diff --git a/cli/lora-quant-fidelity.py b/cli/lora-quant-fidelity.py
index df2c64126..1dc41bfe5 100644
--- a/cli/lora-quant-fidelity.py
+++ b/cli/lora-quant-fidelity.py
@@ -315,8 +315,8 @@ def analyze_module(W_dq, deq_params, mods, calib_rms=None):
                 Dw = D * rms if rms is not None else D
                 with torch.random.fork_rng(devices=[D.device] if D.device.type == 'cuda' else []):
                     torch.manual_seed(0)
-                    U, S, V = torch.svd_lowrank(Dw, q=q, niter=4)
-                Dk = (U * S) @ V.t()
+                    U, S, V = torch.svd_lowrank(Dw, q=min(q + 64, *D.shape), niter=8)
+                Dk = (U[:, :q] * S[:q]) @ V[:, :q].t()
                 if rms is not None:
                     Dk = Dk / rms
                 base16 = W_dq.to(torch.bfloat16).float()
diff --git a/modules/lora/lora_factor_cache.py b/modules/lora/lora_factor_cache.py
index a5ef4041a..6e7e46a0b 100644
--- a/modules/lora/lora_factor_cache.py
+++ b/modules/lora/lora_factor_cache.py
@@ -96,7 +96,7 @@ def begin_pass(wanted_names):
             from safetensors import safe_open
             with safe_open(path, framework='pt', device='cpu') as f:
                 meta = f.metadata() or {}
-                if meta.get('sig') == sig and meta.get('fmt') == '2':
+                if meta.get('sig') == sig and meta.get('fmt') == '3':
                     for k in f.keys():
                         entries[k] = f.get_tensor(k)
             os.utime(path, None) # freshness for LRU eviction
@@ -190,7 +190,7 @@ def flush():
         from safetensors.torch import save_file
         os.makedirs(cache_root, exist_ok=True)
         tmp = state['path'] + '.tmp'
-        save_file(state['store'], tmp, metadata={'sig': state['sig'], 'fmt': '2'})
+        save_file(state['store'], tmp, metadata={'sig': state['sig'], 'fmt': '3'})
         os.replace(tmp, state['path'])
         evict()
     except Exception as e:
diff --git a/modules/lora/lora_sdnq.py b/modules/lora/lora_sdnq.py
index 8d5330e09..b2a91de6c 100644
--- a/modules/lora/lora_sdnq.py
+++ b/modules/lora/lora_sdnq.py
@@ -274,7 +274,9 @@ def apply_hosted(self, network_layer_name, updown, wanted_names, use_previous=Fa
     # svd_lowrank draws random projections; fork so user generation seeds are untouched and re-applies are deterministic
     with torch.random.fork_rng(devices=[D.device] if D.device.type == 'cuda' else []):
         torch.manual_seed(0)
-        U, S, V = torch.svd_lowrank(D, q=q, niter=4)
+        # oversampled sketch with extra power iterations lands within noise of exact svd; only the top q columns are kept
+        U, S, V = torch.svd_lowrank(D, q=min(q + 64, *D.shape), niter=8)
+    U, S, V = U[:, :q], S[:q], V[:, :q]
     energy = float(S.square().sum() / D.square().sum().clamp(min=1e-30)) # captured fraction, in the weighted domain when calibrated
     up_h = (U * S).to(dtype=dtype)
     down_h = V.t()