[CI] update typos config for CI pre-commit and fix some spells (#20919)
Signed-off-by: Peter Pan <Peter.Pan@daocloud.io>
This commit is contained in:
@@ -416,7 +416,7 @@ class RankTensors:
|
||||
# We dequant and use that as hidden_states so the tests are stable.
|
||||
# quantizing and dequantizing yield slightly different results
|
||||
# depending on the hardware. Here we, quantize and dequantize
|
||||
# first - so further quantize and dequantize will yeild the same
|
||||
# first - so further quantize and dequantize will yield the same
|
||||
# values.
|
||||
if config.is_per_tensor_act_quant:
|
||||
a_q, a_scales = ops.scaled_fp8_quant(
|
||||
|
||||
@@ -95,7 +95,7 @@ def run_single_case(m, n, k, topk, num_experts, block_size):
|
||||
topk_weights, topk_ids = torch.topk(router_logits, k=topk, dim=-1)
|
||||
topk_weights = torch.nn.functional.softmax(topk_weights, dim=-1)
|
||||
|
||||
# triton referrence
|
||||
# triton reference
|
||||
out_triton = fused_experts(
|
||||
hidden_states=tokens_bf16,
|
||||
w1=w1,
|
||||
|
||||
@@ -43,7 +43,7 @@ def test_can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch):
|
||||
text_config = hf_config.get_text_config()
|
||||
|
||||
# Ensure at least 2 expert per group
|
||||
# Since `grouped_topk` assums top-2
|
||||
# Since `grouped_topk` assumes top-2
|
||||
n_group = getattr(text_config, 'n_group', None)
|
||||
num_experts = n_group * 2 if n_group is not None else 2
|
||||
|
||||
|
||||
@@ -17,7 +17,7 @@ MODEL_NAME = "ibm-research/PowerMoE-3b"
|
||||
|
||||
# Number of data parallel ranks for external LB testing
|
||||
DP_SIZE = int(os.getenv("DP_SIZE", "2"))
|
||||
# Default tensor parallell size to use
|
||||
# Default tensor parallel size to use
|
||||
TP_SIZE = int(os.getenv("TP_SIZE", "1"))
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user