[pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
This commit is contained in:
parent
cf2b2e593e
commit
69b437fa21
2 changed files with 12 additions and 5 deletions
|
|
@ -357,7 +357,11 @@ _FP16_ACCUM_DENY: frozenset[str] = frozenset()
|
|||
|
||||
|
||||
def _enable_fp16_accumulation(
|
||||
family: Any, logger: Any, *, dtype: Any = None, speed_mode: Optional[str] = None
|
||||
family: Any,
|
||||
logger: Any,
|
||||
*,
|
||||
dtype: Any = None,
|
||||
speed_mode: Optional[str] = None,
|
||||
) -> bool:
|
||||
"""Turn on fp16-accumulated fp16 GEMMs for consumer GPUs, where they run ~2x the
|
||||
fp32-accumulate rate (datacenter HBM parts are not throughput-nerfed, so they keep
|
||||
|
|
|
|||
|
|
@ -357,7 +357,12 @@ def test_apply_tolerates_missing_optims(monkeypatch):
|
|||
# ── fp16 accumulation (consumer fp16-GEMM fast path) ──────────────────────────
|
||||
|
||||
|
||||
def _stub_torch_fp16_accum(monkeypatch, *, consumer = True, with_flag = True):
|
||||
def _stub_torch_fp16_accum(
|
||||
monkeypatch,
|
||||
*,
|
||||
consumer = True,
|
||||
with_flag = True,
|
||||
):
|
||||
torch = types.ModuleType("torch")
|
||||
torch.bfloat16 = "bfloat16"
|
||||
torch.channels_last = "channels_last"
|
||||
|
|
@ -427,9 +432,7 @@ def test_fp16_accum_respects_family_deny_list(monkeypatch):
|
|||
_stub_gguf_accel(monkeypatch)
|
||||
monkeypatch.setattr(ds_mod, "_FP16_ACCUM_DENY", frozenset({"fragile-family"}))
|
||||
fam = types.SimpleNamespace(supports_torch_compile = True, name = "fragile-family")
|
||||
applied = apply_speed_optims(
|
||||
_Pipe(), _target(), is_gguf = True, family = fam, speed_mode = "default"
|
||||
)
|
||||
applied = apply_speed_optims(_Pipe(), _target(), is_gguf = True, family = fam, speed_mode = "default")
|
||||
assert applied["fp16_accum"] is False
|
||||
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue