adv-optm 0.1.7__py3-none-any.whl → 0.1.8__py3-none-any.whl

adv_optm/__init__.py

@@ -16,4 +16,4 @@ __all__ = [
     "Lion_Prodigy_adv",
 ]
 
-__version__ = "0.1.7"
+__version__ = "0.1.8"

adv_optm/optim/AdamW_adv.py

@@ -55,7 +55,7 @@ class AdamW_adv(torch.optim.Optimizer):
             the warmup, `alpha` ramps from 0 to its target value. If `None`,
             the scheduler is disabled. (default: None)
         factored (bool): whether to use the factorization or disable it to use
-            the uncompressed optimizer. (default: True)
+            the uncompressed optimizer. (default: False)
     """
 
     def __init__(
@@ -76,7 +76,7 @@ class AdamW_adv(torch.optim.Optimizer):
         beta3_ema: float = 0.9999,
         alpha: float = 5.0,
         t_alpha: int | None = None,
-        factored: bool = True,
+        factored: bool = False,
     ):
         if not (lr >= 0.0):
             raise ValueError(f"Learning-rate should be >= 0.0. Got {lr}")
@@ -216,7 +216,10 @@ class AdamW_adv(torch.optim.Optimizer):
                 del unpacked_sign_slow
 
                 mt_slow.mul_(beta3_ema).add_(grad_reshaped, alpha=1.0 - beta3_ema)
-                update = mt + (alpha_t * mt_slow) if beta1 > 0 else grad_reshaped + (alpha_t * mt_slow)
+                if beta1 > 0:
+                    update = torch.add(mt, mt_slow, alpha=alpha_t)
+                else:
+                    update = torch.add(grad_reshaped, mt_slow, alpha=alpha_t)
             else:
                 update = mt.clone() if beta1 > 0 else grad_reshaped.clone()
             del grad_reshaped
@@ -262,7 +265,10 @@ class AdamW_adv(torch.optim.Optimizer):
             if self.use_AdEMAMix:
                 exp_avg_slow = state['exp_avg_slow']
                 exp_avg_slow.mul_(beta3_ema).add_(grad, alpha=1 - beta3_ema)
-                update = exp_avg + (alpha_t * exp_avg_slow) if beta1 > 0 else grad + (alpha_t * exp_avg_slow)
+                if beta1 > 0:
+                    update = torch.add(exp_avg, exp_avg_slow, alpha=alpha_t)
+                else:
+                    update = torch.add(grad, exp_avg_slow, alpha=alpha_t)
             else:
                 update = exp_avg.clone() if beta1 > 0 else grad.clone()
 

adv_optm/optim/Adopt_adv.py

@@ -63,7 +63,7 @@ class Adopt_adv(torch.optim.Optimizer):
             the scheduler is disabled and the full `alpha` value is used from
             the start. (default: None)
         factored (bool): whether to use the factorization or disable it to use
-            the uncompressed optimizer. (default: True)
+            the uncompressed optimizer. (default: False)
     """
 
     def __init__(
@@ -84,7 +84,7 @@ class Adopt_adv(torch.optim.Optimizer):
         beta3_ema: float = 0.9999,
         alpha: float = 5.0,
         t_alpha: int | None = None,
-        factored: bool = True,
+        factored: bool = False,
     ):
         if not (lr >= 0.0):
             raise ValueError(f"Learning-rate should be >= 0.0. Got {lr}")
@@ -235,7 +235,7 @@ class Adopt_adv(torch.optim.Optimizer):
 
             if self.use_AdEMAMix:
                 mt_slow.mul_(beta3_ema).add_(normalized_grad, alpha=1.0 - beta3_ema)
-                update = mt + (alpha_t * mt_slow)
+                update = torch.add(mt, m_slow, alpha=alpha_t)
                 update = update.view(p.shape)
             else:
                 update = mt.view(p.shape)
@@ -295,9 +295,9 @@ class Adopt_adv(torch.optim.Optimizer):
 
             if self.use_AdEMAMix:
                 m_slow.mul_(beta3_ema).add_(normalized_grad, alpha=1.0 - beta3_ema)
-                update = m + (alpha_t * m_slow)
+                update = torch.add(m, m_slow, alpha=alpha_t)
             else:
-                update = m
+                update = m.clone()
 
             if self.use_atan2:
                 update.mul_(group['lr'] * 1.2732395447351628)

adv_optm/optim/Lion_Prodigy_adv.py

@@ -33,8 +33,6 @@ class Lion_Prodigy_adv(torch.optim.Optimizer):
             (default: 0.0).
         factored (bool): whether to use the factorization or use the
             uncompressed optimizer. (default: True)
-        variance_reduction (bool): whether to use the variance reduction technique
-            from "Convergence Analysis of the Lion Optimizer" (arXiv:2508.12327v1). (default: False).
         d0 (float):
             Initial D estimate for D-adaptation (default 1e-6). Rarely needs changing.
         d_coef (float):
@@ -66,7 +64,6 @@ class Lion_Prodigy_adv(torch.optim.Optimizer):
         use_cautious: bool = False,
         clip_threshold: float = 0.0,
         factored: bool = True,
-        variance_reduction: bool = False,
         # prodigy parameters
         beta3: float = None,
         d0: float = 1e-6,
@@ -97,7 +94,6 @@ class Lion_Prodigy_adv(torch.optim.Optimizer):
         self.stochastic_rounding = stochastic_rounding
         self.use_cautious = use_cautious
         self.factored = factored
-        self.variance_reduction = variance_reduction
         self.fsdp_in_use = fsdp_in_use
         super().__init__(params, defaults)
         # Global state for accumulating metrics across parameter updates within a single step.
@@ -183,12 +179,8 @@ class Lion_Prodigy_adv(torch.optim.Optimizer):
                 state['mv_m_nmf'] = torch.zeros(d2, device=p.device, dtype=dtype)
                 packed_d2 = (d2 + 7) // 8
                 state['sign'] = torch.zeros((d1, packed_d2), dtype=torch.uint8, device=p.device)
-                if self.variance_reduction:
-                    state['prev_grad'] = torch.zeros((d1, d2), device=p.device, dtype=dtype)
             else: # Fallback to standard Lion
                 state['exp_avg'] = torch.zeros_like(p, device=p.device, dtype=dtype)
-                if self.variance_reduction:
-                    state['prev_grad'] = torch.zeros_like(p, device=p.device, dtype=dtype)
 
         if state['factored']:
             # Factored Path
@@ -215,20 +207,7 @@ class Lion_Prodigy_adv(torch.optim.Optimizer):
             update_for_param = signed_update.view(p.shape).mul(self.dlr)
 
             # Update momentum m_t = β2*m_{t-1} + (1-β2)*lr*g_t
-            if self.variance_reduction:
-                if state['step'] == 1:
-                    exp_avg.copy_(grad_reshaped)
-                else:
-                    # Heuristic Prodigy-STORM update
-                    correction = exp_avg.sub(state['prev_grad'])
-                    grad_alpha = self.d * (1 - self.beta2) + self.beta2
-                    exp_avg.copy_(grad_reshaped).mul_(grad_alpha).add_(correction, alpha=self.beta2)
-                    del correction, grad_alpha
-                state['prev_grad'].copy_(grad_reshaped)
-            else:
-                # Standard Prodigy-Lion
-                alpha = self.d * (1 - self.beta2)
-                exp_avg.mul_(self.beta2).add_(grad_reshaped, alpha=alpha)
+            exp_avg.mul_(self.beta2).add_(grad_reshaped, alpha=self.d * (1 - self.beta2))
             del grad_reshaped
 
             # Compress new momentum m_t and store factors
@@ -254,20 +233,7 @@ class Lion_Prodigy_adv(torch.optim.Optimizer):
             update_for_param = signed_update.mul(self.dlr)
 
             # Update momentum 
-            if self.variance_reduction:
-                if state['step'] == 1:
-                    exp_avg.copy_(grad)
-                else:
-                    # Heuristic Prodigy-STORM update
-                    correction = exp_avg.sub(state['prev_grad'])
-                    grad_alpha = self.d * (1 - self.beta2) + self.beta2
-                    exp_avg.copy_(grad).mul_(grad_alpha).add_(correction, alpha=self.beta2)
-                    del grad_alpha, correction
-                state['prev_grad'].copy_(grad)
-            else:
-                # Standard Prodigy-Lion
-                alpha = self.d * (1 - self.beta2)
-                exp_avg.mul_(self.beta2).add_(grad, alpha=alpha)
+            exp_avg.mul_(self.beta2).add_(grad, alpha=self.d * (1 - self.beta2))
 
         # --- Accumulate Prodigy stats ---
         d0, safeguard_warmup, slice_p = group['d0'], group['safeguard_warmup'], group['slice_p']
@@ -298,7 +264,7 @@ class Lion_Prodigy_adv(torch.optim.Optimizer):
         else:
             p.data.add_(-update_for_param)
 
-            del update_for_param
+        del update_for_param
 
     @torch.no_grad()
     def step(self, closure: Optional[callable] = None):

adv_optm/optim/Lion_adv.py

@@ -33,8 +33,6 @@ class Lion_adv(torch.optim.Optimizer):
             (default: 0.0).
         factored (bool): whether to use the factorization or use the
             uncompressed optimizer. (default: True)
-        variance_reduction (bool): whether to use the variance reduction technique
-            from "Convergence Analysis of the Lion Optimizer" (arXiv:2508.12327v1). (default: False).
     """
 
     def __init__(
@@ -49,7 +47,6 @@ class Lion_adv(torch.optim.Optimizer):
         use_cautious: bool = False,
         clip_threshold: float = 0.0,
         factored: bool = True,
-        variance_reduction: bool = False,
     ):
         if not lr > 0.0:
             raise ValueError(f"Learning rate must be > 0.0, but got {lr}")
@@ -69,7 +66,6 @@ class Lion_adv(torch.optim.Optimizer):
         self.stochastic_rounding = stochastic_rounding
         self.use_cautious = use_cautious
         self.factored = factored
-        self.variance_reduction = variance_reduction
         super().__init__(params, defaults)
 
     @property
@@ -122,12 +118,8 @@ class Lion_adv(torch.optim.Optimizer):
                 state['mv_m_nmf'] = torch.zeros(d2, device=p.device, dtype=dtype)
                 packed_d2 = (d2 + 7) // 8
                 state['sign'] = torch.zeros((d1, packed_d2), dtype=torch.uint8, device=p.device)
-                if self.variance_reduction:
-                    state['prev_grad'] = torch.zeros((d1, d2), device=p.device, dtype=dtype)
             else: # Fallback to standard Lion
                 state['exp_avg'] = torch.zeros_like(p, device=p.device, dtype=dtype)
-                if self.variance_reduction:
-                    state['prev_grad'] = torch.zeros_like(p, device=p.device, dtype=dtype)
 
         state['step'] += 1
         beta1, beta2 = group["betas"]
@@ -157,21 +149,9 @@ class Lion_adv(torch.optim.Optimizer):
             # Parameter update
             update_for_param = signed_update.view(p.shape).mul_(lr)
 
-            # Update momentum
-            if self.variance_reduction:
-                if state['step'] == 1:
-                    exp_avg.copy_(grad_reshaped)
-                else:
-                    # Use the simplified STORM update: m_t = g_t + β₂ * (m_{t-1} - g_{t-1})
-                    correction = exp_avg.sub(state['prev_grad'])
-                    # Calculate the new momentum and store it back into exp_avg
-                    exp_avg.copy_(grad_reshaped).add_(correction, alpha=beta2)
-                    del correction
-                # Update prev_grad for the next iteration
-                state['prev_grad'].copy_(grad_reshaped)
-            else:
-                # Standard Lion momentum update
-                exp_avg.mul_(beta2).add_(grad_reshaped, alpha=1-beta2)
+            # Standard Lion momentum update
+            exp_avg.mul_(beta2).add_(grad_reshaped, alpha=1-beta2)
+            del grad_reshaped
 
             # Compress new momentum m_t and store factors
             state['sign'] = _pack_bools(exp_avg > 0)
@@ -195,21 +175,8 @@ class Lion_adv(torch.optim.Optimizer):
 
             update_for_param = signed_update.mul_(lr)
 
-            # Update momentum 
-            if self.variance_reduction:
-                if state['step'] == 1:
-                    exp_avg.copy_(grad)
-                else:
-                    # Use the simplified STORM update: m_t = g_t + β₂ * (m_{t-1} - g_{t-1})
-                    correction = exp_avg.sub(state['prev_grad'])
-                    # Calculate the new momentum and store it back into exp_avg
-                    exp_avg.copy_(grad).add_(correction, alpha=beta2)
-                    del correction
-                # Update prev_grad for the next iteration
-                state['prev_grad'].copy_(grad)
-            else:
-                # Standard Lion momentum update
-                exp_avg.mul_(beta2).add_(grad, alpha=1-beta2)
+            # Standard Lion momentum update
+            exp_avg.mul_(beta2).add_(grad, alpha=1-beta2)
 
         if group["weight_decay"] != 0:
             if p.dtype == torch.bfloat16 and self.stochastic_rounding:
@@ -225,7 +192,7 @@ class Lion_adv(torch.optim.Optimizer):
         else:
             p.data.add_(-update_for_param)
 
-            del update_for_param
+        del update_for_param
 
     @torch.no_grad()
     def step(self, closure: Optional[callable] = None):

adv_optm/optim/Prodigy_adv.py

@@ -64,7 +64,7 @@ class Prodigy_adv(torch.optim.Optimizer):
             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: True)
+            the uncompressed optimizer. (default: False)
         d0 (float):
             Initial D estimate for D-adaptation (default 1e-6). Rarely needs changing.
         d_coef (float):
@@ -82,6 +82,9 @@ class Prodigy_adv(torch.optim.Optimizer):
         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__(
@@ -103,7 +106,7 @@ class Prodigy_adv(torch.optim.Optimizer):
         t_alpha: int | None = None,
         Simplified_AdEMAMix: bool = False,
         alpha_grad: float = 100.0,
-        factored: bool = True,
+        factored: bool = False,
         # prodigy parameters
         beta3: float = None,
         d0: float = 1e-6,
@@ -112,6 +115,7 @@ class Prodigy_adv(torch.optim.Optimizer):
         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}")
@@ -121,6 +125,8 @@ class Prodigy_adv(torch.optim.Optimizer):
             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 betas[0] == 0.0 and Simplified_AdEMAMix:
             raise ValueError(f"Beta 1 cannot be 0.0 when using Simplified_AdEMAMix. Got {betas[0]}")
         if use_AdEMAMix and Simplified_AdEMAMix:
@@ -132,6 +138,9 @@ class Prodigy_adv(torch.optim.Optimizer):
         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,
@@ -140,7 +149,7 @@ class Prodigy_adv(torch.optim.Optimizer):
             "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,
+            "fsdp_in_use": fsdp_in_use, "prodigy_steps": prodigy_steps,
             "alpha_grad": alpha_grad, 
         }
         self.stochastic_rounding = stochastic_rounding
@@ -293,7 +302,10 @@ class Prodigy_adv(torch.optim.Optimizer):
                 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))
-                update = mt + (alpha_t * mt_slow) if self.beta1 > 0 else grad_reshaped + (alpha_t * mt_slow)
+                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:
@@ -344,7 +356,10 @@ class Prodigy_adv(torch.optim.Optimizer):
             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))
-                update = exp_avg + (alpha_t * exp_avg_slow) if self.beta1 > 0 else grad + (alpha_t * exp_avg_slow)
+                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:
@@ -364,19 +379,27 @@ class Prodigy_adv(torch.optim.Optimizer):
             update.mul_(self.dlr)
 
         # --- Accumulate Prodigy stats ---
-        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()
+        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()
+            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()
+            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
+            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:
@@ -413,29 +436,37 @@ class Prodigy_adv(torch.optim.Optimizer):
     def calculate_d(self):
         """Calculates the new `d` based on the accumulated stats."""
         g_group = self.param_groups[0]
-        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)
-
+        # 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['d_numerator'] = global_d_numerator
-            group['d'] = self.d
-            group['d_max'] = d_max
             group['k'] += 1

{adv_optm-0.1.7.dist-info → adv_optm-0.1.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: adv_optm
-Version: 0.1.7
+Version: 0.1.8
 Summary: A family of highly efficient, lightweight yet powerful optimizers.
 Home-page: https://github.com/Koratahiu/Advanced_Optimizers
 Author: Koratahiu

adv_optm-0.1.8.dist-info/RECORD

@@ -0,0 +1,19 @@
+adv_optm/__init__.py,sha256=csc19AmU_h7daI3bo4hDVBouMqGiHejfipPIOGFAUQ8,306
+adv_optm/optim/AdamW_adv.py,sha256=Had6kzSBI0eEMiL2yI1wa1nEBoPfgwHQGtnRcDJ8tXI,14078
+adv_optm/optim/Adopt_adv.py,sha256=-iAKhPbEnzdL0Mx96h2BBlJB85TyHdkjULRjWvNbTyY,14833
+adv_optm/optim/Lion_Prodigy_adv.py,sha256=kIAGXoMbDNRg5reKXtUC_vQQ2gyM-NXPB-Pv9zSpiE8,12787
+adv_optm/optim/Lion_adv.py,sha256=05j_j6LIzHW5b79DVwMIf1FZHVNB8xnStNVjlOdVkCE,8256
+adv_optm/optim/Prodigy_adv.py,sha256=U4grKRumzDJRYSI-QHmmZZ7ed_67tyiC3OPSXqJVBx8,21759
+adv_optm/optim/Simplified_AdEMAMix.py,sha256=opIZjnGJ03-DDAIHTZyJBMReVfgusGDb8FZSWMU3-UM,9774
+adv_optm/optim/__init__.py,sha256=pcP865H2j1tut2VfTUhzQh7V8TF_tzPjqFnjMfFed2k,382
+adv_optm/util/BF16_Stochastic_Rounding.py,sha256=Q5H0BcogmE4atP65dLoI21HKSf50lRdsBDfeF6v9Tbg,1548
+adv_optm/util/Effective_Shape.py,sha256=TBvIk1V8IuTbbBsxuekJA4e_v8JlR5Nujtut8RTWAm4,318
+adv_optm/util/NNMF.py,sha256=yRf5IP5Sjq0Uf0DxN0Q8NxEGSdD-f1ULziLVDOjY8K4,639
+adv_optm/util/One_Bit_Boolean.py,sha256=Wat49esdwohuN-OHOFMW8D0aOQgV9cP5Rl8z6yfmpos,1068
+adv_optm/util/OrthoGrad.py,sha256=NzInuBQGy_Ja__M1R9XbvqVaQ0fhGbtGgFE9YON7B3I,707
+adv_optm/util/__init__.py,sha256=qoyIF0jcLjs_vSEcsv36clw5LFNBEbifyXrrVxMH-G4,349
+adv_optm-0.1.8.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+adv_optm-0.1.8.dist-info/METADATA,sha256=Ydu5_f_d19hoYMf9zvP3eu9ci8XsLWyDuY99JYJVR9o,5846
+adv_optm-0.1.8.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+adv_optm-0.1.8.dist-info/top_level.txt,sha256=iNfBIIzu-lPrQ7jyC56WBCcbkRwitM2nJ15-MRQ_6fg,9
+adv_optm-0.1.8.dist-info/RECORD,,

adv_optm-0.1.7.dist-info/RECORD

@@ -1,19 +0,0 @@
-adv_optm/__init__.py,sha256=CZ_tjWWk5d5D8q_R0rcr8vvwlZyY_44zyAcIAmN_SDY,306
-adv_optm/optim/AdamW_adv.py,sha256=ZeNzk2tWbyd2QDI5hp4InwG3iuHHfqLrlhr_VmcQfRM,13884
-adv_optm/optim/Adopt_adv.py,sha256=rzBWfFOPrMuC6vwETsw7QPKmVXcv4IJRDCTj-6eU1Qk,14798
-adv_optm/optim/Lion_Prodigy_adv.py,sha256=JMss9X8lRpIU4E34PfFpWMMal_XNvZ8Yuqc6i7R5wIQ,14588
-adv_optm/optim/Lion_adv.py,sha256=BA4bSEhJiQ7BhGLDRn9nuMlBrLVh-OMscbmSTeGgRmI,10137
-adv_optm/optim/Prodigy_adv.py,sha256=gJL2r32R3xGD62jMR55ZyKxRv0yL70XHxj4FzEJbFc4,20196
-adv_optm/optim/Simplified_AdEMAMix.py,sha256=opIZjnGJ03-DDAIHTZyJBMReVfgusGDb8FZSWMU3-UM,9774
-adv_optm/optim/__init__.py,sha256=pcP865H2j1tut2VfTUhzQh7V8TF_tzPjqFnjMfFed2k,382
-adv_optm/util/BF16_Stochastic_Rounding.py,sha256=Q5H0BcogmE4atP65dLoI21HKSf50lRdsBDfeF6v9Tbg,1548
-adv_optm/util/Effective_Shape.py,sha256=TBvIk1V8IuTbbBsxuekJA4e_v8JlR5Nujtut8RTWAm4,318
-adv_optm/util/NNMF.py,sha256=yRf5IP5Sjq0Uf0DxN0Q8NxEGSdD-f1ULziLVDOjY8K4,639
-adv_optm/util/One_Bit_Boolean.py,sha256=Wat49esdwohuN-OHOFMW8D0aOQgV9cP5Rl8z6yfmpos,1068
-adv_optm/util/OrthoGrad.py,sha256=NzInuBQGy_Ja__M1R9XbvqVaQ0fhGbtGgFE9YON7B3I,707
-adv_optm/util/__init__.py,sha256=qoyIF0jcLjs_vSEcsv36clw5LFNBEbifyXrrVxMH-G4,349
-adv_optm-0.1.7.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
-adv_optm-0.1.7.dist-info/METADATA,sha256=BEKyVG9zVdb9WThOw9YtgWZ_zqDmErumpY5Fr-AkbX0,5846
-adv_optm-0.1.7.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-adv_optm-0.1.7.dist-info/top_level.txt,sha256=iNfBIIzu-lPrQ7jyC56WBCcbkRwitM2nJ15-MRQ_6fg,9
-adv_optm-0.1.7.dist-info/RECORD,,