fix: QWEN2MOE support for expert_feed_forward_length

previously, expert ff was taken from n_ff (intermediate size) but it is now properly taken from LLM_KV_EXPERT_FEED_FORWARD_LENGTH n_ff_exp and n_ff_shared_exp are now properly calculated
ggerganov · Jun 10, 2024 · aa8a7cd · aa8a7cd
1 parent 284cec4
commit aa8a7cd
Showing 1 changed file with 11 additions and 4 deletions.
diff --git a/llama.cpp b/llama.cpp
@@ -4247,6 +4247,8 @@ static void llm_load_hparams(
             } break;
         case LLM_ARCH_QWEN2MOE:
             {
+                ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp, false);
+
                 ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
                 switch (hparams.n_layer) {
                     case 24: model.type = e_model::MODEL_A2_7B; break;
@@ -5019,6 +5021,10 @@ static void llm_load_print_meta(llama_model_loader & ml, llama_model & model) {
         LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n",   __func__, hparams.expert_weights_scale);
         LLAMA_LOG_INFO("%s: rope_yarn_log_mul    = %.4f\n",   __func__, hparams.rope_yarn_log_mul);
     }
+
+    if (model.arch == LLM_ARCH_QWEN2MOE) {
+        LLAMA_LOG_INFO("%s: n_ff_exp         = %d\n",     __func__, hparams.n_ff_exp);
+    }
 }
 
 // Returns false if cancelled by progress_callback
@@ -5805,16 +5811,17 @@ static bool llm_load_tensors(
                         GGML_ASSERT(hparams.n_expert_used > 0);
 
                         // MoE branch
-                        auto n_ff_exp = n_ff / hparams.n_expert_used;
+                        auto n_ff_exp = hparams.n_ff_exp ? hparams.n_ff_exp : n_ff / hparams.n_expert_used;
                         layer.ffn_gate_exps = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {  n_embd, n_ff_exp, n_expert});
                         layer.ffn_down_exps = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert});
                         layer.ffn_up_exps   = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {  n_embd, n_ff_exp, n_expert});
 
                         // Shared expert branch
+                        auto n_ff_shared_exp = hparams.n_ff_exp && hparams.n_expert_used ? hparams.n_ff_exp * hparams.n_expert_used : n_ff;
                         layer.ffn_gate_inp_shexp = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_GATE_INP_SHEXP, "weight", i), {n_embd});
-                        layer.ffn_gate_shexp = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd,   n_ff});
-                        layer.ffn_down_shexp = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {  n_ff, n_embd});
-                        layer.ffn_up_shexp   = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd,   n_ff});
+                        layer.ffn_gate_shexp = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd,   n_ff_shared_exp});
+                        layer.ffn_down_shexp = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {  n_ff_shared_exp, n_embd});
+                        layer.ffn_up_shexp   = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd,   n_ff_shared_exp});
                     }
                 } break;
             case LLM_ARCH_PHI2: