mirror of
https://github.com/unslothai/unsloth.git
synced 2026-08-16 12:24:06 +00:00
e99651ba2f
_llama_runtime_fields() returns one entry per field on _InferenceRuntimeFields and get_status splats it. #8007 added chat_template_override to that model while get_status was already passing the same keyword explicitly, so the call raised TypeError: got multiple values for keyword argument 'chat_template_override'. Every status poll with a GGUF loaded returned 500. The chat UI showed 'Failed to get status' plus 'Failed to refresh models', and the model picker rendered empty, which is what took studio-windows-ui-smoke red on main. Assign into the dict before splatting. Pinning the duplicate to None would not help: two values for one keyword is a TypeError whatever the values are. Adds a source-level contract so any future field added to _InferenceRuntimeFields cannot collide with an explicit keyword at a splat site.
91 lines
3.9 KiB
Python
91 lines
3.9 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-only
|
|
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
|
|
"""A runtime field must not be passed twice into an inference response (#8007).
|
|
|
|
`_llama_runtime_fields()` returns one entry per field on `_InferenceRuntimeFields`, and
|
|
callers splat it. Passing any of those same names as an explicit keyword in the same call
|
|
is `TypeError: got multiple values for keyword argument`, whatever the two values are, so
|
|
pinning the duplicate to None does not help -- the key has to be absent from the dict.
|
|
|
|
#8007 added `chat_template_override` to `_InferenceRuntimeFields` while `get_status` was
|
|
already passing it explicitly. Every `/api/inference/status` poll with a GGUF loaded then
|
|
raised, the chat UI showed "Failed to get status", and the model picker rendered empty.
|
|
|
|
Checked at source level: importing the routes module pulls in the whole studio stack, and
|
|
this is a call-shape property that AST can see directly.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import ast
|
|
from pathlib import Path
|
|
|
|
|
|
REPO_ROOT = Path(__file__).resolve().parents[2]
|
|
MODELS_PATH = REPO_ROOT / "studio" / "backend" / "models" / "inference.py"
|
|
ROUTE_PATH = REPO_ROOT / "studio" / "backend" / "routes" / "inference.py"
|
|
|
|
SPLAT_HELPER = "_llama_runtime_fields"
|
|
RUNTIME_MODEL = "_InferenceRuntimeFields"
|
|
|
|
|
|
def _runtime_field_names() -> set[str]:
|
|
"""Field names declared on `_InferenceRuntimeFields`, read without importing it."""
|
|
tree = ast.parse(MODELS_PATH.read_text(encoding = "utf-8"))
|
|
for node in ast.walk(tree):
|
|
if isinstance(node, ast.ClassDef) and node.name == RUNTIME_MODEL:
|
|
# Direct BaseModel subclass, so its own AnnAssign nodes are the whole field set.
|
|
bases = {ast.unparse(b) for b in node.bases}
|
|
assert bases == {"BaseModel"}, (
|
|
f"{RUNTIME_MODEL} now inherits from {sorted(bases)}; this test reads only its "
|
|
"own annotated fields and would miss inherited ones"
|
|
)
|
|
return {
|
|
stmt.target.id
|
|
for stmt in node.body
|
|
if isinstance(stmt, ast.AnnAssign) and isinstance(stmt.target, ast.Name)
|
|
}
|
|
raise AssertionError(f"{RUNTIME_MODEL} not found in {MODELS_PATH}")
|
|
|
|
|
|
def _splat_call_sites() -> list[tuple[int, set[str]]]:
|
|
"""(line, explicit keyword names) for every call that splats the runtime-fields helper."""
|
|
tree = ast.parse(ROUTE_PATH.read_text(encoding = "utf-8"))
|
|
sites = []
|
|
for node in ast.walk(tree):
|
|
if not isinstance(node, ast.Call):
|
|
continue
|
|
splats = any(
|
|
kw.arg is None
|
|
and isinstance(kw.value, ast.Call)
|
|
and isinstance(kw.value.func, ast.Name)
|
|
and kw.value.func.id == SPLAT_HELPER
|
|
for kw in node.keywords
|
|
)
|
|
if splats:
|
|
sites.append((node.lineno, {kw.arg for kw in node.keywords if kw.arg is not None}))
|
|
return sites
|
|
|
|
|
|
def test_runtime_fields_are_not_also_passed_explicitly():
|
|
"""No call may splat the runtime fields and name one of them as a keyword too."""
|
|
runtime_fields = _runtime_field_names()
|
|
sites = _splat_call_sites()
|
|
assert sites, f"no `**{SPLAT_HELPER}(...)` call sites found; did the helper get renamed?"
|
|
|
|
collisions = {
|
|
line: sorted(explicit & runtime_fields)
|
|
for line, explicit in sites
|
|
if explicit & runtime_fields
|
|
}
|
|
assert not collisions, (
|
|
f"{ROUTE_PATH.name} splats **{SPLAT_HELPER}(...) and passes the same field(s) "
|
|
f"explicitly, which raises TypeError at call time: "
|
|
+ "; ".join(f"line {line}: {names}" for line, names in sorted(collisions.items()))
|
|
+ f". Assign into the dict before splatting instead of adding a second keyword."
|
|
)
|
|
|
|
|
|
def test_chat_template_override_is_a_runtime_field():
|
|
"""Guards the premise: if this moves off the model, the regression above changes shape."""
|
|
assert "chat_template_override" in _runtime_field_names()
|