use clamping instead, update number and add comment

Author: CISC

src/llama-graph.cpp

@@ -1006,8 +1006,9 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
         ggml_tensor * weights_sum = ggml_sum_rows(ctx0, weights); // [1, n_tokens]
         cb(weights_sum, "ffn_moe_weights_sum", il);
 
-        weights_sum = ggml_scale_bias(ctx0, weights_sum, 1.0, 1e-20);
-        cb(weights_sum, "ffn_moe_weights_sum_biased", il);
+        // Avoid division by zero, clamp to smallest number representable by F16
+        weights_sum = ggml_clamp(ctx0, weights_sum, 6.103515625e-5, INFINITY);
+        cb(weights_sum, "ffn_moe_weights_sum_clamped", il);
 
         weights = ggml_div(ctx0, weights, weights_sum); // [n_expert_used, n_tokens]
         cb(weights, "ffn_moe_weights_norm", il);