fix: add some changes as per review

Author: Joan Martinez

ggml.c

@@ -5478,9 +5478,9 @@ static struct ggml_tensor * ggml_soft_max_impl(
         GGML_ASSERT(pos->type == mask->type);
     }
 
-    /*if (max_bias > 0.0f) {
+    if (max_bias > 0.0f) {
         GGML_ASSERT(pos);
-    }*/
+    }
 
     bool is_node = false;
 
@@ -12401,7 +12401,6 @@ static void ggml_compute_forward_soft_max_f32(
     float * wp = (float *) params->wdata + (nc + CACHE_LINE_SIZE_F32) * ith;
 
     // when max_bias <= 0.0f, src2 is not used and we default it to src0 to avoid branching
-    //float * pos = src2 ? (float *) src2->data : NULL;
     ggml_fp16_t * pos_f16 = src2 ? (ggml_fp16_t *) src2->data : src0->data;
     float       * pos_f32 = src2 ? (float       *) src2->data : src0->data;
 
@@ -12436,13 +12435,13 @@ static void ggml_compute_forward_soft_max_f32(
 
             if (use_f16) {
                 for (int i = 0; i < nc; ++i) {
-                    wp[i] += slope*GGML_FP16_TO_FP32(pos_f16[i]);
-                    //wp[i] = wp[i] - slope*abs(i1%nc - i);
+                    //wp[i] -= slope*GGML_FP16_TO_FP32(pos_f16[i]);
+                    wp[i] -= slope*abs(i1%nc - i);
                 }
             } else {
                 for (int i = 0; i < nc; ++i) {
-                    wp[i] += slope*pos_f32[i];
-                    //wp[i] = wp[i] - slope*abs(i1%nc - i);
+                    //wp[i] -= slope*pos_f32[i];
+                    wp[i] -= slope*abs(i1%nc - i);
                 }
             }
         }

llama.cpp

@@ -8254,6 +8254,9 @@ struct llm_build_context {
         // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
         struct ggml_tensor * KQ_mask = build_inp_KQ_mask(false);
 
+        // positions of the tokens in the KV cache
+        struct ggml_tensor * KQ_pos = build_inp_KQ_pos(false);
+
         // iterate layers
         for (int il = 0; il < n_layer; ++il) {
             struct ggml_tensor * cur = inpL;
@@ -8322,7 +8325,7 @@ struct llm_build_context {
             struct ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
             cb(kq, "kq", il);
 
-            kq = ggml_soft_max_ext(ctx0, kq, KQ_mask, nullptr, 1.0f/sqrtf(float(n_embd_head)), hparams.f_max_alibi_bias);
+            kq = ggml_soft_max_ext(ctx0, kq, KQ_mask, KQ_pos, 1.0f/sqrtf(float(n_embd_head)), hparams.f_max_alibi_bias);
             cb(kq, "kq_soft_max_ext", il);
 
             struct ggml_tensor * v = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd_gqa, n_tokens)));
@@ -11523,7 +11526,7 @@ static int llama_decode_internal(
         }
 
         // non-causal masks do not use the KV cache
-        if (hparams.causal_attn) {
+        if (hparams.causal_attn || model.arch == LLM_ARCH_JINA_BERT_V2) {
             llama_kv_cache_update(&lctx);
 
             // if we have enough unused cells before the current head ->