bijRddlmZGddeZGddeZGddeZy) )Optimizer1Statec6eZdZ dfd ZxZS)RMSpropc v|dk(r td|r tdt | d||||f|||| | | | y)a Base RMSprop optimizer. Arguments: params (`torch.tensor`): The input parameters to optimize. lr (`float`, defaults to 1e-2): The learning rate. alpha (`float`, defaults to 0.99): The alpha value is the decay rate of the squared gradients of the optimizer. eps (`float`, defaults to 1e-8): The epsilon value prevents division by zero in the optimizer. weight_decay (`float`, defaults to 0.0): The weight decay value for the optimizer. momentum (`float`, defaults to 0): The momentum value speeds up the optimizer by taking bigger steps. centered (`bool`, defaults to `False`): Whether the gradients are normalized by the variance. If `True`, it can help training at the expense of additional compute. 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. percentile_clipping (`int`, defaults to 100): Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability. block_wise (`bool`, defaults to `True`): Whether to independently quantize each block of tensors to reduce outlier effects and improve stability. r)RMSprop with alpha==0.0 is not supported!"Centered RMSprop is not supported!rmspropNNotImplementedErrorsuper__init__)selfparamslralphaeps weight_decaymomentumcentered optim_bitsargs min_8bit_sizepercentile_clipping block_wise __class__s U/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/bitsandbytes/optim/rmsprop.pyr zRMSprop.__init__ s[X A:%&QR R %&JK K     H         ) {Gz?Gz?:0yE>rrF NdT__name__ __module__ __qualname__r __classcell__rs@rrrs0   < < rrc4eZdZ dfd ZxZS) RMSprop8bitc v|dk(r td|r tdt | d||||f||d|| | | y)a 8-bit RMSprop optimizer. Arguments: params (`torch.tensor`): The input parameters to optimize. lr (`float`, defaults to 1e-2): The learning rate. alpha (`float`, defaults to 0.99): The alpha value is the decay rate of the squared gradients of the optimizer. eps (`float`, defaults to 1e-8): The epsilon value prevents division by zero in the optimizer. weight_decay (`float`, defaults to 0.0): The weight decay value for the optimizer. momentum (`float`, defaults to 0): The momentum value speeds up the optimizer by taking bigger steps. centered (`bool`, defaults to `False`): Whether the gradients are normalized by the variance. If `True`, it can help training at the expense of additional compute. 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. percentile_clipping (`int`, defaults to 100): Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability. block_wise (`bool`, defaults to `True`): Whether to independently quantize each block of tensors to reduce outlier effects and improve stability. rrrr Nr rrrrrrrrrrrrrs rr zRMSprop8bit.__init__Is[V A:%&QR R %&JK K     H        r rrr rrFNr"r#Tr$r)s@rr+r+Hs-   ; ; rr+c4eZdZ dfd ZxZS) RMSprop32bitc v|dk(r td|r tdt | d||||f||d|| | | y)a 32-bit RMSprop optimizer. Arguments: params (`torch.tensor`): The input parameters to optimize. lr (`float`, defaults to 1e-2): The learning rate. alpha (`float`, defaults to 0.99): The alpha value is the decay rate of the squared gradients of the optimizer. eps (`float`, defaults to 1e-8): The epsilon value prevents division by zero in the optimizer. weight_decay (`float`, defaults to 0.0): The weight decay value for the optimizer. momentum (`float`, defaults to 0): The momentum value speeds up the optimizer by taking bigger steps. centered (`bool`, defaults to `False`): Whether the gradients are normalized by the variance. If `True`, it can help training at the expense of additional compute. optim_bits (`int`, defaults to 32): The number of bits of the optimizer state. args (`object`, defaults to `None`): An object with additional arguments. min_8bit_size (`int`, defaults to 4096): The minimum number of elements of the parameter tensors for 8-bit optimization. percentile_clipping (`int`, defaults to 100): Adapts clipping threshold automatically by tracking the last 100 gradient norms and clipping the gradient at a certain percentile to improve stability. block_wise (`bool`, defaults to `True`): Whether to independently quantize each block of tensors to reduce outlier effects and improve stability. rrrr r!Nr r.s rr zRMSprop32bit.__init__s[X A:%&QR R %&JK K     H         rr/r$r)s@rr1r1s-   < < rr1N)bitsandbytes.optim.optimizerrrr+r1rrr5s0 9= o= @< /< ~= ?= r