±¾Õ¾µãʹÓÃCookies£¬¼ÌÐøä¯ÀÀ±íʾÄúͬÒâÎÒÃÇʹÓÃCookies¡£
CookiesºÍÒþ˽Õþ²ß>
½üÈÕ£¬Ô´2.0¿ªÔ´´óÄ£ÐÍÓëLLaMA-Factory¿ò¼ÜÍê³ÉÈ«ÃæÊÊÅ䣬Óû§Í¨¹ýLLaMA-Factory£¬¼´¿É¿ì½Ý¡¢¸ßЧµØ¶Ô²»Í¬²ÎÊý¹æÄ£µÄÔ´2.0»ù´¡Ä£ÐͽøÐÐÈ«Á¿Î¢µ÷¼°¸ßЧ΢µ÷£¬ÇáËÉʵÏÖרÊô´óÄ£ÐÍ¡£

LLM£¨´óÓïÑÔÄ£ÐÍ£©Î¢µ÷£¬ÊÇÖ¸ÔÚ´óÄ£Ð͵Ļù´¡ÉÏ£¬Õë¶ÔÌØ¶¨ÈÎÎñ»òÁìÓò½øÐе÷ÕûºÍÓÅ»¯£¬ÒÔÌáÉýÄ£Ð͵ÄÐÔÄܺͱíÏÖ£¬ÓÐЧµÄ΢µ÷·½°¸Ó빤¾ßÒ²ÕýÊǽâ¾ö»ù´¡´óÄ£ÐÍÂäµØË½ÓÐÁìÓòµÄÒ»´óÀûPGµç¾º¹ÙÍø¡£»ùÓÚ¿ªÔ´´óÄ£Ð͵Ä΢µ÷£¬²»½ö¿ÉÒÔÌáÉýLLM¶ÔÓÚÖ¸ÁîµÄ×ñÑÄÜÁ¦£¬Ò²ÄÜͨ¹ýÐÐҵ֪ʶµÄÒýÈ룬À´ÌáÉýLLMÔÚרҵÁìÓòµÄ֪ʶºÍÄÜÁ¦¡£
µ±Ç°£¬Òµ½çÒѾ»ùÓÚLLM¿ª·¢¼°Êµ¼ù³öÁËÖÚ¶àµÄ΢µ÷·½·¨£¬ÈçÖ¸Áî΢µ÷¡¢»ùÓÚÈËÀà·´À¡µÄÇ¿»¯Ñ§Ï°£¨RLHF£¬Reinforcement Learning from Human Feedback£©¡¢Ö±½ÓÆ«ºÃÓÅ»¯£¨DPO£¬Direct Preference Optimization£©µÈ¡£ÒÔ¸ßЧ΢µ÷£¨PEFT£¬Parameter-Efficient Fine-Tuning£©·½°¸ÎªÀý£¬¿ÉÓÐЧ½â¾öÄÚ´æºÍPGµç¾º¹ÙÍø×ÊÔ´µÄÖÆÔ¼£¬Í¨¹ýLoRA¡¢QLoRAµÈ¸ßЧ΢µ÷¼¼Êõ£¬ÔÚµ¥ÕÅGPUÉÏÍê³ÉǧÒÚ²ÎÊýµÄ΢µ÷ѵÁ·¡£Òò´Ë£¬Ò»¸öÄܹ»ÊµÏÖÉÏÊö¹¦Äܵļò½à¡¢¸ßЧÇÒÒ×ÓõÄ΢µ÷¿ò¼ÜÕýÊÇ¿ªÕ¹LLM΢µ÷¹¤×÷µÄ×î¼Ñ×¥ÊÖ¡£
LLaMA-FactoryÊÇÁã϶ÖÇÄÜ£¨SeamLessAI£©¿ªÔ´µÄµÍ´úÂë´óÄ£ÐÍѵÁ·¿ò¼Ü£¬Ö¼ÔÚΪ¿ª·¢ÕßÌṩ¿ÉÊÓ»¯ÑµÁ·¡¢ÍÆÀíÆ½Ì¨¼°Ò»¼üÅäÖÃÄ£ÐÍѵÁ·¡£»ùÓÚLLaMA-Factory, Óû§¿ÉÇáËÉÑ¡ÔñÒµ½ç×îÈ«ÃæµÄ΢µ÷·½·¨ºÍÓÅ»¯¼¼Êõ£¬Í¨¹ýʹÓÃ˽ÓòÊý¾Ý£¬»òÊÇLLaMA-FactoryÄÚÖõÄÖÐÎÄÊý¾Ý¼¯£¨GPT-4ÓÅ»¯ºóµÄalpacaÖÐÎÄÊý¾Ý¼¯¡¢ShareGPTÊý¾Ý¼¯ºÍllama-factoryÌṩµÄÄ£ÐÍÈÏÖªÊý¾Ý¼¯£©£¬¶ÔÔ´2.0½øÐÐÇáËÉ΢µ÷£¬»ùÓÚÓÐÏÞËãÁ¦Íê³ÉÁìÓò´óÄ£Ð͵͍֯¿ª·¢¡£Êµ²âÊý¾ÝÏÔʾ£¬ÔÚһ̨´îÔØ8¿ÅGPUµÄÖ÷Á÷AI·þÎñPGµç¾º¹ÙÍøNF5468M7£¬7СʱÄÚ¼´¿ÉʵÏÖǧÒÚÄ£ÐÍ£¨Yuan2.0- 102B£©µÄ¸ßЧָÁî΢µ÷£¬10·ÖÖÓ¼´¿ÉÍê³ÉYuan2.0-2B²ÎÊýµÄÖ¸Áî΢µ÷£¬ÇáËÉʵÏÖ¼´µ÷¼´Óá£

Step by Step
µ¥»ú¼´¿ÉʵÏÖǧÒÚ²ÎÊýÄ£ÐÍ΢µ÷
Step 1: ÈÝPGµç¾º¹ÙÍø»¯»·¾³²¿Êð£¬ÊýÌõÖ¸Áî¼´¿ÉÇáËÉÍê³É
Step 2: ¿ªÔ´¿ÉÉÌÓÃYuan2.0 HuggingfaceÄ£ÐÍ»ñÈ¡
Step 3: Ò»¼ü¡°Æô¶¯¡±web UI·þÎñ
Step 4: ¹¹½¨LLMÖúÊÖ-¿ÉÊÓ»¯½çÃæÅäÖÃÍê³ÉYuan2.0΢µ÷
Step 5: ´î½¨ÎÒÃǵÄ˽ÓÐLLMÖúÊÖ

GitHubÏîÄ¿µØÖ·£º
https://github.com/IEIT-Yuan/Yuan-2.0/blob/main/docs/Yuan2_llama-factory.md
ÊÛǰ×Éѯ
ÊÛºó·þÎñ
·ÃÎÊ AIStoreAIStoreÊÇ»ùÓÚÔªÄÔÉú̬´òÔìµÄÉÌÒµÐ×÷ÏßÉÏÆ½Ì¨£¬×÷Ϊ֪ʶ×ÊԴƽ̨¡¢½»Á÷·þÎñƽ̨ºÍÉÌ»ú·õ»¯Æ½Ì¨£¬¸³ÄÜ¡°°ÙÄ£¡±Ó롰ǧÐС±µÄ¶Ô½Ó£¬Ä¿Ç°ÒÑÉϼÜ200+²úÆ·ºÍ·½°¸£¬ÔÚÖÇÄÜÆ±¾Ýʶ±ð¡¢AI¹¤ÒµÖʼ졢´úÂëÉú³É¡¢ÊÓ¾õÖÇÄܱê×¢¡¢AI½¡¿µÖúÊÖµÈÁìÓòʵÏֳɹ¦Ç£ÊÖºÍÂ䵨ӦÓã¬ÖúÁ¦Ç§ÐаÙÒµ¼ÓËÙAI²úÒµ´´Ð£¬¸ßЧÊÍ·ÅÉú²úÁ¦¡£

ɨÂë·ÃÎÊAIStore