File size: 21,310 Bytes
357a956 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 9b4534c 9791f0c 357a956 9791f0c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 |
import gradio as gr
import pandas as pd
col=['Layer number', 'Hidden size', 'FFN Hidden size', 'Sequence length', 'Head number', 'Group number',
'dp', 'tp', 'pp', 'cp', 'GPU numbers', 'Batch size', 'FP8', 'Model parameters', 'Model_states', 'Activation', 'Total']
# # global data
# table_data = pd.DataFrame(columns=col)
def Get_GigaByte(memory):
return memory / 1024**3
def Get_BillionParameter(parameter):
return parameter / 1000**3
# model states:
def Compute_Parameters_input(hidden_size, vocab_size, tp):
num_parameters_word_embedding = hidden_size * vocab_size / tp
num_parameters_position_embedding = 0 #args.hidden_size * args.seq_length
return num_parameters_word_embedding + num_parameters_position_embedding
def Compute_Parameters_output(hidden_size, vocab_size, tp):
num_parameters_output_layernorm = 2 * hidden_size
num_parameters_output_embedding = 0 # due to sharedWordEmbedding
return num_parameters_output_layernorm + num_parameters_output_embedding
def Compute_Parameters_attention(hidden_size, kv_hidden_size, is_bias, tp):
# attention:
# layernorm: 2h
num_parameters_attention = 2 * hidden_size
# QKV weight: 3h*h/tp, bias: 3h/tp
# output linear weight: h*h/tp, bias: h
num_parameters_attention_Q_weight = hidden_size * hidden_size / tp
num_parameters_attention_KV_weight = 2 * kv_hidden_size * hidden_size / tp
num_parameters_attention_Linear_weight = hidden_size * hidden_size / tp
num_parameters_attention += num_parameters_attention_Q_weight + num_parameters_attention_KV_weight + num_parameters_attention_Linear_weight
if is_bias == "True":
num_parameters_attention += (hidden_size + 2 * kv_hidden_size) / tp + hidden_size
return num_parameters_attention
def Compute_Parameters_mlp(hidden_size, ffn_size, is_bias, act_func, tp):
# MLP:
# layernorm: 2h
num_parameters_mlp = 2 * hidden_size
# mlp1 weight: h*ffn/tp, bias: ffn/tp
# mlp2 weight: ffn*h/tp, bias: h
if act_func == "True":
num_parameters_mlp += hidden_size * ffn_size * 3 / tp
if is_bias == "True":
num_parameters_mlp += ffn_size * 2 / tp + hidden_size
else:
num_parameters_mlp += hidden_size * ffn_size * 2 / tp
if is_bias == "True":
num_parameters_mlp += ffn_size / tp + hidden_size
return num_parameters_mlp
def Compute_Parameters(vocab_size, layer_num, hidden_size, ffn_size, is_group_query, group_query_num, is_bias, act_func, head_num, tp, pp):
if is_group_query == "False":
group_query_num = head_num
kv_hidden_size = hidden_size / head_num * group_query_num
# input part
num_parameters_input = Compute_Parameters_input(hidden_size, vocab_size, tp)
# middle layers part
num_parameters_attention = Compute_Parameters_attention(hidden_size, kv_hidden_size, is_bias, tp)
num_parameters_mlp = Compute_Parameters_mlp(hidden_size, ffn_size, is_bias, act_func, tp)
num_parameters_in_single_layer = num_parameters_attention + num_parameters_mlp
num_parameters_in_total_layers = num_parameters_in_single_layer * layer_num / pp
# output part
parameters_output = Compute_Parameters_output(hidden_size, vocab_size, tp)
if pp == 1:
num_parameters_total = (
num_parameters_input
+ num_parameters_in_total_layers
+ parameters_output # num_parameters_output_layernorm
)
else:
num_parameters_total = (
num_parameters_input
+ num_parameters_in_total_layers
)
return num_parameters_total
def Compute_Weight(numParametersTotal, is_fp8, is_fp8_init):
if is_fp8 == "False":
weight_memory = 2 * numParametersTotal
elif is_fp8_init == "False":
weight_memory = 4 * numParametersTotal
else:
weight_memory = 2 * numParametersTotal
return weight_memory
def Compute_Gradient(numParametersTotal, g_ty):
if g_ty == "FP32":
gradient_memory = 4 * numParametersTotal
elif g_ty =="BF16":
gradient_memory = 2 * numParametersTotal
return gradient_memory
def Compute_Optimizer_states(numParametersTotal, o_ty, is_dist_opt, dp, cp):
if o_ty == "FP32":
optimizer_memory = 4 * 2 * numParametersTotal
elif o_ty =="BF16":
optimizer_memory = 2 * 2 * numParametersTotal
if is_dist_opt == "True":
optimizer_memory = optimizer_memory / (dp * cp)
return optimizer_memory
def Compute_Master_weight(numParametersTotal, is_dist_opt, dp, cp):
master_weight_memory = 4 * numParametersTotal
if is_dist_opt == "True":
master_weight_memory = master_weight_memory / (dp * cp)
return master_weight_memory
def Compute_Model_states(vocab_size, layer_num, hidden_size, ffn_size, head_num, is_group_query, group_query_num, is_bias, act_func,
dp, tp, pp, cp, is_dist_opt, is_fp8, is_fp8_init, g_ty, o_ty):
numParametersTotal = Compute_Parameters(vocab_size, layer_num, hidden_size, ffn_size, is_group_query, group_query_num, is_bias, act_func, head_num, tp, pp)
weight_memory = Compute_Weight(numParametersTotal, is_fp8, is_fp8_init)
gradient_memory = Compute_Gradient(numParametersTotal, g_ty)
optimizer_memory = Compute_Optimizer_states(numParametersTotal, o_ty, is_dist_opt, dp, cp)
master_weight_memory = Compute_Master_weight(numParametersTotal, is_dist_opt, dp, cp)
return numParametersTotal, weight_memory, gradient_memory, optimizer_memory, master_weight_memory, \
weight_memory + gradient_memory + optimizer_memory + master_weight_memory
# activation memory:
def compute_activation_memory_attention(activation_dtype, seq_length, b, hidden_size, kv_hidden_size, is_sp, tp):
# LN 2bsq
activation_mem_attn_ln = seq_length * b * hidden_size * 2
if is_sp == "False":
activation_mem_attn_ln *= tp
# attention input X, qkv 2bsh/1bsh
activation_mem_attn_qkv = seq_length * b * hidden_size * activation_dtype
if is_sp == "False":
activation_mem_attn_qkv *= tp
# attention q 2bsh
activation_mem_attn_q = seq_length * b * hidden_size * 2
# attention k and v 4bsh
activation_mem_attn_kv = seq_length * b * kv_hidden_size * 2 * 2
# attention proj input 2bsh/1bsh
activation_mem_attn_proj = seq_length * b * hidden_size * activation_dtype
# dropout bsh
activation_mem_attn_dropout = seq_length * b * hidden_size
if is_sp == "False":
activation_mem_attn_dropout *= tp
# bf16: 2+2+2+4+2+1=13bsh
# fp8: 2+1+2+4+1+1=11bsh
activation_memory_attn = (
activation_mem_attn_ln
+ activation_mem_attn_qkv
+ activation_mem_attn_q
+ activation_mem_attn_kv
+ activation_mem_attn_proj
+ activation_mem_attn_dropout
)
return activation_memory_attn
def compute_activation_memory_mlp(activation_dtype, seq_length, b, hidden_size, ffn_size, act_func, is_sp, tp):
# LN 2bsh
activation_mem_mlp_ln = seq_length * b * hidden_size * 2
if is_sp == "False":
activation_mem_mlp_ln *= tp
# FC1 2bsh/1bsh
activation_mem_mlp_fc1 = seq_length * b * hidden_size * activation_dtype
if is_sp == "False":
activation_mem_mlp_fc1 *= tp
# Act 8bsh
if act_func == "Swiglu":
activation_mem_mlp_act = seq_length * b * ffn_size * 2 * 2
else:
activation_mem_mlp_act = seq_length * b * ffn_size * 2
# FC2 8bsh/4bsh
activation_mem_mlp_fc2 = seq_length * b * ffn_size * activation_dtype
# dropout bsh
activation_mem_mlp_dropout = seq_length * b * hidden_size
if is_sp == "False":
activation_mem_mlp_dropout *= tp
# bf16: 2+2+8+8+1=21
# fp8: 2+1+8+4+1=16
activation_memory_mlp = (
activation_mem_mlp_ln
+ activation_mem_mlp_fc1
+ activation_mem_mlp_act
+ activation_mem_mlp_fc2
+ activation_mem_mlp_dropout
)
return activation_memory_mlp
def compute_activation_memory_input(seq_length, b, hidden_size, pp):
# embedding + Dropout
return 8 * seq_length * b * pp + seq_length * b * hidden_size * pp
def compute_activation_memory_output(seq_length, b, hidden_size, vocab_size):
# Inputs to output layer and CE loss(bf16, fp32 * 2).
return 2 * seq_length * b * hidden_size + (2 + 4 + 4) * seq_length * b * vocab_size
def compute_activation_memory_pp(activation_memory, is_ip, vp, pp, num_microbatches):
# Multiply by interleaved PP memory factor.
if is_ip == "True":
interleaved_schedule_memory_penalty = 1 + (pp - 1) / (pp * vp)
activation_memory *= interleaved_schedule_memory_penalty
# If using non-interleaved schedule, number of microbatches in pipeline can be less than pp_size,
# so discount accordingly.
if is_ip == "False" and pp > 1:
if num_microbatches > 1:
activation_memory *= min(1, num_microbatches / pp)
return activation_memory
def compute_activation_memory(vocab_size, seq_length, layer_num, b, b_global, head_num, hidden_size, ffn_size, act_func, is_fp8, is_sp, is_group_query, group_query_num, tp, pp, dp, cp, is_ip, vp):
# Using formula in Table 2 of https://arxiv.org/pdf/2205.05198.pdf.
# We are trying to compute the maximum activation footprint, so all calculations in this function
# are for the first pipeline stage.
# activation dataType
if is_fp8 == "False":
activation_dtype = 2
else:
activation_dtype = 1
# kv_hidden_size
if is_group_query == "False":
group_query_num = head_num
kv_hidden_size = hidden_size / head_num * group_query_num
activation_memory_attn = compute_activation_memory_attention(activation_dtype, seq_length, b, hidden_size, kv_hidden_size, is_sp, tp)
activation_memory_mlp = compute_activation_memory_mlp(activation_dtype, seq_length, b, hidden_size, ffn_size, act_func, is_sp, tp)
activation_memory = activation_memory_attn + activation_memory_mlp
activation_memory *= layer_num
# Now add activation memory required for input embeddings, last LayerNorm and output layer.
# Input to embedding (pp_size microbatches in flight).
activation_memory_input = compute_activation_memory_input(seq_length, b, hidden_size, pp)
activation_memory += activation_memory_input
# get num_microbatches
num_microbatches = b_global / b / dp / cp
activation_memory = compute_activation_memory_pp(activation_memory, is_ip, vp, pp, num_microbatches)
if pp == 1:
# Inputs to output layer and CE loss(fp32).
activation_memory_output = compute_activation_memory_output(seq_length, b, hidden_size, vocab_size)
activation_memory += activation_memory_output
elif pp > 1:
# Sendrecv memory
activation_memory += seq_length * b * hidden_size * 2
# Activation memory is partitioned by TP size due to tensor and sequence model parallelism.
return activation_memory / tp / cp
# compute_btn.click.function
def Compute_ALL_Model_memory(vocab_size, layer_num, hidden_size, ffn_size, seq_length, head_num, is_group_query, group_query_num, is_bias, act_func,
dp, tp, pp, cp, is_sp, is_ip, vp, is_dist_opt, b, b_global, is_fp8, is_fp8_init, g_ty, o_ty, record_df, count):
# get model states
numParameters, weight_memory, gradient_memory, optimizer_memory, master_weight_memory, model_states_memory = Compute_Model_states(vocab_size, layer_num, hidden_size,
ffn_size, head_num, is_group_query, group_query_num, is_bias, act_func, dp, tp, pp, cp, is_dist_opt, is_fp8, is_fp8_init, g_ty, o_ty)
# get activation memory
activation_memory = compute_activation_memory(vocab_size, seq_length, layer_num, b, b_global, head_num, hidden_size, ffn_size, act_func, is_fp8, is_sp, is_group_query, group_query_num, tp, pp, dp, cp, is_ip, vp)
# get model parameters
numParametersTotal = Compute_Parameters(vocab_size, layer_num, hidden_size, ffn_size, is_group_query, group_query_num, is_bias, act_func, head_num, 1, 1)
# get gpu number
gpu_num = dp * tp * pp * cp
# get B/GB
numParametersTotal = round(Get_BillionParameter(numParametersTotal), 3)
numParameters = round(Get_BillionParameter(numParameters), 3)
model_states_memory = round(Get_GigaByte(model_states_memory), 3)
activation_memory = round(Get_GigaByte(activation_memory), 3)
Total = round(model_states_memory + activation_memory, 3)
# record
new_row = pd.DataFrame([[layer_num, hidden_size, ffn_size, seq_length, head_num, group_query_num, dp, tp, pp, cp, gpu_num, b, is_fp8,
numParametersTotal, model_states_memory, activation_memory, Total]],
columns=col)
if count == 1:
record_df = new_row
else:
record_df = record_df._append(new_row, ignore_index=True)
count = count + 1
# return str(gpu_num), str(model_states) + " GB", str(activation) + " GB", str(total) + " GB", table_data
return f"""
GPU numbers = {str(gpu_num)}, \n
Total model parameters = {str(numParametersTotal)} B, \n
Model parameters = {str(numParameters)} B, \n
Model_states = {str(model_states_memory)} GB, \n
Activation = {str(activation_memory)} GB, \n
Total memory consumption = {str(Total)} GB \n
""", record_df, count
def generate_csv(record_df):
# 将 DataFrame 保存为 CSV 文件
csv_filename = "data.csv"
record_df.to_csv(csv_filename, index=False)
# 返回 CSV 文件路径
return csv_filename
# formula string
formula = r"""
> **Note**🔑: In this formula, we assume LLM training with FP32 Gradient and Optimizer state, and bias = False, Zero1 = False, SP = True.
<!-- parameters: -->
$$
P_{input} = \frac{HV}{tp}, \quad
P_{output} = 2H \\\\
P_{attn} = 2H + \frac{2H^2 + 2H_{KV} \times H}{tp}, \quad
P_{MLP} = 2H +
\\begin{cases}
\frac{3H \times FFN}{tp}, & \text{if }GLU\text{ is True} \\\\
\frac{2H \times FFN}{tp}, & \text{if }GLU\text{ is False}
\\end{cases} \\\\
P_{middle} = \frac{(P_{attn} + P_{MLP}) \times L}{pp} \\\\
P = P_{input} + P_{middle} +
\\begin{cases}
P_{output}, & \text{if }pp = 1 \\\\
0, & \text{if }pp > 1
\\end{cases} \\\\
{Total\ Model\ parameters} =
\\begin{cases}
P, & \text{set tp = 1, pp = 1} \\\\
2HV + 2H + (4H + 2H^2 + 2H_{KV} \times H + 3FFN \times H) \times L, & \text{general formula}
\\end{cases} \\\\
{Model\ states} = {Model\ weight} + {Gradient} + {Optimizer\ state} + {Master\ weight} =
\\begin{cases}
18P, & \text{BF16 training} \\\\
18P, & \text{FP8 training with FP8 Init} \\\\
20P, & \text{FP8 training w/o FP8 Init}
\\end{cases} \\\\
$$
***
<!-- activations: -->
$$
A_{input} = (8SB + SBH) \times pp, \quad
A_{output} = 2SBH +
\\begin{cases}
10SBV, & \text{if }pp\text{ = 1} \\\\
0, & \text{if }pp\text{ > 1}
\\end{cases} \\\\
A_{attn} = 5SBH + 4SB \times H_{KV} +
\\begin{cases}
2SBH, & \text{if } FP8 \text{ is True} \\\\
4SBH, & \text{if } FP8 \text{ is False}
\\end{cases} \\\\
A_{MLP} = 3SBH +
\\begin{cases}
SBH + SB \times FFN + 4SB \times FFN, & \text{if }FP8 \text{ is True and }GLU \text{ is True} \\\\
2SBH + 2SB \times FFN + 4SB \times FFN, & \text{if }FP8 \text{ is False and }GLU \text{ is True} \\\\
SBH + SB \times FFN + 2SB \times FFN, & \text{if }FP8 \text{ is True and }GLU \text{ is False} \\\\
2SBH + 2SB \times FFN + 2SB \times FFN, & \text{if }FP8 \text{ is False and }GLU \text{ is False}
\\end{cases} \\\\
A_{middle} = (A_{attn} + A_{MLP}) \times L \\\\
A_{ip} = (A_{input} + A_{middle}) \times
\\begin{cases}
(1 + \frac{pp - 1}{pp \times vp}), & \text{if } Interleaved\ Pipeline \text{ is True} \\\\
min(1, \frac{microbatch}{pp}), & \text{if } Interleaved\ Pipeline \text{ is False and pp > 1} \\\\
1, & \text{other}
\\end{cases} \\\\
Activation =
\\begin{cases}
\frac{A_{ip} + A_{output}}{tp \times cp}, & \text{if pp = 1} \\\\
\frac{A_{ip} + 2BSH}{tp \times cp}, & \text{if pp > 1}
\\end{cases}
$$
***
$$
\\begin{gather}
{GPU\ numbers} = tp \times pp \times dp \times cp\\\\
{Total\ memory\ consumption} = {Model\ states} + Activation
\\end{gather}
$$
"""
with gr.Blocks() as demo:
with gr.Row():
# Text
gr.Markdown(
"""
<div style="text-align: center;">
<h1>GPU memory calculator 🌀</h1>
<p style="font-size:16px;">Here's a GPU memory calculator, it helps you to compute memory comsumption in LLM training. </p>
</div>
"""
)
with gr.Column():
# Input 1.[Model Parameters]
gr.Markdown(
"""
<h1>Model Parameters:</h1>
"""
)
with gr.Accordion("Model Parameters"):
act_func = gr.Radio(["True", "False"], value="True", label="Model type", info="Action Function in MLP, whether to use GLU (Gated Linear Unit). [e.g \"True\" for LlaMA, \"False\" for GPT.]")
vocab_size = gr.Number(label="Vocab size", value=32000)
layer_num = gr.Number(label="Layer number", value=32)
hidden_size = gr.Number(label="Hidden size", value=4096)
ffn_size = gr.Number(label="FFN Hidden size", value=11008)
sequence_len = gr.Number(label="Sequence length", value=1024)
head_num = gr.Number(label="Number of Attention Heads", value=32)
with gr.Row():
is_group_query = gr.Radio(["True", "False"], value="True", label="Use Group Query Attention")
group_query_num = gr.Number(label="Number of Query Groups", value=96)
is_bias = gr.Radio(["True", "False"], value="False", label="Use Bias")
# Input 2.[Parallelism]
gr.Markdown(
"""
<h1>Parallelism config:</h1>
"""
)
with gr.Accordion("Parallelism config"):
dp = gr.Number(label="Data parallelism", value=1)
tp = gr.Number(label="Tensor parallelism", value=2)
pp = gr.Number(label="Pipeline parallelism", value=2)
cp = gr.Number(label="Context parallelism", value=2)
is_sp = gr.Radio(["True", "False"], value="True", label="Sequence parallelism")
with gr.Row():
is_ip = gr.Radio(["True", "False"], value="False", label="Use Interleaved Pipeline")
vp = gr.Number(label="Virtual Pipeline Size")
is_dist_opt = gr.Radio(["True", "False"], value="True", label="Use Distributed Optimizer(Zero1)")
# Input 3.[Training Settings]
gr.Markdown(
"""
<h1>Training Config:</h1>
"""
)
with gr.Accordion("Training Config"):
b = gr.Number(label="Micro Batch size", value=4)
b_global = gr.Number(label="Global Batch size", value=64)
gr.Checkbox(label="True", value=True, info="BF16 Training")
is_fp8 = gr.Radio(["True", "False"], value="True", label="FP8 Training")
is_fp8_init = gr.Radio(["True", "False"], value="True", label="FP8 Initialization(will reduce memory)")
g_ty = gr.Dropdown(["FP32", "BF16"], value="FP32", label="Gradients Dtype")
o_ty = gr.Dropdown(["FP32", "BF16"], value="FP32", label="Optimizer State Dtype")
with gr.Column():
gr.Markdown(
"""
<h1>Output Data:</h1>
"""
)
formula = formula
gr.Markdown(
formula
, latex_delimiters=[{ "left": "$$", "right": "$$", "display": True }]
)
output_text = gr.Textbox(
label="Compute result",
interactive=False,
)
# Button
with gr.Row():
compute_btn = gr.Button("Compute")
download_btn = gr.Button("Download")
record_df = gr.Dataframe(
label="Record Table",
headers=col
)
count = gr.Number(label="Row count", value=1, visible=False)
compute_btn.click(
fn=Compute_ALL_Model_memory,
inputs=[vocab_size, layer_num, hidden_size, ffn_size, sequence_len, head_num, is_group_query, group_query_num, is_bias, act_func,
dp, tp, pp, cp, is_sp, is_ip, vp, is_dist_opt, b, b_global, is_fp8, is_fp8_init, g_ty, o_ty, record_df, count],
outputs=[output_text, record_df, count]
)
output_file=gr.File(label="When you click the download button, the downloaded form will be displayed here.")
# download func
download_btn.click(
fn=generate_csv,
inputs=record_df,
outputs=output_file
)
if __name__ == "__main__":
demo.launch()
|