diff --git a/tinygrad/apps/llm.py b/tinygrad/apps/llm.py
index 4f80d4bf2d..b717d5c9fb 100644
--- a/tinygrad/apps/llm.py
+++ b/tinygrad/apps/llm.py
@@ -116,7 +116,7 @@ class TransformerBlock:
 
   def _feed_forward(self, h: Tensor) -> Tensor:
     h_norm = self.ffn_norm(h)
-    gated  = self.ffn_gate(h_norm).silu() * self.ffn_up(h_norm)
+    gated  = (self.ffn_gate(h_norm).silu() * self.ffn_up(h_norm)).contiguous(arg=(1,))
     return h + self.ffn_down(gated)
 
   def __call__(self, x: Tensor, start_pos: int|UOp):
diff --git a/tinygrad/schedule/rangeify.py b/tinygrad/schedule/rangeify.py
index 68507e9e6e..c8a2a123b7 100644
--- a/tinygrad/schedule/rangeify.py
+++ b/tinygrad/schedule/rangeify.py
@@ -123,7 +123,7 @@ class AddBufferContext:
 def add_store(ctx:AddBufferContext, x:UOp):
   rngs = x.src[1:]
   shape = tuple([r.vmax+1 for r in rngs])
-  buf = UOp(Ops.DEFINE_GLOBAL if prod(shape) > 2000 else Ops.DEFINE_LOCAL, dtype=x.dtype.ptr(size=prod(shape)), arg=ctx.dg)
+  buf = UOp(Ops.DEFINE_GLOBAL if prod(shape) > 65536 else Ops.DEFINE_LOCAL, dtype=x.dtype.ptr(size=prod(shape)), arg=ctx.dg)
   ctx.map[buf] = (buf.op, ctx.dg)
   ctx.dg += 1
   return buf.reshape(shape).index(*rngs).store(x.src[0], *rngs)