Sarvam 30B is also optimized for local execution on Apple Silicon systems using MXFP4 mixed-precision inference. On MacBook Pro M3, the optimized runtime achieves 20 to 40% higher token throughput across common sequence lengths. These improvements make local experimentation significantly more responsive and enable lightweight edge deployments without requiring dedicated accelerators.
Фото: Raghed Waked / Reuters，详情可参考新收录的资料
�@���������r�W�l�X�A�i���X�g�������̂�IT�̕��삾���ł͂����܂����B�����Ɂu�r�W�l�X�A�i���X�g�v�ƌ����Ă����͈͍̔͂L�����̂ŁA�����܂ŏq�ׂĂ����Г��ł�IT�̓����Ŋ����u�r�W�l�X�V�X�e���A�i���X�g�v�����A�Ɩ��v���Z�X���P���s���u�t�@���N�V���i���r�W�l�X�A�i���X�g�v�A�\�t�g�E�F�A�v���_�N�g���T�[�r�X�̉��l�����̂��߂̉��P���s���u�v���_�N�g�A�i���X�g�v�A�f�[�^���͂ɋ��݂����u�r�W�l�X�C���e���W�F���X�A�i���X�g�v�Ȃǂ̂��܂��܂ȗ̈��ɂ����Ă��ꂼ���̖��������r�W�l�X�A�i���X�g�����݂����̂ł��B。业内人士推荐新收录的资料作为进阶阅读
21:13, 9 марта 2026Спорт