Fix chunks being too small with small matrix sizes (#17526)

* debug:"solve minicpmv precision problem"

* “debug minicpmv”

* Apply suggestion from @ngxson

---------

Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>

ggml/src/ggml-cpu/repack.cpp

@@ -1731,12 +1731,13 @@ template <typename BLOC_TYPE, int64_t INTER_SIZE, int64_t NB_COLS, ggml_type PAR
             nchunk0 = (nr0 + min_chunk_size - 1) / min_chunk_size;
         }
 
-        if (nth == 1 || nchunk0 < nth || disable_chunking) {
+        int64_t dr0 = (nr0 + nchunk0 - 1) / nchunk0;
+        // Only increase nchunk0 to nth if it won't make chunks too small
+        if (nth == 1 || ((nchunk0 < nth || disable_chunking) && (nr0 + nth - 1) / nth >= min_chunk_size)) {
             nchunk0 = nth;
+            dr0 = (nr0 + nchunk0 - 1) / nchunk0;
         }
 
-        const int64_t dr0 = (nr0 + nchunk0 - 1) / nchunk0;
-
         // Ensure nchunk doesn't exceed the number of rows divided by minimum chunk size
         // This prevents creating too many tiny chunks that could overlap after alignment
         const int64_t max_nchunk = (nr0 + min_chunk_size - 1) / min_chunk_size;

tools/mtmd/clip.cpp

@@ -1175,10 +1175,11 @@ struct clip_graph {
             cb(K, "resampler_K", -1);
             cb(V, "resampler_V", -1);
 
+            float resampler_kq_scale = 1.0f/ sqrtf(float(d_head));
             embeddings = build_attn(
                 model.mm_model_attn_o_w,
                 model.mm_model_attn_o_b,
-                Q, K, V, nullptr, kq_scale, -1);
+                Q, K, V, nullptr, resampler_kq_scale, -1);
             cb(embeddings, "resampler_attn_out", -1);
         }
         // layernorm