diff --git a/Data/ByteString/FastBuilder/Internal.hs b/Data/ByteString/FastBuilder/Internal.hs
--- a/Data/ByteString/FastBuilder/Internal.hs
+++ b/Data/ByteString/FastBuilder/Internal.hs
@@ -221,12 +221,11 @@
   deriving (Functor)
 
 instance Applicative BuildM where
-  pure = return
+  pure x = BuildM $ \k -> k x
+  {-# INLINE pure #-}
   (<*>) = ap
 
 instance Monad BuildM where
-  return x = BuildM $ \k -> k x
-  {-# INLINE return #-}
   BuildM b >>= f = BuildM $ \k -> b $ \r -> runBuildM (f r) k
   {-# INLINE (>>=) #-}
 
diff --git a/benchmarks/aeson/template.hs b/benchmarks/aeson/template.hs
--- a/benchmarks/aeson/template.hs
+++ b/benchmarks/aeson/template.hs
@@ -10,13 +10,15 @@
 #endif
 
 import Data.Aeson
+import qualified Data.Aeson.Key as Key
+import qualified Data.Aeson.KeyMap as KM
 import qualified Data.ByteString.Lazy as L
-import qualified Data.HashMap.Strict as H
 import Data.Monoid
 import qualified Data.Scientific as Sci
 import qualified Data.Text as T
 import qualified Data.Text.Encoding as T
 import qualified Data.Vector as V
+import GHC.Exts (inline)
 
 import LIB
 
@@ -51,8 +53,8 @@
 fromNumber = either doubleDec integerDec . Sci.floatingOrInteger
 
 fromObject :: Object -> Builder
-fromObject obj = char8 '{' <> H.foldMapWithKey f obj <> char8 '}'
+fromObject obj = char8 '{' <> inline KM.foldMapWithKey f obj <> char8 '}'
   where
       f k v =
-        fromString k <> char8 ':' <> fromValue v
+        fromString (Key.toText k) <> char8 ':' <> fromValue v
         <> char8 ','
diff --git a/fast-builder.cabal b/fast-builder.cabal
--- a/fast-builder.cabal
+++ b/fast-builder.cabal
@@ -1,5 +1,5 @@
 name:                fast-builder
-version:             0.1.4.0
+version:             0.1.5.0
 synopsis:            Fast ByteString Builder
 description:         An efficient implementation of ByteString builder. It should be faster than
                      the standard implementation in most cases.
@@ -36,9 +36,8 @@
   hs-source-dirs:      benchmarks/aeson
   other-modules:       Fast, Bstr
   include-dirs:        benchmarks/aeson
-  build-depends:       base, fast-builder, aeson, criterion, bytestring,
-    scientific, text, vector, deepseq, ghc-prim, template-haskell,
-    unordered-containers >= 0.2.11.0
+  build-depends:       base, fast-builder, aeson >= 2, criterion, bytestring,
+    scientific, text, vector, deepseq, ghc-prim
   ghc-options:         -fsimpl-tick-factor=120
   ghc-options:         -Wall -threaded -g -rtsopts
   default-language:    Haskell2010