diff --git a/Dockerfile b/Dockerfile
index 1846dea..68fc16b 100755
--- a/Dockerfile
+++ b/Dockerfile
@@ -30,6 +30,15 @@ ENV VIRTUAL_ENV=/app/.venv \
 # Copy the cleaned virtual environment
 COPY --from=builder ${VIRTUAL_ENV} ${VIRTUAL_ENV}
 
+# Bake the pre-trained GPT-2 weights into the image so gpt2_infer needs no
+# network access at runtime. HF_HOME points any runtime fallback download at
+# /tmp, the only writable path on Lambda.
+ENV GPT2_MODEL_DIR=/app/hf_models/gpt2 \
+    HF_HOME=/tmp/hf
+COPY scripts/download_gpt2.py /tmp/download_gpt2.py
+RUN HF_HOME=/tmp/hf_build python /tmp/download_gpt2.py "${GPT2_MODEL_DIR}" && \
+    rm -rf /tmp/hf_build /tmp/download_gpt2.py
+
 # Copy evaluation function first (smaller, changes more often)
 COPY evaluation_function ./evaluation_function
 
diff --git a/docs/dev.md b/docs/dev.md
index 7dba2ca..507c258 100644
--- a/docs/dev.md
+++ b/docs/dev.md
@@ -1,11 +1,21 @@
 # langModels
 An evaluation function to aid the learning of language models.
 
-**NOTE: **  All AWS deployments should allocate 1024MB RAM and Storage due to the image size.
+**NOTE:** AWS deployments need at least 2048MB RAM, and storage to match the ~3.4GB image. GPT-2 (`gpt2_infer`) peaks near 1GB in the Python process alone, so 1024MB is no longer enough.
 
 ## Inputs
 *Specific input parameters which can be supplied when the `eval` command is supplied to this function.*
 
+The parameters for each model are documented for teachers in `user.md`.
+
+## Models: build and runtime notes
+
+### `gpt2_infer`
+- `scripts/download_gpt2.py` downloads the weights into the image during `docker build`. They go to `GPT2_MODEL_DIR` (`/app/hf_models/gpt2`), so nothing is fetched at runtime.
+- Locally, the weights download on first use into `HF_HOME` (default `/tmp/hf`).
+- The test that does a real download and generation is skipped unless `RUN_GPT2_TESTS=1` is set, so CI doesn't download about 500MB.
+- Memory: see the note above (2048MB minimum).
+
 ## Outputs
 *Output schema/values for this function*
 
diff --git a/docs/user.md b/docs/user.md
index 624f71b..305bd6f 100644
--- a/docs/user.md
+++ b/docs/user.md
@@ -2,4 +2,16 @@
 
 A series of language models that students can run for inference to explore their performance.
 
-Created from scratch with just a neural network for modeling data from a sine wave. This is to ensure the software all works in integration tests. Following that, more models will be added (Shannon ngrams (letters); Shannon ngrams (words); Bengio neural network model of language with a small context window; basic transformer models; larger transformer models; external LLMs).
\ No newline at end of file
+Created from scratch with just a neural network for modeling data from a sine wave. This is to ensure the software all works in integration tests. Following that, more models will be added (Shannon ngrams (letters); Shannon ngrams (words); Bengio neural network model of language with a small context window; basic transformer models; larger transformer models; external LLMs).
+
+## Models
+
+### `gpt2_infer`
+Pre-trained GPT-2 (124M) from Hugging Face. The student's response is used as the prompt, and GPT-2's sampled continuation is returned as feedback.
+
+| Param | Default | Notes |
+|---|---|---|
+| `word_count` | 20 | New tokens to generate (1–100) |
+| `temperature` | 1.0 | Sampling temperature (0.01–5.0): higher values give more varied text |
+| `top_k` | 50 | Sample only from the k most likely next tokens (1–50257) |
+| `seed` | none | Integer for reproducible output |
diff --git a/evaluation_function/dev.json b/evaluation_function/dev.json
index e2fa18e..a8a60aa 100644
--- a/evaluation_function/dev.json
+++ b/evaluation_function/dev.json
@@ -50,5 +50,13 @@
     "word_count": 12,
     "context_window": 3,
     "dev": true
+  },
+  "gpt2_infer": {
+    "response": "the general sweetness",
+    "answer": 0,
+    "model": "gpt2_infer",
+    "word_count": 20,
+    "temperature": 1.0,
+    "top_k": 50
   }
 }
diff --git a/evaluation_function/evaluation_test.py b/evaluation_function/evaluation_test.py
index 9636787..348f278 100755
--- a/evaluation_function/evaluation_test.py
+++ b/evaluation_function/evaluation_test.py
@@ -1,4 +1,6 @@
+import os
 import unittest
+from unittest import mock
 
 from .evaluation import Params, evaluation_function
 
@@ -35,4 +37,29 @@ def test_evaluation(self):
         if isinstance(feedback, list) and feedback:
             self.assertIsInstance(feedback[0][1], str)
         else:
-            self.assertIsInstance(feedback, str)
\ No newline at end of file
+            self.assertIsInstance(feedback, str)
+
+
+class TestGPT2(unittest.TestCase):
+    """GPT-2 tests. The happy path needs the weights (baked into the image, or a
+    ~500MB Hugging Face download), so it only runs when RUN_GPT2_TESTS is set."""
+
+    @unittest.skipUnless(os.environ.get("RUN_GPT2_TESTS"), "set RUN_GPT2_TESTS=1 to run")
+    def test_generates_continuation(self):
+        prompt = "Once upon a time"
+        result = evaluation_function(
+            prompt, None, {"model": "gpt2_infer", "word_count": 5, "seed": 0}
+        ).to_dict()
+        self.assertTrue(result.get("is_correct"))
+        self.assertTrue(result["feedback"].startswith(prompt))
+        self.assertGreater(len(result["feedback"]), len(prompt))
+
+    def test_load_failure_returns_error_result(self):
+        from .models import gpt2_infer
+
+        with mock.patch.object(gpt2_infer, "_load", side_effect=OSError("no weights")):
+            result = evaluation_function(
+                "hello", None, {"model": "gpt2_infer"}
+            ).to_dict()
+        self.assertFalse(result.get("is_correct"))
+        self.assertIn("no weights", str(result.get("feedback")))
\ No newline at end of file
diff --git a/evaluation_function/models/__init__.py b/evaluation_function/models/__init__.py
index dd1c7c7..454cc2b 100644
--- a/evaluation_function/models/__init__.py
+++ b/evaluation_function/models/__init__.py
@@ -11,6 +11,7 @@
     "shannon_letters_ngram",
     "shannon_words_ngram",
     "bengio_infer",
+    "gpt2_infer",
 ]
 
 __all__ = ["AVAILABLE_MODELS"]
\ No newline at end of file
diff --git a/evaluation_function/models/gpt2_infer.py b/evaluation_function/models/gpt2_infer.py
new file mode 100644
index 0000000..483d40f
--- /dev/null
+++ b/evaluation_function/models/gpt2_infer.py
@@ -0,0 +1,130 @@
+"""
+GPT-2 (124M) inference using the pre-trained Hugging Face ``gpt2`` checkpoint.
+
+The next rung on the ladder after the Bengio neural n-gram model: a full
+transformer language model. The student's response is used as the prompt and
+the model's sampled continuation is returned as feedback.
+
+Weights are baked into the Docker image at build time (see ``Dockerfile``) and
+located via the ``GPT2_MODEL_DIR`` env var. When that directory is missing
+(e.g. local development) the model is downloaded from Hugging Face on first use
+into ``HF_HOME`` (default ``/tmp/hf``, the only writable path on Lambda).
+
+Params:
+- ``word_count``: number of new tokens to generate (default 20, max 100)
+- ``temperature``: sampling temperature (default 1.0)
+- ``top_k``: sample only from the k most likely tokens (default 50)
+- ``seed``: optional integer for reproducible output
+"""
+import os
+import traceback
+from pathlib import Path
+
+from lf_toolkit.evaluation import Result, Params
+
+from evaluation_function.lazy_load import LazyModule
+
+torch = LazyModule("torch")
+transformers = LazyModule("transformers")
+
+DEFAULT_MODEL_ID = "gpt2"
+DEFAULT_PROMPT = "the general"
+DEFAULT_TOKENS = 20
+MAX_TOKENS = 100
+
+# Cached across requests so a warm worker loads the weights only once.
+_MODEL = None
+_TOKENIZER = None
+_DEVICE = None
+
+
+def _load():
+    global _MODEL, _TOKENIZER, _DEVICE
+    if _MODEL is not None:
+        return _MODEL, _TOKENIZER, _DEVICE
+
+    model_dir = os.environ.get("GPT2_MODEL_DIR")
+    if model_dir and Path(model_dir).is_dir():
+        print(f"### Loading GPT-2 from {model_dir} ###")
+        source, kwargs = model_dir, {"local_files_only": True}
+    else:
+        source = os.environ.get("GPT2_MODEL_ID", DEFAULT_MODEL_ID)
+        cache_dir = os.environ.get("HF_HOME", "/tmp/hf")
+        print(f"### Downloading GPT-2 '{source}' into {cache_dir} ###")
+        kwargs = {"cache_dir": cache_dir}
+
+    tokenizer = transformers.AutoTokenizer.from_pretrained(source, **kwargs)
+    model = transformers.AutoModelForCausalLM.from_pretrained(source, **kwargs)
+
+    device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
+    model.to(device).eval()
+
+    _MODEL, _TOKENIZER, _DEVICE = model, tokenizer, device
+    return _MODEL, _TOKENIZER, _DEVICE
+
+
+def _int_param(params, name, default, lo, hi):
+    try:
+        value = int(params.get(name, default))
+    except (TypeError, ValueError):
+        value = default
+    return max(lo, min(hi, value))
+
+
+def _float_param(params, name, default, lo, hi):
+    try:
+        value = float(params.get(name, default))
+    except (TypeError, ValueError):
+        value = default
+    return max(lo, min(hi, value))
+
+
+def complete(prompt, max_new_tokens=DEFAULT_TOKENS, temperature=1.0, top_k=50, seed=None):
+    model, tokenizer, device = _load()
+    if seed is not None:
+        torch.manual_seed(int(seed))
+
+    inputs = tokenizer(prompt, return_tensors="pt").to(device)
+    with torch.no_grad():
+        output = model.generate(
+            **inputs,
+            max_new_tokens=max_new_tokens,
+            do_sample=True,
+            temperature=temperature,
+            top_k=top_k,
+            pad_token_id=tokenizer.eos_token_id,
+        )
+    return tokenizer.decode(output[0], skip_special_tokens=True)
+
+
+def run(response, answer, params: Params) -> Result:
+    # Guard the whole load+infer path so a failed download or corrupt weights
+    # degrade this one request rather than taking down the RPC worker.
+    try:
+        return _run_inference(response, params)
+    except Exception as e:
+        print("### gpt2_infer failed ###")
+        traceback.print_exc()
+        return Result(
+            is_correct=False,
+            feedback_items=[
+                ("general", "Could not run the GPT-2 language model."),
+                ("error", str(e)),
+            ],
+        )
+
+
+def _run_inference(response, params: Params) -> Result:
+    prompt = response.strip() if isinstance(response, str) and response.strip() else DEFAULT_PROMPT
+
+    text = complete(
+        prompt,
+        max_new_tokens=_int_param(params, "word_count", DEFAULT_TOKENS, 1, MAX_TOKENS),
+        temperature=_float_param(params, "temperature", 1.0, 0.01, 5.0),
+        top_k=_int_param(params, "top_k", 50, 1, 50257),
+        seed=params.get("seed"),
+    )
+
+    result = Result(True)
+    result.add_feedback("general", text)
+    return result
diff --git a/poetry.lock b/poetry.lock
index 8fd9e1e..f1af8a7 100644
--- a/poetry.lock
+++ b/poetry.lock
@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 2.4.1 and should not be changed by hand.
+# This file is automatically @generated by Poetry 2.4.3 and should not be changed by hand.
 
 [[package]]
 name = "annotated-types"
@@ -419,7 +419,7 @@ files = [
     {file = "colorama-0.4.6-py2.py3-none-any.whl", hash = "sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6"},
     {file = "colorama-0.4.6.tar.gz", hash = "sha256:08695f5cb7ed6e0531a20572697297273c47b8cae5a63ffc6d6ed5c201be6e44"},
 ]
-markers = {main = "os_name == \"nt\" or sys_platform == \"win32\"", dev = "sys_platform == \"win32\" or platform_system == \"Windows\""}
+markers = {main = "os_name == \"nt\" or sys_platform == \"win32\" or platform_system == \"Windows\"", dev = "sys_platform == \"win32\" or platform_system == \"Windows\""}
 
 [[package]]
 name = "contourpy"
@@ -911,6 +911,37 @@ files = [
     {file = "h11-0.16.0.tar.gz", hash = "sha256:4e35b956cf45792e4caa5885e69fba00bdbc6ffafbfa020300e549b208ee5ff1"},
 ]
 
+[[package]]
+name = "hf-xet"
+version = "1.8.0"
+description = "Fast transfer of large files with the Hugging Face Hub."
+optional = false
+python-versions = ">=3.8"
+groups = ["main"]
+markers = "platform_machine == \"x86_64\" or platform_machine == \"amd64\" or platform_machine == \"arm64\" or platform_machine == \"aarch64\""
+files = [
+    {file = "hf_xet-1.8.0-cp314-cp314t-macosx_10_12_x86_64.whl", hash = "sha256:c74e3751cbcdae9222c5170e8d788e83201a03dc5e5f4308a20f17a5b6023b56"},
+    {file = "hf_xet-1.8.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:56735c2d4f416d7473fc986586b26a8756e067832ca4b7fdaad32a0d6f2ce740"},
+    {file = "hf_xet-1.8.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:3ecfe2934448cca55a4c42732b60f985266aecb241905945219b3d388aceb7ba"},
+    {file = "hf_xet-1.8.0-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:7edeb21c5d381903ee5f4857fc746fa7b52a4b374035e8811ecfb67ee27add0c"},
+    {file = "hf_xet-1.8.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:07a06d7481702616726c2d8578cfb00052a5b3f21beed6a674592bc837a10330"},
+    {file = "hf_xet-1.8.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:f42940869572dc2b05dfe06f4cc398c6291503e086ddc4eddfb7a6ab9409ad01"},
+    {file = "hf_xet-1.8.0-cp314-cp314t-win_amd64.whl", hash = "sha256:23c262e9f5e137b18dadbce60becfeea84156b0d43119c320bffe260d32cd2fa"},
+    {file = "hf_xet-1.8.0-cp314-cp314t-win_arm64.whl", hash = "sha256:62b056aedb89ac809f78940baf465fe553b70c8b25ee56890fbab3039fffec1b"},
+    {file = "hf_xet-1.8.0-cp38-abi3-macosx_10_12_x86_64.whl", hash = "sha256:e00d8f60170cd3dd55d98306e41dbaf28b27ff1bcfefbfa84279d197221be83e"},
+    {file = "hf_xet-1.8.0-cp38-abi3-macosx_11_0_arm64.whl", hash = "sha256:3e8934784425781f0f779b91044df441f3a7af4dab47375d7e3aeaecbb8eecb4"},
+    {file = "hf_xet-1.8.0-cp38-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:29dab7c904cd1914a0439c6e1306c12c7497c8bd32290fbe563ebf51ca4b6d6d"},
+    {file = "hf_xet-1.8.0-cp38-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:8800e3444770f2c630b361bc6d5c6ebb770b7830646cf632fbb79d0e5bbc9e7f"},
+    {file = "hf_xet-1.8.0-cp38-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:a584a7b3268da62f23c3187cc7201ca01f7b610985a36b1b6119c66bccb0383c"},
+    {file = "hf_xet-1.8.0-cp38-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:1ebaf248c9bef8bb0c9d1d8d71db1fdf0cac14f922cc7a936a81adbf12eb5f01"},
+    {file = "hf_xet-1.8.0-cp38-abi3-win_amd64.whl", hash = "sha256:92188598dedc1e1f5609bd718d4db8687a6c89915553912571e2f587892eecec"},
+    {file = "hf_xet-1.8.0-cp38-abi3-win_arm64.whl", hash = "sha256:42b66fdc688eaa4ff3eb77dc9275e5a0f1258ecc9fdd083ddaf43bb49bec925b"},
+    {file = "hf_xet-1.8.0.tar.gz", hash = "sha256:10ed23b51dacf7917d64f335c25048deb470efa755e9989b218fc0531386138e"},
+]
+
+[package.extras]
+tests = ["pytest"]
+
 [[package]]
 name = "httpcore"
 version = "1.0.9"
@@ -958,6 +989,45 @@ http2 = ["h2 (>=3,<5)"]
 socks = ["socksio (==1.*)"]
 zstd = ["zstandard (>=0.18.0)"]
 
+[[package]]
+name = "huggingface-hub"
+version = "0.36.2"
+description = "Client library to download and publish models, datasets and other repos on the huggingface.co hub"
+optional = false
+python-versions = ">=3.8.0"
+groups = ["main"]
+files = [
+    {file = "huggingface_hub-0.36.2-py3-none-any.whl", hash = "sha256:48f0c8eac16145dfce371e9d2d7772854a4f591bcb56c9cf548accf531d54270"},
+    {file = "huggingface_hub-0.36.2.tar.gz", hash = "sha256:1934304d2fb224f8afa3b87007d58501acfda9215b334eed53072dd5e815ff7a"},
+]
+
+[package.dependencies]
+filelock = "*"
+fsspec = ">=2023.5.0"
+hf-xet = {version = ">=1.1.3,<2.0.0", markers = "platform_machine == \"x86_64\" or platform_machine == \"amd64\" or platform_machine == \"arm64\" or platform_machine == \"aarch64\""}
+packaging = ">=20.9"
+pyyaml = ">=5.1"
+requests = "*"
+tqdm = ">=4.42.1"
+typing-extensions = ">=3.7.4.3"
+
+[package.extras]
+all = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "authlib (>=1.3.2)", "fastapi", "fastapi", "gradio (>=4.0.0)", "httpx", "itsdangerous", "jedi", "libcst (>=1.4.0)", "mypy (==1.15.0) ; python_version >= \"3.9\"", "mypy (>=1.14.1,<1.15.0) ; python_version == \"3.8\"", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures (<16.0)", "pytest-vcr", "pytest-xdist", "ruff (>=0.9.0)", "soundfile", "ty", "types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)", "urllib3 (<2.0)"]
+cli = ["InquirerPy (==0.3.4)"]
+dev = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "authlib (>=1.3.2)", "fastapi", "fastapi", "gradio (>=4.0.0)", "httpx", "itsdangerous", "jedi", "libcst (>=1.4.0)", "mypy (==1.15.0) ; python_version >= \"3.9\"", "mypy (>=1.14.1,<1.15.0) ; python_version == \"3.8\"", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures (<16.0)", "pytest-vcr", "pytest-xdist", "ruff (>=0.9.0)", "soundfile", "ty", "types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)", "urllib3 (<2.0)"]
+fastai = ["fastai (>=2.4)", "fastcore (>=1.3.27)", "toml"]
+hf-transfer = ["hf_transfer (>=0.1.4)"]
+hf-xet = ["hf-xet (>=1.1.2,<2.0.0)"]
+inference = ["aiohttp"]
+mcp = ["aiohttp", "mcp (>=1.8.0)", "typer"]
+oauth = ["authlib (>=1.3.2)", "fastapi", "httpx", "itsdangerous"]
+quality = ["libcst (>=1.4.0)", "mypy (==1.15.0) ; python_version >= \"3.9\"", "mypy (>=1.14.1,<1.15.0) ; python_version == \"3.8\"", "ruff (>=0.9.0)", "ty"]
+tensorflow = ["graphviz", "pydot", "tensorflow"]
+tensorflow-testing = ["keras (<3.0)", "tensorflow"]
+testing = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "authlib (>=1.3.2)", "fastapi", "fastapi", "gradio (>=4.0.0)", "httpx", "itsdangerous", "jedi", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures (<16.0)", "pytest-vcr", "pytest-xdist", "soundfile", "urllib3 (<2.0)"]
+torch = ["safetensors[torch]", "torch"]
+typing = ["types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)"]
+
 [[package]]
 name = "idna"
 version = "3.11"
@@ -2577,6 +2647,89 @@ files = [
     {file = "pywin32_ctypes-0.2.3-py3-none-any.whl", hash = "sha256:8a1513379d709975552d202d942d9837758905c8d01eb82b8bcc30918929e7b8"},
 ]
 
+[[package]]
+name = "pyyaml"
+version = "6.0.3"
+description = "YAML parser and emitter for Python"
+optional = false
+python-versions = ">=3.8"
+groups = ["main"]
+files = [
+    {file = "PyYAML-6.0.3-cp38-cp38-macosx_10_13_x86_64.whl", hash = "sha256:c2514fceb77bc5e7a2f7adfaa1feb2fb311607c9cb518dbc378688ec73d8292f"},
+    {file = "PyYAML-6.0.3-cp38-cp38-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:9c57bb8c96f6d1808c030b1687b9b5fb476abaa47f0db9c0101f5e9f394e97f4"},
+    {file = "PyYAML-6.0.3-cp38-cp38-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:efd7b85f94a6f21e4932043973a7ba2613b059c4a000551892ac9f1d11f5baf3"},
+    {file = "PyYAML-6.0.3-cp38-cp38-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:22ba7cfcad58ef3ecddc7ed1db3409af68d023b7f940da23c6c2a1890976eda6"},
+    {file = "PyYAML-6.0.3-cp38-cp38-musllinux_1_2_x86_64.whl", hash = "sha256:6344df0d5755a2c9a276d4473ae6b90647e216ab4757f8426893b5dd2ac3f369"},
+    {file = "PyYAML-6.0.3-cp38-cp38-win32.whl", hash = "sha256:3ff07ec89bae51176c0549bc4c63aa6202991da2d9a6129d7aef7f1407d3f295"},
+    {file = "PyYAML-6.0.3-cp38-cp38-win_amd64.whl", hash = "sha256:5cf4e27da7e3fbed4d6c3d8e797387aaad68102272f8f9752883bc32d61cb87b"},
+    {file = "pyyaml-6.0.3-cp310-cp310-macosx_10_13_x86_64.whl", hash = "sha256:214ed4befebe12df36bcc8bc2b64b396ca31be9304b8f59e25c11cf94a4c033b"},
+    {file = "pyyaml-6.0.3-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:02ea2dfa234451bbb8772601d7b8e426c2bfa197136796224e50e35a78777956"},
+    {file = "pyyaml-6.0.3-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:b30236e45cf30d2b8e7b3e85881719e98507abed1011bf463a8fa23e9c3e98a8"},
+    {file = "pyyaml-6.0.3-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:66291b10affd76d76f54fad28e22e51719ef9ba22b29e1d7d03d6777a9174198"},
+    {file = "pyyaml-6.0.3-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9c7708761fccb9397fe64bbc0395abcae8c4bf7b0eac081e12b809bf47700d0b"},
+    {file = "pyyaml-6.0.3-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:418cf3f2111bc80e0933b2cd8cd04f286338bb88bdc7bc8e6dd775ebde60b5e0"},
+    {file = "pyyaml-6.0.3-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:5e0b74767e5f8c593e8c9b5912019159ed0533c70051e9cce3e8b6aa699fcd69"},
+    {file = "pyyaml-6.0.3-cp310-cp310-win32.whl", hash = "sha256:28c8d926f98f432f88adc23edf2e6d4921ac26fb084b028c733d01868d19007e"},
+    {file = "pyyaml-6.0.3-cp310-cp310-win_amd64.whl", hash = "sha256:bdb2c67c6c1390b63c6ff89f210c8fd09d9a1217a465701eac7316313c915e4c"},
+    {file = "pyyaml-6.0.3-cp311-cp311-macosx_10_13_x86_64.whl", hash = "sha256:44edc647873928551a01e7a563d7452ccdebee747728c1080d881d68af7b997e"},
+    {file = "pyyaml-6.0.3-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:652cb6edd41e718550aad172851962662ff2681490a8a711af6a4d288dd96824"},
+    {file = "pyyaml-6.0.3-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:10892704fc220243f5305762e276552a0395f7beb4dbf9b14ec8fd43b57f126c"},
+    {file = "pyyaml-6.0.3-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:850774a7879607d3a6f50d36d04f00ee69e7fc816450e5f7e58d7f17f1ae5c00"},
+    {file = "pyyaml-6.0.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:b8bb0864c5a28024fac8a632c443c87c5aa6f215c0b126c449ae1a150412f31d"},
+    {file = "pyyaml-6.0.3-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:1d37d57ad971609cf3c53ba6a7e365e40660e3be0e5175fa9f2365a379d6095a"},
+    {file = "pyyaml-6.0.3-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:37503bfbfc9d2c40b344d06b2199cf0e96e97957ab1c1b546fd4f87e53e5d3e4"},
+    {file = "pyyaml-6.0.3-cp311-cp311-win32.whl", hash = "sha256:8098f252adfa6c80ab48096053f512f2321f0b998f98150cea9bd23d83e1467b"},
+    {file = "pyyaml-6.0.3-cp311-cp311-win_amd64.whl", hash = "sha256:9f3bfb4965eb874431221a3ff3fdcddc7e74e3b07799e0e84ca4a0f867d449bf"},
+    {file = "pyyaml-6.0.3-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:7f047e29dcae44602496db43be01ad42fc6f1cc0d8cd6c83d342306c32270196"},
+    {file = "pyyaml-6.0.3-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:fc09d0aa354569bc501d4e787133afc08552722d3ab34836a80547331bb5d4a0"},
+    {file = "pyyaml-6.0.3-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:9149cad251584d5fb4981be1ecde53a1ca46c891a79788c0df828d2f166bda28"},
+    {file = "pyyaml-6.0.3-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:5fdec68f91a0c6739b380c83b951e2c72ac0197ace422360e6d5a959d8d97b2c"},
+    {file = "pyyaml-6.0.3-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ba1cc08a7ccde2d2ec775841541641e4548226580ab850948cbfda66a1befcdc"},
+    {file = "pyyaml-6.0.3-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:8dc52c23056b9ddd46818a57b78404882310fb473d63f17b07d5c40421e47f8e"},
+    {file = "pyyaml-6.0.3-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:41715c910c881bc081f1e8872880d3c650acf13dfa8214bad49ed4cede7c34ea"},
+    {file = "pyyaml-6.0.3-cp312-cp312-win32.whl", hash = "sha256:96b533f0e99f6579b3d4d4995707cf36df9100d67e0c8303a0c55b27b5f99bc5"},
+    {file = "pyyaml-6.0.3-cp312-cp312-win_amd64.whl", hash = "sha256:5fcd34e47f6e0b794d17de1b4ff496c00986e1c83f7ab2fb8fcfe9616ff7477b"},
+    {file = "pyyaml-6.0.3-cp312-cp312-win_arm64.whl", hash = "sha256:64386e5e707d03a7e172c0701abfb7e10f0fb753ee1d773128192742712a98fd"},
+    {file = "pyyaml-6.0.3-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:8da9669d359f02c0b91ccc01cac4a67f16afec0dac22c2ad09f46bee0697eba8"},
+    {file = "pyyaml-6.0.3-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:2283a07e2c21a2aa78d9c4442724ec1eb15f5e42a723b99cb3d822d48f5f7ad1"},
+    {file = "pyyaml-6.0.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ee2922902c45ae8ccada2c5b501ab86c36525b883eff4255313a253a3160861c"},
+    {file = "pyyaml-6.0.3-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:a33284e20b78bd4a18c8c2282d549d10bc8408a2a7ff57653c0cf0b9be0afce5"},
+    {file = "pyyaml-6.0.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0f29edc409a6392443abf94b9cf89ce99889a1dd5376d94316ae5145dfedd5d6"},
+    {file = "pyyaml-6.0.3-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:f7057c9a337546edc7973c0d3ba84ddcdf0daa14533c2065749c9075001090e6"},
+    {file = "pyyaml-6.0.3-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:eda16858a3cab07b80edaf74336ece1f986ba330fdb8ee0d6c0d68fe82bc96be"},
+    {file = "pyyaml-6.0.3-cp313-cp313-win32.whl", hash = "sha256:d0eae10f8159e8fdad514efdc92d74fd8d682c933a6dd088030f3834bc8e6b26"},
+    {file = "pyyaml-6.0.3-cp313-cp313-win_amd64.whl", hash = "sha256:79005a0d97d5ddabfeeea4cf676af11e647e41d81c9a7722a193022accdb6b7c"},
+    {file = "pyyaml-6.0.3-cp313-cp313-win_arm64.whl", hash = "sha256:5498cd1645aa724a7c71c8f378eb29ebe23da2fc0d7a08071d89469bf1d2defb"},
+    {file = "pyyaml-6.0.3-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:8d1fab6bb153a416f9aeb4b8763bc0f22a5586065f86f7664fc23339fc1c1fac"},
+    {file = "pyyaml-6.0.3-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:34d5fcd24b8445fadc33f9cf348c1047101756fd760b4dacb5c3e99755703310"},
+    {file = "pyyaml-6.0.3-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:501a031947e3a9025ed4405a168e6ef5ae3126c59f90ce0cd6f2bfc477be31b7"},
+    {file = "pyyaml-6.0.3-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:b3bc83488de33889877a0f2543ade9f70c67d66d9ebb4ac959502e12de895788"},
+    {file = "pyyaml-6.0.3-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c458b6d084f9b935061bc36216e8a69a7e293a2f1e68bf956dcd9e6cbcd143f5"},
+    {file = "pyyaml-6.0.3-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:7c6610def4f163542a622a73fb39f534f8c101d690126992300bf3207eab9764"},
+    {file = "pyyaml-6.0.3-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:5190d403f121660ce8d1d2c1bb2ef1bd05b5f68533fc5c2ea899bd15f4399b35"},
+    {file = "pyyaml-6.0.3-cp314-cp314-win_amd64.whl", hash = "sha256:4a2e8cebe2ff6ab7d1050ecd59c25d4c8bd7e6f400f5f82b96557ac0abafd0ac"},
+    {file = "pyyaml-6.0.3-cp314-cp314-win_arm64.whl", hash = "sha256:93dda82c9c22deb0a405ea4dc5f2d0cda384168e466364dec6255b293923b2f3"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:02893d100e99e03eda1c8fd5c441d8c60103fd175728e23e431db1b589cf5ab3"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:c1ff362665ae507275af2853520967820d9124984e0f7466736aea23d8611fba"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6adc77889b628398debc7b65c073bcb99c4a0237b248cacaf3fe8a557563ef6c"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:a80cb027f6b349846a3bf6d73b5e95e782175e52f22108cfa17876aaeff93702"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:00c4bdeba853cc34e7dd471f16b4114f4162dc03e6b7afcc2128711f0eca823c"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:66e1674c3ef6f541c35191caae2d429b967b99e02040f5ba928632d9a7f0f065"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:16249ee61e95f858e83976573de0f5b2893b3677ba71c9dd36b9cf8be9ac6d65"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-win_amd64.whl", hash = "sha256:4ad1906908f2f5ae4e5a8ddfce73c320c2a1429ec52eafd27138b7f1cbe341c9"},
+    {file = "pyyaml-6.0.3-cp314-cp314t-win_arm64.whl", hash = "sha256:ebc55a14a21cb14062aa4162f906cd962b28e2e9ea38f9b4391244cd8de4ae0b"},
+    {file = "pyyaml-6.0.3-cp39-cp39-macosx_10_13_x86_64.whl", hash = "sha256:b865addae83924361678b652338317d1bd7e79b1f4596f96b96c77a5a34b34da"},
+    {file = "pyyaml-6.0.3-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:c3355370a2c156cffb25e876646f149d5d68f5e0a3ce86a5084dd0b64a994917"},
+    {file = "pyyaml-6.0.3-cp39-cp39-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3c5677e12444c15717b902a5798264fa7909e41153cdf9ef7ad571b704a63dd9"},
+    {file = "pyyaml-6.0.3-cp39-cp39-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:5ed875a24292240029e4483f9d4a4b8a1ae08843b9c54f43fcc11e404532a8a5"},
+    {file = "pyyaml-6.0.3-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0150219816b6a1fa26fb4699fb7daa9caf09eb1999f3b70fb6e786805e80375a"},
+    {file = "pyyaml-6.0.3-cp39-cp39-musllinux_1_2_aarch64.whl", hash = "sha256:fa160448684b4e94d80416c0fa4aac48967a969efe22931448d853ada8baf926"},
+    {file = "pyyaml-6.0.3-cp39-cp39-musllinux_1_2_x86_64.whl", hash = "sha256:27c0abcb4a5dac13684a37f76e701e054692a9b2d3064b70f5e4eb54810553d7"},
+    {file = "pyyaml-6.0.3-cp39-cp39-win32.whl", hash = "sha256:1ebe39cb5fc479422b83de611d14e2c0d3bb2a18bbcb01f229ab3cfbd8fee7a0"},
+    {file = "pyyaml-6.0.3-cp39-cp39-win_amd64.whl", hash = "sha256:2e71d11abed7344e42a8849600193d15b6def118602c4c176f748e4583246007"},
+    {file = "pyyaml-6.0.3.tar.gz", hash = "sha256:d76623373421df22fb4cf8817020cbb7ef15c725b9d5e45f17e189bfc384190f"},
+]
+
 [[package]]
 name = "rapidfuzz"
 version = "3.14.3"
@@ -2696,7 +2849,7 @@ version = "2025.11.3"
 description = "Alternative regular expression module, to replace re."
 optional = false
 python-versions = ">=3.9"
-groups = ["dev"]
+groups = ["main", "dev"]
 files = [
     {file = "regex-2025.11.3-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:2b441a4ae2c8049106e8b39973bfbddfb25a179dda2bdb99b0eeb60c40a6a3af"},
     {file = "regex-2025.11.3-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:2fa2eed3f76677777345d2f81ee89f5de2f5745910e805f7af7386a920fa7313"},
@@ -2995,6 +3148,48 @@ botocore = ">=1.37.4,<2.0a0"
 [package.extras]
 crt = ["botocore[crt] (>=1.37.4,<2.0a0)"]
 
+[[package]]
+name = "safetensors"
+version = "0.8.0"
+description = ""
+optional = false
+python-versions = ">=3.10"
+groups = ["main"]
+files = [
+    {file = "safetensors-0.8.0-cp310-abi3-macosx_10_12_x86_64.whl", hash = "sha256:c554f85858e05226d3c2828e32395e677434685d6d94594a41643361c5e837f0"},
+    {file = "safetensors-0.8.0-cp310-abi3-macosx_11_0_arm64.whl", hash = "sha256:c80201d22cbf405b80647a60ada77bba06c8fba2da2743ba1e89cdcc39a81f25"},
+    {file = "safetensors-0.8.0-cp310-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7a46e5ff292c356d6991e60942ba7f79817682d3a2cef0702136448cb9c4d235"},
+    {file = "safetensors-0.8.0-cp310-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:4124502b78f03534117c848f87a39b8f31e577b15eff423bf8bfb95f2a8c30d0"},
+    {file = "safetensors-0.8.0-cp310-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:7bc0a787ba8a35be368ee3574edfa2b1ad389eebd0a72e482ae275490e3f6c98"},
+    {file = "safetensors-0.8.0-cp310-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:040070828e36dc8e122178bbbd5830ff9e97920affb84cbe0f46442497bed358"},
+    {file = "safetensors-0.8.0-cp310-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fd6f3f93c9a0a7cc2788ee63fb763353d4bd2e89b0751bc78fcf7dda00bea774"},
+    {file = "safetensors-0.8.0-cp310-abi3-manylinux_2_31_riscv64.whl", hash = "sha256:fcdd41ec4628fee5799f807c73c353629130fbd942aa23d83c623dd6c9d52d78"},
+    {file = "safetensors-0.8.0-cp310-abi3-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:8e9f537aa183a38ace122d27303dcd986b26bd2a7591f9181d7f0c396f4677ca"},
+    {file = "safetensors-0.8.0-cp310-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:87eec7ffed2b809f05a398a8becb7d013f19f7837cd15d9748580d6cf30dbaf4"},
+    {file = "safetensors-0.8.0-cp310-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:4a95ae2b05d7726d751da4ebf626a2ca782b706e101bd894c95bc2450b1cffcc"},
+    {file = "safetensors-0.8.0-cp310-abi3-musllinux_1_2_i686.whl", hash = "sha256:3ae091f16662658bdc019a4ff6cb4c085bb7d725eb5978b183ffd265863b6d2d"},
+    {file = "safetensors-0.8.0-cp310-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:8e080062fcde23be189565e1c3305d16751a218ecf9412c8601e64204eb6f846"},
+    {file = "safetensors-0.8.0-cp310-abi3-win32.whl", hash = "sha256:2ddf52eac562eda224f99acfa7889d02968c1fd59a5b011ae7d8137c37e9c02d"},
+    {file = "safetensors-0.8.0-cp310-abi3-win_amd64.whl", hash = "sha256:096ec1a98435df7beb08853bb5aa9081a84f23d0adc67ed1a0a10550f608373f"},
+    {file = "safetensors-0.8.0-cp310-abi3-win_arm64.whl", hash = "sha256:f7838e5135a406ad3e02efdcb8cf2e5397d368b0154537c4fec682dbc544d452"},
+    {file = "safetensors-0.8.0.tar.gz", hash = "sha256:fabaf3e0f18a6618d9b36560682562157f77c2b71fcffc7b432be2baed9d753d"},
+]
+
+[package.extras]
+all = ["safetensors[convert]", "safetensors[jax]", "safetensors[numpy]", "safetensors[paddlepaddle]", "safetensors[quality]", "safetensors[testing]", "safetensors[torch]"]
+convert = ["huggingface-hub (>=1.4)", "safetensors[torch]"]
+dev = ["safetensors[all]", "safetensors[pinned-tf]"]
+jax = ["flax (>=0.6.3)", "jax (>=0.3.25)", "jaxlib (>=0.3.25)", "safetensors[numpy]"]
+mlx = ["mlx (>=0.0.9)"]
+numpy = ["numpy (>=1.24.6)"]
+paddlepaddle = ["paddlepaddle (>=2.4.1)", "safetensors[numpy]"]
+pinned-tf = ["safetensors[numpy]", "tensorflow (==2.18.0)"]
+quality = ["ruff"]
+tensorflow = ["safetensors[numpy]", "tensorflow (>=2.11.0)"]
+testing = ["fsspec (>=2024.6.0)", "h5py (>=3.7.0)", "hypothesis (>=6.70.2)", "pytest (>=9.0)", "pytest-benchmark (>=5.2)", "s3fs (>=2024.6.0)", "safetensors[numpy]", "setuptools-rust (>=1.12.0)"]
+tf-nightly = ["safetensors[numpy]", "tf-nightly"]
+torch = ["safetensors[numpy]", "torch (>=2.4)"]
+
 [[package]]
 name = "secretstorage"
 version = "3.4.0"
@@ -3167,6 +3362,48 @@ mpmath = ">=1.1.0,<1.4"
 [package.extras]
 dev = ["hypothesis (>=6.70.0)", "pytest (>=7.1.0)"]
 
+[[package]]
+name = "tokenizers"
+version = "0.22.2"
+description = ""
+optional = false
+python-versions = ">=3.9"
+groups = ["main"]
+files = [
+    {file = "tokenizers-0.22.2-cp39-abi3-macosx_10_12_x86_64.whl", hash = "sha256:544dd704ae7238755d790de45ba8da072e9af3eea688f698b137915ae959281c"},
+    {file = "tokenizers-0.22.2-cp39-abi3-macosx_11_0_arm64.whl", hash = "sha256:1e418a55456beedca4621dbab65a318981467a2b188e982a23e117f115ce5001"},
+    {file = "tokenizers-0.22.2-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2249487018adec45d6e3554c71d46eb39fa8ea67156c640f7513eb26f318cec7"},
+    {file = "tokenizers-0.22.2-cp39-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:25b85325d0815e86e0bac263506dd114578953b7b53d7de09a6485e4a160a7dd"},
+    {file = "tokenizers-0.22.2-cp39-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:bfb88f22a209ff7b40a576d5324bf8286b519d7358663db21d6246fb17eea2d5"},
+    {file = "tokenizers-0.22.2-cp39-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:1c774b1276f71e1ef716e5486f21e76333464f47bece56bbd554485982a9e03e"},
+    {file = "tokenizers-0.22.2-cp39-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:df6c4265b289083bf710dff49bc51ef252f9d5be33a45ee2bed151114a56207b"},
+    {file = "tokenizers-0.22.2-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:369cc9fc8cc10cb24143873a0d95438bb8ee257bb80c71989e3ee290e8d72c67"},
+    {file = "tokenizers-0.22.2-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:29c30b83d8dcd061078b05ae0cb94d3c710555fbb44861139f9f83dcca3dc3e4"},
+    {file = "tokenizers-0.22.2-cp39-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:37ae80a28c1d3265bb1f22464c856bd23c02a05bb211e56d0c5301a435be6c1a"},
+    {file = "tokenizers-0.22.2-cp39-abi3-musllinux_1_2_i686.whl", hash = "sha256:791135ee325f2336f498590eb2f11dc5c295232f288e75c99a36c5dbce63088a"},
+    {file = "tokenizers-0.22.2-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:38337540fbbddff8e999d59970f3c6f35a82de10053206a7562f1ea02d046fa5"},
+    {file = "tokenizers-0.22.2-cp39-abi3-win32.whl", hash = "sha256:a6bf3f88c554a2b653af81f3204491c818ae2ac6fbc09e76ef4773351292bc92"},
+    {file = "tokenizers-0.22.2-cp39-abi3-win_amd64.whl", hash = "sha256:c9ea31edff2968b44a88f97d784c2f16dc0729b8b143ed004699ebca91f05c48"},
+    {file = "tokenizers-0.22.2-cp39-abi3-win_arm64.whl", hash = "sha256:9ce725d22864a1e965217204946f830c37876eee3b2ba6fc6255e8e903d5fcbc"},
+    {file = "tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:753d47ebd4542742ef9261d9da92cd545b2cacbb48349a1225466745bb866ec4"},
+    {file = "tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:e10bf9113d209be7cd046d40fbabbaf3278ff6d18eb4da4c500443185dc1896c"},
+    {file = "tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:64d94e84f6660764e64e7e0b22baa72f6cd942279fdbb21d46abd70d179f0195"},
+    {file = "tokenizers-0.22.2-pp310-pypy310_pp73-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:f01a9c019878532f98927d2bacb79bbb404b43d3437455522a00a30718cdedb5"},
+    {file = "tokenizers-0.22.2-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:319f659ee992222f04e58f84cbf407cfa66a65fe3a8de44e8ad2bc53e7d99012"},
+    {file = "tokenizers-0.22.2-pp39-pypy39_pp73-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:1e50f8554d504f617d9e9d6e4c2c2884a12b388a97c5c77f0bc6cf4cd032feee"},
+    {file = "tokenizers-0.22.2-pp39-pypy39_pp73-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:1a62ba2c5faa2dd175aaeed7b15abf18d20266189fb3406c5d0550dd34dd5f37"},
+    {file = "tokenizers-0.22.2-pp39-pypy39_pp73-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:143b999bdc46d10febb15cbffb4207ddd1f410e2c755857b5a0797961bbdc113"},
+    {file = "tokenizers-0.22.2.tar.gz", hash = "sha256:473b83b915e547aa366d1eee11806deaf419e17be16310ac0a14077f1e28f917"},
+]
+
+[package.dependencies]
+huggingface-hub = ">=0.16.4,<2.0"
+
+[package.extras]
+dev = ["tokenizers[testing]"]
+docs = ["setuptools-rust", "sphinx", "sphinx-rtd-theme"]
+testing = ["datasets", "numpy", "pytest", "pytest-asyncio", "requests", "ruff", "ty"]
+
 [[package]]
 name = "tomlkit"
 version = "0.13.3"
@@ -3238,7 +3475,7 @@ version = "4.67.1"
 description = "Fast, Extensible Progress Meter"
 optional = false
 python-versions = ">=3.7"
-groups = ["dev"]
+groups = ["main", "dev"]
 files = [
     {file = "tqdm-4.67.1-py3-none-any.whl", hash = "sha256:26445eca388f82e72884e0d580d5464cd801a3ea01e63e5601bdff9ba6a48de2"},
     {file = "tqdm-4.67.1.tar.gz", hash = "sha256:f8aef9c52c08c13a65f30ea34f4e5aac3fd1a34959879d7e59e63027286627f2"},
@@ -3254,6 +3491,81 @@ notebook = ["ipywidgets (>=6)"]
 slack = ["slack-sdk"]
 telegram = ["requests"]
 
+[[package]]
+name = "transformers"
+version = "4.57.6"
+description = "State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow"
+optional = false
+python-versions = ">=3.9.0"
+groups = ["main"]
+files = [
+    {file = "transformers-4.57.6-py3-none-any.whl", hash = "sha256:4c9e9de11333ddfe5114bc872c9f370509198acf0b87a832a0ab9458e2bd0550"},
+    {file = "transformers-4.57.6.tar.gz", hash = "sha256:55e44126ece9dc0a291521b7e5492b572e6ef2766338a610b9ab5afbb70689d3"},
+]
+
+[package.dependencies]
+filelock = "*"
+huggingface-hub = ">=0.34.0,<1.0"
+numpy = ">=1.17"
+packaging = ">=20.0"
+pyyaml = ">=5.1"
+regex = "!=2019.12.17"
+requests = "*"
+safetensors = ">=0.4.3"
+tokenizers = ">=0.22.0,<=0.23.0"
+tqdm = ">=4.27"
+
+[package.extras]
+accelerate = ["accelerate (>=0.26.0)"]
+all = ["Pillow (>=10.0.1,<=15.0)", "Pillow (>=10.0.1,<=15.0)", "accelerate (>=0.26.0)", "accelerate (>=0.26.0)", "av", "codecarbon (>=2.8.1)", "flax (>=0.4.1,<=0.7.0)", "jax (>=0.4.1,<=0.4.13)", "jaxlib (>=0.4.1,<=0.4.13)", "jinja2 (>=3.1.0)", "kenlm", "keras-nlp (>=0.3.1,<0.14.0)", "kernels (>=0.6.1,<=0.9)", "librosa", "mistral-common[opencv] (>=1.6.3)", "num2words", "onnxconverter-common", "optax (>=0.0.8,<=0.1.4)", "optuna", "phonemizer", "protobuf", "pyctcdecode (>=0.4.0)", "ray[tune] (>=2.7.0)", "scipy (<1.13.0)", "sentencepiece (>=0.1.91,!=0.1.92)", "tensorflow (>2.9,<2.16)", "tensorflow-text (<2.16)", "tf2onnx", "timm (!=1.0.18,<=1.0.19)", "tokenizers (>=0.22.0,<=0.23.0)", "torch (>=2.2)", "torchaudio", "torchvision"]
+audio = ["kenlm", "librosa", "phonemizer", "pyctcdecode (>=0.4.0)"]
+benchmark = ["optimum-benchmark (>=0.3.0)"]
+chat-template = ["jinja2 (>=3.1.0)"]
+codecarbon = ["codecarbon (>=2.8.1)"]
+deepspeed = ["accelerate (>=0.26.0)", "deepspeed (>=0.9.3)"]
+deepspeed-testing = ["GitPython (<3.1.19)", "accelerate (>=0.26.0)", "accelerate (>=0.26.0)", "beautifulsoup4", "cookiecutter (==1.7.3)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "deepspeed (>=0.9.3)", "dill (<0.3.5)", "evaluate (>=0.2.0)", "faiss-cpu", "fastapi", "libcst", "mistral-common[opencv] (>=1.6.3)", "nltk (<=3.8.1)", "openai (>=1.98.0)", "optuna", "parameterized (>=0.9)", "protobuf", "psutil", "pydantic (>=2)", "pydantic (>=2)", "pytest (>=7.2.0)", "pytest-asyncio", "pytest-order", "pytest-rerunfailures (<16.0)", "pytest-rich", "pytest-timeout", "pytest-xdist", "rjieba", "rouge-score (!=0.0.7,!=0.0.8,!=0.1,!=0.1.1)", "ruff (==0.13.1)", "sacrebleu (>=1.4.12,<2.0.0)", "sacremoses", "sentencepiece (>=0.1.91,!=0.1.92)", "sentencepiece (>=0.1.91,!=0.1.92)", "starlette", "tensorboard", "timeout-decorator", "torch (>=2.2)", "uvicorn"]
+dev = ["GitPython (<3.1.19)", "GitPython (<3.1.19)", "Pillow (>=10.0.1,<=15.0)", "Pillow (>=10.0.1,<=15.0)", "accelerate (>=0.26.0)", "accelerate (>=0.26.0)", "accelerate (>=0.26.0)", "av", "beautifulsoup4", "codecarbon (>=2.8.1)", "cookiecutter (==1.7.3)", "cookiecutter (==1.7.3)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "dill (<0.3.5)", "evaluate (>=0.2.0)", "faiss-cpu", "fastapi", "flax (>=0.4.1,<=0.7.0)", "fugashi (>=1.0)", "ipadic (>=1.0.0,<2.0)", "jax (>=0.4.1,<=0.4.13)", "jaxlib (>=0.4.1,<=0.4.13)", "jinja2 (>=3.1.0)", "kenlm", "keras-nlp (>=0.3.1,<0.14.0)", "kernels (>=0.6.1,<=0.9)", "libcst", "libcst", "librosa", "mistral-common[opencv] (>=1.6.3)", "mistral-common[opencv] (>=1.6.3)", "nltk (<=3.8.1)", "num2words", "onnxconverter-common", "openai (>=1.98.0)", "optax (>=0.0.8,<=0.1.4)", "optuna", "pandas (<2.3.0)", "parameterized (>=0.9)", "phonemizer", "protobuf", "psutil", "pyctcdecode (>=0.4.0)", "pydantic (>=2)", "pydantic (>=2)", "pytest (>=7.2.0)", "pytest-asyncio", "pytest-order", "pytest-rerunfailures (<16.0)", "pytest-rich", "pytest-timeout", "pytest-xdist", "ray[tune] (>=2.7.0)", "rhoknp (>=1.1.0,<1.3.1)", "rich", "rjieba", "rouge-score (!=0.0.7,!=0.0.8,!=0.1,!=0.1.1)", "ruff (==0.13.1)", "ruff (==0.13.1)", "sacrebleu (>=1.4.12,<2.0.0)", "sacremoses", "scikit-learn", "scipy (<1.13.0)", "sentencepiece (>=0.1.91,!=0.1.92)", "sentencepiece (>=0.1.91,!=0.1.92)", "starlette", "sudachidict_core (>=20220729)", "sudachipy (>=0.6.6)", "tensorboard", "tensorflow (>2.9,<2.16)", "tensorflow-text (<2.16)", "tf2onnx", "timeout-decorator", "timm (!=1.0.18,<=1.0.19)", "tokenizers (>=0.22.0,<=0.23.0)", "torch (>=2.2)", "torch (>=2.2)", "torchaudio", "torchvision", "unidic (>=1.0.2)", "unidic_lite (>=1.0.7)", "urllib3 (<2.0.0)", "uvicorn"]
+dev-tensorflow = ["GitPython (<3.1.19)", "GitPython (<3.1.19)", "Pillow (>=10.0.1,<=15.0)", "accelerate (>=0.26.0)", "beautifulsoup4", "cookiecutter (==1.7.3)", "cookiecutter (==1.7.3)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "dill (<0.3.5)", "evaluate (>=0.2.0)", "faiss-cpu", "fastapi", "kenlm", "keras-nlp (>=0.3.1,<0.14.0)", "libcst", "libcst", "librosa", "mistral-common[opencv] (>=1.6.3)", "nltk (<=3.8.1)", "onnxconverter-common", "onnxconverter-common", "onnxruntime (>=1.4.0)", "onnxruntime-tools (>=1.4.2)", "openai (>=1.98.0)", "pandas (<2.3.0)", "parameterized (>=0.9)", "phonemizer", "protobuf", "psutil", "pyctcdecode (>=0.4.0)", "pydantic (>=2)", "pydantic (>=2)", "pytest (>=7.2.0)", "pytest-asyncio", "pytest-order", "pytest-rerunfailures (<16.0)", "pytest-rich", "pytest-timeout", "pytest-xdist", "rich", "rjieba", "rouge-score (!=0.0.7,!=0.0.8,!=0.1,!=0.1.1)", "ruff (==0.13.1)", "ruff (==0.13.1)", "sacrebleu (>=1.4.12,<2.0.0)", "sacremoses", "scikit-learn", "sentencepiece (>=0.1.91,!=0.1.92)", "sentencepiece (>=0.1.91,!=0.1.92)", "starlette", "tensorboard", "tensorflow (>2.9,<2.16)", "tensorflow-text (<2.16)", "tf2onnx", "tf2onnx", "timeout-decorator", "tokenizers (>=0.22.0,<=0.23.0)", "torch (>=2.2)", "urllib3 (<2.0.0)", "uvicorn"]
+dev-torch = ["GitPython (<3.1.19)", "GitPython (<3.1.19)", "Pillow (>=10.0.1,<=15.0)", "Pillow (>=10.0.1,<=15.0)", "accelerate (>=0.26.0)", "accelerate (>=0.26.0)", "beautifulsoup4", "codecarbon (>=2.8.1)", "cookiecutter (==1.7.3)", "cookiecutter (==1.7.3)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "dill (<0.3.5)", "evaluate (>=0.2.0)", "faiss-cpu", "fastapi", "fugashi (>=1.0)", "ipadic (>=1.0.0,<2.0)", "kenlm", "kernels (>=0.6.1,<=0.9)", "libcst", "libcst", "librosa", "mistral-common[opencv] (>=1.6.3)", "nltk (<=3.8.1)", "num2words", "onnxruntime (>=1.4.0)", "onnxruntime-tools (>=1.4.2)", "openai (>=1.98.0)", "optuna", "pandas (<2.3.0)", "parameterized (>=0.9)", "phonemizer", "protobuf", "psutil", "pyctcdecode (>=0.4.0)", "pydantic (>=2)", "pydantic (>=2)", "pytest (>=7.2.0)", "pytest-asyncio", "pytest-order", "pytest-rerunfailures (<16.0)", "pytest-rich", "pytest-timeout", "pytest-xdist", "ray[tune] (>=2.7.0)", "rhoknp (>=1.1.0,<1.3.1)", "rich", "rjieba", "rouge-score (!=0.0.7,!=0.0.8,!=0.1,!=0.1.1)", "ruff (==0.13.1)", "ruff (==0.13.1)", "sacrebleu (>=1.4.12,<2.0.0)", "sacremoses", "scikit-learn", "sentencepiece (>=0.1.91,!=0.1.92)", "sentencepiece (>=0.1.91,!=0.1.92)", "starlette", "sudachidict_core (>=20220729)", "sudachipy (>=0.6.6)", "tensorboard", "timeout-decorator", "timm (!=1.0.18,<=1.0.19)", "tokenizers (>=0.22.0,<=0.23.0)", "torch (>=2.2)", "torch (>=2.2)", "torchaudio", "torchvision", "unidic (>=1.0.2)", "unidic_lite (>=1.0.7)", "urllib3 (<2.0.0)", "uvicorn"]
+flax = ["flax (>=0.4.1,<=0.7.0)", "jax (>=0.4.1,<=0.4.13)", "jaxlib (>=0.4.1,<=0.4.13)", "optax (>=0.0.8,<=0.1.4)", "scipy (<1.13.0)"]
+flax-speech = ["kenlm", "librosa", "phonemizer", "pyctcdecode (>=0.4.0)"]
+ftfy = ["ftfy"]
+hf-xet = ["hf_xet"]
+hub-kernels = ["kernels (>=0.6.1,<=0.9)"]
+integrations = ["kernels (>=0.6.1,<=0.9)", "optuna", "ray[tune] (>=2.7.0)"]
+ja = ["fugashi (>=1.0)", "ipadic (>=1.0.0,<2.0)", "rhoknp (>=1.1.0,<1.3.1)", "sudachidict_core (>=20220729)", "sudachipy (>=0.6.6)", "unidic (>=1.0.2)", "unidic_lite (>=1.0.7)"]
+mistral-common = ["mistral-common[opencv] (>=1.6.3)"]
+modelcreation = ["cookiecutter (==1.7.3)"]
+natten = ["natten (>=0.14.6,<0.15.0)"]
+num2words = ["num2words"]
+onnx = ["onnxconverter-common", "onnxruntime (>=1.4.0)", "onnxruntime-tools (>=1.4.2)", "tf2onnx"]
+onnxruntime = ["onnxruntime (>=1.4.0)", "onnxruntime-tools (>=1.4.2)"]
+open-telemetry = ["opentelemetry-api", "opentelemetry-exporter-otlp", "opentelemetry-sdk"]
+optuna = ["optuna"]
+quality = ["GitPython (<3.1.19)", "datasets (>=2.15.0)", "libcst", "pandas (<2.3.0)", "rich", "ruff (==0.13.1)", "urllib3 (<2.0.0)"]
+ray = ["ray[tune] (>=2.7.0)"]
+retrieval = ["datasets (>=2.15.0)", "faiss-cpu"]
+ruff = ["ruff (==0.13.1)"]
+sagemaker = ["sagemaker (>=2.31.0)"]
+sentencepiece = ["protobuf", "sentencepiece (>=0.1.91,!=0.1.92)"]
+serving = ["accelerate (>=0.26.0)", "fastapi", "openai (>=1.98.0)", "pydantic (>=2)", "starlette", "torch (>=2.2)", "uvicorn"]
+sigopt = ["sigopt"]
+sklearn = ["scikit-learn"]
+speech = ["kenlm", "librosa", "phonemizer", "pyctcdecode (>=0.4.0)", "torchaudio"]
+testing = ["GitPython (<3.1.19)", "accelerate (>=0.26.0)", "beautifulsoup4", "cookiecutter (==1.7.3)", "datasets (>=2.15.0)", "datasets (>=2.15.0)", "dill (<0.3.5)", "evaluate (>=0.2.0)", "faiss-cpu", "fastapi", "libcst", "mistral-common[opencv] (>=1.6.3)", "nltk (<=3.8.1)", "openai (>=1.98.0)", "parameterized (>=0.9)", "psutil", "pydantic (>=2)", "pydantic (>=2)", "pytest (>=7.2.0)", "pytest-asyncio", "pytest-order", "pytest-rerunfailures (<16.0)", "pytest-rich", "pytest-timeout", "pytest-xdist", "rjieba", "rouge-score (!=0.0.7,!=0.0.8,!=0.1,!=0.1.1)", "ruff (==0.13.1)", "sacrebleu (>=1.4.12,<2.0.0)", "sacremoses", "sentencepiece (>=0.1.91,!=0.1.92)", "starlette", "tensorboard", "timeout-decorator", "torch (>=2.2)", "uvicorn"]
+tf = ["keras-nlp (>=0.3.1,<0.14.0)", "onnxconverter-common", "tensorflow (>2.9,<2.16)", "tensorflow-text (<2.16)", "tf2onnx"]
+tf-cpu = ["keras (>2.9,<2.16)", "keras-nlp (>=0.3.1,<0.14.0)", "onnxconverter-common", "tensorflow-cpu (>2.9,<2.16)", "tensorflow-probability (<0.24)", "tensorflow-text (<2.16)", "tf2onnx"]
+tf-speech = ["kenlm", "librosa", "phonemizer", "pyctcdecode (>=0.4.0)"]
+tiktoken = ["blobfile", "tiktoken"]
+timm = ["timm (!=1.0.18,<=1.0.19)"]
+tokenizers = ["tokenizers (>=0.22.0,<=0.23.0)"]
+torch = ["accelerate (>=0.26.0)", "torch (>=2.2)"]
+torch-speech = ["kenlm", "librosa", "phonemizer", "pyctcdecode (>=0.4.0)", "torchaudio"]
+torch-vision = ["Pillow (>=10.0.1,<=15.0)", "torchvision"]
+torchhub = ["filelock", "huggingface-hub (>=0.34.0,<1.0)", "importlib_metadata", "numpy (>=1.17)", "packaging (>=20.0)", "protobuf", "regex (!=2019.12.17)", "requests", "sentencepiece (>=0.1.91,!=0.1.92)", "tokenizers (>=0.22.0,<=0.23.0)", "torch (>=2.2)", "tqdm (>=4.27)"]
+video = ["av"]
+vision = ["Pillow (>=10.0.1,<=15.0)"]
+
 [[package]]
 name = "trove-classifiers"
 version = "2025.9.11.17"
@@ -3635,4 +3947,4 @@ cffi = ["cffi (>=1.17,<2.0) ; platform_python_implementation != \"PyPy\" and pyt
 [metadata]
 lock-version = "2.1"
 python-versions = ">=3.11,<3.14"
-content-hash = "2b3f834cb3845ccd3ea44cc1826d0a256dbc8063820113360002af8538b5d645"
+content-hash = "b8c950e065998bb3b74ec876708a7d27af7710d0d3b874862d4803c3d6216f12"
diff --git a/pyproject.toml b/pyproject.toml
index f38aafa..9d685c4 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -29,6 +29,7 @@ lf_toolkit = { git = "https://github.com/lambda-feedback/toolkit-python.git", ta
 numpy = "^2.3.4"
 lmdb = "^1.7.5"
 sentencepiece = "^0.2.1"
+transformers = "^4.57"
 
 [tool.poetry.group.overrides.dependencies]
 # Override sympy because lf-toolkit pins sympy==1.12, 
diff --git a/scripts/download_gpt2.py b/scripts/download_gpt2.py
new file mode 100644
index 0000000..70486ca
--- /dev/null
+++ b/scripts/download_gpt2.py
@@ -0,0 +1,46 @@
+"""Download the pre-trained GPT-2 weights and tokenizer into a local directory.
+
+Used at Docker build time to bake the model into the image so that
+``evaluation_function.models.gpt2_infer`` needs no network access at runtime.
+
+Usage:
+    python scripts/download_gpt2.py [TARGET_DIR] [--model-id gpt2]
+
+TARGET_DIR defaults to $GPT2_MODEL_DIR.
+"""
+import argparse
+import os
+import sys
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    parser.add_argument(
+        "target_dir",
+        nargs="?",
+        default=os.environ.get("GPT2_MODEL_DIR"),
+        help="Directory to save the model into (default: $GPT2_MODEL_DIR)",
+    )
+    parser.add_argument(
+        "--model-id",
+        default=os.environ.get("GPT2_MODEL_ID", "gpt2"),
+        help="Hugging Face model id (default: $GPT2_MODEL_ID or 'gpt2')",
+    )
+    args = parser.parse_args()
+
+    if not args.target_dir:
+        parser.error("no target directory given and GPT2_MODEL_DIR is not set")
+
+    from transformers import AutoModelForCausalLM, AutoTokenizer
+
+    print(f"Downloading '{args.model_id}' into {args.target_dir}")
+    AutoTokenizer.from_pretrained(args.model_id).save_pretrained(args.target_dir)
+    AutoModelForCausalLM.from_pretrained(
+        args.model_id, use_safetensors=True
+    ).save_pretrained(args.target_dir)
+    print("Done")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())