Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Slot-clamp lint: accept the allow marker on any line of a wrapped assignment (#17384)
Daniel Handb0367a328ab
diff --git a/scripts/lint_no_parallel_clamp.py b/scripts/lint_no_parallel_clamp.pyindex a3a3e0e2f05..50353c0e691 100644--- a/scripts/lint_no_parallel_clamp.py+++ b/scripts/lint_no_parallel_clamp.py@@ -25,7 +25,7 @@ Parameter defaults, class-body annotations, `self.<attr>` assignments and`max(1, ...)` / `getattr(..., 1)` floors are structurally distinct and pass.A genuine capability or resource limit is allowed with a trailing-`# allow-slot-clamp: <reason>` comment on the assignment line.+`# allow-slot-clamp: <reason>` comment on the assignment (any line it spans).Exit codes: 0 = clean, 1 = findings, 2 = usage error.Run from repo root: python3 scripts/lint_no_parallel_clamp.py@@ -113,7 +113,10 @@ def _findings_for_tree(what = "pinned to a single slot by an expression"else:continue- if ALLOW_MARKER in lines[node.lineno - 1]:+ # The formatter moves a trailing comment inside the parentheses when it wraps a long+ # assignment, so the marker counts on any line the statement spans.+ span = lines[node.lineno - 1 : (node.end_lineno or node.lineno)]+ if any(ALLOW_MARKER in line for line in span):continuefound.append((filename, node.lineno, what))return found@@ -151,6 +154,11 @@ _SELF_TEST_CASES: tuple[tuple[str, int], ...] = (("def load():\n n_parallel: int = 1 # allow-slot-clamp: ok\n", 0),("def load(x):\n n_parallel: int = x\n", 0),("def load():\n n_parallel = 1 # allow-slot-clamp: no --kv-unified\n", 0),+ (+ "def load():\n n_parallel = (\n 1 # allow-slot-clamp: wrapped by the formatter\n )\n",+ 0,+ ),+ ("def load():\n n_parallel = (\n 1\n )\n", 1),("def load():\n n_parallel = _mtp_clamped_slots\n", 1),("def load(n):\n n_parallel = min(n, 1)\n", 1),("def load(n, mtp):\n n_parallel = 1 if mtp else n\n", 1),diff --git a/tests/test_lint_no_parallel_clamp.py b/tests/test_lint_no_parallel_clamp.pyindex 3e9337aa66f..bceea5a54b3 100644--- a/tests/test_lint_no_parallel_clamp.py+++ b/tests/test_lint_no_parallel_clamp.py@@ -34,10 +34,16 @@ CLAMPS = ("def load():\n n_parallel = _mtp_clamped_slots\n","async def load():\n n_parallel = 1\n","def load():\n if mtp:\n n_parallel = 1\n",+ # Wrapped by the formatter and unmarked: still a clamp.+ "def load():\n n_parallel = (\n 1\n )\n",+ # A marker on the next statement does not excuse this one.+ "def load():\n n_parallel = 1\n x = 2 # allow-slot-clamp: elsewhere\n",)ALLOWED = ("def load():\n n_parallel = 1 # allow-slot-clamp: no --kv-unified\n",+ # The formatter moves the trailing marker inside the parentheses when it wraps the line.+ "def load():\n n_parallel = (\n 1 # allow-slot-clamp: --fit on keeps layers over slots\n )\n","def load():\n n_parallel: int = 1 # allow-slot-clamp: no --kv-unified\n","def load(fit):\n n_parallel = fit.slots\n","def load(n):\n n_parallel = min(n, 1) # allow-slot-clamp: no --kv-unified\n",