diff --git a/src/paddlefleet/transformer/moe/fp8_utils.py b/src/paddlefleet/transformer/moe/fp8_utils.py index e229e40c6..08f60cb1f 100644 --- a/src/paddlefleet/transformer/moe/fp8_utils.py +++ b/src/paddlefleet/transformer/moe/fp8_utils.py @@ -410,7 +410,9 @@ def fwd_gate_up_fp8( """ if self.moe_grouped_gemm: + raise RuntimeError("test ci") self.m_indices = self.gen_m_indices(tokens_per_expert) + raise RuntimeError("test ci") # concat w1, shape is [num_groups, n, k] w1_t_quant, w1_t_scale = fused_stack_quant(expert_w1, transpose=True) w1_t_quant = w1_t_quant.reshape([num_expert, -1, w1_t_quant.shape[-1]])