mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-02-25 07:34:10 +00:00
Use fused mul - unary op also for MoE models
This commit is contained in:
@@ -8326,22 +8326,24 @@ static struct ggml_tensor * llm_build_moe_ffn(
|
||||
ggml_tensor * gate = llm_build_lora_mm_id(lctx, ctx, gate_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens]
|
||||
cb(gate, "ffn_moe_gate", il);
|
||||
|
||||
switch (type_op) {
|
||||
case LLM_FFN_SILU:
|
||||
{
|
||||
gate = ggml_silu(ctx, gate);
|
||||
cb(gate, "ffn_moe_silu", il);
|
||||
} break;
|
||||
case LLM_FFN_GELU:
|
||||
{
|
||||
gate = ggml_gelu(ctx, gate);
|
||||
cb(gate, "ffn_moe_gelu", il);
|
||||
} break;
|
||||
default:
|
||||
GGML_ABORT("fatal error");
|
||||
}
|
||||
// This is equivalent to the commented out code below
|
||||
ggml_tensor * par = ggml_fused_mul_unary(ctx, gate, up, type_op == LLM_FFN_SILU ? GGML_UNARY_OP_SILU : GGML_UNARY_OP_GELU);
|
||||
|
||||
ggml_tensor * par = ggml_mul(ctx, up, gate); // [n_ff, n_expert_used, n_tokens]
|
||||
//switch (type_op) {
|
||||
// case LLM_FFN_SILU:
|
||||
// {
|
||||
// gate = ggml_silu(ctx, gate);
|
||||
// cb(gate, "ffn_moe_silu", il);
|
||||
// } break;
|
||||
// case LLM_FFN_GELU:
|
||||
// {
|
||||
// gate = ggml_gelu(ctx, gate);
|
||||
// cb(gate, "ffn_moe_gelu", il);
|
||||
// } break;
|
||||
// default:
|
||||
// GGML_ABORT("fatal error");
|
||||
//}
|
||||
//ggml_tensor * par = ggml_mul(ctx, up, gate); // [n_ff, n_expert_used, n_tokens]
|
||||
cb(par, "ffn_moe_gate_par", il);
|
||||
|
||||
ggml_tensor * experts = llm_build_lora_mm_id(lctx, ctx, down_exps, par, selected_experts); // [n_embd, n_expert_used, n_tokens]
|
||||
|
||||
Reference in New Issue
Block a user