handle no-mmap as well

Author: rgerganov

examples/gguf-hash/gguf-hash.cpp

@@ -365,7 +365,14 @@ static hash_exit_code_t gguf_hash(const hash_params & hash_params) {
 
         if (hash_params.fnv) {
             uint64_t hash = fnv_hash((const uint8_t *)raw_data, n_bytes);
-            printf("%016lx  %s\n", hash, tensor_layer_name.c_str());
+            char hex_result[17];
+            for (int  offset = 0; offset < 8; offset++) {
+                unsigned int shift_bits_by = (8 * (8 - offset - 1));
+                snprintf( ( hex_result + (2*offset)), sizeof(hex_result) - (2*offset), "%02x", (unsigned char) (hash >> shift_bits_by)&0xff);
+            }
+
+            printf("%-8s  %-s  %s\n", "fnv", hex_result, tensor_layer_name.c_str());
+
             char hash_key[128];
             snprintf(hash_key, sizeof(hash_key), "%s_hash", name);
             gguf_set_val_u64(ctx_out, hash_key, hash);

ggml/src/ggml-rpc/ggml-rpc.cpp

@@ -1143,7 +1143,7 @@ bool rpc_server::set_tensor_hash(const rpc_msg_set_tensor_hash_req & request, rp
          || request.tensor.data + request.offset >= p1
          || size > (p1 - request.tensor.data - request.offset)) {
             GGML_LOG_ERROR("[%s] tensor data region (data=0x%" PRIx64 ", offset=%" PRIu64 ", size=%zu, hash=0x%" PRIx64 ") out of buffer bounds [0x%zx, 0x%zx)\n",
-                           __func__, in_tensor->data, offset, size, *hash, p0, p1);
+                           __func__, request.tensor.data, request.offset, size, request.hash, p0, p1);
             return false;
         }
     }

src/llama-model-loader.cpp

@@ -1079,12 +1079,14 @@ bool llama_model_loader::load_all_data(
                         buffer_idx %= n_buffers;
                     }
                 } else {
-                    read_buf.resize(n_size);
-                    file->seek(weight->offs, SEEK_SET);
-                    file->read_raw(read_buf.data(), n_size);
-                    ggml_backend_tensor_set(cur, read_buf.data(), 0, n_size);
-                    if (check_tensors && !ggml_validate_row_data(cur->type, read_buf.data(), n_size)) {
-                        throw std::runtime_error(format("tensor '%s' has invalid data", ggml_get_name(cur)));
+                    if (!check_tensors && !rpc_load_tensor(cur)) {
+                        read_buf.resize(n_size);
+                        file->seek(weight->offs, SEEK_SET);
+                        file->read_raw(read_buf.data(), n_size);
+                        ggml_backend_tensor_set(cur, read_buf.data(), 0, n_size);
+                        if (check_tensors && !ggml_validate_row_data(cur->type, read_buf.data(), n_size)) {
+                            throw std::runtime_error(format("tensor '%s' has invalid data", ggml_get_name(cur)));
+                        }
                     }
                 }
             }