File size: 10,140 Bytes
96ba80e | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 | from collections import OrderedDict
from quantize.int_linear import QuantLinear
import torch
from quantize.int_matmul import QuantMatMul
from models.transformation import *
from pprint import pprint
from parallel_utils import map_layers_to_multi_gpus
import os
from tqdm import tqdm
import torch.nn as nn
import numpy as np
from datautils import get_loaders
from accelerate.utils import get_balanced_memory
from accelerate import dispatch_model,infer_auto_device_map
import copy
def get_max_memory_map(ratio=0.95):
max_memory = {}
for i in range(torch.cuda.device_count()):
total_mem = torch.cuda.get_device_properties(i).total_memory
mem_in_gib = int(total_mem * ratio / (1024 ** 3)) # 转换为 GiB
max_memory[i] = f"{mem_in_gib}GiB"
return max_memory
@torch.no_grad()
def evaluate(lm, args, logger):
results = {}
assert not (args.multigpu and args.parallelize)
if args.multigpu:
if "llama" in args.net.lower() or "vicuna" in args.net.lower() or "qwen" in args.net.lower():
map_layers_to_multi_gpus(lm.model.model.layers)
input_device = lm.model.model.layers[0].device
output_device = lm.model.model.layers[-1].device
assert input_device == output_device
lm._device = input_device
lm.model.model.embed_tokens.to(input_device)
lm.model.model.norm.to(output_device)
lm.model.lm_head.to(output_device)
else:
raise NotImplementedError("Only support for llama/vicuna/qwen multigpu now")
elif args.parallelize:
balanced_mem = get_balanced_memory(
lm.model,
max_memory=get_max_memory_map(0.95),
no_split_module_classes=["LlamaDecoderLayer","QuantLlamaDecoderLayer","Qwen3MoeDecoderLayer"]
)
logger.info(f"mem is {balanced_mem}")
device_map = infer_auto_device_map(
lm.model,
max_memory=balanced_mem,
no_split_module_classes=["LlamaDecoderLayer","QuantLlamaDecoderLayer","Qwen3MoeDecoderLayer"]
)
lm.model = dispatch_model(lm.model,device_map=device_map)
else:
if "llama" in args.net.lower() or "vicuna" in args.net.lower() or "qwen" in args.net.lower():
lm.model = lm.model.to(lm.device)
else:
raise NotImplementedError("Only support for llama/vicuna/qwen single gpu now")
if args.eval_ppl:
logger.info(f"model seqlen is {lm.seqlen}")
datasets = args.test_datasets.split(",")
for dataset in datasets:
# for dataset in ["wikitext2"]:
cache_testloader = f'{args.cache_dir}/testloader_{args.net}_{dataset}_all.cache'
if os.path.exists(cache_testloader):
testloader = torch.load(cache_testloader)
logger.info(f"load calibration from {cache_testloader}")
else:
dataloader, testloader = get_loaders(
dataset,
seed=args.seed,
model=args.model,
seqlen=lm.seqlen,
args=args
)
torch.save(testloader, cache_testloader)
if "c4" in dataset:
testenc = testloader
else:
testenc = testloader.input_ids
nsamples = testenc.numel() // lm.seqlen
use_cache = lm.model.config.use_cache
lm.model.config.use_cache = False
lm.model.eval()
nlls = []
all_hidden_states = []
output_hidden_states = False
for i in tqdm(range(nsamples)):
batch = testenc[:, (i * lm.seqlen) : ((i + 1) * lm.seqlen)].to(lm.device)
outputs = lm.model.model(batch,output_hidden_states=output_hidden_states)
hidden_states = outputs[0]
if hasattr(lm.model.lm_head,"bias") and lm.model.lm_head.bias is not None:
lm.model.lm_head.bias = torch.nn.Parameter(lm.model.lm_head.bias.to(lm.model.lm_head.weight.device))
logits = lm.model.lm_head(hidden_states)
shift_logits = logits[:, :-1, :]
shift_labels = testenc[:, (i * lm.seqlen) : ((i + 1) * lm.seqlen)][
:, 1:
].to(lm.model.lm_head.weight.device)
loss_fct = nn.CrossEntropyLoss()
loss = loss_fct(
shift_logits.view(-1, shift_logits.size(-1)),
shift_labels.view(-1),
)
neg_log_likelihood = loss.float() * lm.seqlen
nlls.append(neg_log_likelihood)
if i == args.limit:
break
ppl = torch.exp(torch.stack(nlls).sum() / (nsamples * lm.seqlen))
logger.info(f'{dataset} : {ppl.item()}')
lm.model.config.use_cache = use_cache
results[dataset] = ppl.item()
if args.tasks != "":
args.tasks = args.tasks.split(",")
import lm_eval
from lm_eval import utils as lm_eval_utils
from lm_eval.api.registry import ALL_TASKS
from lm_eval.models.huggingface import HFLM
print(f"use lm_eval in {lm_eval}")
task_manager = lm_eval.tasks.TaskManager(include_path="./datasets_local/lm_eval_configs/tasks", include_defaults=True)
hflm = HFLM(pretrained=lm.model,tokenizer=lm.tokenizer, batch_size=args.lm_eval_batch_size)
t_results = lm_eval.simple_evaluate(hflm, tasks=args.tasks, batch_size=args.lm_eval_batch_size,task_manager=task_manager)['results']
metric_vals = {task: round(result.get('acc_norm,none', result['acc,none']), 4) for task, result in t_results.items()}
logger.info(metric_vals)
pprint(metric_vals)
if args.eval_ppl is True:
metric_vals.update(results)
reported_metric_vals = {}
for k,v in metric_vals.items():
if "mmlu" in k:
if k == "mmlu":
reported_metric_vals[k] = v
else:
reported_metric_vals[k] = v
import pandas as pd
if os.path.exists(f"{args.output_dir}/results.csv"):
df = pd.read_csv(f"{args.output_dir}/results.csv")
new_df = pd.DataFrame(reported_metric_vals,index=[0])
df[new_df.columns] = new_df
else:
df = pd.DataFrame(reported_metric_vals,index=[0])
if args.eval_ppl:
new_columns = ['wikitext2','c4']
else:
new_columns = []
new_columns += args.tasks
if len(args.tasks) >= 5:
df["avg-5"] = df[['piqa','arc_easy', 'arc_challenge','hellaswag', 'winogrande']].mean(axis=1)
if len(args.tasks) >= 6:
df["avg-6"] = df[['piqa','arc_easy', 'arc_challenge','hellaswag', 'winogrande', 'boolq']].mean(axis=1)
if 'mmlu' in args.tasks and len(args.tasks) >= 7:
df["avg-7"] = df[['piqa','arc_easy', 'arc_challenge','hellaswag', 'winogrande', 'boolq','mmlu']].mean(axis=1)
logger.info(df)
df.to_csv(f"{args.output_dir}/results.csv",index=False)
model = lm.model
if "llama" in args.net.lower() or "vicuna" in args.net.lower() or "qwen" in args.net.lower():
model.model.embed_tokens = model.model.embed_tokens.cpu()
model.model.norm = model.model.norm.cpu()
else:
raise ValueError("Only support for opt/llama/Llama-2/falcon/mixtral now")
return results
def get_slider_parameters(sub_layers, use_list=["scale","alpha","shift"]):
params = []
print(f"get {use_list} parameters!")
for sub_layer_idx in range(len(sub_layers)):
for n, m in sub_layers[sub_layer_idx].named_parameters():
if any(n.find(t) > -1 for t in use_list) and not n.find('bound_factor') > -1:
params.append(m)
# print(n)
# print(params)
return iter(params)
def get_lwc_parameters(sub_layers):
params = []
print("get lwc parameters!")
for sub_layer_idx in range(len(sub_layers)):
for n, m in sub_layers[sub_layer_idx].named_parameters():
if n.find('bound_factor') > -1:
params.append(m)
return iter(params)
def try_delete_object(object,logger,name=None):
try:
del object
except Exception as e:
logger.info(f"del tensor occurs {e}, skip!")
else:
logger.info(f"del tensor {name} successfully!")
def cleanup_memory(verbos=True,logger=None) -> None:
"""Run GC and clear GPU memory."""
import gc
import inspect
caller_name = ''
try:
caller_name = f' (from {inspect.stack()[1].function})'
except (ValueError, KeyError):
pass
def total_reserved_mem() -> int:
return sum(torch.cuda.memory_reserved(device=i) for i in range(torch.cuda.device_count()))
memory_before = total_reserved_mem()
# gc.collect and empty cache are necessary to clean up GPU memory if the model was distributed
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
memory_after = total_reserved_mem()
if verbos and logger:
logger.info(
f"GPU memory{caller_name}: {memory_before / (1024 ** 3):.2f} -> {memory_after / (1024 ** 3):.2f} GB"
f" ({(memory_after - memory_before) / (1024 ** 3):.2f} GB)"
)
def slider_state_dict(model, destination=None, prefix='', keep_vars=False):
if destination is None:
destination = OrderedDict()
for name, param in model.named_parameters():
if name.find('smooth') > -1 or name.find('bound_factor') > -1 or name.find('lora_') > -1 or name.find('Q_') > -1 :
destination[prefix + name] = param if keep_vars else param.detach()
return destination
def register_scales_and_zeros(model):
for name, module in model.named_modules():
if isinstance(module, QuantLinear):
module.weight_quantizer.register_scales_and_zeros()
|