bitsandbytes/main/en/reference/optim/lion.md

Lion

Lion (Evolved Sign Momentum) is a unique optimizer that uses the sign of the gradient to determine the update direction of the momentum. This makes Lion more memory-efficient and faster than AdamW which tracks and store the first and second-order moments.

Lion[[api-class]][[bitsandbytes.optim.Lion]]

bitsandbytes.optim.Lion[[bitsandbytes.optim.Lion]]

Source

__init__bitsandbytes.optim.Lion.__init__https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/bitsandbytes/optim/lion.py#L9[{"name": "params", "val": ""}, {"name": "lr", "val": " = 0.0001"}, {"name": "betas", "val": " = (0.9, 0.99)"}, {"name": "weight_decay", "val": " = 0"}, {"name": "optim_bits", "val": " = 32"}, {"name": "args", "val": " = None"}, {"name": "min_8bit_size", "val": " = 4096"}, {"name": "is_paged", "val": " = False"}]- params (torch.tensor) -- The input parameters to optimize.

• lr (float, defaults to 1e-4) -- The learning rate.
• betas (tuple(float, float), defaults to (0.9, 0.999)) -- The beta values are the decay rates of the first and second-order moment of the optimizer.
• weight_decay (float, defaults to 0) -- The weight decay value for the optimizer.
• optim_bits (int, defaults to 32) -- The number of bits of the optimizer state.
• args (object, defaults to None) -- An object with additional arguments.
• min_8bit_size (int, defaults to 4096) -- The minimum number of elements of the parameter tensors for 8-bit optimization.
• is_paged (bool, defaults to False) -- Whether the optimizer is a paged optimizer or not.0

Base Lion optimizer.

Parameters:

params (torch.tensor) : The input parameters to optimize.

lr (float, defaults to 1e-4) : The learning rate.

betas (tuple(float, float), defaults to (0.9, 0.999)) : The beta values are the decay rates of the first and second-order moment of the optimizer.

weight_decay (float, defaults to 0) : The weight decay value for the optimizer.

optim_bits (int, defaults to 32) : The number of bits of the optimizer state.

args (object, defaults to None) : An object with additional arguments.

min_8bit_size (int, defaults to 4096) : The minimum number of elements of the parameter tensors for 8-bit optimization.

is_paged (bool, defaults to False) : Whether the optimizer is a paged optimizer or not.

Lion8bit[[bitsandbytes.optim.Lion8bit]]

bitsandbytes.optim.Lion8bit[[bitsandbytes.optim.Lion8bit]]

Source

__init__bitsandbytes.optim.Lion8bit.__init__https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/bitsandbytes/optim/lion.py#L56[{"name": "params", "val": ""}, {"name": "lr", "val": " = 0.0001"}, {"name": "betas", "val": " = (0.9, 0.99)"}, {"name": "weight_decay", "val": " = 0"}, {"name": "args", "val": " = None"}, {"name": "min_8bit_size", "val": " = 4096"}, {"name": "is_paged", "val": " = False"}]- params (torch.tensor) -- The input parameters to optimize.

• lr (float, defaults to 1e-4) -- The learning rate.
• betas (tuple(float, float), defaults to (0.9, 0.999)) -- The beta values are the decay rates of the first and second-order moment of the optimizer.
• weight_decay (float, defaults to 0) -- The weight decay value for the optimizer.
• args (object, defaults to None) -- An object with additional arguments.
• min_8bit_size (int, defaults to 4096) -- The minimum number of elements of the parameter tensors for 8-bit optimization.
• is_paged (bool, defaults to False) -- Whether the optimizer is a paged optimizer or not.0

8-bit Lion optimizer.

Parameters:

params (torch.tensor) : The input parameters to optimize.

lr (float, defaults to 1e-4) : The learning rate.

betas (tuple(float, float), defaults to (0.9, 0.999)) : The beta values are the decay rates of the first and second-order moment of the optimizer.

weight_decay (float, defaults to 0) : The weight decay value for the optimizer.

args (object, defaults to None) : An object with additional arguments.

min_8bit_size (int, defaults to 4096) : The minimum number of elements of the parameter tensors for 8-bit optimization.

is_paged (bool, defaults to False) : Whether the optimizer is a paged optimizer or not.

Lion32bit[[bitsandbytes.optim.Lion32bit]]

bitsandbytes.optim.Lion32bit[[bitsandbytes.optim.Lion32bit]]

Source

__init__bitsandbytes.optim.Lion32bit.__init__https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/bitsandbytes/optim/lion.py#L100[{"name": "params", "val": ""}, {"name": "lr", "val": " = 0.0001"}, {"name": "betas", "val": " = (0.9, 0.99)"}, {"name": "weight_decay", "val": " = 0"}, {"name": "args", "val": " = None"}, {"name": "min_8bit_size", "val": " = 4096"}, {"name": "is_paged", "val": " = False"}]- params (torch.tensor) -- The input parameters to optimize.

• lr (float, defaults to 1e-4) -- The learning rate.
• betas (tuple(float, float), defaults to (0.9, 0.999)) -- The beta values are the decay rates of the first and second-order moment of the optimizer.
• weight_decay (float, defaults to 0) -- The weight decay value for the optimizer.
• args (object, defaults to None) -- An object with additional arguments.
• min_8bit_size (int, defaults to 4096) -- The minimum number of elements of the parameter tensors for 8-bit optimization.
• is_paged (bool, defaults to False) -- Whether the optimizer is a paged optimizer or not.0

32-bit Lion optimizer.

Parameters:

params (torch.tensor) : The input parameters to optimize.

lr (float, defaults to 1e-4) : The learning rate.

betas (tuple(float, float), defaults to (0.9, 0.999)) : The beta values are the decay rates of the first and second-order moment of the optimizer.

weight_decay (float, defaults to 0) : The weight decay value for the optimizer.

args (object, defaults to None) : An object with additional arguments.

min_8bit_size (int, defaults to 4096) : The minimum number of elements of the parameter tensors for 8-bit optimization.

is_paged (bool, defaults to False) : Whether the optimizer is a paged optimizer or not.

PagedLion[[bitsandbytes.optim.PagedLion]]

bitsandbytes.optim.PagedLion[[bitsandbytes.optim.PagedLion]]

Source

__init__bitsandbytes.optim.PagedLion.__init__https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/bitsandbytes/optim/lion.py#L144[{"name": "params", "val": ""}, {"name": "lr", "val": " = 0.0001"}, {"name": "betas", "val": " = (0.9, 0.99)"}, {"name": "weight_decay", "val": " = 0"}, {"name": "optim_bits", "val": " = 32"}, {"name": "args", "val": " = None"}, {"name": "min_8bit_size", "val": " = 4096"}]- params (torch.tensor) -- The input parameters to optimize.

• lr (float, defaults to 1e-4) -- The learning rate.
• betas (tuple(float, float), defaults to (0.9, 0.999)) -- The beta values are the decay rates of the first and second-order moment of the optimizer.
• weight_decay (float, defaults to 0) -- The weight decay value for the optimizer.
• optim_bits (int, defaults to 32) -- The number of bits of the optimizer state.
• args (object, defaults to None) -- An object with additional arguments.
• min_8bit_size (int, defaults to 4096) -- The minimum number of elements of the parameter tensors for 8-bit optimization.0

Paged Lion optimizer.

Parameters:

params (torch.tensor) : The input parameters to optimize.

lr (float, defaults to 1e-4) : The learning rate.

betas (tuple(float, float), defaults to (0.9, 0.999)) : The beta values are the decay rates of the first and second-order moment of the optimizer.

weight_decay (float, defaults to 0) : The weight decay value for the optimizer.

optim_bits (int, defaults to 32) : The number of bits of the optimizer state.

args (object, defaults to None) : An object with additional arguments.

min_8bit_size (int, defaults to 4096) : The minimum number of elements of the parameter tensors for 8-bit optimization.

PagedLion8bit[[bitsandbytes.optim.PagedLion8bit]]

bitsandbytes.optim.PagedLion8bit[[bitsandbytes.optim.PagedLion8bit]]

Source

__init__bitsandbytes.optim.PagedLion8bit.__init__https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/bitsandbytes/optim/lion.py#L188[{"name": "params", "val": ""}, {"name": "lr", "val": " = 0.0001"}, {"name": "betas", "val": " = (0.9, 0.99)"}, {"name": "weight_decay", "val": " = 0"}, {"name": "args", "val": " = None"}, {"name": "min_8bit_size", "val": " = 4096"}]- params (torch.tensor) -- The input parameters to optimize.

• lr (float, defaults to 1e-4) -- The learning rate.
• betas (tuple(float, float), defaults to (0.9, 0.999)) -- The beta values are the decay rates of the first and second-order moment of the optimizer.
• weight_decay (float, defaults to 0) -- The weight decay value for the optimizer.
• args (object, defaults to None) -- An object with additional arguments.
• min_8bit_size (int, defaults to 4096) -- The minimum number of elements of the parameter tensors for 8-bit optimization.0

Paged 8-bit Lion optimizer.

Parameters:

params (torch.tensor) : The input parameters to optimize.

lr (float, defaults to 1e-4) : The learning rate.

betas (tuple(float, float), defaults to (0.9, 0.999)) : The beta values are the decay rates of the first and second-order moment of the optimizer.

weight_decay (float, defaults to 0) : The weight decay value for the optimizer.

args (object, defaults to None) : An object with additional arguments.

min_8bit_size (int, defaults to 4096) : The minimum number of elements of the parameter tensors for 8-bit optimization.

PagedLion32bit[[bitsandbytes.optim.PagedLion32bit]]

bitsandbytes.optim.PagedLion32bit[[bitsandbytes.optim.PagedLion32bit]]

Source

__init__bitsandbytes.optim.PagedLion32bit.__init__https://github.com/bitsandbytes-foundation/bitsandbytes/blob/main/bitsandbytes/optim/lion.py#L229[{"name": "params", "val": ""}, {"name": "lr", "val": " = 0.0001"}, {"name": "betas", "val": " = (0.9, 0.99)"}, {"name": "weight_decay", "val": " = 0"}, {"name": "args", "val": " = None"}, {"name": "min_8bit_size", "val": " = 4096"}]- params (torch.tensor) -- The input parameters to optimize.

• lr (float, defaults to 1e-4) -- The learning rate.
• betas (tuple(float, float), defaults to (0.9, 0.999)) -- The beta values are the decay rates of the first and second-order moment of the optimizer.
• weight_decay (float, defaults to 0) -- The weight decay value for the optimizer.
• args (object, defaults to None) -- An object with additional arguments.
• min_8bit_size (int, defaults to 4096) -- The minimum number of elements of the parameter tensors for 8-bit optimization.0

Paged 32-bit Lion optimizer.

Parameters:

params (torch.tensor) : The input parameters to optimize.

lr (float, defaults to 1e-4) : The learning rate.

betas (tuple(float, float), defaults to (0.9, 0.999)) : The beta values are the decay rates of the first and second-order moment of the optimizer.

weight_decay (float, defaults to 0) : The weight decay value for the optimizer.

args (object, defaults to None) : An object with additional arguments.

min_8bit_size (int, defaults to 4096) : The minimum number of elements of the parameter tensors for 8-bit optimization.