diff --git a/Cargo.lock b/Cargo.lock
index fb9f76c..0fd2e30 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -24,7 +24,9 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5a15f179cd60c4584b8a8c596927aadc462e27f2ca70c04e0071964a73ba7a75"
 dependencies = [
  "cfg-if",
+ "getrandom 0.3.4",
  "once_cell",
+ "serde",
  "version_check",
  "zerocopy",
 ]
@@ -104,7 +106,7 @@ version = "1.1.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "40c48f72fd53cd289104fc64099abca73db4166ad86ea0b4341abe65af83dadc"
 dependencies = [
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -115,7 +117,7 @@ checksum = "291e6a250ff86cd4a820112fb8898808a366d8f9f58ce16d1f538353ad55747d"
 dependencies = [
  "anstyle",
  "once_cell_polyfill",
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -243,7 +245,7 @@ dependencies = [
  "bitflags",
  "cexpr",
  "clang-sys",
- "itertools 0.12.1",
+ "itertools 0.13.0",
  "log",
  "prettyplease",
  "proc-macro2",
@@ -299,6 +301,15 @@ version = "0.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "37b2a672a2cb129a2e41c10b1224bb368f9f37a2b16b612598138befd7b37eb5"
 
+[[package]]
+name = "castaway"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dec551ab6e7578819132c713a93c022a05d60159dc86e7a7050223577484c55a"
+dependencies = [
+ "rustversion",
+]
+
 [[package]]
 name = "cc"
 version = "1.2.61"
@@ -420,17 +431,31 @@ version = "1.0.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1d07550c9036bf2ae0c684c4297d503f838287c83c53686d05370d0e139ae570"
 
+[[package]]
+name = "compact_str"
+version = "0.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9dfdd1c2274d9aa354115b09dc9a901d6c5576818cdf70d14cae2bdb47df00ab"
+dependencies = [
+ "castaway",
+ "cfg-if",
+ "itoa",
+ "rustversion",
+ "ryu",
+ "serde",
+ "static_assertions",
+]
+
 [[package]]
 name = "console"
-version = "0.15.11"
+version = "0.16.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "054ccb5b10f9f2cbf51eb355ca1d05c2d279ce1804688d0db74b4733a5aeafd8"
+checksum = "d64e8af5551369d19cf50138de61f1c42074ab970f74e99be916646777f8fc87"
 dependencies = [
  "encode_unicode",
  "libc",
- "once_cell",
  "unicode-width",
- "windows-sys 0.59.0",
+ "windows-sys",
 ]
 
 [[package]]
@@ -578,6 +603,12 @@ dependencies = [
  "memchr",
 ]
 
+[[package]]
+name = "daachorse"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6f55d7153ba3b507595872a3874803f07a8a81d1e888abed8e5db7da0597d6e2"
+
 [[package]]
 name = "darling"
 version = "0.20.11"
@@ -613,6 +644,15 @@ dependencies = [
  "syn",
 ]
 
+[[package]]
+name = "dary_heap"
+version = "0.3.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8b1e3a325bc115f096c8b77bbf027a7c2592230e70be2d985be950d3d5e60ebe"
+dependencies = [
+ "serde",
+]
+
 [[package]]
 name = "der"
 version = "0.8.0"
@@ -721,7 +761,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "39cab71617ae0d63f51a36d69f866391735b51691dbda63cf6f96d042b63efeb"
 dependencies = [
  "libc",
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -914,13 +954,14 @@ dependencies = [
 
 [[package]]
 name = "getrandom"
-version = "0.2.17"
+version = "0.3.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ff2abc00be7fca6ebc474524697ae276ad847ad0a6b3faa4bcb027e9a4614ad0"
+checksum = "899def5c37c4fd7b2664648c28120ecec138e4d395b459e5ca34f9cce2dd77fd"
 dependencies = [
  "cfg-if",
  "libc",
- "wasi",
+ "r-efi 5.3.0",
+ "wasip2",
 ]
 
 [[package]]
@@ -931,7 +972,7 @@ checksum = "0de51e6874e94e7bf76d726fc5d13ba782deca734ff60d5bb2fb2607c7406555"
 dependencies = [
  "cfg-if",
  "libc",
- "r-efi",
+ "r-efi 6.0.0",
  "wasip2",
  "wasip3",
 ]
@@ -1165,14 +1206,14 @@ dependencies = [
 
 [[package]]
 name = "indicatif"
-version = "0.17.11"
+version = "0.18.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "183b3088984b400f4cfac3620d5e076c84da5364016b4f49473de574b2586235"
+checksum = "25470f23803092da7d239834776d653104d551bc4d7eacaf31e6837854b8e9eb"
 dependencies = [
  "console",
- "number_prefix",
  "portable-atomic",
  "unicode-width",
+ "unit-prefix",
  "web-time",
 ]
 
@@ -1184,27 +1225,18 @@ checksum = "a6cb138bb79a146c1bd460005623e142ef0181e3d0219cb493e02f7d08a35695"
 
 [[package]]
 name = "itertools"
-version = "0.11.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b1c173a5686ce8bfa551b3563d0c2170bf24ca44da99c7ca4bfdab5418c3fe57"
-dependencies = [
- "either",
-]
-
-[[package]]
-name = "itertools"
-version = "0.12.1"
+version = "0.13.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ba291022dbbd398a455acf126c1e341954079855bc60dfdda641363bd6922569"
+checksum = "413ee7dfc52ee1a4949ceeb7dbc8a33f2d6c088194d9f922fb8318faf1f01186"
 dependencies = [
  "either",
 ]
 
 [[package]]
 name = "itertools"
-version = "0.13.0"
+version = "0.14.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "413ee7dfc52ee1a4949ceeb7dbc8a33f2d6c088194d9f922fb8318faf1f01186"
+checksum = "2b192c782037fadd9cfa75548310488aabdbf3d2da73885b31bd0abd03351285"
 dependencies = [
  "either",
 ]
@@ -1387,7 +1419,7 @@ checksum = "50b7e5b27aa02a74bac8c3f23f448f8d87ff11f92d3aac1a6ed369ee08cc56c1"
 dependencies = [
  "libc",
  "wasi",
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -1472,7 +1504,7 @@ version = "0.50.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7957b9740744892f114936ab4a57b3f487491bbeafaf8083688b16841a4240e5"
 dependencies = [
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -1508,12 +1540,6 @@ dependencies = [
  "autocfg",
 ]
 
-[[package]]
-name = "number_prefix"
-version = "0.4.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "830b246a0e5f20af87141b25c173cd1b609bd7779a4617d6ec582abaf90870f3"
-
 [[package]]
 name = "object"
 version = "0.37.3"
@@ -1777,6 +1803,12 @@ dependencies = [
  "proc-macro2",
 ]
 
+[[package]]
+name = "r-efi"
+version = "5.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "69cdb34c158ceb288df11e18b4bd39de994f6657d83847bdffdbd7f346754b0f"
+
 [[package]]
 name = "r-efi"
 version = "6.0.0"
@@ -1785,20 +1817,19 @@ checksum = "f8dcc9c7d52a811697d2151c701e0d08956f92b0e24136cf4cf27b57a6a0d9bf"
 
 [[package]]
 name = "rand"
-version = "0.8.6"
+version = "0.9.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5ca0ecfa931c29007047d1bc58e623ab12e5590e8c7cc53200d5202b69266d8a"
+checksum = "44c5af06bb1b7d3216d91932aed5265164bf384dc89cd6ba05cf59a35f5f76ea"
 dependencies = [
- "libc",
  "rand_chacha",
  "rand_core",
 ]
 
 [[package]]
 name = "rand_chacha"
-version = "0.3.1"
+version = "0.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e6c10a63a0fa32252be49d21e7709d4d4baf8d231c2dbce1eaa8141b9b127d88"
+checksum = "d3022b5f1df60f26e1ffddd6c66e8aa15de382ae63b3a0c1bfc0e4d3e3f325cb"
 dependencies = [
  "ppv-lite86",
  "rand_core",
@@ -1806,11 +1837,11 @@ dependencies = [
 
 [[package]]
 name = "rand_core"
-version = "0.6.4"
+version = "0.9.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ec0be4795e2f6a28069bec0b5ff3e2ac9bafc99e6a9a7dc3547996c5c816922c"
+checksum = "76afc826de14238e6e8c374ddcc1fa19e374fd8dd986b0d2af0d02377261d83c"
 dependencies = [
- "getrandom 0.2.17",
+ "getrandom 0.3.4",
 ]
 
 [[package]]
@@ -1825,12 +1856,12 @@ dependencies = [
 
 [[package]]
 name = "rayon-cond"
-version = "0.3.0"
+version = "0.4.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "059f538b55efd2309c9794130bc149c6a553db90e9d99c2030785c82f0bd7df9"
+checksum = "2964d0cf57a3e7a06e8183d14a8b527195c706b7983549cd5462d5aa3747438f"
 dependencies = [
  "either",
- "itertools 0.11.0",
+ "itertools 0.14.0",
  "rayon",
 ]
 
@@ -1901,7 +1932,7 @@ dependencies = [
  "serde_json",
  "symphonia",
  "tempfile",
- "thiserror 2.0.18",
+ "thiserror",
  "tokenizers",
  "tracing",
 ]
@@ -1949,7 +1980,7 @@ dependencies = [
  "serde",
  "serde_json",
  "tempfile",
- "thiserror 2.0.18",
+ "thiserror",
  "toml",
  "tracing",
 ]
@@ -1960,7 +1991,7 @@ version = "0.2.0"
 dependencies = [
  "rmlx-core",
  "rmlx-mlx",
- "thiserror 2.0.18",
+ "thiserror",
  "tracing",
 ]
 
@@ -1976,7 +2007,7 @@ dependencies = [
  "safetensors",
  "serde_json",
  "tempfile",
- "thiserror 2.0.18",
+ "thiserror",
  "tracing",
 ]
 
@@ -1992,7 +2023,7 @@ dependencies = [
  "serde",
  "serde_json",
  "tempfile",
- "thiserror 2.0.18",
+ "thiserror",
  "tracing",
 ]
 
@@ -2008,7 +2039,7 @@ dependencies = [
  "serde_json",
  "sha2",
  "tempfile",
- "thiserror 2.0.18",
+ "thiserror",
  "time",
  "toml",
  "tracing",
@@ -2023,7 +2054,7 @@ dependencies = [
  "rmlx-core",
  "rmlx-loader",
  "safetensors",
- "thiserror 2.0.18",
+ "thiserror",
  "tracing",
 ]
 
@@ -2048,7 +2079,7 @@ dependencies = [
  "serde_json",
  "symphonia",
  "tempfile",
- "thiserror 2.0.18",
+ "thiserror",
  "tokenizers",
  "tracing",
  "twox-hash",
@@ -2061,7 +2092,7 @@ dependencies = [
  "criterion",
  "rmlx-core",
  "rmlx-kv-quant",
- "thiserror 2.0.18",
+ "thiserror",
  "tracing",
 ]
 
@@ -2098,7 +2129,7 @@ dependencies = [
  "serde",
  "serde_json",
  "tempfile",
- "thiserror 2.0.18",
+ "thiserror",
  "tokenizers",
  "tokio",
  "tower",
@@ -2163,7 +2194,7 @@ dependencies = [
  "errno",
  "libc",
  "linux-raw-sys",
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -2213,7 +2244,7 @@ version = "0.1.29"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "91c1b7e4904c873ef0710c1f407dde2e6287de2bebc1bbbf7d430bb7cbffd939"
 dependencies = [
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -2388,7 +2419,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3a766e1110788c36f4fa1c2b71b387a7815aa65f88ce0229841826633d93723e"
 dependencies = [
  "libc",
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -2409,6 +2440,12 @@ dependencies = [
  "unicode-segmentation",
 ]
 
+[[package]]
+name = "static_assertions"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a2eb9349b6444b326872e140eb1cf5e7c522154d69e7a0ffb0fb81c06b37543f"
+
 [[package]]
 name = "strength_reduce"
 version = "0.2.4"
@@ -2528,16 +2565,7 @@ dependencies = [
  "getrandom 0.4.2",
  "once_cell",
  "rustix",
- "windows-sys 0.61.2",
-]
-
-[[package]]
-name = "thiserror"
-version = "1.0.69"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b6aaf5339b578ea85b50e080feb250a3e8ae8cfcdff9a461c9ec2904bc923f52"
-dependencies = [
- "thiserror-impl 1.0.69",
+ "windows-sys",
 ]
 
 [[package]]
@@ -2546,18 +2574,7 @@ version = "2.0.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4288b5bcbc7920c07a1149a35cf9590a2aa808e0bc1eafaade0b80947865fbc4"
 dependencies = [
- "thiserror-impl 2.0.18",
-]
-
-[[package]]
-name = "thiserror-impl"
-version = "1.0.69"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4fee6c4efc90059e10f81e6d42c60a18f76588c3d74cb83a0b242a2b6c7504c1"
-dependencies = [
- "proc-macro2",
- "quote",
- "syn",
+ "thiserror-impl",
 ]
 
 [[package]]
@@ -2648,17 +2665,19 @@ dependencies = [
 
 [[package]]
 name = "tokenizers"
-version = "0.20.4"
+version = "0.23.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3b08cc37428a476fc9e20ac850132a513a2e1ce32b6a31addf2b74fa7033b905"
+checksum = "44e5bea67576e04b6ff8564c5d9e09c2ef0cf476502245f2f120e497769d3112"
 dependencies = [
- "aho-corasick",
+ "ahash",
+ "compact_str",
+ "daachorse",
+ "dary_heap",
  "derive_builder",
  "esaxx-rs",
- "getrandom 0.2.17",
+ "getrandom 0.3.4",
  "indicatif",
- "itertools 0.12.1",
- "lazy_static",
+ "itertools 0.14.0",
  "log",
  "macro_rules_attribute",
  "monostate",
@@ -2672,7 +2691,7 @@ dependencies = [
  "serde",
  "serde_json",
  "spm_precompiled",
- "thiserror 1.0.69",
+ "thiserror",
  "unicode-normalization-alignments",
  "unicode-segmentation",
  "unicode_categories",
@@ -2691,7 +2710,7 @@ dependencies = [
  "signal-hook-registry",
  "socket2",
  "tokio-macros",
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -2792,7 +2811,7 @@ checksum = "050686193eb999b4bb3bc2acfa891a13da00f79734704c4b8b4ef1a10b368a3c"
 dependencies = [
  "crossbeam-channel",
  "symlink",
- "thiserror 2.0.18",
+ "thiserror",
  "time",
  "tracing-subscriber",
 ]
@@ -2921,6 +2940,12 @@ version = "0.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "39ec24b3121d976906ece63c9daad25b85969647682eee313cb5779fdd69e14e"
 
+[[package]]
+name = "unit-prefix"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "81e544489bf3d8ef66c953931f56617f423cd4b5494be343d9b9d3dda037b9a3"
+
 [[package]]
 name = "ureq"
 version = "3.3.0"
@@ -3155,7 +3180,7 @@ version = "0.1.11"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c2a7b1c03c876122aa43f3020e6c3c3ee5c05081c9a00739faf7503aeba10d22"
 dependencies = [
- "windows-sys 0.61.2",
+ "windows-sys",
 ]
 
 [[package]]
@@ -3223,15 +3248,6 @@ dependencies = [
  "windows-link",
 ]
 
-[[package]]
-name = "windows-sys"
-version = "0.59.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1e38bc4d79ed67fd075bcc251a1c39b32a1776bbe92e5bef1f0bf1f8c531853b"
-dependencies = [
- "windows-targets",
-]
-
 [[package]]
 name = "windows-sys"
 version = "0.61.2"
@@ -3241,70 +3257,6 @@ dependencies = [
  "windows-link",
 ]
 
-[[package]]
-name = "windows-targets"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9b724f72796e036ab90c1021d4780d4d3d648aca59e491e6b98e725b84e99973"
-dependencies = [
- "windows_aarch64_gnullvm",
- "windows_aarch64_msvc",
- "windows_i686_gnu",
- "windows_i686_gnullvm",
- "windows_i686_msvc",
- "windows_x86_64_gnu",
- "windows_x86_64_gnullvm",
- "windows_x86_64_msvc",
-]
-
-[[package]]
-name = "windows_aarch64_gnullvm"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "32a4622180e7a0ec044bb555404c800bc9fd9ec262ec147edd5989ccd0c02cd3"
-
-[[package]]
-name = "windows_aarch64_msvc"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "09ec2a7bb152e2252b53fa7803150007879548bc709c039df7627cabbd05d469"
-
-[[package]]
-name = "windows_i686_gnu"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8e9b5ad5ab802e97eb8e295ac6720e509ee4c243f69d781394014ebfe8bbfa0b"
-
-[[package]]
-name = "windows_i686_gnullvm"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0eee52d38c090b3caa76c563b86c3a4bd71ef1a819287c19d586d7334ae8ed66"
-
-[[package]]
-name = "windows_i686_msvc"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "240948bc05c5e7c6dabba28bf89d89ffce3e303022809e73deaefe4f6ec56c66"
-
-[[package]]
-name = "windows_x86_64_gnu"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "147a5c80aabfbf0c7d901cb5895d1de30ef2907eb21fbbab29ca94c5b08b1a78"
-
-[[package]]
-name = "windows_x86_64_gnullvm"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "24d5b23dc417412679681396f2b49f3de8c1473deb516bd34410872eff51ed0d"
-
-[[package]]
-name = "windows_x86_64_msvc"
-version = "0.52.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec"
-
 [[package]]
 name = "winnow"
 version = "1.0.3"
diff --git a/Cargo.toml b/Cargo.toml
index f8fe03c..7d5a688 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -82,7 +82,7 @@ tower = { version = "0.5", default-features = false }
 
 # Templates / tokenizer
 minijinja  = { version = "2", features = ["json", "preserve_order"] }
-tokenizers = "0.20"
+tokenizers = "0.23"
 
 # MLX bindings — disabled until tensor compute is needed (S1.4 Gemma4 graph).
 # See FINDINGS.md "MLX bindings" for the build-tooling situation.
diff --git a/crates/rmlx-audio/src/tts.rs b/crates/rmlx-audio/src/tts.rs
index ceb9d51..409387e 100644
--- a/crates/rmlx-audio/src/tts.rs
+++ b/crates/rmlx-audio/src/tts.rs
@@ -1928,7 +1928,13 @@ impl TtsTokenizer {
                                 })
                                 .collect();
                             if !special_tokens.is_empty() {
-                                inner.add_special_tokens(&special_tokens);
+                                // tokenizers 0.21+ takes the tokens by value
+                                // (`impl IntoIterator<Item = AddedToken>`) and
+                                // returns the added count as a `Result`; the
+                                // local Vec is not used afterwards.
+                                inner
+                                    .add_special_tokens(special_tokens)
+                                    .map_err(|e| TtsError::Tokenizer(e.to_string()))?;
                             }
                         }
                     }
diff --git a/crates/rmlx-models/src/decode_loop_tests.rs b/crates/rmlx-models/src/decode_loop_tests.rs
index d8a0a89..9a4b782 100644
--- a/crates/rmlx-models/src/decode_loop_tests.rs
+++ b/crates/rmlx-models/src/decode_loop_tests.rs
@@ -1,12 +1,12 @@
 use super::*;
 
 use std::cell::Cell;
-use std::collections::HashMap;
+use std::fmt::Write as _;
+use std::str::FromStr as _;
 use std::sync::{Mutex, MutexGuard};
 
 use rmlx_core::error::Error;
 use rmlx_kv_quant::KvQuant;
-use tokenizers::models::wordlevel::WordLevel;
 use tokenizers::Tokenizer;
 
 /// Serializes the MLX-touching tests *in this module* against each other. MLX
@@ -55,18 +55,25 @@ fn id_of(a: &Array) -> u32 {
 }
 
 /// A minimal `WordLevel` tokenizer over ids `0..n` mapped to `t<id>`.
+///
+/// Built from a HuggingFace tokenizer JSON string rather than the
+/// `WordLevelBuilder::vocab` API: tokenizers 0.21+ switched the builder's
+/// vocab map to `ahash::AHashMap`, which is not a direct dependency of this
+/// crate. The JSON path is behaviour-identical and dependency-free.
 #[allow(clippy::expect_used)]
 fn tiny_tokenizer(n: u32) -> Tokenizer {
-    let mut vocab: HashMap<String, u32> = HashMap::new();
+    let mut vocab = String::from("{");
     for i in 0..n {
-        vocab.insert(format!("t{i}"), i);
+        if i > 0 {
+            vocab.push(',');
+        }
+        let _ = write!(vocab, "\"t{i}\":{i}");
     }
-    let wl = WordLevel::builder()
-        .vocab(vocab)
-        .unk_token("<unk>".to_string())
-        .build()
-        .expect("build wordlevel");
-    Tokenizer::new(wl)
+    vocab.push('}');
+    let json = format!(
+        r#"{{"version":"1.0","truncation":null,"padding":null,"added_tokens":[],"normalizer":null,"pre_tokenizer":null,"post_processor":null,"decoder":null,"model":{{"type":"WordLevel","vocab":{vocab},"unk_token":"<unk>"}}}}"#
+    );
+    Tokenizer::from_str(&json).expect("build wordlevel tokenizer")
 }
 
 fn greedy_cfg() -> SamplerConfig {