1
00:00:00,000 --> 00:00:16,000
你好,千文。介绍一下你自己吧。给我讲一个笑话吧。

2
00:00:30,000 --> 00:00:32,560
剧员说不用咖啡是热的

3
00:00:32,560 --> 00:00:34,640
电脑说那好吧

4
00:00:34,640 --> 00:00:36,600
我这就把咖啡也热一下

5
00:00:36,600 --> 00:00:39,540
希望这个笑话能让你会心一笑

6
00:00:39,540 --> 00:00:43,120
大家好啊

7
00:00:43,120 --> 00:00:46,120
刚才给这是给大家展示了一下

8
00:00:46,120 --> 00:00:49,340
这个项目的一个效果

9
00:00:49,340 --> 00:00:51,120
那么今天呢

10
00:00:51,120 --> 00:00:53,000
我们讲一个非常有趣

11
00:00:53,000 --> 00:00:56,200
并且我感觉也比较新颖的一个项目

12
00:00:56,200 --> 00:00:58,400
叫千万Audio Agent

13
00:00:58,400 --> 00:01:05,040
大家看一下这个名字就非常清楚这个项目讲了个什么事

14
00:01:05,040 --> 00:01:07,040
首先是audio

15
00:01:07,040 --> 00:01:09,280
重要的是audio和agent

16
00:01:09,280 --> 00:01:11,420
千万必须要介绍

17
00:01:11,420 --> 00:01:13,440
肯定是人家阿里搞的

18
00:01:13,440 --> 00:01:17,760
感谢官方的分享

19
00:01:17,760 --> 00:01:20,180
然后audio和agent

20
00:01:20,180 --> 00:01:21,560
结合在一起

21
00:01:21,560 --> 00:01:25,180
大家实际看到了刚才我给大家展示的效果

22
00:01:25,180 --> 00:01:27,600
可以进行语音交互

23
00:01:27,600 --> 00:01:30,260
以及自然的问答

24
00:01:30,260 --> 00:01:31,540
特别真实

25
00:01:31,540 --> 00:01:33,120
特别流畅

26
00:01:33,120 --> 00:01:35,820
而且大家也可以看到视频中

27
00:01:35,820 --> 00:01:39,180
那么这个项目也支持这个打断功能

28
00:01:39,180 --> 00:01:43,100
你随时可以新的问题提出可以打断

29
00:01:43,100 --> 00:01:45,520
那么下面我们就看一下

30
00:01:45,520 --> 00:01:47,280
这个项目的一些具体介绍

31
00:01:47,280 --> 00:01:48,780
以及安装注意事项

32
00:01:48,780 --> 00:01:52,060
先说这个介绍吧

33
00:01:52,060 --> 00:01:54,760
因为安装注意事项那还是比较复杂的

34
00:01:54,760 --> 00:01:56,960
然后说这个项目呢

35
00:01:56,960 --> 00:01:58,920
官方说是真正的交流啊

36
00:01:58,920 --> 00:02:00,660
不该再说完一句话后

37
00:02:00,660 --> 00:02:02,020
陷入漫长的等待

38
00:02:02,020 --> 00:02:03,400
也不该因为agent

39
00:02:03,400 --> 00:02:04,240
他正在查资料

40
00:02:04,240 --> 00:02:05,900
调用工具和处理任务

41
00:02:05,900 --> 00:02:08,420
正常对话就此诈听啊

42
00:02:08,420 --> 00:02:09,080
其实他

43
00:02:09,080 --> 00:02:10,980
我是这么理解的啊

44
00:02:10,980 --> 00:02:12,080
可能不是很准确

45
00:02:12,080 --> 00:02:12,920
我认为就是

46
00:02:12,920 --> 00:02:16,340
一个一个真正的对话的一个agent啊

47
00:02:16,340 --> 00:02:17,740
不需要再打字啊

48
00:02:17,740 --> 00:02:20,920
那么通过这种对话的交流啊

49
00:02:20,920 --> 00:02:23,820
就可以实现调用agent啊

50
00:02:23,820 --> 00:02:26,220
调用工具啊

51
00:02:26,220 --> 00:02:30,380
实现某种电脑上的那具体的操作啊

52
00:02:30,380 --> 00:02:33,020
然后我们接着看啊

53
00:02:33,020 --> 00:02:33,800
做了这个

54
00:02:33,800 --> 00:02:36,340
然后官方就做了这个agent啊

55
00:02:36,340 --> 00:02:37,720
让agent持续交流

56
00:02:37,720 --> 00:02:38,380
持续工作

57
00:02:38,380 --> 00:02:40,760
持续实时对话进行交流啊

58
00:02:40,760 --> 00:02:42,180
这个我觉得还是

59
00:02:42,180 --> 00:02:44,640
比较新颖啊

60
00:02:44,640 --> 00:02:45,440
比较特别的

61
00:02:45,440 --> 00:02:46,880
而且大家也看到我

62
00:02:46,880 --> 00:02:49,520
我们实际这个展示这个效果啊

63
00:02:49,520 --> 00:02:52,360
我感觉这个效果还是非常好的啊

64
00:02:52,360 --> 00:02:55,180
也没有打断很流畅很连贯啊

65
00:02:55,180 --> 00:02:56,660
就是这么样一个效果

66
00:02:56,660 --> 00:02:58,020
然后下面呢

67
00:02:58,020 --> 00:03:01,200
就是它的一个日志

68
00:03:01,200 --> 00:03:02,520
就是更新日志

69
00:03:02,520 --> 00:03:04,400
大家可以具体看一看

70
00:03:04,400 --> 00:03:06,560
然后这个呢

71
00:03:06,560 --> 00:03:09,060
就是一个实际的展示视频

72
00:03:09,060 --> 00:03:11,480
那么大家有兴趣的话

73
00:03:11,480 --> 00:03:14,540
我就不在这给大家一块看了

74
00:03:14,540 --> 00:03:15,720
大家这一分多钟

75
00:03:15,720 --> 00:03:16,900
大家自己去慢慢看

76
00:03:16,900 --> 00:03:19,280
然后看一下特色

77
00:03:19,280 --> 00:03:20,520
说双双

78
00:03:20,520 --> 00:03:23,120
全双公实施语音交互

79
00:03:23,120 --> 00:03:25,220
自然打断持续多轮对话

80
00:03:25,220 --> 00:03:26,480
刚才的演示

81
00:03:26,480 --> 00:03:29,660
我已经充分了给大家展示了个

82
00:03:29,660 --> 00:03:31,720
什么叫实时运营交互

83
00:03:31,720 --> 00:03:33,080
什么叫打断

84
00:03:33,080 --> 00:03:34,220
什么叫持续多轮

85
00:03:34,220 --> 00:03:35,440
刚才演示中

86
00:03:35,440 --> 00:03:37,120
已经给大家介绍了

87
00:03:37,120 --> 00:03:38,880
已经给大家展示出来了

88
00:03:38,880 --> 00:03:41,520
然后一键选择你办事的agent

89
00:03:41,520 --> 00:03:43,180
服用所有的工具

90
00:03:43,180 --> 00:03:44,900
mcp以及技能

91
00:03:44,900 --> 00:03:49,060
这个项目还是扩展性还是很好的

92
00:03:49,060 --> 00:03:52,880
就这个扩展性的说法

93
00:03:52,880 --> 00:03:54,580
就是在于它可以

94
00:03:54,580 --> 00:03:56,440
使用现有的工具

95
00:03:56,440 --> 00:03:57,720
mcp及技能

96
00:03:57,720 --> 00:03:59,580
这个扩展性非常好

97
00:03:59,580 --> 00:04:01,500
而兼容性也非常好

98
00:04:01,500 --> 00:04:03,140
扩展和兼容性都非常好

99
00:04:03,140 --> 00:04:05,760
然后前台对话与后台任务

100
00:04:05,760 --> 00:04:07,020
并加起去可以

101
00:04:07,020 --> 00:04:10,480
随时追问任务技能和取消任务

102
00:04:10,480 --> 00:04:12,600
这个我也尝试过

103
00:04:12,600 --> 00:04:13,420
完全OK的

104
00:04:13,420 --> 00:04:16,440
它甚至可以开子任务

105
00:04:16,440 --> 00:04:18,920
但是它的子任务

106
00:04:18,920 --> 00:04:20,620
我目前认为

107
00:04:20,620 --> 00:04:23,680
可能他还是在在调用其他agent中的

108
00:04:23,680 --> 00:04:26,020
的任务去执行相关的

109
00:04:26,320 --> 00:04:28,060
呃任务工作啊

110
00:04:28,560 --> 00:04:30,720
支持创建多个独立任务啊

111
00:04:30,720 --> 00:04:33,460
然后由后台的agent一步执行啊

112
00:04:33,460 --> 00:04:35,220
并持续追踪任务状态

113
00:04:35,220 --> 00:04:36,680
就是我说的这个意思啊

114
00:04:36,980 --> 00:04:38,860
其实用后台的这个agent

115
00:04:38,860 --> 00:04:40,520
一步执行啊

116
00:04:40,520 --> 00:04:43,460
多个独立的子agent的任务啊

117
00:04:43,460 --> 00:04:44,320
就可以这么理解

118
00:04:44,320 --> 00:04:45,180
可能比较绕嘴

119
00:04:45,180 --> 00:04:46,720
还是其实呃

120
00:04:46,720 --> 00:04:48,380
总体来说就这么样一个流程

121
00:04:48,380 --> 00:04:49,480
就是说呃

122
00:04:49,480 --> 00:04:53,280
前往audio agent去调用后台的agent

123
00:04:53,280 --> 00:04:55,680
那么这个后台的agent有多种方案

124
00:04:55,680 --> 00:04:57,980
那么你可以选取一种agent的方案

125
00:04:57,980 --> 00:04:59,160
然后这个agent的方案

126
00:04:59,160 --> 00:05:02,520
可以一步的去使用其中的自agent

127
00:05:02,520 --> 00:05:04,040
去执行任务啊

128
00:05:04,040 --> 00:05:06,160
就是这么样一个流程

129
00:05:06,160 --> 00:05:07,480
大家可以理解一下

130
00:05:07,480 --> 00:05:08,860
我说了这个意思啊

131
00:05:08,860 --> 00:05:10,620
任务结束自动回答

132
00:05:10,620 --> 00:05:12,180
回答到当前对话

133
00:05:12,180 --> 00:05:14,940
就是说他的一步子任务啊

134
00:05:14,940 --> 00:05:17,560
完成之后可以就给你一个反馈啊

135
00:05:17,560 --> 00:05:18,560
就是一步的啊

136
00:05:18,560 --> 00:05:21,860
支持继续追问和修改

137
00:05:21,860 --> 00:05:25,240
那么你也可以对这个一部任务进行继续的追问修改

138
00:05:25,240 --> 00:05:26,300
支持web UI

139
00:05:26,300 --> 00:05:28,760
终端TUI和桌面球

140
00:05:28,760 --> 00:05:33,360
这个目前的话web UI和终端UI的话

141
00:05:33,360 --> 00:05:34,420
我没怎么用

142
00:05:34,420 --> 00:05:36,820
但是我用的是个悬浮球

143
00:05:36,820 --> 00:05:38,360
我认为这个悬浮球非常好用

144
00:05:38,360 --> 00:05:41,400
我就没用web UI和中端TUI

145
00:05:41,400 --> 00:05:43,340
其实这两个其实也没什么

146
00:05:43,340 --> 00:05:45,620
其实这三个没什么区别

147
00:05:45,620 --> 00:05:46,860
啊啊

148
00:05:46,860 --> 00:05:48,620
我觉得这个中单球比较方便

149
00:05:48,620 --> 00:05:50,820
所以也不占地方顺手就可以

150
00:05:51,380 --> 00:05:52,220
拈来啊

151
00:05:52,220 --> 00:05:53,120
比较舒服

152
00:05:53,120 --> 00:05:54,620
所以我就用这个了啊

153
00:05:54,620 --> 00:05:57,020
支持当心用户长期更新化的

154
00:05:57,020 --> 00:05:59,020
跨会画机翼啊

155
00:05:59,020 --> 00:06:00,620
这个我机翼我倒是

156
00:06:00,620 --> 00:06:02,380
呃没没充分利用上啊

157
00:06:02,380 --> 00:06:05,620
我等于后期研究研究再再给大家展示啊

158
00:06:07,020 --> 00:06:09,720
那么这样这个就是探口价值不拖啊

159
00:06:09,720 --> 00:06:10,720
那么用户问了

160
00:06:10,720 --> 00:06:13,220
然后通过这个audio runtime啊

161
00:06:13,220 --> 00:06:16,360
通过上下文去执行这个Audio Runtime

162
00:06:16,360 --> 00:06:17,520
这个Agent

163
00:06:17,520 --> 00:06:18,800
Agent Runtime

164
00:06:18,800 --> 00:06:20,700
然后给反馈

165
00:06:20,700 --> 00:06:22,100
最终给了用户

166
00:06:22,100 --> 00:06:24,160
其实就这么样一个流程

167
00:06:24,160 --> 00:06:25,680
其实这个流程中

168
00:06:25,680 --> 00:06:27,940
我们看了三个主要主体用户

169
00:06:27,940 --> 00:06:31,520
千万Audio Runtime

170
00:06:31,520 --> 00:06:32,500
还有Agent

171
00:06:32,500 --> 00:06:35,480
这么三个主要的角色

172
00:06:35,480 --> 00:06:37,060
主体

173
00:06:37,060 --> 00:06:37,920
任务主体

174
00:06:37,920 --> 00:06:42,080
那么这样就是一个简单的架构图

175
00:06:42,080 --> 00:06:45,080
还有一个具体的一个架构图

176
00:06:45,080 --> 00:06:46,080
这个具体架构图

177
00:06:46,080 --> 00:06:47,580
接入参考架构

178
00:06:47,580 --> 00:06:49,340
那么这个架构图的话

179
00:06:49,340 --> 00:06:52,040
大家就自己去详细看吧

180
00:06:52,040 --> 00:06:53,840
因为这个上面文字比较多

181
00:06:53,840 --> 00:06:55,520
我不希望视频太长

182
00:06:55,520 --> 00:06:58,820
但流程其实就跟上面这个大差不诈

183
00:06:58,820 --> 00:07:01,120
只不过是可能多了很多的细分

184
00:07:01,120 --> 00:07:04,620
下面就说到了比较重要的东西了

185
00:07:04,620 --> 00:07:06,380
就是它这个后端agent

186
00:07:06,380 --> 00:07:09,120
就是我们刚才讲的后端agent这里

187
00:07:09,120 --> 00:07:12,060
就刚才特色的后端agent

188
00:07:12,060 --> 00:07:15,060
那么这个后端的agent呢

189
00:07:15,060 --> 00:07:17,060
你可以选到没有啊

190
00:07:17,060 --> 00:07:22,060
你也可以用目前一支的全部主流啊

191
00:07:22,060 --> 00:07:28,060
记得是全部主流的这种agent工具啊

192
00:07:28,060 --> 00:07:30,060
可能是agent工具啊

193
00:07:30,060 --> 00:07:32,060
包括我就不念了

194
00:07:32,060 --> 00:07:35,060
反正就是市面最主流的全部都有啊

195
00:07:35,060 --> 00:07:40,060
这里面呢我用的是hermes

196
00:07:40,060 --> 00:07:41,420
Hermes啊

197
00:07:41,420 --> 00:07:43,520
我觉得OpenCloud我还挺想用的

198
00:07:43,520 --> 00:07:46,900
但是呃我我我喜欢用的Hermes

199
00:07:46,900 --> 00:07:50,500
因为我觉得现在Hermes的速度特别快啊

200
00:07:50,500 --> 00:07:53,320
因为最近公新了这个呃

201
00:07:53,320 --> 00:07:58,100
181920这个版本之后这个Hermes的嗯

202
00:08:09,203 --> 00:08:12,203
至于说其他的比如说Codex啊

203
00:08:12,203 --> 00:08:15,443
Claw Code啊还有这个OpenClaw呢

204
00:08:15,443 --> 00:08:17,963
我因为他是巧接啊

205
00:08:17,963 --> 00:08:20,363
巧接的话他的我觉得他的反应

206
00:08:20,363 --> 00:08:23,163
反应速度可能没有远胜速度那么快啊

207
00:08:23,163 --> 00:08:26,043
所以我这里面最推荐的还是用Hermes啊

208
00:08:26,043 --> 00:08:30,163
大家可以啊就是用自己喜欢的

209
00:08:30,163 --> 00:08:34,003
然后如果你是用这个这个Web UI啊

210
00:08:34,003 --> 00:08:36,163
你可以用这个这个安装啊

211
00:08:36,163 --> 00:08:39,163
安装一下这个NUDEJS安装一下啊

212
00:08:39,163 --> 00:08:41,323
大家注意这个node.js版本就行了

213
00:08:42,163 --> 00:08:43,363
然后呢

214
00:08:43,363 --> 00:08:45,463
是配这个模型啊

215
00:08:45,463 --> 00:08:49,803
这个这个如果你是用这个node.js这个这个安装方案的话

216
00:08:49,803 --> 00:08:50,803
那你就要需要

217
00:08:51,203 --> 00:08:55,263
嗯去去在这个webui或者tui里面去配置一下

218
00:08:55,703 --> 00:08:57,603
嗯相关的这个模型

219
00:08:58,023 --> 00:08:58,723
k啊

220
00:08:59,203 --> 00:09:01,403
呃包括这个云前天模型

221
00:09:01,403 --> 00:09:02,263
还有这个

222
00:09:02,903 --> 00:09:06,003
agent 他的这个后台agent 的这个方案啊

223
00:09:06,503 --> 00:09:08,203
嗯还有这个

224
00:09:08,203 --> 00:09:09,863
后台模型啊

225
00:09:09,863 --> 00:09:11,903
因为他这里面模型非常多啊

226
00:09:11,903 --> 00:09:13,903
因为你看这就有

227
00:09:15,203 --> 00:09:16,363
语音模型啊

228
00:09:16,363 --> 00:09:19,003
语音前的语音模型用用于交互的啊

229
00:09:19,003 --> 00:09:21,203
就是语音交互的啊

230
00:09:21,203 --> 00:09:21,703
模型

231
00:09:21,703 --> 00:09:23,943
然后这个后台模型呢

232
00:09:23,943 --> 00:09:25,303
就是他交互的话

233
00:09:25,303 --> 00:09:29,003
你不是有一个一个思考处理的过程吗

234
00:09:29,003 --> 00:09:30,003
那么就是这个东西

235
00:09:30,003 --> 00:09:30,963
其实啊

236
00:09:30,963 --> 00:09:32,563
这个就负责语音交互啊

237
00:09:32,563 --> 00:09:33,963
这个就负责这个

238
00:09:35,363 --> 00:09:36,303
文字反馈啊

239
00:09:36,303 --> 00:09:37,403
你可以这么想反馈吧

240
00:09:37,403 --> 00:09:39,643
包括工具工具调用

241
00:09:39,643 --> 00:09:42,063
也不是工具工具调用

242
00:09:42,063 --> 00:09:44,063
其实也是后台agent就投作

243
00:09:44,063 --> 00:09:45,643
就是一个反馈吧

244
00:09:45,643 --> 00:09:47,263
问答问答反馈

245
00:09:47,263 --> 00:09:49,543
就像你正常使用模型的问答反馈

246
00:09:49,543 --> 00:09:52,763
然后包括这些设置

247
00:09:52,763 --> 00:09:54,203
包括你可以用这个

248
00:09:54,203 --> 00:09:57,063
这个tui可以听tui

249
00:09:57,063 --> 00:09:58,763
如重点来了

250
00:09:58,763 --> 00:09:59,503
大家重点

251
00:09:59,503 --> 00:10:01,663
如果你想用本地部署

252
00:10:01,663 --> 00:10:04,543
那你就需要s2s

253
00:10:04,543 --> 00:10:06,163
这个本地部署方案

254
00:10:06,163 --> 00:10:10,203
然后桌面版的这个TYF

255
00:10:10,203 --> 00:10:14,163
后台的Agent

256
00:10:14,163 --> 00:10:18,163
各样化记忆的话这个东西

257
00:10:18,163 --> 00:10:20,163
我还真没没去试

258
00:10:20,163 --> 00:10:22,163
但是我觉得这个东西其实没必要去试

259
00:10:22,163 --> 00:10:23,163
为什么呢

260
00:10:23,163 --> 00:10:27,163
因为这个东西跟OpenCloud和Hermes的话是一样的

261
00:10:27,163 --> 00:10:31,163
那么都有这些角色都有那记忆

262
00:10:31,163 --> 00:10:35,163
我认为没什么任何巨大的区别

263
00:10:35,163 --> 00:10:38,863
巨大的区别顺手啊顺手拈来啊顺手就可以上手的事

264
00:10:38,863 --> 00:10:42,463
所以你必要去具体去演示这个这个东西了啊

265
00:10:42,463 --> 00:10:44,943
然后注意事项大家看一下啊

266
00:10:44,943 --> 00:10:48,163
然后圆满圆满的开发啊

267
00:10:48,163 --> 00:10:51,183
圆满由nojs去做一些开发工作啊

268
00:10:51,183 --> 00:10:52,483
总体来说的话

269
00:10:52,483 --> 00:10:53,903
这个项目啊

270
00:10:53,903 --> 00:10:55,343
我认为还是比较吸引的

271
00:10:55,343 --> 00:10:58,043
这种全创工的这种交互啊

272
00:10:58,043 --> 00:11:01,863
打断而且准确率非常好啊

273
00:11:01,863 --> 00:11:03,163
非常好啊

274
00:11:03,163 --> 00:11:04,763
是这样的啊

275
00:11:04,763 --> 00:11:07,523
我还要补充几点啊

276
00:11:07,523 --> 00:11:08,903
补充几点呢

277
00:11:08,903 --> 00:11:12,243
是在于这个安装和使用的问题啊

278
00:11:12,243 --> 00:11:13,683
我找一下那个问题啊

279
00:11:13,683 --> 00:11:14,703
就这里吧

280
00:11:14,703 --> 00:11:18,083
就是说这个这个语音模型呢

281
00:11:18,083 --> 00:11:20,403
它默认啊

282
00:11:20,403 --> 00:11:26,283
你可以接阿里百列千万的这个这个API啊

283
00:11:26,283 --> 00:11:29,343
那么APIAPI会一定的免费额度

284
00:11:29,343 --> 00:11:34,383
那如果你想用本地的这个前台语音模型

285
00:11:34,383 --> 00:11:35,883
那你只能用

286
00:11:35,883 --> 00:11:37,023
刚才我说这个

287
00:11:37,023 --> 00:11:37,903
s

288
00:11:37,903 --> 00:11:39,583
s2s啊

289
00:11:39,583 --> 00:11:40,543
刚才那个

290
00:11:40,543 --> 00:11:42,383
我找一下那个位置啊

291
00:11:43,143 --> 00:11:44,823
啊

292
00:11:44,823 --> 00:11:46,963
就这个s2s啊

293
00:11:46,963 --> 00:11:48,303
去去接这个

294
00:11:48,303 --> 00:11:49,503
这个本地的

295
00:11:49,503 --> 00:11:51,063
云音模型啊

296
00:11:51,063 --> 00:11:52,463
本地的云音模型

297
00:11:52,463 --> 00:11:54,043
那么这个问题就解决了

298
00:11:54,043 --> 00:11:55,863
而且这个本地的云音模型

299
00:11:55,863 --> 00:11:57,163
它又有两种方案

300
00:11:57,163 --> 00:11:58,463
两种模型可以做

301
00:11:58,463 --> 00:11:59,223
比如说

302
00:11:59,223 --> 00:12:00,043
你可以走

303
00:12:00,043 --> 00:12:02,183
GUF了和

304
00:12:02,183 --> 00:12:02,743
这个

305
00:12:02,743 --> 00:12:03,703
这个

306
00:12:03,703 --> 00:12:04,103
这个

307
00:12:04,103 --> 00:12:05,743
这个原本

308
00:12:06,343 --> 00:12:07,603
他是这个模型啊

309
00:12:07,603 --> 00:12:08,203
就反正

310
00:12:08,603 --> 00:12:09,303
嗯

311
00:12:10,303 --> 00:12:12,003
就是两种也是两种方案啊

312
00:12:12,003 --> 00:12:14,043
但是根据你机器的情况啊

313
00:12:14,043 --> 00:12:15,163
去做本地的部署

314
00:12:15,163 --> 00:12:17,603
这这是这个这个云模型啊

315
00:12:18,243 --> 00:12:20,843
那么这两个本地部署的模型

316
00:12:20,843 --> 00:12:21,603
包括API

317
00:12:21,603 --> 00:12:23,363
我都全部试了一遍啊

318
00:12:23,363 --> 00:12:25,863
我认为可能API的

319
00:12:26,363 --> 00:12:27,463
效果会

320
00:12:27,703 --> 00:12:28,363
够好一点

321
00:12:28,363 --> 00:12:30,043
但是其实本地的效果

322
00:12:30,043 --> 00:12:31,703
我觉得也非常OK啊

323
00:12:31,703 --> 00:12:32,643
就正常的

324
00:12:32,643 --> 00:12:36,643
正常的识别以及正常的反馈都非常快啊

325
00:12:36,643 --> 00:12:39,643
而且它的显存的话

326
00:12:39,643 --> 00:12:44,643
应该从两三个G到最高占用六个多G吧

327
00:12:44,643 --> 00:12:46,643
就是这么样的显存占用

328
00:12:46,643 --> 00:12:49,643
如果是用本地的云云前台模型的话

329
00:12:49,643 --> 00:12:54,643
那如果你用的是这个这个后台模型啊

330
00:12:54,643 --> 00:12:55,643
就是说后台模型的话

331
00:12:55,643 --> 00:12:57,643
你也可以用本地部署

332
00:12:57,643 --> 00:12:59,643
因为你后台模型你可以接

333
00:12:59,643 --> 00:13:04,243
这个用open cloud去接这个后台本地的这个模型啊

334
00:13:04,243 --> 00:13:05,723
也可以就说

335
00:13:05,723 --> 00:13:09,363
就说如果你不用这个后台

336
00:13:09,363 --> 00:13:12,363
就是这个就就选择这个无模式啊

337
00:13:12,363 --> 00:13:14,523
无无无这个后台agent的话

338
00:13:14,523 --> 00:13:18,243
那么你就可以选用自己再接一个这个这个本地的这个

339
00:13:18,243 --> 00:13:23,183
比如说千万3.6那一些模型啊

340
00:13:23,183 --> 00:13:23,723
就可以了

341
00:13:23,723 --> 00:13:24,923
所以这个项目

342
00:13:24,923 --> 00:13:28,883
完全可以实现啊

343
00:13:28,883 --> 00:13:30,283
本地化啊

344
00:13:30,283 --> 00:13:31,403
完全本地化

345
00:13:31,403 --> 00:13:33,083
而且他的本地化

346
00:13:33,083 --> 00:13:34,643
我感觉他的效果

347
00:13:34,643 --> 00:13:37,963
其实就说能达到APR

348
00:13:37,963 --> 00:13:40,843
我就认为80%90%以上的效果

349
00:13:40,843 --> 00:13:42,123
是没有任何问题的

350
00:13:42,123 --> 00:13:45,663
我认为这个效果还是很好的

351
00:13:45,663 --> 00:13:47,423
所以对于这种

352
00:13:47,423 --> 00:13:52,203
语音全双公实施对话的agent

353
00:13:52,203 --> 00:13:54,063
我还是比较看好的

354
00:13:54,063 --> 00:13:55,743
也花了很长的时间

355
00:13:55,743 --> 00:13:56,923
一两个周的时间

356
00:13:56,923 --> 00:13:59,223
去对这个项目详细的了解一下

357
00:13:59,223 --> 00:14:02,523
那么今天就把这个项目推荐给大家

358
00:14:02,523 --> 00:14:04,363
还有一点我想补充的话

359
00:14:04,363 --> 00:14:08,263
就是说这个项目的部署难度是

360
00:14:08,263 --> 00:14:11,623
如果是就是说你要接API的话

361
00:14:11,623 --> 00:14:13,663
那么难度还不是很高

362
00:14:13,663 --> 00:14:17,383
但是如果说你要走原生本地

363
00:14:17,383 --> 00:14:21,663
本地去全部本地离线去做部署的话

364
00:14:21,663 --> 00:14:24,883
这个项目难度还是有一定的难度

365
00:14:24,883 --> 00:14:26,923
如果按照10分满分的话

366
00:14:26,923 --> 00:14:30,563
我认为他的部署难度可能在5分左右

367
00:14:30,563 --> 00:14:33,243
但是如果说像我这样老鼠的话

368
00:14:33,243 --> 00:14:36,843
部署的话应该也就那么回事

369
00:14:36,843 --> 00:14:37,823
也就非常简单

370
00:14:37,823 --> 00:14:38,503
也就部署完了

371
00:14:38,503 --> 00:14:40,443
其实对于老鼠的话来说

372
00:14:40,443 --> 00:14:42,583
这个项目并部署并没有难度

373
00:14:42,583 --> 00:14:43,963
并没有难度

374
00:14:43,963 --> 00:14:48,823
但是对于刚接触部署的新同学来说的话

375
00:14:48,823 --> 00:14:51,623
难度我认为是至少得有5分左右

376
00:14:51,623 --> 00:14:53,283
满分10分的话只有5分

377
00:14:53,283 --> 00:14:56,103
所以这个项目推荐给大家啊

378
00:14:56,103 --> 00:15:00,043
如果说大家在使用或安装过程中有任何的问题啊

379
00:15:00,043 --> 00:15:02,943
欢迎大家平台随意我或者加我微啊

380
00:15:02,943 --> 00:15:05,383
我的联系方式都在视频下方啊

381
00:15:05,383 --> 00:15:07,043
感谢大家观看

382
00:15:07,043 --> 00:15:08,243
我们下期再见