Fix: model fails to load when chat template uses HuggingFace generation tags (abetlen#2226)

* chat-format: ignore HuggingFace's {% generation %} chat-template tag

HuggingFace's transformers chat-template extension adds {% generation %}
and {% endgeneration %} tags so trainers can mark generation spans for
loss masking. The tags ship in GGUF tokenizer.chat_template metadata
(SmolLM3 et al), but jinja2's default environment doesn't recognize
them, so Llama() raises TemplateSyntaxError at init for any affected
GGUF, even when the caller passes an explicit chat_format override.

Register a minimal Jinja extension that treats both tags as inert
wrappers: the body between them renders as-is, the markers themselves
emit nothing. No behavioral change for templates that don't use the
tags.

Prior art: PR abetlen#2082 attempted the same approach but referenced an
unimported 'nodes' module and didn't consume the body or closing tag.

* fix: simplify generation tag handling

* refactor: rename generation tag extension

---------

Co-authored-by: abetlen <abetlen@gmail.com>

Author: tobocop2

CHANGELOG.md

@@ -8,6 +8,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 - feat: Update llama.cpp to ggml-org/llama.cpp@d749821db
+- fix: model fails to load when chat template uses HuggingFace generation tags by @tobocop2 in #2226
 - docs: add contributing guide by @abetlen in #2229
 - chore: Migrate llama.cpp submodule URL to ggml-org/llama.cpp by @shalinib-ibm in #2034
 - fix: Enable unified KV cache for embedding contexts to preserve full per-sequence context in batch embedding calls by @SanjanaB123 in #2217

llama_cpp/llama_chat_format.py

@@ -24,6 +24,7 @@
 )
 
 import jinja2
+from jinja2.ext import Extension
 from jinja2.sandbox import ImmutableSandboxedEnvironment
 
 import numpy as np
@@ -192,6 +193,15 @@ def __call__(
 
 
 class Jinja2ChatFormatter(ChatFormatter):
+    class IgnoreGenerationTags(Extension):
+        """Pass-through for HuggingFace's ``{% generation %}`` chat-template tag."""
+
+        tags = {"generation"}
+
+        def parse(self, parser: jinja2.parser.Parser):
+            parser.stream.skip(1)
+            return parser.parse_statements(("name:endgeneration",), drop_needle=True)
+
     def __init__(
         self,
         template: str,
@@ -213,6 +223,7 @@ def __init__(
             loader=jinja2.BaseLoader(),
             trim_blocks=True,
             lstrip_blocks=True,
+            extensions=[Jinja2ChatFormatter.IgnoreGenerationTags],
         ).from_string(self.template)
 
     @staticmethod