print(f"{col}: {df[col].iloc[index]}")
|