@lobehub/chat 1.49.14 → 1.49.16
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +50 -0
- package/changelog/v1.json +18 -0
- package/package.json +1 -1
- package/src/config/aiModels/fireworksai.ts +168 -91
- package/src/config/aiModels/perplexity.ts +25 -32
- package/src/config/modelProviders/fireworksai.ts +156 -100
- package/src/config/modelProviders/perplexity.ts +26 -32
- package/src/libs/agent-runtime/openrouter/__snapshots__/index.test.ts.snap +7 -7
package/CHANGELOG.md
CHANGED
@@ -2,6 +2,56 @@
|
|
2
2
|
|
3
3
|
# Changelog
|
4
4
|
|
5
|
+
### [Version 1.49.16](https://github.com/lobehub/lobe-chat/compare/v1.49.15...v1.49.16)
|
6
|
+
|
7
|
+
<sup>Released on **2025-02-03**</sup>
|
8
|
+
|
9
|
+
#### 💄 Styles
|
10
|
+
|
11
|
+
- **misc**: Update perplexity models.
|
12
|
+
|
13
|
+
<br/>
|
14
|
+
|
15
|
+
<details>
|
16
|
+
<summary><kbd>Improvements and Fixes</kbd></summary>
|
17
|
+
|
18
|
+
#### Styles
|
19
|
+
|
20
|
+
- **misc**: Update perplexity models, closes [#5624](https://github.com/lobehub/lobe-chat/issues/5624) ([58a86fc](https://github.com/lobehub/lobe-chat/commit/58a86fc))
|
21
|
+
|
22
|
+
</details>
|
23
|
+
|
24
|
+
<div align="right">
|
25
|
+
|
26
|
+
[](#readme-top)
|
27
|
+
|
28
|
+
</div>
|
29
|
+
|
30
|
+
### [Version 1.49.15](https://github.com/lobehub/lobe-chat/compare/v1.49.14...v1.49.15)
|
31
|
+
|
32
|
+
<sup>Released on **2025-02-03**</sup>
|
33
|
+
|
34
|
+
#### 💄 Styles
|
35
|
+
|
36
|
+
- **misc**: Update Fireworks check model and fix check error.
|
37
|
+
|
38
|
+
<br/>
|
39
|
+
|
40
|
+
<details>
|
41
|
+
<summary><kbd>Improvements and Fixes</kbd></summary>
|
42
|
+
|
43
|
+
#### Styles
|
44
|
+
|
45
|
+
- **misc**: Update Fireworks check model and fix check error, closes [#5680](https://github.com/lobehub/lobe-chat/issues/5680) ([64ea539](https://github.com/lobehub/lobe-chat/commit/64ea539))
|
46
|
+
|
47
|
+
</details>
|
48
|
+
|
49
|
+
<div align="right">
|
50
|
+
|
51
|
+
[](#readme-top)
|
52
|
+
|
53
|
+
</div>
|
54
|
+
|
5
55
|
### [Version 1.49.14](https://github.com/lobehub/lobe-chat/compare/v1.49.13...v1.49.14)
|
6
56
|
|
7
57
|
<sup>Released on **2025-02-03**</sup>
|
package/changelog/v1.json
CHANGED
@@ -1,4 +1,22 @@
|
|
1
1
|
[
|
2
|
+
{
|
3
|
+
"children": {
|
4
|
+
"improvements": [
|
5
|
+
"Update perplexity models."
|
6
|
+
]
|
7
|
+
},
|
8
|
+
"date": "2025-02-03",
|
9
|
+
"version": "1.49.16"
|
10
|
+
},
|
11
|
+
{
|
12
|
+
"children": {
|
13
|
+
"improvements": [
|
14
|
+
"Update Fireworks check model and fix check error."
|
15
|
+
]
|
16
|
+
},
|
17
|
+
"date": "2025-02-03",
|
18
|
+
"version": "1.49.15"
|
19
|
+
},
|
2
20
|
{
|
3
21
|
"children": {
|
4
22
|
"fixes": [
|
package/package.json
CHANGED
@@ -1,6 +1,6 @@
|
|
1
1
|
{
|
2
2
|
"name": "@lobehub/chat",
|
3
|
-
"version": "1.49.
|
3
|
+
"version": "1.49.16",
|
4
4
|
"description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
|
5
5
|
"keywords": [
|
6
6
|
"framework",
|
@@ -1,56 +1,30 @@
|
|
1
1
|
import { AIChatModelCard } from '@/types/aiModel';
|
2
2
|
|
3
3
|
const fireworksaiChatModels: AIChatModelCard[] = [
|
4
|
-
{
|
5
|
-
abilities: {
|
6
|
-
functionCall: true,
|
7
|
-
},
|
8
|
-
contextWindowTokens: 8192,
|
9
|
-
description:
|
10
|
-
'Fireworks 公司最新推出的 Firefunction-v2 是一款性能卓越的函数调用模型,基于 Llama-3 开发,并通过大量优化,特别适用于函数调用、对话及指令跟随等场景。',
|
11
|
-
displayName: 'Firefunction V2',
|
12
|
-
enabled: true,
|
13
|
-
id: 'accounts/fireworks/models/firefunction-v2',
|
14
|
-
type: 'chat',
|
15
|
-
},
|
16
|
-
{
|
17
|
-
abilities: {
|
18
|
-
functionCall: true,
|
19
|
-
},
|
20
|
-
contextWindowTokens: 32_768,
|
21
|
-
description: 'Fireworks 开源函数调用模型,提供卓越的指令执行能力和开放可定制的特性。',
|
22
|
-
displayName: 'Firefunction V1',
|
23
|
-
id: 'accounts/fireworks/models/firefunction-v1',
|
24
|
-
type: 'chat',
|
25
|
-
},
|
26
|
-
{
|
27
|
-
abilities: {
|
28
|
-
vision: true,
|
29
|
-
},
|
30
|
-
contextWindowTokens: 4096,
|
31
|
-
description:
|
32
|
-
'fireworks-ai/FireLLaVA-13b 是一款视觉语言模型,可以同时接收图像和文本输入,经过高质量数据训练,适合多模态任务。',
|
33
|
-
displayName: 'FireLLaVA-13B',
|
34
|
-
enabled: true,
|
35
|
-
id: 'accounts/fireworks/models/firellava-13b',
|
36
|
-
type: 'chat',
|
37
|
-
},
|
38
4
|
{
|
39
5
|
contextWindowTokens: 131_072,
|
40
6
|
description:
|
41
|
-
|
42
|
-
displayName: 'Llama 3.
|
7
|
+
'Llama 3.3 70B Instruct 是 Llama 3.1 70B 的 12 月更新版本。该模型在 Llama 3.1 70B(于 2024 年 7 月发布)的基础上进行了改进,增强了工具调用、多语言文本支持、数学和编程能力。该模型在推理、数学和指令遵循方面达到了行业领先水平,并且能够提供与 3.1 405B 相似的性能,同时在速度和成本上具有显著优势。',
|
8
|
+
displayName: 'Llama 3.3 70B Instruct',
|
43
9
|
enabled: true,
|
44
|
-
id: 'accounts/fireworks/models/llama-
|
10
|
+
id: 'accounts/fireworks/models/llama-v3p3-70b-instruct',
|
11
|
+
pricing: {
|
12
|
+
input: 0.9,
|
13
|
+
output: 0.9,
|
14
|
+
},
|
45
15
|
type: 'chat',
|
46
16
|
},
|
47
17
|
{
|
48
18
|
contextWindowTokens: 131_072,
|
49
19
|
description:
|
50
|
-
'Llama 3.2 3B
|
20
|
+
'Llama 3.2 3B Instruct 是 Meta 推出的轻量级多语言模型。该模型专为高效运行而设计,相较于更大型的模型,具有显著的延迟和成本优势。其典型应用场景包括查询和提示重写,以及写作辅助。',
|
51
21
|
displayName: 'Llama 3.2 3B Instruct',
|
52
22
|
enabled: true,
|
53
23
|
id: 'accounts/fireworks/models/llama-v3p2-3b-instruct',
|
24
|
+
pricing: {
|
25
|
+
input: 0.1,
|
26
|
+
output: 0.1,
|
27
|
+
},
|
54
28
|
type: 'chat',
|
55
29
|
},
|
56
30
|
{
|
@@ -59,10 +33,14 @@ const fireworksaiChatModels: AIChatModelCard[] = [
|
|
59
33
|
},
|
60
34
|
contextWindowTokens: 131_072,
|
61
35
|
description:
|
62
|
-
'Meta
|
36
|
+
'Meta 推出的指令微调图像推理模型,拥有 110 亿参数。该模型针对视觉识别、图像推理、图片字幕生成以及图片相关的常规问答进行了优化。它能够理解视觉数据,如图表和图形,并通过生成文本描述图像细节,弥合视觉与语言之间的鸿沟。',
|
63
37
|
displayName: 'Llama 3.2 11B Vision Instruct',
|
64
38
|
enabled: true,
|
65
39
|
id: 'accounts/fireworks/models/llama-v3p2-11b-vision-instruct',
|
40
|
+
pricing: {
|
41
|
+
input: 0.2,
|
42
|
+
output: 0.2,
|
43
|
+
},
|
66
44
|
type: 'chat',
|
67
45
|
},
|
68
46
|
{
|
@@ -71,91 +49,132 @@ const fireworksaiChatModels: AIChatModelCard[] = [
|
|
71
49
|
},
|
72
50
|
contextWindowTokens: 131_072,
|
73
51
|
description:
|
74
|
-
'Meta
|
52
|
+
'Meta 推出的指令微调图像推理模型,拥有 900 亿参数。该模型针对视觉识别、图像推理、图片字幕生成以及图片相关的常规问答进行了优化。它能够理解视觉数据,如图表和图形,并通过生成文本描述图像细节,弥合视觉与语言之间的鸿沟。注意:该模型目前作为无服务器模型进行实验性提供。如果用于生产环境,请注意 Fireworks 可能会在短时间内取消部署该模型。',
|
75
53
|
displayName: 'Llama 3.2 90B Vision Instruct',
|
76
54
|
enabled: true,
|
77
55
|
id: 'accounts/fireworks/models/llama-v3p2-90b-vision-instruct',
|
56
|
+
pricing: {
|
57
|
+
input: 0.9,
|
58
|
+
output: 0.9,
|
59
|
+
},
|
78
60
|
type: 'chat',
|
79
61
|
},
|
80
62
|
{
|
81
63
|
contextWindowTokens: 131_072,
|
82
64
|
description:
|
83
|
-
'Llama 3.1 8B
|
65
|
+
'Meta Llama 3.1 系列是多语言大语言模型(LLM)集合,包含 8B、70B 和 405B 三种参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型(8B、70B、405B)专为多语言对话应用优化,并在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。',
|
84
66
|
displayName: 'Llama 3.1 8B Instruct',
|
85
|
-
enabled: true,
|
86
67
|
id: 'accounts/fireworks/models/llama-v3p1-8b-instruct',
|
68
|
+
pricing: {
|
69
|
+
input: 0.2,
|
70
|
+
output: 0.2,
|
71
|
+
},
|
87
72
|
type: 'chat',
|
88
73
|
},
|
89
74
|
{
|
75
|
+
abilities: {
|
76
|
+
functionCall: true,
|
77
|
+
},
|
90
78
|
contextWindowTokens: 131_072,
|
91
79
|
description:
|
92
|
-
'Llama 3.1 70B
|
80
|
+
'Meta Llama 3.1 系列是多语言大语言模型(LLM)集合,包含 8B、70B 和 405B 三种参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型(8B、70B、405B)专为多语言对话应用优化,并在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。',
|
93
81
|
displayName: 'Llama 3.1 70B Instruct',
|
94
|
-
enabled: true,
|
95
82
|
id: 'accounts/fireworks/models/llama-v3p1-70b-instruct',
|
83
|
+
pricing: {
|
84
|
+
input: 0.9,
|
85
|
+
output: 0.9,
|
86
|
+
},
|
96
87
|
type: 'chat',
|
97
88
|
},
|
98
89
|
{
|
90
|
+
abilities: {
|
91
|
+
functionCall: true,
|
92
|
+
},
|
99
93
|
contextWindowTokens: 131_072,
|
100
94
|
description:
|
101
|
-
'Llama 3.1 405B
|
95
|
+
'Meta Llama 3.1 系列是多语言大语言模型(LLM)集合,包含 8B、70B 和 405B 参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型(8B、70B、405B)专为多语言对话场景优化,在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。405B 是 Llama 3.1 家族中能力最强的模型。该模型采用 FP8 进行推理,与参考实现高度匹配。',
|
102
96
|
displayName: 'Llama 3.1 405B Instruct',
|
103
|
-
enabled: true,
|
104
97
|
id: 'accounts/fireworks/models/llama-v3p1-405b-instruct',
|
98
|
+
pricing: {
|
99
|
+
input: 3,
|
100
|
+
output: 3,
|
101
|
+
},
|
105
102
|
type: 'chat',
|
106
103
|
},
|
107
104
|
{
|
108
105
|
contextWindowTokens: 8192,
|
109
|
-
description:
|
106
|
+
description:
|
107
|
+
'Meta 开发并发布了 Meta Llama 3 系列大语言模型(LLM),这是一个包含 8B 和 70B 参数规模的预训练和指令微调生成文本模型的集合。Llama 3 指令微调模型专为对话应用场景优化,并在常见的行业基准测试中优于许多现有的开源聊天模型。',
|
110
108
|
displayName: 'Llama 3 8B Instruct',
|
111
109
|
id: 'accounts/fireworks/models/llama-v3-8b-instruct',
|
110
|
+
pricing: {
|
111
|
+
input: 0.2,
|
112
|
+
output: 0.2,
|
113
|
+
},
|
112
114
|
type: 'chat',
|
113
115
|
},
|
114
116
|
{
|
115
117
|
contextWindowTokens: 8192,
|
116
|
-
description:
|
118
|
+
description:
|
119
|
+
'Meta 开发并发布了 Meta Llama 3 系列大语言模型(LLM),该系列包含 8B 和 70B 参数规模的预训练和指令微调生成文本模型。Llama 3 指令微调模型专为对话应用场景优化,并在常见的行业基准测试中优于许多现有的开源聊天模型。',
|
117
120
|
displayName: 'Llama 3 70B Instruct',
|
118
121
|
id: 'accounts/fireworks/models/llama-v3-70b-instruct',
|
122
|
+
pricing: {
|
123
|
+
input: 0.9,
|
124
|
+
output: 0.9,
|
125
|
+
},
|
119
126
|
type: 'chat',
|
120
127
|
},
|
121
128
|
{
|
122
129
|
contextWindowTokens: 8192,
|
123
130
|
description:
|
124
|
-
'Llama 3 8B
|
131
|
+
'Meta Llama 3 指令微调模型专为对话应用场景优化,并在常见的行业基准测试中优于许多现有的开源聊天模型。Llama 3 8B Instruct(HF 版本)是 Llama 3 8B Instruct 的原始 FP16 版本,其结果应与官方 Hugging Face 实现一致。',
|
125
132
|
displayName: 'Llama 3 8B Instruct (HF version)',
|
126
133
|
id: 'accounts/fireworks/models/llama-v3-8b-instruct-hf',
|
134
|
+
pricing: {
|
135
|
+
input: 0.2,
|
136
|
+
output: 0.2,
|
137
|
+
},
|
127
138
|
type: 'chat',
|
128
139
|
},
|
129
140
|
{
|
130
|
-
contextWindowTokens:
|
141
|
+
contextWindowTokens: 32_768,
|
131
142
|
description:
|
132
|
-
'
|
133
|
-
displayName: '
|
134
|
-
|
143
|
+
'24B 参数模型,具备与更大型模型相当的最先进能力。',
|
144
|
+
displayName: 'Mistral Small 3 Instruct',
|
145
|
+
enabled: true,
|
146
|
+
id: 'accounts/fireworks/models/mistral-small-24b-instruct-2501',
|
147
|
+
pricing: {
|
148
|
+
input: 0.9,
|
149
|
+
output: 0.9,
|
150
|
+
},
|
135
151
|
type: 'chat',
|
136
152
|
},
|
137
153
|
{
|
138
154
|
contextWindowTokens: 32_768,
|
139
|
-
description:
|
155
|
+
description:
|
156
|
+
'Mixtral MoE 8x7B Instruct 是 Mixtral MoE 8x7B 的指令微调版本,已启用聊天完成功能 API。',
|
140
157
|
displayName: 'Mixtral MoE 8x7B Instruct',
|
141
|
-
enabled: true,
|
142
158
|
id: 'accounts/fireworks/models/mixtral-8x7b-instruct',
|
159
|
+
pricing: {
|
160
|
+
input: 0.5,
|
161
|
+
output: 0.5,
|
162
|
+
},
|
143
163
|
type: 'chat',
|
144
164
|
},
|
145
165
|
{
|
166
|
+
abilities: {
|
167
|
+
functionCall: true,
|
168
|
+
},
|
146
169
|
contextWindowTokens: 65_536,
|
147
170
|
description:
|
148
|
-
'Mixtral MoE 8x22B
|
171
|
+
'Mixtral MoE 8x22B Instruct v0.1 是 Mixtral MoE 8x22B v0.1 的指令微调版本,已启用聊天完成功能 API。',
|
149
172
|
displayName: 'Mixtral MoE 8x22B Instruct',
|
150
|
-
enabled: true,
|
151
173
|
id: 'accounts/fireworks/models/mixtral-8x22b-instruct',
|
152
|
-
|
153
|
-
|
154
|
-
|
155
|
-
|
156
|
-
description: 'Mixtral MoE 8x7B 指令模型(HF 版本),性能与官方实现一致,适合多种高效任务场景。',
|
157
|
-
displayName: 'Mixtral MoE 8x7B Instruct (HF version)',
|
158
|
-
id: 'accounts/fireworks/models/mixtral-8x7b-instruct-hf',
|
174
|
+
pricing: {
|
175
|
+
input: 1.2,
|
176
|
+
output: 1.2,
|
177
|
+
},
|
159
178
|
type: 'chat',
|
160
179
|
},
|
161
180
|
{
|
@@ -164,65 +183,123 @@ const fireworksaiChatModels: AIChatModelCard[] = [
|
|
164
183
|
},
|
165
184
|
contextWindowTokens: 32_064,
|
166
185
|
description:
|
167
|
-
'Phi-3-Vision-128K-Instruct
|
186
|
+
'Phi-3-Vision-128K-Instruct 是一个轻量级的、最先进的开放多模态模型,基于包括合成数据和筛选后的公开网站数据集构建,重点关注文本和视觉方面的高质量、推理密集型数据。该模型属于 Phi-3 模型家族,其多模态版本支持 128K 上下文长度(以标记为单位)。该模型经过严格的增强过程,包括监督微调和直接偏好优化,以确保精确的指令遵循和强大的安全措施。',
|
168
187
|
displayName: 'Phi 3.5 Vision Instruct',
|
169
188
|
enabled: true,
|
170
189
|
id: 'accounts/fireworks/models/phi-3-vision-128k-instruct',
|
190
|
+
pricing: {
|
191
|
+
input: 0.2,
|
192
|
+
output: 0.2,
|
193
|
+
},
|
171
194
|
type: 'chat',
|
172
195
|
},
|
173
196
|
{
|
174
197
|
contextWindowTokens: 32_768,
|
175
|
-
description:
|
176
|
-
|
198
|
+
description:
|
199
|
+
'MythoMix 的改进版,可能是其更为完善的变体,是 MythoLogic-L2 和 Huginn 的合并,采用了高度实验性的张量类型合并技术。由于其独特的性质,该模型在讲故事和角色扮演方面表现出色。',
|
200
|
+
displayName: 'MythoMax L2 13b',
|
201
|
+
id: 'accounts/fireworks/models/mythomax-l2-13b',
|
202
|
+
pricing: {
|
203
|
+
input: 0.2,
|
204
|
+
output: 0.2,
|
205
|
+
},
|
206
|
+
type: 'chat',
|
207
|
+
},
|
208
|
+
{
|
209
|
+
contextWindowTokens: 131_072,
|
210
|
+
description:
|
211
|
+
'Deepseek 提供的强大 Mixture-of-Experts (MoE) 语言模型,总参数量为 671B,每个标记激活 37B 参数。',
|
212
|
+
displayName: 'Deepseek V3',
|
177
213
|
enabled: true,
|
178
|
-
id: 'accounts/fireworks/models/
|
214
|
+
id: 'accounts/fireworks/models/deepseek-v3',
|
215
|
+
pricing: {
|
216
|
+
input: 0.9,
|
217
|
+
output: 0.9,
|
218
|
+
},
|
179
219
|
type: 'chat',
|
180
220
|
},
|
181
221
|
{
|
182
|
-
|
222
|
+
abilities: {
|
223
|
+
reasoning: true,
|
224
|
+
},
|
225
|
+
contextWindowTokens: 163_840,
|
183
226
|
description:
|
184
|
-
'
|
185
|
-
displayName: '
|
227
|
+
'DeepSeek-R1 是一款最先进的大型语言模型,经过强化学习和冷启动数据的优化,具有出色的推理、数学和编程性能。',
|
228
|
+
displayName: 'Deepseek R1',
|
186
229
|
enabled: true,
|
187
|
-
id: 'accounts/fireworks/models/
|
230
|
+
id: 'accounts/fireworks/models/deepseek-r1',
|
231
|
+
pricing: {
|
232
|
+
input: 8,
|
233
|
+
output: 8,
|
234
|
+
},
|
188
235
|
type: 'chat',
|
189
236
|
},
|
190
237
|
{
|
191
238
|
contextWindowTokens: 32_768,
|
192
239
|
description:
|
193
|
-
'
|
194
|
-
displayName: '
|
195
|
-
enabled:
|
196
|
-
id: 'accounts/fireworks/models/
|
240
|
+
'Qwen QwQ 模型专注于推动 AI 推理,并展示了开放模型在推理能力上与闭源前沿模型匹敌的力量。QwQ-32B-Preview 是一个实验性发布版本,在 GPQA、AIME、MATH-500 和 LiveCodeBench 基准测试中,在分析和推理能力上可与 o1 相媲美,并超越 GPT-4o 和 Claude 3.5 Sonnet。注意:该模型目前作为无服务器模型进行实验性提供。如果用于生产环境,请注意 Fireworks 可能会在短时间内取消部署该模型。',
|
241
|
+
displayName: 'Qwen Qwq 32b Preview',
|
242
|
+
enabled: true,
|
243
|
+
id: 'accounts/fireworks/models/qwen-qwq-32b-preview',
|
244
|
+
pricing: {
|
245
|
+
input: 0.9,
|
246
|
+
output: 0.9,
|
247
|
+
},
|
197
248
|
type: 'chat',
|
198
249
|
},
|
199
250
|
{
|
200
251
|
contextWindowTokens: 32_768,
|
201
|
-
description:
|
202
|
-
|
252
|
+
description:
|
253
|
+
'Qwen2.5 是由 Qwen 团队和阿里云开发的一系列仅解码语言模型,提供 0.5B、1.5B、3B、7B、14B、32B 和 72B 不同参数规模,并包含基础版和指令微调版。',
|
254
|
+
displayName: 'Qwen2.5 72B Instruct',
|
203
255
|
enabled: true,
|
204
|
-
id: 'accounts/
|
256
|
+
id: 'accounts/fireworks/models/qwen2p5-72b-instruct',
|
257
|
+
pricing: {
|
258
|
+
input: 0.9,
|
259
|
+
output: 0.9,
|
260
|
+
},
|
205
261
|
type: 'chat',
|
206
262
|
},
|
207
263
|
{
|
208
|
-
|
209
|
-
|
210
|
-
|
211
|
-
|
264
|
+
abilities: {
|
265
|
+
vision: true,
|
266
|
+
},
|
267
|
+
contextWindowTokens: 32_768,
|
268
|
+
description:
|
269
|
+
'Qwen-VL 模型的 72B 版本是阿里巴巴最新迭代的成果,代表了近一年的创新。',
|
270
|
+
displayName: 'Qwen2 VL 72B Instruct',
|
271
|
+
enabled: true,
|
272
|
+
id: 'accounts/fireworks/models/qwen2-vl-72b-instruct',
|
273
|
+
pricing: {
|
274
|
+
input: 0.9,
|
275
|
+
output: 0.9,
|
276
|
+
},
|
212
277
|
type: 'chat',
|
213
278
|
},
|
214
279
|
{
|
215
|
-
contextWindowTokens:
|
216
|
-
description:
|
217
|
-
|
218
|
-
|
280
|
+
contextWindowTokens: 32_768,
|
281
|
+
description:
|
282
|
+
'Qwen2.5-Coder 是最新一代专为代码设计的 Qwen 大型语言模型(前称为 CodeQwen)。注意:该模型目前作为无服务器模型进行实验性提供。如果用于生产环境,请注意 Fireworks 可能会在短时间内取消部署该模型。',
|
283
|
+
displayName: 'Qwen2.5-Coder-32B-Instruct',
|
284
|
+
enabled: true,
|
285
|
+
id: 'accounts/fireworks/models/qwen2p5-coder-32b-instruct',
|
286
|
+
pricing: {
|
287
|
+
input: 0.9,
|
288
|
+
output: 0.9,
|
289
|
+
},
|
219
290
|
type: 'chat',
|
220
291
|
},
|
221
292
|
{
|
222
|
-
contextWindowTokens:
|
223
|
-
description:
|
224
|
-
|
225
|
-
|
293
|
+
contextWindowTokens: 32_768,
|
294
|
+
description:
|
295
|
+
'Yi-Large 是顶尖的大型语言模型之一,在 LMSYS 基准测试排行榜上,其表现仅次于 GPT-4、Gemini 1.5 Pro 和 Claude 3 Opus。它在多语言能力方面表现卓越,特别是在西班牙语、中文、日语、德语和法语方面。Yi-Large 还具有用户友好性,采用与 OpenAI 相同的 API 定义,便于集成。',
|
296
|
+
displayName: 'Yi-Large',
|
297
|
+
enabled: true,
|
298
|
+
id: 'accounts/yi-01-ai/models/yi-large',
|
299
|
+
pricing: {
|
300
|
+
input: 3,
|
301
|
+
output: 3,
|
302
|
+
},
|
226
303
|
type: 'chat',
|
227
304
|
},
|
228
305
|
];
|
@@ -2,62 +2,55 @@ import { AIChatModelCard } from '@/types/aiModel';
|
|
2
2
|
|
3
3
|
const perplexityChatModels: AIChatModelCard[] = [
|
4
4
|
{
|
5
|
-
contextWindowTokens:
|
5
|
+
contextWindowTokens: 127_072,
|
6
6
|
description:
|
7
|
-
'
|
8
|
-
displayName: '
|
9
|
-
enabled: true,
|
10
|
-
id: 'llama-3.1-sonar-small-128k-online',
|
11
|
-
type: 'chat',
|
12
|
-
},
|
13
|
-
{
|
14
|
-
contextWindowTokens: 128_000,
|
15
|
-
description:
|
16
|
-
'Llama 3.1 Sonar Large Online 模型,具备70B参数,支持约127,000个标记的上下文长度,适用于高容量和多样化聊天任务。',
|
17
|
-
displayName: 'Llama 3.1 Sonar Large Online',
|
7
|
+
'由 DeepSeek 推理模型提供支持的新 API 产品。',
|
8
|
+
displayName: 'Sonar Reasoning',
|
18
9
|
enabled: true,
|
19
|
-
id: '
|
10
|
+
id: 'sonar-reasoning',
|
20
11
|
type: 'chat',
|
21
12
|
},
|
22
13
|
{
|
23
|
-
contextWindowTokens:
|
14
|
+
contextWindowTokens: 200_000,
|
24
15
|
description:
|
25
|
-
'
|
26
|
-
displayName: '
|
16
|
+
'支持搜索上下文的高级搜索产品,支持高级查询和跟进。',
|
17
|
+
displayName: 'Sonar Pro',
|
27
18
|
enabled: true,
|
28
|
-
id: '
|
19
|
+
id: 'sonar-pro',
|
29
20
|
type: 'chat',
|
30
21
|
},
|
31
22
|
{
|
32
|
-
contextWindowTokens:
|
23
|
+
contextWindowTokens: 127_072,
|
33
24
|
description:
|
34
|
-
'
|
35
|
-
displayName: '
|
25
|
+
'基于搜索上下文的轻量级搜索产品,比 Sonar Pro 更快、更便宜。',
|
26
|
+
displayName: 'Sonar',
|
36
27
|
enabled: true,
|
37
|
-
id: '
|
28
|
+
id: 'sonar',
|
38
29
|
type: 'chat',
|
39
30
|
},
|
31
|
+
// The following will be deprecated on 02-22
|
40
32
|
{
|
41
|
-
contextWindowTokens:
|
33
|
+
contextWindowTokens: 127_072,
|
42
34
|
description:
|
43
|
-
'Llama 3.1 Sonar
|
44
|
-
displayName: 'Llama 3.1 Sonar
|
45
|
-
|
46
|
-
id: 'llama-3.1-sonar-large-128k-chat',
|
35
|
+
'Llama 3.1 Sonar Small Online 模型,具备8B参数,支持约127,000个标记的上下文长度,专为在线聊天设计,能高效处理各种文本交互。',
|
36
|
+
displayName: 'Llama 3.1 Sonar Small Online',
|
37
|
+
id: 'llama-3.1-sonar-small-128k-online',
|
47
38
|
type: 'chat',
|
48
39
|
},
|
49
40
|
{
|
50
|
-
contextWindowTokens:
|
41
|
+
contextWindowTokens: 127_072,
|
51
42
|
description:
|
52
|
-
'Llama 3.1
|
53
|
-
|
43
|
+
'Llama 3.1 Sonar Large Online 模型,具备70B参数,支持约127,000个标记的上下文长度,适用于高容量和多样化聊天任务。',
|
44
|
+
displayName: 'Llama 3.1 Sonar Large Online',
|
45
|
+
id: 'llama-3.1-sonar-large-128k-online',
|
54
46
|
type: 'chat',
|
55
47
|
},
|
56
48
|
{
|
57
|
-
contextWindowTokens:
|
49
|
+
contextWindowTokens: 127_072,
|
58
50
|
description:
|
59
|
-
'Llama 3.1
|
60
|
-
|
51
|
+
'Llama 3.1 Sonar Huge Online 模型,具备405B参数,支持约127,000个标记的上下文长度,设计用于复杂的在线聊天应用。',
|
52
|
+
displayName: 'Llama 3.1 Sonar Huge Online',
|
53
|
+
id: 'llama-3.1-sonar-huge-128k-online',
|
61
54
|
type: 'chat',
|
62
55
|
},
|
63
56
|
];
|
@@ -4,214 +4,270 @@ import { ModelProviderCard } from '@/types/llm';
|
|
4
4
|
// ref: https://fireworks.ai/pricing
|
5
5
|
const FireworksAI: ModelProviderCard = {
|
6
6
|
chatModels: [
|
7
|
-
{
|
8
|
-
contextWindowTokens: 8192,
|
9
|
-
description:
|
10
|
-
'Fireworks 公司最新推出的 Firefunction-v2 是一款性能卓越的函数调用模型,基于 Llama-3 开发,并通过大量优化,特别适用于函数调用、对话及指令跟随等场景。',
|
11
|
-
displayName: 'Firefunction V2',
|
12
|
-
enabled: true,
|
13
|
-
functionCall: true,
|
14
|
-
id: 'accounts/fireworks/models/firefunction-v2',
|
15
|
-
},
|
16
|
-
{
|
17
|
-
contextWindowTokens: 32_768,
|
18
|
-
description: 'Fireworks 开源函数调用模型,提供卓越的指令执行能力和开放可定制的特性。',
|
19
|
-
displayName: 'Firefunction V1',
|
20
|
-
functionCall: true,
|
21
|
-
id: 'accounts/fireworks/models/firefunction-v1',
|
22
|
-
},
|
23
|
-
{
|
24
|
-
contextWindowTokens: 4096,
|
25
|
-
description:
|
26
|
-
'fireworks-ai/FireLLaVA-13b 是一款视觉语言模型,可以同时接收图像和文本输入,经过高质量数据训练,适合多模态任务。',
|
27
|
-
displayName: 'FireLLaVA-13B',
|
28
|
-
enabled: true,
|
29
|
-
functionCall: false,
|
30
|
-
id: 'accounts/fireworks/models/firellava-13b',
|
31
|
-
vision: true,
|
32
|
-
},
|
33
7
|
{
|
34
8
|
contextWindowTokens: 131_072,
|
35
9
|
description:
|
36
|
-
'Llama 3.
|
37
|
-
displayName: 'Llama 3.
|
10
|
+
'Llama 3.3 70B Instruct 是 Llama 3.1 70B 的 12 月更新版本。该模型在 Llama 3.1 70B(于 2024 年 7 月发布)的基础上进行了改进,增强了工具调用、多语言文本支持、数学和编程能力。该模型在推理、数学和指令遵循方面达到了行业领先水平,并且能够提供与 3.1 405B 相似的性能,同时在速度和成本上具有显著优势。',
|
11
|
+
displayName: 'Llama 3.3 70B Instruct',
|
38
12
|
enabled: true,
|
39
|
-
id: 'accounts/fireworks/models/llama-
|
13
|
+
id: 'accounts/fireworks/models/llama-v3p3-70b-instruct',
|
14
|
+
pricing: {
|
15
|
+
input: 0.9,
|
16
|
+
output: 0.9,
|
17
|
+
},
|
40
18
|
},
|
41
19
|
{
|
42
20
|
contextWindowTokens: 131_072,
|
43
21
|
description:
|
44
|
-
'Llama 3.2 3B
|
22
|
+
'Llama 3.2 3B Instruct 是 Meta 推出的轻量级多语言模型。该模型专为高效运行而设计,相较于更大型的模型,具有显著的延迟和成本优势。其典型应用场景包括查询和提示重写,以及写作辅助。',
|
45
23
|
displayName: 'Llama 3.2 3B Instruct',
|
46
24
|
enabled: true,
|
47
25
|
id: 'accounts/fireworks/models/llama-v3p2-3b-instruct',
|
26
|
+
pricing: {
|
27
|
+
input: 0.1,
|
28
|
+
output: 0.1,
|
29
|
+
},
|
48
30
|
},
|
49
31
|
{
|
50
32
|
contextWindowTokens: 131_072,
|
51
33
|
description:
|
52
|
-
'Meta
|
34
|
+
'Meta 推出的指令微调图像推理模型,拥有 110 亿参数。该模型针对视觉识别、图像推理、图片字幕生成以及图片相关的常规问答进行了优化。它能够理解视觉数据,如图表和图形,并通过生成文本描述图像细节,弥合视觉与语言之间的鸿沟。',
|
53
35
|
displayName: 'Llama 3.2 11B Vision Instruct',
|
54
36
|
enabled: true,
|
55
37
|
id: 'accounts/fireworks/models/llama-v3p2-11b-vision-instruct',
|
38
|
+
pricing: {
|
39
|
+
input: 0.2,
|
40
|
+
output: 0.2,
|
41
|
+
},
|
56
42
|
vision: true,
|
57
43
|
},
|
58
44
|
{
|
59
45
|
contextWindowTokens: 131_072,
|
60
46
|
description:
|
61
|
-
'Meta
|
47
|
+
'Meta 推出的指令微调图像推理模型,拥有 900 亿参数。该模型针对视觉识别、图像推理、图片字幕生成以及图片相关的常规问答进行了优化。它能够理解视觉数据,如图表和图形,并通过生成文本描述图像细节,弥合视觉与语言之间的鸿沟。注意:该模型目前作为无服务器模型进行实验性提供。如果用于生产环境,请注意 Fireworks 可能会在短时间内取消部署该模型。',
|
62
48
|
displayName: 'Llama 3.2 90B Vision Instruct',
|
63
49
|
enabled: true,
|
64
50
|
id: 'accounts/fireworks/models/llama-v3p2-90b-vision-instruct',
|
51
|
+
pricing: {
|
52
|
+
input: 0.9,
|
53
|
+
output: 0.9,
|
54
|
+
},
|
65
55
|
vision: true,
|
66
56
|
},
|
67
57
|
{
|
68
58
|
contextWindowTokens: 131_072,
|
69
59
|
description:
|
70
|
-
'Llama 3.1 8B
|
60
|
+
'Meta Llama 3.1 系列是多语言大语言模型(LLM)集合,包含 8B、70B 和 405B 三种参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型(8B、70B、405B)专为多语言对话应用优化,并在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。',
|
71
61
|
displayName: 'Llama 3.1 8B Instruct',
|
72
|
-
enabled: true,
|
73
|
-
functionCall: false,
|
74
62
|
id: 'accounts/fireworks/models/llama-v3p1-8b-instruct',
|
63
|
+
pricing: {
|
64
|
+
input: 0.2,
|
65
|
+
output: 0.2,
|
66
|
+
},
|
75
67
|
},
|
76
68
|
{
|
77
69
|
contextWindowTokens: 131_072,
|
78
70
|
description:
|
79
|
-
'Llama 3.1 70B
|
71
|
+
'Meta Llama 3.1 系列是多语言大语言模型(LLM)集合,包含 8B、70B 和 405B 三种参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型(8B、70B、405B)专为多语言对话应用优化,并在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。',
|
80
72
|
displayName: 'Llama 3.1 70B Instruct',
|
81
|
-
|
82
|
-
functionCall: false,
|
73
|
+
functionCall: true,
|
83
74
|
id: 'accounts/fireworks/models/llama-v3p1-70b-instruct',
|
75
|
+
pricing: {
|
76
|
+
input: 0.9,
|
77
|
+
output: 0.9,
|
78
|
+
},
|
84
79
|
},
|
85
80
|
{
|
86
81
|
contextWindowTokens: 131_072,
|
87
82
|
description:
|
88
|
-
'Llama 3.1 405B
|
83
|
+
'Meta Llama 3.1 系列是多语言大语言模型(LLM)集合,包含 8B、70B 和 405B 参数规模的预训练和指令微调生成模型。Llama 3.1 指令微调文本模型(8B、70B、405B)专为多语言对话场景优化,在常见的行业基准测试中优于许多现有的开源和闭源聊天模型。405B 是 Llama 3.1 家族中能力最强的模型。该模型采用 FP8 进行推理,与参考实现高度匹配。',
|
89
84
|
displayName: 'Llama 3.1 405B Instruct',
|
90
|
-
|
91
|
-
functionCall: false,
|
85
|
+
functionCall: true,
|
92
86
|
id: 'accounts/fireworks/models/llama-v3p1-405b-instruct',
|
87
|
+
pricing: {
|
88
|
+
input: 3,
|
89
|
+
output: 3,
|
90
|
+
},
|
93
91
|
},
|
94
92
|
{
|
95
93
|
contextWindowTokens: 8192,
|
96
|
-
description:
|
94
|
+
description:
|
95
|
+
'Meta 开发并发布了 Meta Llama 3 系列大语言模型(LLM),这是一个包含 8B 和 70B 参数规模的预训练和指令微调生成文本模型的集合。Llama 3 指令微调模型专为对话应用场景优化,并在常见的行业基准测试中优于许多现有的开源聊天模型。',
|
97
96
|
displayName: 'Llama 3 8B Instruct',
|
98
|
-
functionCall: false,
|
99
97
|
id: 'accounts/fireworks/models/llama-v3-8b-instruct',
|
98
|
+
pricing: {
|
99
|
+
input: 0.2,
|
100
|
+
output: 0.2,
|
101
|
+
},
|
100
102
|
},
|
101
103
|
{
|
102
104
|
contextWindowTokens: 8192,
|
103
|
-
description:
|
105
|
+
description:
|
106
|
+
'Meta 开发并发布了 Meta Llama 3 系列大语言模型(LLM),该系列包含 8B 和 70B 参数规模的预训练和指令微调生成文本模型。Llama 3 指令微调模型专为对话应用场景优化,并在常见的行业基准测试中优于许多现有的开源聊天模型。',
|
104
107
|
displayName: 'Llama 3 70B Instruct',
|
105
|
-
functionCall: false,
|
106
108
|
id: 'accounts/fireworks/models/llama-v3-70b-instruct',
|
109
|
+
pricing: {
|
110
|
+
input: 0.9,
|
111
|
+
output: 0.9,
|
112
|
+
},
|
107
113
|
},
|
108
114
|
{
|
109
115
|
contextWindowTokens: 8192,
|
110
116
|
description:
|
111
|
-
'Llama 3 8B
|
117
|
+
'Meta Llama 3 指令微调模型专为对话应用场景优化,并在常见的行业基准测试中优于许多现有的开源聊天模型。Llama 3 8B Instruct(HF 版本)是 Llama 3 8B Instruct 的原始 FP16 版本,其结果应与官方 Hugging Face 实现一致。',
|
112
118
|
displayName: 'Llama 3 8B Instruct (HF version)',
|
113
|
-
functionCall: false,
|
114
119
|
id: 'accounts/fireworks/models/llama-v3-8b-instruct-hf',
|
120
|
+
pricing: {
|
121
|
+
input: 0.2,
|
122
|
+
output: 0.2,
|
123
|
+
},
|
115
124
|
},
|
116
125
|
{
|
117
|
-
contextWindowTokens:
|
126
|
+
contextWindowTokens: 32_768,
|
118
127
|
description:
|
119
|
-
'
|
120
|
-
displayName: '
|
121
|
-
|
122
|
-
id: 'accounts/fireworks/models/
|
128
|
+
'24B 参数模型,具备与更大型模型相当的最先进能力。',
|
129
|
+
displayName: 'Mistral Small 3 Instruct',
|
130
|
+
enabled: true,
|
131
|
+
id: 'accounts/fireworks/models/mistral-small-24b-instruct-2501',
|
132
|
+
pricing: {
|
133
|
+
input: 0.9,
|
134
|
+
output: 0.9,
|
135
|
+
},
|
123
136
|
},
|
124
137
|
{
|
125
138
|
contextWindowTokens: 32_768,
|
126
|
-
description:
|
139
|
+
description:
|
140
|
+
'Mixtral MoE 8x7B Instruct 是 Mixtral MoE 8x7B 的指令微调版本,已启用聊天完成功能 API。',
|
127
141
|
displayName: 'Mixtral MoE 8x7B Instruct',
|
128
|
-
enabled: true,
|
129
|
-
functionCall: false,
|
130
142
|
id: 'accounts/fireworks/models/mixtral-8x7b-instruct',
|
143
|
+
pricing: {
|
144
|
+
input: 0.5,
|
145
|
+
output: 0.5,
|
146
|
+
},
|
131
147
|
},
|
132
148
|
{
|
133
149
|
contextWindowTokens: 65_536,
|
134
150
|
description:
|
135
|
-
'Mixtral MoE 8x22B
|
151
|
+
'Mixtral MoE 8x22B Instruct v0.1 是 Mixtral MoE 8x22B v0.1 的指令微调版本,已启用聊天完成功能 API。',
|
136
152
|
displayName: 'Mixtral MoE 8x22B Instruct',
|
137
|
-
|
138
|
-
functionCall: false,
|
153
|
+
functionCall: true,
|
139
154
|
id: 'accounts/fireworks/models/mixtral-8x22b-instruct',
|
140
|
-
|
141
|
-
|
142
|
-
|
143
|
-
|
144
|
-
'Mixtral MoE 8x7B 指令模型(HF 版本),性能与官方实现一致,适合多种高效任务场景。',
|
145
|
-
displayName: 'Mixtral MoE 8x7B Instruct (HF version)',
|
146
|
-
functionCall: false,
|
147
|
-
id: 'accounts/fireworks/models/mixtral-8x7b-instruct-hf',
|
155
|
+
pricing: {
|
156
|
+
input: 1.2,
|
157
|
+
output: 1.2,
|
158
|
+
},
|
148
159
|
},
|
149
160
|
{
|
150
161
|
contextWindowTokens: 32_064,
|
151
162
|
description:
|
152
|
-
'Phi-3-Vision-128K-Instruct
|
163
|
+
'Phi-3-Vision-128K-Instruct 是一个轻量级的、最先进的开放多模态模型,基于包括合成数据和筛选后的公开网站数据集构建,重点关注文本和视觉方面的高质量、推理密集型数据。该模型属于 Phi-3 模型家族,其多模态版本支持 128K 上下文长度(以标记为单位)。该模型经过严格的增强过程,包括监督微调和直接偏好优化,以确保精确的指令遵循和强大的安全措施。',
|
153
164
|
displayName: 'Phi 3.5 Vision Instruct',
|
154
165
|
enabled: true,
|
155
|
-
functionCall: false,
|
156
166
|
id: 'accounts/fireworks/models/phi-3-vision-128k-instruct',
|
167
|
+
pricing: {
|
168
|
+
input: 0.2,
|
169
|
+
output: 0.2,
|
170
|
+
},
|
157
171
|
vision: true,
|
158
172
|
},
|
159
173
|
{
|
160
174
|
contextWindowTokens: 32_768,
|
161
|
-
description:
|
162
|
-
|
175
|
+
description:
|
176
|
+
'MythoMix 的改进版,可能是其更为完善的变体,是 MythoLogic-L2 和 Huginn 的合并,采用了高度实验性的张量类型合并技术。由于其独特的性质,该模型在讲故事和角色扮演方面表现出色。',
|
177
|
+
displayName: 'MythoMax L2 13b',
|
178
|
+
id: 'accounts/fireworks/models/mythomax-l2-13b',
|
179
|
+
pricing: {
|
180
|
+
input: 0.2,
|
181
|
+
output: 0.2,
|
182
|
+
},
|
183
|
+
},
|
184
|
+
{
|
185
|
+
contextWindowTokens: 131_072,
|
186
|
+
description:
|
187
|
+
'Deepseek 提供的强大 Mixture-of-Experts (MoE) 语言模型,总参数量为 671B,每个标记激活 37B 参数。',
|
188
|
+
displayName: 'Deepseek V3',
|
189
|
+
enabled: true,
|
190
|
+
id: 'accounts/fireworks/models/deepseek-v3',
|
191
|
+
pricing: {
|
192
|
+
input: 0.9,
|
193
|
+
output: 0.9,
|
194
|
+
},
|
195
|
+
},
|
196
|
+
{
|
197
|
+
contextWindowTokens: 163_840,
|
198
|
+
description:
|
199
|
+
'DeepSeek-R1 是一款最先进的大型语言模型,经过强化学习和冷启动数据的优化,具有出色的推理、数学和编程性能。',
|
200
|
+
displayName: 'Deepseek R1',
|
201
|
+
enabled: true,
|
202
|
+
id: 'accounts/fireworks/models/deepseek-r1',
|
203
|
+
pricing: {
|
204
|
+
input: 8,
|
205
|
+
output: 8,
|
206
|
+
},
|
207
|
+
},
|
208
|
+
{
|
209
|
+
contextWindowTokens: 32_768,
|
210
|
+
description:
|
211
|
+
'Qwen QwQ 模型专注于推动 AI 推理,并展示了开放模型在推理能力上与闭源前沿模型匹敌的力量。QwQ-32B-Preview 是一个实验性发布版本,在 GPQA、AIME、MATH-500 和 LiveCodeBench 基准测试中,在分析和推理能力上可与 o1 相媲美,并超越 GPT-4o 和 Claude 3.5 Sonnet。注意:该模型目前作为无服务器模型进行实验性提供。如果用于生产环境,请注意 Fireworks 可能会在短时间内取消部署该模型。',
|
212
|
+
displayName: 'Qwen Qwq 32b Preview',
|
163
213
|
enabled: true,
|
164
214
|
id: 'accounts/fireworks/models/qwen-qwq-32b-preview',
|
215
|
+
pricing: {
|
216
|
+
input: 0.9,
|
217
|
+
output: 0.9,
|
218
|
+
},
|
165
219
|
},
|
166
220
|
{
|
167
221
|
contextWindowTokens: 32_768,
|
168
222
|
description:
|
169
|
-
'Qwen2.5
|
223
|
+
'Qwen2.5 是由 Qwen 团队和阿里云开发的一系列仅解码语言模型,提供 0.5B、1.5B、3B、7B、14B、32B 和 72B 不同参数规模,并包含基础版和指令微调版。',
|
170
224
|
displayName: 'Qwen2.5 72B Instruct',
|
171
225
|
enabled: true,
|
172
|
-
functionCall: false,
|
173
226
|
id: 'accounts/fireworks/models/qwen2p5-72b-instruct',
|
227
|
+
pricing: {
|
228
|
+
input: 0.9,
|
229
|
+
output: 0.9,
|
230
|
+
},
|
174
231
|
},
|
175
232
|
{
|
176
233
|
contextWindowTokens: 32_768,
|
177
234
|
description:
|
178
|
-
'
|
179
|
-
displayName: 'Qwen2
|
180
|
-
enabled:
|
181
|
-
id: 'accounts/fireworks/models/
|
235
|
+
'Qwen-VL 模型的 72B 版本是阿里巴巴最新迭代的成果,代表了近一年的创新。',
|
236
|
+
displayName: 'Qwen2 VL 72B Instruct',
|
237
|
+
enabled: true,
|
238
|
+
id: 'accounts/fireworks/models/qwen2-vl-72b-instruct',
|
239
|
+
pricing: {
|
240
|
+
input: 0.9,
|
241
|
+
output: 0.9,
|
242
|
+
},
|
243
|
+
vision: true,
|
182
244
|
},
|
183
245
|
{
|
184
246
|
contextWindowTokens: 32_768,
|
185
|
-
description:
|
186
|
-
|
247
|
+
description:
|
248
|
+
'Qwen2.5-Coder 是最新一代专为代码设计的 Qwen 大型语言模型(前称为 CodeQwen)。注意:该模型目前作为无服务器模型进行实验性提供。如果用于生产环境,请注意 Fireworks 可能会在短时间内取消部署该模型。',
|
249
|
+
displayName: 'Qwen2.5-Coder-32B-Instruct',
|
187
250
|
enabled: true,
|
188
|
-
|
189
|
-
|
190
|
-
|
191
|
-
|
192
|
-
|
193
|
-
description: 'StarCoder 7B 模型,针对80多种编程语言训练,拥有出色的编程填充能力和语境理解。',
|
194
|
-
displayName: 'StarCoder 7B',
|
195
|
-
functionCall: false,
|
196
|
-
id: 'accounts/fireworks/models/starcoder-7b',
|
251
|
+
id: 'accounts/fireworks/models/qwen2p5-coder-32b-instruct',
|
252
|
+
pricing: {
|
253
|
+
input: 0.9,
|
254
|
+
output: 0.9,
|
255
|
+
},
|
197
256
|
},
|
198
257
|
{
|
199
|
-
contextWindowTokens:
|
258
|
+
contextWindowTokens: 32_768,
|
200
259
|
description:
|
201
|
-
'
|
202
|
-
displayName: '
|
203
|
-
|
204
|
-
id: 'accounts/
|
205
|
-
|
206
|
-
|
207
|
-
|
208
|
-
|
209
|
-
displayName: 'MythoMax L2 13b',
|
210
|
-
functionCall: false,
|
211
|
-
id: 'accounts/fireworks/models/mythomax-l2-13b',
|
260
|
+
'Yi-Large 是顶尖的大型语言模型之一,在 LMSYS 基准测试排行榜上,其表现仅次于 GPT-4、Gemini 1.5 Pro 和 Claude 3 Opus。它在多语言能力方面表现卓越,特别是在西班牙语、中文、日语、德语和法语方面。Yi-Large 还具有用户友好性,采用与 OpenAI 相同的 API 定义,便于集成。',
|
261
|
+
displayName: 'Yi-Large',
|
262
|
+
enabled: true,
|
263
|
+
id: 'accounts/yi-01-ai/models/yi-large',
|
264
|
+
pricing: {
|
265
|
+
input: 3,
|
266
|
+
output: 3,
|
267
|
+
},
|
212
268
|
},
|
213
269
|
],
|
214
|
-
checkModel: 'accounts/fireworks/models/
|
270
|
+
checkModel: 'accounts/fireworks/models/llama-v3p2-3b-instruct',
|
215
271
|
description:
|
216
272
|
'Fireworks AI 是一家领先的高级语言模型服务商,专注于功能调用和多模态处理。其最新模型 Firefunction V2 基于 Llama-3,优化用于函数调用、对话及指令跟随。视觉语言模型 FireLLaVA-13B 支持图像和文本混合输入。其他 notable 模型包括 Llama 系列和 Mixtral 系列,提供高效的多语言指令跟随与生成支持。',
|
217
273
|
id: 'fireworksai',
|
@@ -4,59 +4,53 @@ import { ModelProviderCard } from '@/types/llm';
|
|
4
4
|
const Perplexity: ModelProviderCard = {
|
5
5
|
chatModels: [
|
6
6
|
{
|
7
|
-
contextWindowTokens:
|
7
|
+
contextWindowTokens: 127_072,
|
8
8
|
description:
|
9
|
-
'
|
10
|
-
displayName: '
|
9
|
+
'由 DeepSeek 推理模型提供支持的新 API 产品。',
|
10
|
+
displayName: 'Sonar Reasoning',
|
11
11
|
enabled: true,
|
12
|
-
id: '
|
12
|
+
id: 'sonar-reasoning',
|
13
13
|
},
|
14
14
|
{
|
15
|
-
contextWindowTokens:
|
15
|
+
contextWindowTokens: 200_000,
|
16
16
|
description:
|
17
|
-
'
|
18
|
-
displayName: '
|
17
|
+
'支持搜索上下文的高级搜索产品,支持高级查询和跟进。',
|
18
|
+
displayName: 'Sonar Pro',
|
19
19
|
enabled: true,
|
20
|
-
id: '
|
20
|
+
id: 'sonar-pro',
|
21
21
|
},
|
22
22
|
{
|
23
|
-
contextWindowTokens:
|
23
|
+
contextWindowTokens: 127_072,
|
24
24
|
description:
|
25
|
-
'
|
26
|
-
displayName: '
|
25
|
+
'基于搜索上下文的轻量级搜索产品,比 Sonar Pro 更快、更便宜。',
|
26
|
+
displayName: 'Sonar',
|
27
27
|
enabled: true,
|
28
|
-
id: '
|
28
|
+
id: 'sonar',
|
29
29
|
},
|
30
|
+
// The following will be deprecated on 02-22
|
30
31
|
{
|
31
|
-
contextWindowTokens:
|
32
|
+
contextWindowTokens: 127_072,
|
32
33
|
description:
|
33
|
-
'Llama 3.1 Sonar Small
|
34
|
-
displayName: 'Llama 3.1 Sonar Small
|
35
|
-
|
36
|
-
id: 'llama-3.1-sonar-small-128k-chat',
|
37
|
-
},
|
38
|
-
{
|
39
|
-
contextWindowTokens: 128_000,
|
40
|
-
description:
|
41
|
-
'Llama 3.1 Sonar Large Chat 模型,具备70B参数,支持约127,000个标记的上下文长度,适合于复杂的离线聊天任务。',
|
42
|
-
displayName: 'Llama 3.1 Sonar Large Chat',
|
43
|
-
enabled: true,
|
44
|
-
id: 'llama-3.1-sonar-large-128k-chat',
|
34
|
+
'Llama 3.1 Sonar Small Online 模型,具备8B参数,支持约127,000个标记的上下文长度,专为在线聊天设计,能高效处理各种文本交互。',
|
35
|
+
displayName: 'Llama 3.1 Sonar Small Online',
|
36
|
+
id: 'llama-3.1-sonar-small-128k-online',
|
45
37
|
},
|
46
38
|
{
|
47
|
-
contextWindowTokens:
|
39
|
+
contextWindowTokens: 127_072,
|
48
40
|
description:
|
49
|
-
'Llama 3.1
|
50
|
-
|
41
|
+
'Llama 3.1 Sonar Large Online 模型,具备70B参数,支持约127,000个标记的上下文长度,适用于高容量和多样化聊天任务。',
|
42
|
+
displayName: 'Llama 3.1 Sonar Large Online',
|
43
|
+
id: 'llama-3.1-sonar-large-128k-online',
|
51
44
|
},
|
52
45
|
{
|
53
|
-
contextWindowTokens:
|
46
|
+
contextWindowTokens: 127_072,
|
54
47
|
description:
|
55
|
-
'Llama 3.1
|
56
|
-
|
48
|
+
'Llama 3.1 Sonar Huge Online 模型,具备405B参数,支持约127,000个标记的上下文长度,设计用于复杂的在线聊天应用。',
|
49
|
+
displayName: 'Llama 3.1 Sonar Huge Online',
|
50
|
+
id: 'llama-3.1-sonar-huge-128k-online',
|
57
51
|
},
|
58
52
|
],
|
59
|
-
checkModel: '
|
53
|
+
checkModel: 'sonar',
|
60
54
|
description:
|
61
55
|
'Perplexity 是一家领先的对话生成模型提供商,提供多种先进的Llama 3.1模型,支持在线和离线应用,特别适用于复杂的自然语言处理任务。',
|
62
56
|
id: 'perplexity',
|
@@ -196,7 +196,7 @@ _These are extended-context endpoints for [Hermes 3 405B Instruct](/models/nousr
|
|
196
196
|
"contextWindowTokens": 127072,
|
197
197
|
"description": "Llama 3.1 Sonar is Perplexity's latest model family. It surpasses their earlier Sonar models in cost-efficiency, speed, and performance. The model is built upon the Llama 3.1 405B and has internet access.",
|
198
198
|
"displayName": "Perplexity: Llama 3.1 Sonar 405B Online",
|
199
|
-
"enabled":
|
199
|
+
"enabled": false,
|
200
200
|
"functionCall": false,
|
201
201
|
"id": "perplexity/llama-3.1-sonar-huge-128k-online",
|
202
202
|
"maxTokens": undefined,
|
@@ -304,7 +304,7 @@ Note: This model is experimental and not suited for production use-cases. It may
|
|
304
304
|
|
305
305
|
This is the online version of the [offline chat model](/models/perplexity/llama-3.1-sonar-large-128k-chat). It is focused on delivering helpful, up-to-date, and factual responses. #online",
|
306
306
|
"displayName": "Perplexity: Llama 3.1 Sonar 70B Online",
|
307
|
-
"enabled":
|
307
|
+
"enabled": false,
|
308
308
|
"functionCall": false,
|
309
309
|
"id": "perplexity/llama-3.1-sonar-large-128k-online",
|
310
310
|
"maxTokens": undefined,
|
@@ -316,7 +316,7 @@ This is the online version of the [offline chat model](/models/perplexity/llama-
|
|
316
316
|
|
317
317
|
This is a normal offline LLM, but the [online version](/models/perplexity/llama-3.1-sonar-large-128k-online) of this model has Internet access.",
|
318
318
|
"displayName": "Perplexity: Llama 3.1 Sonar 70B",
|
319
|
-
"enabled":
|
319
|
+
"enabled": false,
|
320
320
|
"functionCall": false,
|
321
321
|
"id": "perplexity/llama-3.1-sonar-large-128k-chat",
|
322
322
|
"maxTokens": undefined,
|
@@ -328,7 +328,7 @@ This is a normal offline LLM, but the [online version](/models/perplexity/llama-
|
|
328
328
|
|
329
329
|
This is the online version of the [offline chat model](/models/perplexity/llama-3.1-sonar-small-128k-chat). It is focused on delivering helpful, up-to-date, and factual responses. #online",
|
330
330
|
"displayName": "Perplexity: Llama 3.1 Sonar 8B Online",
|
331
|
-
"enabled":
|
331
|
+
"enabled": false,
|
332
332
|
"functionCall": false,
|
333
333
|
"id": "perplexity/llama-3.1-sonar-small-128k-online",
|
334
334
|
"maxTokens": undefined,
|
@@ -340,7 +340,7 @@ This is the online version of the [offline chat model](/models/perplexity/llama-
|
|
340
340
|
|
341
341
|
This is a normal offline LLM, but the [online version](/models/perplexity/llama-3.1-sonar-small-128k-online) of this model has Internet access.",
|
342
342
|
"displayName": "Perplexity: Llama 3.1 Sonar 8B",
|
343
|
-
"enabled":
|
343
|
+
"enabled": false,
|
344
344
|
"functionCall": false,
|
345
345
|
"id": "perplexity/llama-3.1-sonar-small-128k-chat",
|
346
346
|
"maxTokens": undefined,
|
@@ -354,7 +354,7 @@ It has demonstrated strong performance compared to leading closed-source models
|
|
354
354
|
|
355
355
|
To read more about the model release, [click here](https://ai.meta.com/blog/meta-llama-3/). Usage of this model is subject to [Meta's Acceptable Use Policy](https://llama.meta.com/llama3/use-policy/).",
|
356
356
|
"displayName": "Meta: Llama 3.1 70B Instruct",
|
357
|
-
"enabled":
|
357
|
+
"enabled": true,
|
358
358
|
"functionCall": false,
|
359
359
|
"id": "meta-llama/llama-3.1-70b-instruct",
|
360
360
|
"maxTokens": undefined,
|
@@ -384,7 +384,7 @@ It has demonstrated strong performance compared to leading closed-source models
|
|
384
384
|
|
385
385
|
To read more about the model release, [click here](https://ai.meta.com/blog/meta-llama-3/). Usage of this model is subject to [Meta's Acceptable Use Policy](https://llama.meta.com/llama3/use-policy/).",
|
386
386
|
"displayName": "Meta: Llama 3.1 8B Instruct",
|
387
|
-
"enabled":
|
387
|
+
"enabled": true,
|
388
388
|
"functionCall": false,
|
389
389
|
"id": "meta-llama/llama-3.1-8b-instruct",
|
390
390
|
"maxTokens": undefined,
|