llguidance 1.3.0__tar.gz → 1.4.0__tar.gz

{llguidance-1.3.0 → llguidance-1.4.0}/.github/workflows/rust.yml

@@ -5,6 +5,16 @@ on:
     branches: [ "main" ]
   pull_request:
     branches: [ "main" ]
+  workflow_dispatch:
+    inputs:
+      commit_id:
+        description: 'Branch or Commit ID (optional)'
+        required: false
+        type: string
+  schedule:
+    # * is a special character in YAML so we quote this string
+    # Run at 09:10 UTC every day
+    - cron:  '10 09 * * *'
 
 env:
   CARGO_TERM_COLOR: always
@@ -15,7 +25,10 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
-    - uses: actions/checkout@v4
+    - name: Checkout repo at ${{ github.event_name == 'workflow_dispatch' && inputs.commit_id || github.sha }}
+      uses: actions/checkout@v4
+      with:
+        ref: ${{ github.event_name == 'workflow_dispatch' && inputs.commit_id || github.sha }}
 
     - name: Build parser
       run: cargo build --verbose

{llguidance-1.3.0 → llguidance-1.4.0}/CHANGELOG.md

@@ -4,6 +4,16 @@ All notable changes to this project will be documented in this file. Dates are d
 
 If a release doesn't introduce any interesting changes (build fixes etc.), it's skipped.
 
+#### [v1.4.0](https://github.com/guidance-ai/llguidance/compare/v1.3.0...v1.4.0) 2025-11-18
+
+- Update pyo3 dependency to latest version to support python 3.14t [`#255`](https://github.com/guidance-ai/llguidance/pull/255)
+  - Thank you [@ngoldbaum](https://github.com/ngoldbaum)!
+- Fix tokenization of special tokens for ByteTokenizer [`#249`](https://github.com/guidance-ai/llguidance/pull/249)
+  - Thank you [@daulet](https://github.com/daulet)!
+- Bump minimum python version to 3.10 [`#252`](https://github.com/guidance-ai/llguidance/pull/252)
+- Update MSRV in README [`30d1515`](https://github.com/guidance-ai/llguidance/commit/30d15150e2c6bae9e52dea58c72574c160f4697a)
+  - Thank you for pointing out the error [@nteodosio](https://github.com/nteodosio)!
+
 #### [v1.3.0](https://github.com/guidance-ai/llguidance/compare/v1.2.0...v1.3.0) 2025-10-20
 
 - change semantics of JSON separators (now regex patterns instead of text literals) [`#246`](https://github.com/guidance-ai/llguidance/pull/246)

{llguidance-1.3.0 → llguidance-1.4.0}/Cargo.lock

@@ -1241,7 +1241,7 @@ checksum = "241eaef5fd12c88705a01fc1066c48c4b36e0dd4377dcdc7ec3942cea7a69956"
 
 [[package]]
 name = "llguidance"
-version = "1.3.0"
+version = "1.4.0"
 dependencies = [
  "anyhow",
  "derivre",
@@ -1260,7 +1260,7 @@ dependencies = [
 
 [[package]]
 name = "llguidance_py"
-version = "1.3.0"
+version = "1.4.0"
 dependencies = [
  "anyhow",
  "bytemuck",
@@ -1669,12 +1669,11 @@ dependencies = [
 
 [[package]]
 name = "pyo3"
-version = "0.24.2"
+version = "0.27.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e5203598f366b11a02b13aa20cab591229ff0a89fd121a308a5df751d5fc9219"
+checksum = "37a6df7eab65fc7bee654a421404947e10a0f7085b6951bf2ea395f4659fb0cf"
 dependencies = [
  "anyhow",
- "cfg-if",
  "indoc",
  "libc",
  "memoffset",
@@ -1688,19 +1687,18 @@ dependencies = [
 
 [[package]]
 name = "pyo3-build-config"
-version = "0.24.2"
+version = "0.27.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "99636d423fa2ca130fa5acde3059308006d46f98caac629418e53f7ebb1e9999"
+checksum = "f77d387774f6f6eec64a004eac0ed525aab7fa1966d94b42f743797b3e395afb"
 dependencies = [
- "once_cell",
  "target-lexicon",
 ]
 
 [[package]]
 name = "pyo3-ffi"
-version = "0.24.2"
+version = "0.27.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "78f9cf92ba9c409279bc3305b5409d90db2d2c22392d443a87df3a1adad59e33"
+checksum = "2dd13844a4242793e02df3e2ec093f540d948299a6a77ea9ce7afd8623f542be"
 dependencies = [
  "libc",
  "pyo3-build-config",
@@ -1708,9 +1706,9 @@ dependencies = [
 
 [[package]]
 name = "pyo3-macros"
-version = "0.24.2"
+version = "0.27.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0b999cb1a6ce21f9a6b147dcf1be9ffedf02e0043aec74dc390f3007047cecd9"
+checksum = "eaf8f9f1108270b90d3676b8679586385430e5c0bb78bb5f043f95499c821a71"
 dependencies = [
  "proc-macro2",
  "pyo3-macros-backend",
@@ -1720,9 +1718,9 @@ dependencies = [
 
 [[package]]
 name = "pyo3-macros-backend"
-version = "0.24.2"
+version = "0.27.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "822ece1c7e1012745607d5cf0bcb2874769f0f7cb34c4cde03b9358eb9ef911a"
+checksum = "70a3b2274450ba5288bc9b8c1b69ff569d1d61189d4bff38f8d22e03d17f932b"
 dependencies = [
  "heck",
  "proc-macro2",
@@ -2074,6 +2072,7 @@ dependencies = [
  "llguidance",
  "rstest",
  "serde",
+ "serde-json-fmt",
  "serde_json",
  "toktrie_hf_downloader",
  "toktrie_hf_tokenizers",
@@ -2132,6 +2131,17 @@ dependencies = [
  "serde_derive",
 ]
 
+[[package]]
+name = "serde-json-fmt"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a4a33b7a5f52a26d520099339add40c48baf2e5ada194c8cc1b18cafa2b5e419"
+dependencies = [
+ "serde",
+ "serde_json",
+ "smartstring",
+]
+
 [[package]]
 name = "serde_derive"
 version = "1.0.219"
@@ -2186,6 +2196,17 @@ version = "1.15.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "67b1b7a3b5fe4f1376887184045fcf45c69e92af734b7aaddc05fb777b6fbd03"
 
+[[package]]
+name = "smartstring"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3fb72c633efbaa2dd666986505016c32c3044395ceaf881518399d2f4127ee29"
+dependencies = [
+ "autocfg",
+ "static_assertions",
+ "version_check",
+]
+
 [[package]]
 name = "socket2"
 version = "0.5.10"
@@ -2478,7 +2499,7 @@ dependencies = [
 
 [[package]]
 name = "toktrie"
-version = "1.3.0"
+version = "1.4.0"
 dependencies = [
  "anyhow",
  "bytemuck",
@@ -2489,7 +2510,7 @@ dependencies = [
 
 [[package]]
 name = "toktrie_hf_downloader"
-version = "1.3.0"
+version = "1.4.0"
 dependencies = [
  "anyhow",
  "hf-hub",
@@ -2500,7 +2521,7 @@ dependencies = [
 
 [[package]]
 name = "toktrie_hf_tokenizers"
-version = "1.3.0"
+version = "1.4.0"
 dependencies = [
  "anyhow",
  "log",
@@ -2512,7 +2533,7 @@ dependencies = [
 
 [[package]]
 name = "toktrie_tiktoken"
-version = "1.3.0"
+version = "1.4.0"
 dependencies = [
  "anyhow",
  "log",

{llguidance-1.3.0 → llguidance-1.4.0}/PKG-INFO

@@ -1,11 +1,11 @@
 Metadata-Version: 2.4
 Name: llguidance
-Version: 1.3.0
+Version: 1.4.0
 License-File: LICENSE
 Summary: Bindings for the Low-level Guidance (llguidance) Rust library for use within Guidance
 Author: Michal Moskal
 License-Expression: MIT
-Requires-Python: >=3.9
+Requires-Python: >=3.10
 Description-Content-Type: text/markdown; charset=UTF-8; variant=GFM
 Project-URL: repository, https://github.com/microsoft/llguidance
 Project-URL: issue_tracker, https://github.com/microsoft/llguidance/issues
@@ -111,14 +111,14 @@ Thus, with 16 cores and a 10ms forward pass, llguidance can handle batch sizes u
 
 ## Building
 
-- [install rust](https://www.rust-lang.org/tools/install); 1.75 or later
+- [install rust](https://www.rust-lang.org/tools/install); 1.87 or later
 
 If you just need the C or Rust library (`llguidance`), 
 check the [parser](./parser/README.md) directory.
 
 For Python bindings:
 
-- install python 3.9 or later; very likely you'll need a virtual env/conda
+- install python 3.10 or later; very likely you'll need a virtual env/conda
 - run `./scripts/install-deps.sh`
 - to build and after any changes, run `./scripts/test-guidance.sh`
 

{llguidance-1.3.0 → llguidance-1.4.0}/README.md

@@ -99,14 +99,14 @@ Thus, with 16 cores and a 10ms forward pass, llguidance can handle batch sizes u
 
 ## Building
 
-- [install rust](https://www.rust-lang.org/tools/install); 1.75 or later
+- [install rust](https://www.rust-lang.org/tools/install); 1.87 or later
 
 If you just need the C or Rust library (`llguidance`), 
 check the [parser](./parser/README.md) directory.
 
 For Python bindings:
 
-- install python 3.9 or later; very likely you'll need a virtual env/conda
+- install python 3.10 or later; very likely you'll need a virtual env/conda
 - run `./scripts/install-deps.sh`
 - to build and after any changes, run `./scripts/test-guidance.sh`
 

{llguidance-1.3.0 → llguidance-1.4.0}/parser/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "llguidance"
-version = "1.3.0"
+version = "1.4.0"
 edition = "2021"
 license = "MIT"
 description = "Super-fast Structured Outputs"

{llguidance-1.3.0 → llguidance-1.4.0}/pyproject.toml

@@ -1,8 +1,8 @@
 [project]
 name = "llguidance"
-version = "1.3.0"
+version = "1.4.0"
 description = "Bindings for the Low-level Guidance (llguidance) Rust library for use within Guidance"
-requires-python = ">=3.9"
+requires-python = ">=3.10"
 license = "MIT"
 authors = [{name = "Michal Moskal"}]
 dependencies = []

{llguidance-1.3.0 → llguidance-1.4.0}/python/mypy.ini

@@ -2,4 +2,4 @@
 strict = True
 files = llguidance, torch_tests
 exclude = tmp
-python_version = 3.9
+python_version = 3.10

{llguidance-1.3.0 → llguidance-1.4.0}/python/torch_tests/test_matcher.py

@@ -178,11 +178,11 @@ def test_par_errors() -> None:
     with pytest.raises(RuntimeError, match="Already borrowed"):
         fill_next_token_bitmask_par(exec, [(g0, 0), (g1, 1), (g1, 2)], mask)
 
-    with pytest.raises(TypeError, match="cannot be converted"):
+    with pytest.raises(TypeError, match="cannot be cast"):
         l = [1, (g1, 0), (g1, 1)]
         fill_next_token_bitmask_par(exec, l, mask)  # type: ignore
 
-    with pytest.raises(TypeError, match="cannot be converted"):
+    with pytest.raises(TypeError, match="cannot be cast"):
         l = [(tokenizer(), 0)]
         fill_next_token_bitmask_par(exec, l, mask)  # type: ignore
 

{llguidance-1.3.0 → llguidance-1.4.0}/python_ext/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "llguidance_py"
-version = "1.3.0"
+version = "1.4.0"
 edition = "2021"
 license = "MIT"
 description = "Super-fast Structured Outputs"
@@ -12,7 +12,7 @@ llguidance = { workspace = true }
 toktrie_hf_tokenizers = { workspace = true }
 toktrie_tiktoken = { workspace = true }
 bytemuck = "1.21.0"
-pyo3 = {version = "0.24.1", features = ["extension-module", "abi3-py39", "anyhow"]}
+pyo3 = {version = "0.27.1", features = ["extension-module", "abi3-py39", "anyhow"]}
 serde = { version = "1.0.217", features = ["derive"] }
 serde_json = "1.0.138"
 rayon = "1.10.0"

{llguidance-1.3.0 → llguidance-1.4.0}/python_ext/src/llinterpreter.rs

@@ -150,7 +150,7 @@ impl LLInterpreter {
 
     fn compute_mask(&mut self, py: Python<'_>) -> PyResult<(Option<Cow<'_, [u8]>>, String)> {
         let r = py
-            .allow_threads(|| self.inner.compute_mask())
+            .detach(|| self.inner.compute_mask())
             .map_err(val_error)?
             .clone();
         let mask = if r.is_stop() || r.unconditional_splice().is_some() {

{llguidance-1.3.0 → llguidance-1.4.0}/python_ext/src/llmatcher.rs

@@ -58,7 +58,7 @@ impl LLExecutor {
 
         let mut mut_refs = vec![];
         for ent in interpreters.iter() {
-            let tupl = ent.downcast::<PyTuple>()?;
+            let tupl = ent.cast::<PyTuple>()?;
             if tupl.len() != 2 {
                 return Err(PyValueError::new_err("Expecting (LLMatcher, int) tuple"));
             }
@@ -87,7 +87,7 @@ impl LLExecutor {
 
         use rayon::prelude::*;
 
-        py.allow_threads(|| {
+        py.detach(|| {
             self.pool.install(|| {
                 mut_refs2.into_par_iter().for_each(|(interp, idx)| {
                     interp.unsafe_compute_mask_ptr_inner(
@@ -115,7 +115,7 @@ impl LLExecutor {
 
         let mut mut_refs = vec![];
         for ent in interpreters.iter() {
-            let tupl = ent.downcast::<PyTuple>()?;
+            let tupl = ent.cast::<PyTuple>()?;
             if tupl.len() != 3 {
                 return Err(PyValueError::new_err(
                     "Expecting (LLMatcher, int, List[int]) tuple",
@@ -151,7 +151,7 @@ impl LLExecutor {
 
         use rayon::prelude::*;
 
-        py.allow_threads(|| {
+        py.detach(|| {
             self.pool.install(|| {
                 mut_refs2
                     .into_par_iter()
@@ -247,7 +247,7 @@ fn new_matcher(
     let logger = Logger::new(0, std::cmp::max(0, log_level) as u32);
     // constructing a grammar can take on the order of 100ms
     // for very large grammars, so we drop the GIL here
-    let inner = py.allow_threads(|| {
+    let inner = py.detach(|| {
         let mut r = fact.create_parser_from_init_ext(
             GrammarInit::Serialized(grammar),
             logger,
@@ -306,7 +306,7 @@ impl LLMatcher {
         py: Python<'_>,
     ) -> String {
         match extract_grammar(grammar) {
-            Ok((_, grammar)) => py.allow_threads(|| {
+            Ok((_, grammar)) => py.detach(|| {
                 GrammarInit::Serialized(grammar)
                     .validate(
                         tokenizer.map(|t| t.factory().tok_env().clone()),
@@ -328,7 +328,7 @@ impl LLMatcher {
         py: Python<'_>,
     ) -> (bool, Vec<String>) {
         match extract_grammar(grammar) {
-            Ok((_, grammar)) => py.allow_threads(|| {
+            Ok((_, grammar)) => py.detach(|| {
                 GrammarInit::Serialized(grammar)
                     .validate(
                         tokenizer.map(|t| t.factory().tok_env().clone()),
@@ -428,7 +428,7 @@ impl LLMatcher {
         py: Python<'_>,
     ) -> PyResult<()> {
         self.validate_mask_ptr(trg_ptr, trg_bytes)?;
-        py.allow_threads(|| self.unsafe_compute_mask_ptr_inner(trg_ptr, trg_bytes));
+        py.detach(|| self.unsafe_compute_mask_ptr_inner(trg_ptr, trg_bytes));
         Ok(())
     }
 
@@ -440,14 +440,14 @@ impl LLMatcher {
         py: Python<'_>,
     ) -> PyResult<()> {
         self.validate_mask_ptr(trg_ptr, trg_bytes)?;
-        py.allow_threads(|| {
+        py.detach(|| {
             self.unsafe_compute_mask_ptr_inner_with_draft_tokens(trg_ptr, trg_bytes, draft_tokens)
         });
         Ok(())
     }
 
     fn compute_logit_bias(&mut self, py: Python<'_>) -> Cow<'_, [u8]> {
-        py.allow_threads(|| {
+        py.detach(|| {
             let m = self.compute_mask_or_eos();
             let mut res = vec![0u8; m.len()];
             m.iter_set_entries(|i| res[i] = 200);
@@ -456,7 +456,7 @@ impl LLMatcher {
     }
 
     fn compute_bitmask(&mut self, py: Python<'_>) -> Cow<'_, [u8]> {
-        py.allow_threads(|| {
+        py.detach(|| {
             let m = self.compute_mask_or_eos();
             Cow::Owned(bytemuck::cast_slice(m.as_slice()).to_vec())
         })

{llguidance-1.3.0 → llguidance-1.4.0}/python_ext/src/py.rs

@@ -320,7 +320,7 @@ impl TokenizerEnv for PyTokenizer {
 
     fn tokenize_bytes(&self, utf8bytes: &[u8]) -> Vec<TokenId> {
         self.tok_trie.tokenize_with_greedy_fallback(utf8bytes, |s| {
-            Python::with_gil(|py| {
+            Python::attach(|py| {
                 let r = self.tokenizer_fun.call1(py, (s,)).unwrap();
                 r.extract::<Vec<TokenId>>(py).unwrap()
             })

{llguidance-1.3.0 → llguidance-1.4.0}/python_ext/src/pyjson.rs

@@ -67,7 +67,7 @@ impl Serialize for SerializePyObject<'_> {
         extract!(String);
         extract!(bool);
 
-        if let Ok(x) = self.v.downcast::<PyFloat>() {
+        if let Ok(x) = self.v.cast::<PyFloat>() {
             return x.value().serialize(serializer);
         }
 
@@ -78,7 +78,7 @@ impl Serialize for SerializePyObject<'_> {
             return serializer.serialize_unit();
         }
 
-        if let Ok(x) = self.v.downcast::<PyDict>() {
+        if let Ok(x) = self.v.cast::<PyDict>() {
             let mut map = serializer.serialize_map(Some(x.len()))?;
             for (key, value) in x {
                 if let Ok(key) = key.str() {
@@ -94,7 +94,7 @@ impl Serialize for SerializePyObject<'_> {
             return map.end();
         }
 
-        if let Ok(x) = self.v.downcast::<PyList>() {
+        if let Ok(x) = self.v.cast::<PyList>() {
             let mut seq = serializer.serialize_seq(Some(x.len()))?;
             for element in x {
                 seq.serialize_element(&SerializePyObject { v: element })?
@@ -102,7 +102,7 @@ impl Serialize for SerializePyObject<'_> {
             return seq.end();
         }
 
-        if let Ok(x) = self.v.downcast::<PyTuple>() {
+        if let Ok(x) = self.v.cast::<PyTuple>() {
             let mut seq = serializer.serialize_seq(Some(x.len()))?;
             for element in x {
                 seq.serialize_element(&SerializePyObject { v: element })?

{llguidance-1.3.0 → llguidance-1.4.0}/sample_parser/Cargo.toml

@@ -16,6 +16,7 @@ clap = { version = "4.5.31", features = ["derive"] }
 
 [dev-dependencies]
 rstest = "0.25.0"
+serde-json-fmt = "0.1.0"
 
 [[bin]]
 name = "sample_parser"