adv-optm 0.1.9__py3-none-any.whl → 1.0.1__py3-none-any.whl

adv_optm/optim/Prodigy_adv.py

@@ -1,478 +1,475 @@
-import torch
-import torch.distributed as dist
-
-import math
-
-from ..util.BF16_Stochastic_Rounding import add_stochastic_
-from ..util.Effective_Shape import _get_effective_shape
-from ..util.NNMF import _nnmf,_unnmf
-from ..util.OrthoGrad import _orthogonalize_gradient
-from ..util.One_Bit_Boolean import _pack_bools, _unpack_bools
-
-class Prodigy_adv(torch.optim.Optimizer):
-    """
-    Implements a factored Prodigy/AdamW algorithm.
-    This is an advanced version of Prodigy with optional features like
-    low-rank factorization of optimizer states (SMMF), OrthoGrad, etc.
-
-    Args:
-        params (iterable): iterable of parameters to optimize or dicts defining
-            parameter groups
-        lr (float): learning rate (default: 1)
-        betas (tuple[float, float]): coefficients used for computing running
-            averages of gradient and its square (default: (0.9, 0.999))
-        eps (float): term added to the denominator to improve
-            numerical stability (default: 1e-8)
-        weight_decay (float): weight decay (L2 penalty) (default: 0)
-        vector_reshape (bool): whether to reshape 1D vectors into 2D
-            matrices to apply low-rank compression (default: True).
-        stochastic_rounding (bool): whether to use stochastic
-            rounding for BF16 parameter updates (default: True).
-        use_atan2 (bool): whether to use the atan2 update rule. (default: False)
-        use_grams (bool): whether to use Grams-style updates. (default: False)
-        use_cautious (bool):  whether to use cautious masking to align the gradient's
-            direction with the first moment's.  (default: False)
-        use_orthograd (bool): whether to use OrthoGrad.  (default: False)
-        use_AdEMAMix (bool): whether to enable the AdEMAMix feature. This adds
-            a second, slow-moving average of the momentum (`mt_slow`) which is
-            combined with the primary momentum (`mt`) to stabilize updates,
-            especially in noisy, small-batch settings. If `False`, the
-            optimizer behaves as standard AdamW. (default: False)
-        beta3_ema (float): The decay rate for the slow exponential moving average of
-            the momentum (only used when `use_AdEMAMix` is `True`). A higher
-            value (e.g., 0.9999) gives the EMA a longer memory, making it more
-            stable but slower to adapt. A lower value (e.g., 0.999) is often
-            better for shorter training runs. (default: 0.9999)
-        alpha (float): The mixing coefficient that scales the slow momentum term
-            before it is added to the fast momentum term (`update = mt + alpha * mt_slow`).
-            A higher value increases the stabilizing influence of the slow
-            momentum. (default: 5.0)
-        t_alpha (Optional[int]): The number of steps for a linear warmup of the
-            `alpha` parameter (only used when `use_AdEMAMix` is `True`). This is
-            highly recommended to prevent instability at the beginning of training,
-            as it gradually introduces the stabilizing slow momentum term. During
-            the warmup, `alpha` ramps from 0 to its target value. If `None`,
-            the scheduler is disabled.
-        Simplified_AdEMAMix (bool): whether to use the Simplified AdEMAMix update rule.
-            This changes the EMA to accumulator and the update numerator to `alpha_grad * grad + mt`, which can be
-            more responsive, especially for small batch sizes. Enabling this will
-            automatically disable `use_AdEMAMix`, `use_cautious`, `use_grams`,
-            and `use_atan2`. (default: False)
-        alpha_grad (float): Mixing coefficient for the Simplified AdEMAMix update rule
-            (only used when `Simplified_AdEMAMix` is `True`). Controls the weight of the
-            current gradient. For small batch sizes, use high values (e.g., 10-100) to be
-            more responsive. For large batch sizes, use low values (e.g., 0-1) for
-            stability. (default: 100.0)
-        factored (bool): whether to use the factorization or disable it to use
-            the uncompressed optimizer. (default: False)
-        d0 (float):
-            Initial D estimate for D-adaptation (default 1e-6). Rarely needs changing.
-        d_coef (float):
-            Coefficient in the expression for the estimate of d (default 1.0).
-            Values such as 0.5 and 2.0 typically work as well. 
-            Changing this parameter is the preferred way to tune the method.
-        growth_rate (float):
-            prevent the D estimate from growing faster than this multiplicative rate.
-            Default is inf, for unrestricted. Values like 1.02 give a kind of learning
-            rate warmup effect.
-        fsdp_in_use (bool):
-            If you're using sharded parameters, this should be set to True. The optimizer
-            will attempt to auto-detect this, but if you're using an implementation other
-            than PyTorch's builtin version, the auto-detection won't work.
-        slice_p (int): Reduce memory usage by calculating LR adaptation statistics on only every 
-            pth entry of each tensor. For values greater than 1 this an an approximation to standard 
-            Prodigy. Values ~11 are reasonable (default 11).
-        prodigy_steps (int): If greater than zero, disable Prodigy's stepsize adjustments
-            after the specified optimiser step and release all state memory required by Prodigy
-            (default: 0).
-    """
-
-    def __init__(
-        self,
-        params,
-        lr: float = 1,
-        betas: tuple[float, float] = (0.9, 0.999),
-        eps: float = 1e-8,
-        weight_decay: float = 0.0,
-        vector_reshape: bool = True,
-        stochastic_rounding: bool = True,
-        use_atan2: bool = False,
-        use_cautious: bool = False,
-        use_grams: bool = False,
-        use_orthograd: bool = False,
-        use_AdEMAMix: bool = False,
-        beta3_ema: float = 0.9999,
-        alpha: float = 5.0,
-        t_alpha: int | None = None,
-        Simplified_AdEMAMix: bool = False,
-        alpha_grad: float = 100.0,
-        factored: bool = False,
-        # prodigy parameters
-        beta3: float = None,
-        d0: float = 1e-6,
-        d_coef: float = 1,
-        growth_rate: float = float('inf'),
-        safeguard_warmup: bool = False,
-        fsdp_in_use: bool = False,
-        slice_p: int = 11,
-        prodigy_steps: int = 0,
-    ):
-        if not (lr >= 0.0):
-            raise ValueError(f"Learning-rate should be >= 0.0. Got {lr}")
-        if not (0.0 <= betas[0] < 1.0 and 0.0 <= betas[1] < 1.0):
-            raise ValueError(f"Betas should be in [0.0, 1.0). Got {betas}")
-        if not (eps >= 0.0):
-            raise ValueError(f"Epsilon should be >= 0.0. Got {eps}")
-        if not (weight_decay >= 0.0):
-            raise ValueError(f"Weight-decay should be >= 0.0. Got {weight_decay}")
-        if not (prodigy_steps >= 0):
-            raise ValueError(f"prodigy_steps should be >= 0. Got {prodigy_steps}")
-        if use_cautious and use_grams:
-            print("Warning: use_cautious is incompatible with use_grams, Disabling use_cautious.")
-            use_cautious = False
-        if betas[0] == 0.0 and Simplified_AdEMAMix:
-            raise ValueError(f"Beta1 cannot be 0.0 when using Simplified_AdEMAMix. Got {betas[0]}")
-        if use_AdEMAMix and Simplified_AdEMAMix:
-            print("Warning: use_AdEMAMix is incompatible with Simplified_AdEMAMix, Disabling use_AdEMAMix.")
-        if use_grams and Simplified_AdEMAMix:
-            print("Warning: use_grams is incompatible with Simplified_AdEMAMix, Disabling use_grams.")
-        if use_cautious and Simplified_AdEMAMix:
-            print("Warning: use_cautious is incompatible with Simplified_AdEMAMix, Disabling use_cautious.")
-        if use_atan2 and Simplified_AdEMAMix:
-            print("Warning: use_atan2 is incompatible with Simplified_AdEMAMix. Disabling use_atan2.")
-            use_atan2 = False
-
-        defaults = {
-            "lr": lr, "betas": betas, "eps": eps, "weight_decay": weight_decay,
-            "vector_reshape": vector_reshape, "use_atan2": use_atan2,
-            "use_orthograd": use_orthograd,
-            "beta3_ema": beta3_ema, "alpha": alpha, "t_alpha": t_alpha,
-            "beta3": beta3, "d": d0, "d0": d0, "d_max": d0, "d_numerator": 0.0, "d_coef": d_coef,
-            "growth_rate": growth_rate, "safeguard_warmup": safeguard_warmup, "k": 0, "slice_p": slice_p,
-            "fsdp_in_use": fsdp_in_use, "prodigy_steps": prodigy_steps,
-            "alpha_grad": alpha_grad, 
-        }
-        self.stochastic_rounding = stochastic_rounding
-        self.use_cautious = use_cautious and not Simplified_AdEMAMix
-        self.use_grams = use_grams and not Simplified_AdEMAMix
-        self.use_AdEMAMix = use_AdEMAMix and not Simplified_AdEMAMix
-        self.Simplified_AdEMAMix = Simplified_AdEMAMix
-        self.factored = factored
-        self.fsdp_in_use = fsdp_in_use
-        super().__init__(params, defaults)
-        self.init_step()
-
-    @property
-    def supports_fused_back_pass(self):
-        return True
-
-    @property
-    def supports_memory_efficient_fp16(self):
-        return True
-
-    @property
-    def supports_flat_params(self):
-        return False
-
-    def init_step(self):
-        """Resets accumulators and calculates dlr for the upcoming step."""
-        self.d_denom = 0.0
-        
-        g_group = self.param_groups[0]
-        self.beta1, self.beta2 = g_group['betas']
-        self.beta3 = g_group['beta3']
-        if self.beta3 is None:
-            self.beta3 = math.sqrt(self.beta2)
-        
-        k = g_group['k']
-        self.d = g_group['d']
-        lr = g_group['lr']
-
-        self.dlr = self.d * lr
-
-        self.d_numerator = g_group.get('d_numerator', 0.0) * self.beta3
-
-    @torch.no_grad()
-    def step_parameter(self, p: torch.Tensor, group: dict, i: int | None = None):
-        if p.grad is None:
-            return
-
-        if hasattr(p, "_fsdp_flattened"):
-            self.fsdp_in_use = True
-
-        grad = p.grad
-        if grad.dtype != torch.float32 and self.factored:
-            grad = grad.float()
-        if group["use_orthograd"]:
-            grad = _orthogonalize_gradient(p, grad)
-        state = self.state[p]
-
-        # State Initialization
-        if len(state) == 0:
-            state['step'] = 0
-
-            should_factor = (
-                self.factored and
-                not (len(p.shape) == 1 and not group['vector_reshape'])
-            )
-
-            state['factored'] = should_factor
-
-            slice_p = group['slice_p']
-
-            dtype = torch.float32 if self.factored else p.dtype
-            device = p.device
-
-            if state['factored']:
-                state['effective_shape'] = _get_effective_shape(p.numel())
-                d1, d2 = state['effective_shape']
-
-                # First moment (m)
-                if self.beta1 > 0:
-                    state['mu_m_nmf'] = torch.zeros(d1, device=device, dtype=dtype) 
-                    state['mv_m_nmf'] = torch.zeros(d2, device=device, dtype=dtype)
-                    if not self.use_grams:
-                        packed_d2 = (d2 + 7) // 8
-                        state['sign'] = torch.zeros((d1, packed_d2), dtype=torch.uint8, device=device)
-                if self.use_AdEMAMix:
-                    state['mu_m_slow_nmf'] = torch.zeros(d1, device=p.device, dtype=dtype) 
-                    state['mv_m_slow_nmf'] = torch.zeros(d2, device=p.device, dtype=dtype)
-                    packed_d2 = (d2 + 7) // 8
-                    state['sign_slow'] = torch.zeros((d1, packed_d2), dtype=torch.uint8, device=p.device)
-                # Second moment (v)
-                state['mu_v_nmf'] = torch.zeros(d1, device=device, dtype=dtype) 
-                state['mv_v_nmf'] = torch.zeros(d2, device=device, dtype=dtype)
-            else:  # Fallback to standard AdamW for non-factored tensors
-                if self.beta1 > 0:
-                    state['exp_avg'] = torch.zeros_like(p, device=device, dtype=dtype)
-                if self.use_AdEMAMix:
-                    state['exp_avg_slow'] = torch.zeros_like(p, dtype=dtype)
-                state['exp_avg_sq'] = torch.zeros_like(p, device=device, dtype=dtype)
-
-            state['s'] = torch.zeros_like(p.flatten()[::slice_p]).detach()
-            if p.any():
-                state['p0'] = p.flatten()[::slice_p].detach().clone()
-            else:
-                state['p0'] = torch.tensor(0, device=device, dtype=p.dtype)
-
-        if self.use_AdEMAMix:
-            beta3_ema = group['beta3_ema']
-            alpha = group['alpha']
-            t_alpha = group['t_alpha']
-            current_step = state['step'] + 1
-            alpha_t = alpha
-            if t_alpha is not None and t_alpha > 0 and current_step < t_alpha:
-                alpha_t = min(current_step * alpha / t_alpha, alpha)
-        if self.Simplified_AdEMAMix:
-            alpha_grad = group["alpha_grad"]
-
-        if state['factored']:
-            d1, d2 = state['effective_shape']
-
-            grad_reshaped = grad.view(d1, d2)
-
-            # Reconstruct momentum from previous step's factors
-            if self.beta1 > 0:
-                mt = _unnmf((state['mu_m_nmf'], state['mv_m_nmf']))
-                if not self.use_grams:
-                    unpacked_sign = _unpack_bools(state['sign'], original_m=d2)
-                    torch.where(unpacked_sign, mt, -mt, out=mt)
-                    del unpacked_sign
-                # Update momentum in full-size
-                if self.Simplified_AdEMAMix:
-                    mt.mul_(self.beta1).add_(grad_reshaped, alpha=self.d)
-                else:
-                    mt.mul_(self.beta1).add_(grad_reshaped, alpha=self.d * (1.0 - self.beta1))
-                if self.use_grams:
-                    mt.copy_(grad_reshaped.sign() * mt.abs())
-                elif self.use_cautious:
-                    mask = (mt * grad_reshaped > 0).to(grad_reshaped.dtype)
-                    mask.div_(mask.mean().clamp_(min=1e-3))
-                    mt.mul_(mask)
-                    del mask
-
-            vt = _unnmf((state['mu_v_nmf'], state['mv_v_nmf']))
-            vt.mul_(self.beta2).addcmul_(grad_reshaped, grad_reshaped, value=self.d * self.d * (1.0 - self.beta2))
-
-            if self.use_AdEMAMix:
-                mt_slow = _unnmf((state['mu_m_slow_nmf'], state['mv_m_slow_nmf']))
-                if state['sign_slow'].dtype != torch.uint8:
-                    state['sign_slow'] = state['sign_slow'].to(torch.uint8)
-                unpacked_sign_slow = _unpack_bools(state['sign_slow'], original_m=d2)
-                torch.where(unpacked_sign_slow, mt_slow, -mt_slow, out=mt_slow)
-                del unpacked_sign_slow
-                mt_slow.mul_(beta3_ema).add_(grad_reshaped, alpha=self.d * (1.0 - beta3_ema))
-                if self.beta1 > 0:
-                    update = torch.add(mt, mt_slow, alpha=alpha_t)
-                else:
-                    update = torch.add(grad_reshaped, mt_slow, alpha=alpha_t)
-            elif self.Simplified_AdEMAMix:
-                update = torch.add(mt, grad_reshaped, alpha=alpha_grad * self.d)
-            else:
-                update = mt.clone() if self.beta1 > 0 else grad_reshaped.clone()
-            del grad_reshaped
-
-            if group['use_atan2']:
-                a = 1.2732395
-                denom = vt.sqrt()
-                update.atan2_(denom).mul_(a)
-            else:
-                denom = vt.sqrt()
-                update.div_(denom.add_(self.d * group['eps']))
-            del denom
-
-            update = update.view(p.shape).mul_(self.dlr)
-
-            # Compress updated moments and store new factors
-            if self.beta1 > 0:
-                if not self.use_grams:
-                    state['sign'] = _pack_bools(mt > 0)
-                _nnmf(mt.abs(), out=(state['mu_m_nmf'], state['mv_m_nmf']))
-                del mt
-            if self.use_AdEMAMix:
-                state['sign_slow'] = _pack_bools(mt_slow > 0)
-                _nnmf(mt_slow.abs(), out=(state['mu_m_slow_nmf'], state['mv_m_slow_nmf']))
-                del mt_slow
-            _nnmf(vt, out=(state['mu_v_nmf'], state['mv_v_nmf']))
-            del vt
-
-        else:  # Standard AdamW logic for non-factored tensors
-            exp_avg_sq = state['exp_avg_sq']
-
-            if self.beta1 > 0:
-                exp_avg = state['exp_avg']
-                if self.Simplified_AdEMAMix:
-                    exp_avg.mul_(self.beta1).add_(grad, alpha=self.d)
-                else:
-                    exp_avg.mul_(self.beta1).add_(grad, alpha=self.d * (1.0 - self.beta1))
-                if self.use_grams:
-                    exp_avg = grad.sign() * exp_avg.abs()
-                elif self.use_cautious:
-                    mask = (exp_avg * grad > 0).to(grad.dtype)
-                    mask.div_(mask.mean().clamp_(min=1e-3))
-                    exp_avg.mul_(mask)
-                    del mask
-
-            if self.use_AdEMAMix:
-                exp_avg_slow = state['exp_avg_slow']
-                exp_avg_slow.mul_(beta3_ema).add_(grad, alpha=self.d * (1.0 - beta3_ema))
-                if self.beta1 > 0:
-                    update = torch.add(exp_avg, exp_avg_slow, alpha=alpha_t)
-                else:
-                    update = torch.add(grad, exp_avg_slow, alpha=alpha_t)
-            elif self.Simplified_AdEMAMix:
-                update = torch.add(exp_avg, grad, alpha=alpha_grad * self.d)
-            else:
-                update = exp_avg.clone() if self.beta1 > 0 else grad.clone()
-
-            exp_avg_sq.mul_(self.beta2).addcmul_(grad, grad.conj(), value=self.d * self.d * (1.0 - self.beta2))
-
-            if group['use_atan2']:
-                a = 1.2732395
-                denom = exp_avg_sq.sqrt()
-                update.atan2_(denom).mul_(a)
-            else:
-                denom = exp_avg_sq.sqrt()
-                update.div_(denom.add_(self.d * group['eps']))
-            del denom
-
-            update.mul_(self.dlr)
-
-        # --- Accumulate Prodigy stats ---
-        prodigy_steps = group['prodigy_steps']
-        if prodigy_steps <= 0 or group['k'] < prodigy_steps:
-            d0, safeguard_warmup, slice_p = group['d0'], group['safeguard_warmup'], group['slice_p']
-            s, p0 = state['s'], state['p0']
-            grad_flat = grad.flatten().float()
-            p_flat = p.data.flatten().float()
-            p0 = p0.float()
-
-            self.d_numerator += (self.d / d0) * self.dlr * torch.dot(grad_flat[::slice_p], p0.data - p_flat[::slice_p]).item()
-
-            alpha = ((self.d / d0) * self.d) if safeguard_warmup else ((self.d / d0) * self.dlr)
-            s.mul_(self.beta3).add_(grad_flat[::slice_p], alpha=alpha)
-            self.d_denom += s.abs().sum().item()
-
-            del s, p0, grad_flat, p_flat, alpha
-        else:
-            # Free memory if prodigy_steps is reached
-            if 's' in state:
-                del state['s']
-            if 'p0' in state:
-                del state['p0']
-
-        # Decoupled weight decay
-        if group["weight_decay"] != 0:
-            if p.dtype == torch.bfloat16 and self.stochastic_rounding:
-                add_stochastic_(p.data, p.data, alpha=-group["weight_decay"] * self.dlr)
-            else:
-                p.data.add_(p.data, alpha=-group["weight_decay"] * self.dlr)
-
-        if p.dtype == torch.bfloat16 and self.stochastic_rounding:
-            add_stochastic_(p.data, -update)
-        else:
-            p.data.add_(-update)
-        del update
-
-        state['step'] += 1
-
-    @torch.no_grad()
-    def step(self, closure=None):
-        """Performs a single optimization step."""
-        loss = None
-        if closure is not None:
-            with torch.enable_grad():
-                loss = closure()
-
-        for group in self.param_groups:
-            for i, p in enumerate(group['params']):
-                self.step_parameter(p, group, i)
-
-
-        self.calculate_d()
-        self.init_step()
-        return loss
-
-    def calculate_d(self):
-        """Calculates the new `d` based on the accumulated stats."""
-        g_group = self.param_groups[0]
-        
-        # Only perform d-adaptation if prodigy_steps has not been reached
-        prodigy_active = not (g_group.get('prodigy_steps', 0) > 0 and g_group['k'] >= g_group['prodigy_steps'])
-
-        if prodigy_active:
-            d_max, d_coef, growth_rate = g_group['d_max'], g_group['d_coef'], g_group['growth_rate']
-            
-            if self.fsdp_in_use and dist.is_available() and dist.is_initialized():
-                # Use the device of the first parameter to avoid hardcoding '.cuda()'
-                device = self.param_groups[0]['params'][0].device
-                dist_tensor = torch.tensor([self.d_numerator, self.d_denom], device=device)
-                dist.all_reduce(dist_tensor, op=dist.ReduceOp.SUM)
-                global_d_numerator = dist_tensor[0].item()
-                global_d_denom = dist_tensor[1].item()
-            else:
-                global_d_numerator = self.d_numerator
-                global_d_denom = self.d_denom
-
-            d_hat = self.d
-            if global_d_denom > 0:
-                if self.Simplified_AdEMAMix and g_group['alpha_grad'] > 0:
-                    # A simple and effective hack to make prodigy compatible with Simplified_AdEMAMix large step sizes
-                    # by diving by alpha_grad we make sure that d_numerator that was influenced by (alpha_grad * grad)
-                    # are now normalized by /alpha_grad. this is a heuristic way since the update is also influenced by
-                    # the increasing and decaying accumulator but it's effective and it worked for me (for Lora/Finetune).
-                    global_d_numerator /= g_group['alpha_grad']
-                d_hat = d_coef * global_d_numerator / global_d_denom
-                if self.d == g_group['d0']:
-                    self.d = max(self.d, d_hat)
-                d_max = max(d_max, d_hat)
-                self.d = min(d_max, self.d * growth_rate)
-
-            for group in self.param_groups:
-                group['d_numerator'] = global_d_numerator
-                group['d'] = self.d
-                group['d_max'] = d_max
-        
-        # Increment step counter for all groups, regardless of whether d was updated
-        for group in self.param_groups:
-            group['k'] += 1
+import torch
+import torch.distributed as dist
+
+import math
+
+from ..util.BF16_Stochastic_Rounding import add_stochastic_
+from ..util.Effective_Shape import _get_effective_shape
+from ..util.NNMF import _nnmf,_unnmf
+from ..util.OrthoGrad import _orthogonalize_gradient
+from ..util.One_Bit_Boolean import _pack_bools, _unpack_bools
+
+class Prodigy_adv(torch.optim.Optimizer):
+    """
+    Implements a factored Prodigy/AdamW algorithm.
+    This is an advanced version of Prodigy with optional features like
+    low-rank factorization of optimizer states (SMMF), OrthoGrad, etc.
+
+    Args:
+        params (iterable): iterable of parameters to optimize or dicts defining
+            parameter groups
+        lr (float): learning rate (default: 1)
+        betas (tuple[float, float]): coefficients used for computing running
+            averages of gradient and its square (default: (0.9, 0.999))
+        eps (float): term added to the denominator to improve
+            numerical stability (default: 1e-8)
+        weight_decay (float): weight decay (L2 penalty) (default: 0)
+        vector_reshape (bool): whether to reshape 1D vectors into 2D
+            matrices to apply low-rank compression (default: True).
+        stochastic_rounding (bool): whether to use stochastic
+            rounding for BF16 parameter updates (default: True).
+        use_atan2 (bool): whether to use the atan2 update rule. (default: False)
+        grams_moment (bool): whether to use Grams-style updates. (default: False)
+        cautious_mask (bool):  whether to use cautious masking to align the gradient's
+            direction with the first moment's.  (default: False)
+        use_orthograd (bool): whether to use OrthoGrad.  (default: False)
+        use_AdEMAMix (bool): whether to enable the AdEMAMix feature. This adds
+            a second, slow-moving average of the momentum (`mt_slow`) which is
+            combined with the primary momentum (`mt`) to stabilize updates,
+            especially in noisy, small-batch settings. If `False`, the
+            optimizer behaves as standard AdamW. (default: False)
+        beta3_ema (float): The decay rate for the slow exponential moving average of
+            the momentum (only used when `use_AdEMAMix` is `True`). A higher
+            value (e.g., 0.9999) gives the EMA a longer memory, making it more
+            stable but slower to adapt. A lower value (e.g., 0.999) is often
+            better for shorter training runs. (default: 0.9999)
+        alpha (float): The mixing coefficient that scales the slow momentum term
+            before it is added to the fast momentum term (`update = mt + alpha * mt_slow`).
+            A higher value increases the stabilizing influence of the slow
+            momentum. (default: 5.0)
+        t_alpha (Optional[int]): The number of steps for a linear warmup of the
+            `alpha` parameter (only used when `use_AdEMAMix` is `True`). This is
+            highly recommended to prevent instability at the beginning of training,
+            as it gradually introduces the stabilizing slow momentum term. During
+            the warmup, `alpha` ramps from 0 to its target value. If `None`,
+            the scheduler is disabled.
+        Simplified_AdEMAMix (bool): whether to use the Simplified AdEMAMix update rule.
+            This changes the EMA to accumulator and the update numerator to `alpha_grad * grad + mt`, which can be
+            more responsive, especially for small batch sizes. Enabling this will
+            automatically disable `use_AdEMAMix`, `cautious_mask`, `grams_moment`,
+            and `use_atan2`. (default: False)
+        alpha_grad (float): Mixing coefficient for the Simplified AdEMAMix update rule
+            (only used when `Simplified_AdEMAMix` is `True`). Controls the weight of the
+            current gradient. For small batch sizes, use high values (e.g., 10-100) to be
+            more responsive. For large batch sizes, use low values (e.g., 0-1) for
+            stability. (default: 100.0)
+        nnmf_factor (bool): whether to use the factorization or disable it to use
+            the uncompressed optimizer. (default: False)
+        d0 (float):
+            Initial D estimate for D-adaptation (default 1e-6). Rarely needs changing.
+        d_coef (float):
+            Coefficient in the expression for the estimate of d (default 1.0).
+            Values such as 0.5 and 2.0 typically work as well. 
+            Changing this parameter is the preferred way to tune the method.
+        growth_rate (float):
+            prevent the D estimate from growing faster than this multiplicative rate.
+            Default is inf, for unrestricted. Values like 1.02 give a kind of learning
+            rate warmup effect.
+        fsdp_in_use (bool):
+            If you're using sharded parameters, this should be set to True. The optimizer
+            will attempt to auto-detect this, but if you're using an implementation other
+            than PyTorch's builtin version, the auto-detection won't work.
+        slice_p (int): Reduce memory usage by calculating LR adaptation statistics on only every 
+            pth entry of each tensor. For values greater than 1 this an an approximation to standard 
+            Prodigy. Values ~11 are reasonable (default 11).
+        prodigy_steps (int): If greater than zero, disable Prodigy's stepsize adjustments
+            after the specified optimiser step and release all state memory required by Prodigy
+            (default: 0).
+    """
+
+    def __init__(
+        self,
+        params,
+        lr: float = 1,
+        betas: tuple[float, float] = (0.9, 0.999),
+        eps: float = 1e-8,
+        weight_decay: float = 0.0,
+        vector_reshape: bool = True,
+        stochastic_rounding: bool = True,
+        use_atan2: bool = False,
+        cautious_mask: bool = False,
+        grams_moment: bool = False,
+        use_orthograd: bool = False,
+        use_AdEMAMix: bool = False,
+        beta3_ema: float = 0.9999,
+        alpha: float = 5.0,
+        t_alpha: int | None = None,
+        Simplified_AdEMAMix: bool = False,
+        alpha_grad: float = 100.0,
+        nnmf_factor: bool = False,
+        # prodigy parameters
+        beta3: float = None,
+        d0: float = 1e-6,
+        d_coef: float = 1,
+        growth_rate: float = float('inf'),
+        safeguard_warmup: bool = False,
+        fsdp_in_use: bool = False,
+        slice_p: int = 11,
+        prodigy_steps: int = 0,
+    ):
+        if not (lr >= 0.0):
+            raise ValueError(f"Learning-rate should be >= 0.0. Got {lr}")
+        if not (0.0 <= betas[0] < 1.0 and 0.0 <= betas[1] < 1.0):
+            raise ValueError(f"Betas should be in [0.0, 1.0). Got {betas}")
+        if not (eps >= 0.0):
+            raise ValueError(f"Epsilon should be >= 0.0. Got {eps}")
+        if not (weight_decay >= 0.0):
+            raise ValueError(f"Weight-decay should be >= 0.0. Got {weight_decay}")
+        if not (prodigy_steps >= 0):
+            raise ValueError(f"prodigy_steps should be >= 0. Got {prodigy_steps}")
+        if cautious_mask and grams_moment:
+            print("Warning: cautious is incompatible with grams, Disabling cautious.")
+            cautious_mask = False
+        if betas[0] == 0.0 and Simplified_AdEMAMix:
+            raise ValueError(f"Beta1 cannot be 0.0 when using Simplified_AdEMAMix. Got {betas[0]}")
+        if use_AdEMAMix and Simplified_AdEMAMix:
+            print("Warning: use_AdEMAMix is incompatible with Simplified_AdEMAMix, Disabling use_AdEMAMix.")
+        if grams_moment and Simplified_AdEMAMix:
+            print("Warning: grams is incompatible with Simplified_AdEMAMix, Disabling grams.")
+        if cautious_mask and Simplified_AdEMAMix:
+            print("Warning: cautious is incompatible with Simplified_AdEMAMix, Disabling cautious.")
+        if use_atan2 and Simplified_AdEMAMix:
+            print("Warning: use_atan2 is incompatible with Simplified_AdEMAMix. Disabling use_atan2.")
+            use_atan2 = False
+        if Simplified_AdEMAMix and alpha_grad > 0:
+            # scales d_coef by alpha_grad, this force prodigy to behave well with Simplified_AdEMAMix
+            d_coef = d_coef/alpha_grad
+
+        defaults = {
+            "lr": lr, "betas": betas, "eps": eps, "weight_decay": weight_decay,
+            "vector_reshape": vector_reshape, "use_atan2": use_atan2,
+            "use_orthograd": use_orthograd,
+            "beta3_ema": beta3_ema, "alpha": alpha, "t_alpha": t_alpha,
+            "beta3": beta3, "d": d0, "d0": d0, "d_max": d0, "d_numerator": 0.0, "d_coef": d_coef,
+            "growth_rate": growth_rate, "safeguard_warmup": safeguard_warmup, "k": 0, "slice_p": slice_p,
+            "fsdp_in_use": fsdp_in_use, "prodigy_steps": prodigy_steps,
+            "alpha_grad": alpha_grad, 
+        }
+        self.stochastic_rounding = stochastic_rounding
+        self.cautious_mask = cautious_mask and not Simplified_AdEMAMix
+        self.grams_moment = grams_moment and not Simplified_AdEMAMix
+        self.use_AdEMAMix = use_AdEMAMix and not Simplified_AdEMAMix
+        self.Simplified_AdEMAMix = Simplified_AdEMAMix
+        self.factored = nnmf_factor
+        self.fsdp_in_use = fsdp_in_use
+        super().__init__(params, defaults)
+        self.init_step()
+
+    @property
+    def supports_fused_back_pass(self):
+        return True
+
+    @property
+    def supports_memory_efficient_fp16(self):
+        return True
+
+    @property
+    def supports_flat_params(self):
+        return False
+
+    def init_step(self):
+        """Resets accumulators and calculates dlr for the upcoming step."""
+        self.d_denom = 0.0
+        
+        g_group = self.param_groups[0]
+        self.beta1, self.beta2 = g_group['betas']
+        self.beta3 = g_group['beta3']
+        if self.beta3 is None:
+            self.beta3 = math.sqrt(self.beta2)
+        
+        k = g_group['k']
+        self.d = g_group['d']
+        lr = g_group['lr']
+
+        self.dlr = self.d * lr
+
+        self.d_numerator = g_group.get('d_numerator', 0.0) * self.beta3
+
+    @torch.no_grad()
+    def step_parameter(self, p: torch.Tensor, group: dict, i: int | None = None):
+        if p.grad is None:
+            return
+
+        if hasattr(p, "_fsdp_flattened"):
+            self.fsdp_in_use = True
+
+        grad = p.grad
+        if grad.dtype != torch.float32 and self.factored:
+            grad = grad.float()
+        if group["use_orthograd"]:
+            grad = _orthogonalize_gradient(p, grad)
+        state = self.state[p]
+
+        # State Initialization
+        if len(state) == 0:
+            state['step'] = 0
+
+            should_factor = (
+                self.factored and
+                not (len(p.shape) == 1 and not group['vector_reshape'])
+            )
+
+            state['factored'] = should_factor
+
+            slice_p = group['slice_p']
+
+            dtype = torch.float32 if self.factored else p.dtype
+            device = p.device
+
+            if state['factored']:
+                state['effective_shape'] = _get_effective_shape(p.numel())
+                d1, d2 = state['effective_shape']
+
+                # First moment (m)
+                if self.beta1 > 0:
+                    state['mu_m_nmf'] = torch.zeros(d1, device=device, dtype=dtype) 
+                    state['mv_m_nmf'] = torch.zeros(d2, device=device, dtype=dtype)
+                    if not self.grams_moment:
+                        packed_d2 = (d2 + 7) // 8
+                        state['sign'] = torch.zeros((d1, packed_d2), dtype=torch.uint8, device=device)
+                if self.use_AdEMAMix:
+                    state['mu_m_slow_nmf'] = torch.zeros(d1, device=p.device, dtype=dtype) 
+                    state['mv_m_slow_nmf'] = torch.zeros(d2, device=p.device, dtype=dtype)
+                    packed_d2 = (d2 + 7) // 8
+                    state['sign_slow'] = torch.zeros((d1, packed_d2), dtype=torch.uint8, device=p.device)
+                # Second moment (v)
+                state['mu_v_nmf'] = torch.zeros(d1, device=device, dtype=dtype) 
+                state['mv_v_nmf'] = torch.zeros(d2, device=device, dtype=dtype)
+            else:  # Fallback to standard AdamW for non-factored tensors
+                if self.beta1 > 0:
+                    state['exp_avg'] = torch.zeros_like(p, device=device, dtype=dtype)
+                if self.use_AdEMAMix:
+                    state['exp_avg_slow'] = torch.zeros_like(p, dtype=dtype)
+                state['exp_avg_sq'] = torch.zeros_like(p, device=device, dtype=dtype)
+
+            state['s'] = torch.zeros_like(p.flatten()[::slice_p]).detach()
+            if p.any():
+                state['p0'] = p.flatten()[::slice_p].detach().clone()
+            else:
+                state['p0'] = torch.tensor(0, device=device, dtype=p.dtype)
+
+        if self.use_AdEMAMix:
+            beta3_ema = group['beta3_ema']
+            alpha = group['alpha']
+            t_alpha = group['t_alpha']
+            current_step = state['step'] + 1
+            alpha_t = alpha
+            if t_alpha is not None and t_alpha > 0 and current_step < t_alpha:
+                alpha_t = min(current_step * alpha / t_alpha, alpha)
+        if self.Simplified_AdEMAMix:
+            alpha_grad = group["alpha_grad"]
+
+        if state['factored']:
+            d1, d2 = state['effective_shape']
+
+            grad_reshaped = grad.view(d1, d2)
+
+            # Reconstruct momentum from previous step's factors
+            if self.beta1 > 0:
+                mt = _unnmf((state['mu_m_nmf'], state['mv_m_nmf']))
+                if not self.grams_moment:
+                    unpacked_sign = _unpack_bools(state['sign'], original_m=d2)
+                    torch.where(unpacked_sign, mt, -mt, out=mt)
+                    del unpacked_sign
+                # Update momentum in full-size
+                if self.Simplified_AdEMAMix:
+                    mt.mul_(self.beta1).add_(grad_reshaped, alpha=self.d)
+                else:
+                    mt.mul_(self.beta1).add_(grad_reshaped, alpha=self.d * (1.0 - self.beta1))
+                if self.grams_moment:
+                    mt.copy_(grad_reshaped.sign() * mt.abs())
+                elif self.cautious_mask:
+                    mask = (mt * grad_reshaped > 0).to(grad_reshaped.dtype)
+                    mask.div_(mask.mean().clamp_(min=1e-3))
+                    mt.mul_(mask)
+                    del mask
+
+            vt = _unnmf((state['mu_v_nmf'], state['mv_v_nmf']))
+            vt.mul_(self.beta2).addcmul_(grad_reshaped, grad_reshaped, value=self.d * self.d * (1.0 - self.beta2))
+
+            if self.use_AdEMAMix:
+                mt_slow = _unnmf((state['mu_m_slow_nmf'], state['mv_m_slow_nmf']))
+                if state['sign_slow'].dtype != torch.uint8:
+                    state['sign_slow'] = state['sign_slow'].to(torch.uint8)
+                unpacked_sign_slow = _unpack_bools(state['sign_slow'], original_m=d2)
+                torch.where(unpacked_sign_slow, mt_slow, -mt_slow, out=mt_slow)
+                del unpacked_sign_slow
+                mt_slow.mul_(beta3_ema).add_(grad_reshaped, alpha=self.d * (1.0 - beta3_ema))
+                if self.beta1 > 0:
+                    update = torch.add(mt, mt_slow, alpha=alpha_t)
+                else:
+                    update = torch.add(grad_reshaped, mt_slow, alpha=alpha_t)
+            elif self.Simplified_AdEMAMix:
+                update = torch.add(mt, grad_reshaped, alpha=alpha_grad * self.d)
+            else:
+                update = mt.clone() if self.beta1 > 0 else grad_reshaped.clone()
+            del grad_reshaped
+
+            if group['use_atan2']:
+                a = 1.2732395
+                denom = vt.sqrt()
+                update.atan2_(denom).mul_(a)
+            else:
+                denom = vt.sqrt()
+                update.div_(denom.add_(self.d * group['eps']))
+            del denom
+
+            update = update.view(p.shape).mul_(self.dlr)
+
+            # Compress updated moments and store new factors
+            if self.beta1 > 0:
+                if not self.grams_moment:
+                    state['sign'] = _pack_bools(mt > 0)
+                _nnmf(mt.abs(), out=(state['mu_m_nmf'], state['mv_m_nmf']))
+                del mt
+            if self.use_AdEMAMix:
+                state['sign_slow'] = _pack_bools(mt_slow > 0)
+                _nnmf(mt_slow.abs(), out=(state['mu_m_slow_nmf'], state['mv_m_slow_nmf']))
+                del mt_slow
+            _nnmf(vt, out=(state['mu_v_nmf'], state['mv_v_nmf']))
+            del vt
+
+        else:  # Standard AdamW logic for non-factored tensors
+            exp_avg_sq = state['exp_avg_sq']
+
+            if self.beta1 > 0:
+                exp_avg = state['exp_avg']
+                if self.Simplified_AdEMAMix:
+                    exp_avg.mul_(self.beta1).add_(grad, alpha=self.d)
+                else:
+                    exp_avg.mul_(self.beta1).add_(grad, alpha=self.d * (1.0 - self.beta1))
+                if self.grams_moment:
+                    exp_avg = grad.sign() * exp_avg.abs()
+                elif self.cautious_mask:
+                    mask = (exp_avg * grad > 0).to(grad.dtype)
+                    mask.div_(mask.mean().clamp_(min=1e-3))
+                    exp_avg.mul_(mask)
+                    del mask
+
+            if self.use_AdEMAMix:
+                exp_avg_slow = state['exp_avg_slow']
+                exp_avg_slow.mul_(beta3_ema).add_(grad, alpha=self.d * (1.0 - beta3_ema))
+                if self.beta1 > 0:
+                    update = torch.add(exp_avg, exp_avg_slow, alpha=alpha_t)
+                else:
+                    update = torch.add(grad, exp_avg_slow, alpha=alpha_t)
+            elif self.Simplified_AdEMAMix:
+                update = torch.add(exp_avg, grad, alpha=alpha_grad * self.d)
+            else:
+                update = exp_avg.clone() if self.beta1 > 0 else grad.clone()
+
+            exp_avg_sq.mul_(self.beta2).addcmul_(grad, grad.conj(), value=self.d * self.d * (1.0 - self.beta2))
+
+            if group['use_atan2']:
+                a = 1.2732395
+                denom = exp_avg_sq.sqrt()
+                update.atan2_(denom).mul_(a)
+            else:
+                denom = exp_avg_sq.sqrt()
+                update.div_(denom.add_(self.d * group['eps']))
+            del denom
+
+            update.mul_(self.dlr)
+
+        # --- Accumulate Prodigy stats ---
+        prodigy_steps = group['prodigy_steps']
+        if prodigy_steps <= 0 or group['k'] < prodigy_steps:
+            d0, safeguard_warmup, slice_p = group['d0'], group['safeguard_warmup'], group['slice_p']
+            s, p0 = state['s'], state['p0']
+            grad_flat = grad.flatten().float()
+            p_flat = p.data.flatten().float()
+            p0 = p0.float()
+
+            self.d_numerator += (self.d / d0) * self.dlr * torch.dot(grad_flat[::slice_p], p0.data - p_flat[::slice_p]).item()
+
+            alpha = ((self.d / d0) * self.d) if safeguard_warmup else ((self.d / d0) * self.dlr)
+            s.mul_(self.beta3).add_(grad_flat[::slice_p], alpha=alpha)
+            self.d_denom += s.abs().sum().item()
+
+            del s, p0, grad_flat, p_flat, alpha
+        else:
+            # Free memory if prodigy_steps is reached
+            if 's' in state:
+                del state['s']
+            if 'p0' in state:
+                del state['p0']
+
+        # Decoupled weight decay
+        if group["weight_decay"] != 0:
+            if p.dtype == torch.bfloat16 and self.stochastic_rounding:
+                add_stochastic_(p.data, p.data, alpha=-group["weight_decay"] * self.dlr)
+            else:
+                p.data.add_(p.data, alpha=-group["weight_decay"] * self.dlr)
+
+        if p.dtype == torch.bfloat16 and self.stochastic_rounding:
+            add_stochastic_(p.data, -update)
+        else:
+            p.data.add_(-update)
+        del update
+
+        state['step'] += 1
+
+    @torch.no_grad()
+    def step(self, closure=None):
+        """Performs a single optimization step."""
+        loss = None
+        if closure is not None:
+            with torch.enable_grad():
+                loss = closure()
+
+        for group in self.param_groups:
+            for i, p in enumerate(group['params']):
+                self.step_parameter(p, group, i)
+
+
+        self.calculate_d()
+        self.init_step()
+        return loss
+
+    def calculate_d(self):
+        """Calculates the new `d` based on the accumulated stats."""
+        g_group = self.param_groups[0]
+        
+        # Only perform d-adaptation if prodigy_steps has not been reached
+        prodigy_active = not (g_group.get('prodigy_steps', 0) > 0 and g_group['k'] >= g_group['prodigy_steps'])
+
+        if prodigy_active:
+            d_max, d_coef, growth_rate = g_group['d_max'], g_group['d_coef'], g_group['growth_rate']
+            
+            if self.fsdp_in_use and dist.is_available() and dist.is_initialized():
+                # Use the device of the first parameter to avoid hardcoding '.cuda()'
+                device = self.param_groups[0]['params'][0].device
+                dist_tensor = torch.tensor([self.d_numerator, self.d_denom], device=device)
+                dist.all_reduce(dist_tensor, op=dist.ReduceOp.SUM)
+                global_d_numerator = dist_tensor[0].item()
+                global_d_denom = dist_tensor[1].item()
+            else:
+                global_d_numerator = self.d_numerator
+                global_d_denom = self.d_denom
+
+            d_hat = self.d
+            if global_d_denom > 0:
+                d_hat = d_coef * global_d_numerator / global_d_denom
+                if self.d == g_group['d0']:
+                    self.d = max(self.d, d_hat)
+                d_max = max(d_max, d_hat)
+                self.d = min(d_max, self.d * growth_rate)
+
+            for group in self.param_groups:
+                group['d_numerator'] = global_d_numerator
+                group['d'] = self.d
+                group['d_max'] = d_max
+        
+        # Increment step counter for all groups, regardless of whether d was updated
+        for group in self.param_groups:
+            group['k'] += 1