bitsandbytes-rocm/bitsandbytes/optim/adam.py

332 lines
12 KiB
Python
Raw Normal View History

# Copyright (c) Facebook, Inc. and its affiliates.
#
# This source code is licensed under the MIT license found in the
2021-10-06 02:16:20 +00:00
# LICENSE file in the root directory of this source tree.
2021-10-31 15:38:38 +00:00
import math
import os
2021-10-21 23:04:18 +00:00
import torch
2021-10-31 15:38:38 +00:00
import torch.distributed as dist
2021-10-21 22:53:20 +00:00
import bitsandbytes.functional as F
from bitsandbytes.optim.optimizer import Optimizer2State
2021-10-06 02:16:20 +00:00
class Adam(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=0,
amsgrad=False,
optim_bits=32,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
):
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
optim_bits,
args,
min_8bit_size,
percentile_clipping,
block_wise,
)
2021-10-06 02:16:20 +00:00
class Adam8bit(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=0,
amsgrad=False,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
):
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
8,
args,
min_8bit_size,
percentile_clipping,
block_wise,
)
2021-10-06 02:16:20 +00:00
class Adam32bit(Optimizer2State):
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=0,
amsgrad=False,
args=None,
min_8bit_size=4096,
percentile_clipping=100,
block_wise=True,
):
super().__init__(
"adam",
params,
lr,
betas,
eps,
weight_decay,
32,
args,
min_8bit_size,
percentile_clipping,
block_wise,
)
2021-10-06 02:16:20 +00:00
2021-10-21 17:20:41 +00:00
class AnalysisAdam(torch.optim.Optimizer):
2021-10-21 22:53:20 +00:00
"""Adam that performs 8-bit vs 32-bit error analysis.
2021-10-21 17:20:41 +00:00
This implementation is modified from torch.optim.Adam based on:
`Fixed Weight Decay Regularization in Adam`
(see https://arxiv.org/abs/1711.05101)
It has been proposed in `Adam: A Method for Stochastic Optimization`_.
Arguments:
params (iterable): iterable of parameters to optimize or dicts defining
parameter groups
lr (float, optional): learning rate (default: 1e-3)
betas (Tuple[float, float], optional): coefficients used for computing
running averages of gradient and its square (default: (0.9, 0.999))
eps (float, optional): term added to the denominator to improve
numerical stability (default: 1e-8)
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
amsgrad (boolean, optional): whether to use the AMSGrad variant of this
algorithm from the paper `On the Convergence of Adam and Beyond`_
2021-10-21 23:04:18 +00:00
.. _Adam: A Method for Stochastic Optimization:
2021-10-21 17:20:41 +00:00
https://arxiv.org/abs/1412.6980
.. _On the Convergence of Adam and Beyond:
https://openreview.net/forum?id=ryQu7f-RZ
"""
def __init__(
self,
params,
lr=1e-3,
betas=(0.9, 0.999),
eps=1e-8,
weight_decay=0,
amsgrad=False,
bnb_analysis="dynamic-blockwise",
savedir=None,
2021-10-21 17:20:41 +00:00
):
defaults = dict(
lr=lr,
betas=betas,
eps=eps,
weight_decay=weight_decay,
amsgrad=amsgrad,
2021-10-21 17:20:41 +00:00
)
super().__init__(params, defaults)
2021-10-21 17:20:41 +00:00
self.analysis = bnb_analysis
self.savedir = savedir
@property
def supports_memory_efficient_fp16(self):
return True
@property
def supports_flat_params(self):
return True
def step(self, closure=None):
"""Performs a single optimization step.
Arguments:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
"""
loss = None
if closure is not None:
loss = closure()
for group in self.param_groups:
for p_id, p in enumerate(group["params"]):
if p.grad is None:
continue
grad = p.grad.data
if grad.dtype in {torch.float16, torch.bfloat16}:
grad = grad.float()
if grad.is_sparse:
raise RuntimeError(
"Adam does not support sparse gradients, please consider SparseAdam instead"
)
amsgrad = group.get("amsgrad", False)
assert not amsgrad
p_data_fp32 = p.data
if p.data.dtype in {torch.float16, torch.bfloat16}:
p_data_fp32 = p_data_fp32.float()
state = self.state[p]
# State initialization
if len(state) == 0:
state["step"] = 0
# Exponential moving average of gradient values
state["exp_avg"] = torch.zeros_like(p_data_fp32)
# Exponential moving average of squared gradient values
state["exp_avg_sq"] = torch.zeros_like(p_data_fp32)
state["abserrors"] = torch.zeros(
(256, 256), device=p_data_fp32.device
)
state["relerrors"] = torch.zeros(
(256, 256), device=p_data_fp32.device
)
state["counts"] = torch.zeros(
(256, 256), device=p_data_fp32.device
)
2021-10-21 17:20:41 +00:00
if amsgrad:
# Maintains max of all exp. moving avg. of sq. grad. values
state["max_exp_avg_sq"] = torch.zeros_like(p_data_fp32)
else:
state["exp_avg"] = state["exp_avg"].to(p_data_fp32)
state["exp_avg_sq"] = state["exp_avg_sq"].to(p_data_fp32)
if amsgrad:
state["max_exp_avg_sq"] = state["max_exp_avg_sq"].to(
p_data_fp32
)
state["step"] += 1
beta1, beta2 = group["betas"]
bias_correction1 = 1 - beta1 ** state["step"]
bias_correction2 = 1 - beta2 ** state["step"]
step_size = (
group["lr"] * math.sqrt(bias_correction2) / bias_correction1
)
e = state["abserrors"]
rele = state["relerrors"]
counts = state["counts"]
2021-10-21 17:20:41 +00:00
if group["weight_decay"] != 0:
p_data_fp32.add_(
p_data_fp32, alpha=-group["weight_decay"] * group["lr"]
)
exp_avg, exp_avg_sq = state["exp_avg"], state["exp_avg_sq"]
if amsgrad:
max_exp_avg_sq = state["max_exp_avg_sq"]
# Decay the first and second moment running average coefficient
exp_avg.mul_(beta1).add_(grad, alpha=1 - beta1)
exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=1 - beta2)
denom = exp_avg_sq.sqrt().add_(group["eps"])
update_fp32 = exp_avg / denom
2021-10-21 17:20:41 +00:00
if (
p_data_fp32.numel() <= 8192
or p_data_fp32.numel() > 50000 * 1000
):
2021-10-21 17:20:41 +00:00
# embedding layer or too small
p_data_fp32 += -step_size * update_fp32
2021-10-21 17:20:41 +00:00
else:
if self.analysis == "dynamic-blockwise":
2021-10-21 17:20:41 +00:00
code1 = F.create_dynamic_map(signed=True).to(p.device)
code2 = F.create_dynamic_map(signed=False).to(p.device)
C1, S1 = F.quantize_blockwise(exp_avg, code=code1)
state1 = F.dequantize_blockwise(C1, S1)
C2, S2 = F.quantize_blockwise(exp_avg_sq, code=code2)
state2 = F.dequantize_blockwise(C2, S2)
elif self.analysis == "dynamic":
2021-10-21 17:20:41 +00:00
code1 = F.create_dynamic_map(signed=True).to(p.device)
code2 = F.create_dynamic_map(signed=False).to(p.device)
C1, S1 = F.quantize(exp_avg, code=code1)
state1 = F.dequantize(C1, S1)
C2, S2 = F.quantize(exp_avg_sq, code=code2)
state2 = F.dequantize(C2, S2)
elif self.analysis == "linear":
2021-10-21 17:20:41 +00:00
code1 = F.create_linear_map(signed=True).to(p.device)
code2 = F.create_linear_map(signed=False).to(p.device)
C1, S1 = F.quantize(exp_avg, code=code1)
state1 = F.dequantize(C1, S1)
C2, S2 = F.quantize(exp_avg_sq, code=code2)
state2 = F.dequantize(C2, S2)
elif self.analysis == "quantile":
2021-10-21 17:20:41 +00:00
code1 = F.estimate_quantiles(exp_avg)
code2 = F.estimate_quantiles(exp_avg_sq)
C1 = F.quantize_no_absmax(exp_avg, code=code1)
state1 = F.dequantize_no_absmax(C1, code1)
C2 = F.quantize_no_absmax(exp_avg_sq, code=code2)
state2 = F.dequantize_no_absmax(C2, code2)
elif self.analysis == "my-quantization-routine":
2021-10-21 23:04:18 +00:00
pass
2021-10-21 22:53:20 +00:00
# 1. get code
# 2. quantize
# 3. dequantize
# Error will be calculated automatically!
2021-10-21 17:20:41 +00:00
else:
raise ValueError(
f"Invalid analysis value: {self.analysis}!"
)
2021-10-21 17:20:41 +00:00
denom = state2.sqrt().add_(group["eps"])
update_8bit = state1 / denom
2021-10-21 17:20:41 +00:00
abserr = torch.abs(update_8bit - update_fp32)
relerr = abserr / torch.abs(update_fp32 + 1e-6)
2021-10-21 17:20:41 +00:00
C1, C2 = C1.int(), C2.int()
F.histogram_scatter_add_2d(e, C1.int(), C2.int(), abserr)
F.histogram_scatter_add_2d(rele, C1.int(), C2.int(), relerr)
F.histogram_scatter_add_2d(
counts, C1.int(), C2.int(), torch.ones_like(abserr)
)
2021-10-21 17:20:41 +00:00
p_data_fp32 += -step_size * update_fp32
2021-10-21 17:20:41 +00:00
if not dist.is_initialized() or dist.get_rank() == 0:
if self.savedir != "" and state["step"] % 100 == 0:
if not os.path.exists(self.savedir):
os.makedirs(self.savedir)
shapestr = "_".join(
[str(dim) for dim in p_data_fp32.shape]
)
pathe = os.path.join(
self.savedir, f"{p_id}_{shapestr}_abserr.pkl"
)
pathrele = os.path.join(
self.savedir, f"{p_id}_{shapestr}_relerr.pkl"
)
pathcounts = os.path.join(
self.savedir, f"{p_id}_{shapestr}_counts.pkl"
)
2021-10-21 17:20:41 +00:00
torch.save(e, pathe)
torch.save(rele, pathrele)
torch.save(counts, pathcounts)
if p.data.dtype in {torch.float16, torch.bfloat16}:
p.data.copy_(p_data_fp32)
return loss