fix(reasoning): honor Poolside template thinking defaults (#32540)

Co-authored-by: Jiminator <69131491+Jiminator@users.noreply.github.com>
This commit is contained in:
Willow Lopez
2026-07-29 21:02:19 +00:00
committed by GitHub
co-authored by Jiminator
parent d0e69d3881
commit ffd4705baa
2 changed files with 219 additions and 8 deletions
+84 -7
View File
@@ -26,6 +26,7 @@ from typing import Callable, Optional, Tuple
import jinja2
import jinja2.ext
import jinja2.nodes
import jinja2.sandbox
logger = logging.getLogger(__name__)
@@ -71,6 +72,72 @@ class ReasoningToggleConfig:
return self.special_case == "always"
class _GenerationTagExtension(jinja2.ext.Extension):
"""Parse-only support for the ``{% generation %}`` blocks emitted by
transformers chat templates (assistant token masking). Needed so the
toggle-default parser below can build an AST for templates that use it."""
tags = {"generation"}
def parse(self, parser):
lineno = next(parser.stream).lineno
body = parser.parse_statements(("name:endgeneration",), drop_needle=True)
return jinja2.nodes.Scope(body).set_lineno(lineno)
def _has_toggle_default_assignment(
ctx: TemplateDetectionContext, param: str, default: bool
) -> bool:
"""True if the template applies Jinja's ``default`` filter (or its ``d``
alias) to ``param`` with working-toggle semantics and the given default,
e.g. ``{%- set enable_thinking = enable_thinking | default(false) -%}``
(Laguna-XS-2.1) or ``default(true)`` (Laguna-S-2.1).
Matches on the parsed ``Filter`` node, so comments, ``{% raw %}`` blocks,
and string literals cannot shadow a live assignment. Boolean-mode handling:
``default(x, false)`` is equivalent to ``default(x)``; ``default(false,
true)`` still maps False -> False and True -> True, so it is a working
default-off toggle; only ``default(true, true)`` collapses the toggle
(an explicit False is replaced by True) and is not matched.
"""
try:
env = jinja2.Environment(
extensions=[jinja2.ext.loopcontrols, _GenerationTagExtension]
)
tree = env.parse(ctx.template)
except jinja2.TemplateError:
return False
except Exception:
# e.g. RecursionError on a pathologically nested template; detection
# must never take down server startup (mirrors
# detect_inline_system_support).
return False
for node in tree.find_all(jinja2.nodes.Filter):
if node.name not in ("default", "d") or node.dyn_args or node.dyn_kwargs:
continue
if not isinstance(node.node, jinja2.nodes.Name) or node.node.name != param:
continue
args = list(node.args)
kwargs = {kw.key: kw.value for kw in node.kwargs}
if not args or not isinstance(args[0], jinja2.nodes.Const):
continue
default_value = args[0].value
if not isinstance(default_value, bool):
continue
boolean_arg = args[1] if len(args) > 1 else kwargs.get("boolean")
if boolean_arg is not None and not isinstance(boolean_arg, jinja2.nodes.Const):
continue
boolean_mode = bool(boolean_arg.value) if boolean_arg is not None else False
if boolean_mode and default_value:
# default(true, true): any falsy value is replaced by True, so an
# explicit enable_thinking=false still renders thinking on -- the
# assignment is not a working toggle.
continue
if default_value is default:
return True
return False
# ---------------------------------------------------------------------------
# Reasoning mode rules (detect toggle config from template)
# ---------------------------------------------------------------------------
@@ -103,7 +170,8 @@ REASONING_MODE_RULES = (
r"{%\s*if\s+not\s+enable_thinking\s+is\s+defined\s*%}.*?"
r"{%\s*set\s+enable_thinking\s*=\s*(?:false|False)\s*%}",
re.DOTALL,
),
)
or _has_toggle_default_assignment(ctx, "enable_thinking", False),
),
DetectionRule(
name="nemotron_3_super_low_effort",
@@ -134,7 +202,8 @@ REASONING_MODE_RULES = (
or ctx.has_pattern(
r"enable_thinking\s+is\s+not\s+defined\s+or\s+enable_thinking"
)
or ctx.has_pattern(r"namespace\([^)]*enable_thinking\s*=\s*true"),
or ctx.has_pattern(r"namespace\([^)]*enable_thinking\s*=\s*true")
or _has_toggle_default_assignment(ctx, "enable_thinking", True),
),
DetectionRule(
name="explicit_thinking_default_false",
@@ -143,7 +212,8 @@ REASONING_MODE_RULES = (
r"{%\s*if\s+not\s+thinking\s+is\s+defined\s*%}.*?"
r"{%\s*set\s+thinking\s*=\s*(?:false|False)\s*%}",
re.DOTALL,
),
)
or _has_toggle_default_assignment(ctx, "thinking", False),
),
DetectionRule(
name="thinking_default_true",
@@ -160,7 +230,8 @@ REASONING_MODE_RULES = (
r"thinking\s+is\s+defined\s+and\s+(?:thinking\s+is\s+false|not\s+thinking)"
)
or ctx.has_pattern(r"thinking\s+is\s+not\s+defined\s+or\s+thinking")
or ctx.has_pattern(r"namespace\([^)]*thinking\s*=\s*true"),
or ctx.has_pattern(r"namespace\([^)]*thinking\s*=\s*true")
or _has_toggle_default_assignment(ctx, "thinking", True),
),
)
@@ -269,10 +340,16 @@ def _is_hunyuan(ctx):
def _is_poolside_v1(ctx):
has_poolside_tool_format = (
ctx.has_text("unescaped XML-like object")
and ctx.has_text("<tool_call>function-name")
and ctx.has_text("<arg_key>")
ctx.has_text("<arg_key>")
and ctx.has_text("<arg_value>")
and (
# Laguna-XS.2 spells out the tool-call format in prose;
# Laguna-S-2.1 does not, so also key on the tool preamble both
# template families share. The Poolside identity must not depend
# on the enable_thinking default, which differs between families.
ctx.has_text("unescaped XML-like object")
or ctx.has_text("All available function signatures are listed below")
)
)
return has_poolside_tool_format or (
ctx.reasoning_config
@@ -90,6 +90,126 @@ class TestTemplateManagerReasoningDetection(unittest.TestCase):
)
self.assertEqual(parser, "interns1")
def test_poolside_v1_detects_variant_template_defaults(self):
tool_format = """
return an unescaped XML-like object with function name and arguments
within '<tool_call>' and '</tool_call>' tags
<tool_call>function-name
<arg_key>argument-key</arg_key>
<arg_value>value-of-argument-key</arg_value>
</tool_call>
"""
for default, expected in (("false", False), ("true", True)):
with self.subTest(default=default):
template = (
"{%- set enable_thinking = enable_thinking | default("
f"{default}) -%}}\n{tool_format}"
)
_, config, parser = self._detect(template, ["<tool_call>"])
self.assertEqual(
config,
ReasoningToggleConfig(
toggle_param="enable_thinking", default_enabled=expected
),
)
self.assertEqual(parser, "poolside_v1")
def test_poolside_v1_laguna_s21_shaped_template(self):
# Laguna-S-2.1's real template defaults enable_thinking on and lacks
# the XS-style prose describing the tool-call format; only the tool
# preamble and the <arg_key>/<arg_value> markup identify the family.
template = (
"{%- set enable_thinking = enable_thinking | default(true) -%}\n"
"You may call functions to assist with the user query.\n"
"All available function signatures are listed below:\n"
"{{- '<tool_call>' + function_data.name -}}\n"
'{{- "<arg_key>" ~ k ~ "</arg_key>" -}}'
'{{- "<arg_value>" ~ v ~ "</arg_value>" -}}\n'
"{{- '</tool_call>' -}}"
)
_, config, parser = self._detect(template, ["<tool_call>"])
self.assertEqual(
config,
ReasoningToggleConfig(toggle_param="enable_thinking", default_enabled=True),
)
self.assertEqual(parser, "poolside_v1")
def test_enable_thinking_default_filter_variants(self):
cases = [
# Jinja's documented alias for the default filter.
("{%- set enable_thinking = enable_thinking | d(true) -%}", True),
# No whitespace around the pipe.
("{% set enable_thinking = enable_thinking|default(false) %}", False),
# An explicit boolean=false second argument is equivalent to the
# one-argument form.
(
"{%- set enable_thinking = enable_thinking"
" | default(true, false) -%}",
True,
),
# Boolean mode with a false default still maps False -> False and
# True -> True, so it is a working default-off toggle.
(
"{%- set enable_thinking = enable_thinking"
" | default(false, true) -%}",
False,
),
(
"{%- set enable_thinking = enable_thinking"
" | default(false, boolean=true) -%}",
False,
),
# The filter also counts outside a set-assignment (gemma-4 style).
("{%- if enable_thinking | default(false) -%}x{%- endif -%}", False),
# A commented-out assignment must not shadow the live one.
(
"{# {%- set enable_thinking = enable_thinking | default(false) -%} #}\n"
"{%- set enable_thinking = enable_thinking | default(true) -%}",
True,
),
# Neither must a raw block or a string literal.
(
"{% raw %}{% set enable_thinking = enable_thinking"
" | default(false) %}{% endraw %}\n"
"{%- set enable_thinking = enable_thinking | default(true) -%}",
True,
),
]
for template, expected in cases:
with self.subTest(template=template):
_, config, _ = self._detect(template, [])
self.assertEqual(
config,
ReasoningToggleConfig(
toggle_param="enable_thinking", default_enabled=expected
),
)
def test_pathological_template_does_not_raise(self):
# A template jinja cannot parse (RecursionError from deep nesting) must
# degrade to no detection, not crash template loading.
template = (
"{% if a %}" * 5000 + "x" + "{% endif %}" * 5000 + "\n"
"{%- set enable_thinking = enable_thinking | default(true) -%}"
)
_, config = detect_reasoning_pattern(template)
self.assertIsNone(config)
def test_enable_thinking_collapsing_default_not_a_toggle(self):
# default(true, true) / default(true, boolean=true) replace any falsy
# value with True, so an explicit enable_thinking=false still renders
# thinking on; the assignment is not a working toggle and must stay
# undetected.
for template in (
"{%- set enable_thinking = enable_thinking | default(true, true) -%}",
"{%- set enable_thinking = enable_thinking"
" | default(true, boolean=true) -%}",
):
with self.subTest(template=template):
_, config, _ = self._detect(template, [])
self.assertIsNone(config)
def test_nemotron_detects_uppercase_true_assignment(self):
template = """
{% set enable_thinking = enable_thinking if enable_thinking is defined else True %}
@@ -225,7 +345,7 @@ class TestTemplateDetectionRuleMatrix(unittest.TestCase):
"</tool_call>",
["<tool_call>"],
"poolside_v1",
None,
"enable_thinking",
),
(
"lfm2_not_deepseek_r1_from_history_cleanup",
@@ -379,6 +499,20 @@ class TestToolCallParserDetection(unittest.TestCase):
tcp = detect_tool_call_parser(template, tok, config, force)
return rp, tcp
def test_poolside_s21_shape_resolves_poolside_tool_call_parser(self):
# Regression: with default(true) the S-2.1 shape must not fall through
# to the qwen tool-call parser, which cannot read Poolside's XML-KV
# tool format.
template = (
"{%- set enable_thinking = enable_thinking | default(true) -%}\n"
"All available function signatures are listed below:\n"
"<tool_call>{{ name }}<arg_key>{{ k }}</arg_key>"
"<arg_value>{{ v }}</arg_value></tool_call>"
)
force, config = detect_reasoning_pattern(template)
result = detect_tool_call_parser(template, _DummyTokenizer([]), config, force)
self.assertEqual(result, "poolside_v1")
def test_qwen3_detects_qwen_tool_call_parser(self):
rp, tcp = self._detect_all("Qwen/Qwen3-0.6B")
self.assertEqual(rp, "qwen3")