Commit
•
97d7816
1
Parent(s):
c19bc51
Update README.md
Browse files
README.md
CHANGED
@@ -1,190 +1,49 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
4 |
|
5 |
-
|
6 |
|
7 |
-
|
8 |
-
|
9 |
-
> **! Important Note !**
|
10 |
-
>
|
11 |
-
> The following is the HF transformers implementation of the RWKV-5 Eagle 7B model. **This is meant to be used with the huggingface transformers**
|
12 |
-
>
|
13 |
-
> For the full model weights on its own, to use with other RWKV libraries, refer to [here](https://huggingface.co/RWKV/v5-Eagle-7B)
|
14 |
-
>
|
15 |
-
> This is not an instruct tune model! (soon...)
|
16 |
-
|
17 |
-
- [HF Demo](https://huggingface.co/spaces/BlinkDL/RWKV-Gradio-2)
|
18 |
-
- [Our wiki](https://wiki.rwkv.com)
|
19 |
-
- [pth model weights](https://huggingface.co/RWKV/v5-Eagle-7B)
|
20 |
-
|
21 |
-
#### Running on CPU via HF transformers
|
22 |
-
|
23 |
-
```python
|
24 |
-
import torch
|
25 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer
|
26 |
-
|
27 |
-
def generate_prompt(instruction, input=""):
|
28 |
-
instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
|
29 |
-
input = input.strip().replace('\r\n','\n').replace('\n\n','\n')
|
30 |
-
if input:
|
31 |
-
return f"""Instruction: {instruction}
|
32 |
-
|
33 |
-
Input: {input}
|
34 |
-
|
35 |
-
Response:"""
|
36 |
-
else:
|
37 |
-
return f"""User: hi
|
38 |
-
|
39 |
-
Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
|
40 |
-
|
41 |
-
User: {instruction}
|
42 |
-
|
43 |
-
Assistant:"""
|
44 |
-
|
45 |
-
|
46 |
-
model = AutoModelForCausalLM.from_pretrained("RWKV/HF_v5-Eagle-7B", trust_remote_code=True).to(torch.float32)
|
47 |
-
tokenizer = AutoTokenizer.from_pretrained("RWKV/HF_v5-Eagle-7B", trust_remote_code=True)
|
48 |
-
|
49 |
-
text = "请介绍北京的旅游景点"
|
50 |
-
prompt = generate_prompt(text)
|
51 |
-
|
52 |
-
inputs = tokenizer(prompt, return_tensors="pt")
|
53 |
-
output = model.generate(inputs["input_ids"], max_new_tokens=333, do_sample=True, temperature=1.0, top_p=0.3, top_k=0, )
|
54 |
-
print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))
|
55 |
-
```
|
56 |
-
|
57 |
-
output:
|
58 |
-
|
59 |
-
```shell
|
60 |
-
User: hi
|
61 |
-
|
62 |
-
Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
|
63 |
-
|
64 |
-
User: 请介绍北京的旅游景点
|
65 |
-
|
66 |
-
Assistant: 北京是中国的首都,拥有众多的旅游景点,以下是其中一些著名的景点:
|
67 |
-
1. 故宫:位于北京市中心,是明清两代的皇宫,内有大量的文物和艺术品。
|
68 |
-
2. 天安门广场:是中国最著名的广场之一,是中国人民政治协商会议的旧址,也是中国人民政治协商会议的中心。
|
69 |
-
3. 颐和园:是中国古代皇家园林之一,有着悠久的历史和丰富的文化内涵。
|
70 |
-
4. 长城:是中国古代的一道长城,全长约万里,是中国最著名的旅游景点之一。
|
71 |
-
5. 北京大学:是中国著名的高等教育机构之一,有着悠久的历史和丰富的文化内涵。
|
72 |
-
6. 北京动物园:是中国最大的动物园之一,有着丰富的动物资源和丰富的文化内涵。
|
73 |
-
7. 故宫博物院:是中国最著名的博物馆之一,收藏了大量的文物和艺术品,是中国最重要的文化遗产之一。
|
74 |
-
8. 天坛:是中国古代皇家
|
75 |
```
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))
|
111 |
```
|
112 |
-
|
113 |
-
output:
|
114 |
-
|
115 |
-
```shell
|
116 |
-
User: hi
|
117 |
-
|
118 |
-
Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
|
119 |
-
|
120 |
-
User: 介绍一下大熊猫
|
121 |
-
|
122 |
-
Assistant: 大熊猫是一种中国特有的哺乳动物,也是中国的国宝之一。它们的外貌特征是圆形的黑白相间的身体,有着黑色的毛发和白色的耳朵。大熊猫的食物主要是竹子,它们会在竹林中寻找竹子,并且会将竹子放在竹笼中进行储存。大熊猫的寿命约为20至30年,但由于栖息地的丧失和人类活动的
|
123 |
-
```
|
124 |
-
|
125 |
-
#### Batch Inference
|
126 |
-
|
127 |
-
```python
|
128 |
-
import torch
|
129 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer
|
130 |
-
|
131 |
-
def generate_prompt(instruction, input=""):
|
132 |
-
instruction = instruction.strip().replace('\r\n', '\n').replace('\n\n', '\n')
|
133 |
-
input = input.strip().replace('\r\n', '\n').replace('\n\n', '\n')
|
134 |
-
if input:
|
135 |
-
return f"""Instruction: {instruction}
|
136 |
-
|
137 |
-
Input: {input}
|
138 |
-
|
139 |
-
Response:"""
|
140 |
-
else:
|
141 |
-
return f"""User: hi
|
142 |
-
|
143 |
-
Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
|
144 |
-
|
145 |
-
User: {instruction}
|
146 |
-
|
147 |
-
Assistant:"""
|
148 |
-
|
149 |
-
model = AutoModelForCausalLM.from_pretrained("RWKV/HF_v5-Eagle-7B", trust_remote_code=True).to(torch.float32)
|
150 |
-
tokenizer = AutoTokenizer.from_pretrained("RWKV/HF_v5-Eagle-7B", trust_remote_code=True)
|
151 |
-
|
152 |
-
texts = ["请介绍北京的旅游景点", "介绍一下大熊猫", "乌兰察布"]
|
153 |
-
prompts = [generate_prompt(text) for text in texts]
|
154 |
-
|
155 |
-
inputs = tokenizer(prompts, return_tensors="pt", padding=True)
|
156 |
-
outputs = model.generate(inputs["input_ids"], max_new_tokens=128, do_sample=True, temperature=1.0, top_p=0.3, top_k=0, )
|
157 |
-
|
158 |
-
for output in outputs:
|
159 |
-
print(tokenizer.decode(output.tolist(), skip_special_tokens=True))
|
160 |
-
|
161 |
-
```
|
162 |
-
|
163 |
-
output:
|
164 |
-
|
165 |
-
```shell
|
166 |
-
User: hi
|
167 |
-
|
168 |
-
Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
|
169 |
-
|
170 |
-
User: 请介绍北京的旅游景点
|
171 |
-
|
172 |
-
Assistant: 北京是中国的首都,拥有丰富的旅游资源和历史文化遗产。以下是一些北京的旅游景点:
|
173 |
-
1. 故宫:位于北京市中心,是明清两代的皇宫,是中国最大的古代宫殿建筑群之一。
|
174 |
-
2. 天安门广场:位于北京市中心,是中国最著名的城市广场之一,也是中国最大的城市广场。
|
175 |
-
3. 颐和
|
176 |
-
User: hi
|
177 |
-
|
178 |
-
Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
|
179 |
-
|
180 |
-
User: 介绍一下大熊猫
|
181 |
-
|
182 |
-
Assistant: 大熊猫是一种生活在中国中部地区的哺乳动物,也是中国的国宝之一。它们的外貌特征是圆形的黑白相间的身体,有着黑色的毛发和圆圆的眼睛。大熊猫是一种濒危物种,目前只有在野外的几个保护区才能看到它们的身影。大熊猫的食物主要是竹子,它们会在竹子上寻找食物,并且可以通
|
183 |
-
User: hi
|
184 |
-
|
185 |
-
Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.
|
186 |
-
|
187 |
-
User: 乌兰察布
|
188 |
-
|
189 |
-
Assistant: 乌兰察布是中国新疆维吾尔自治区的一个县级市,位于新疆维吾尔自治区中部,是新疆的第二大城市。乌兰察布市是新疆的第一大城市,也是新疆的重要城市之一。乌兰察布市是新疆的经济中心,也是新疆的重要交通枢纽之一。乌兰察布市的人口约为2.5万人,其中汉族占绝大多数。乌
|
190 |
-
```
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- teknium/OpenHermes-2.5
|
5 |
+
language:
|
6 |
+
- en
|
7 |
---
|
8 |
+
This is an RWKV-v5 ("Eagle") RNN language model pretrained on 7B of multi-lingual data and fine-tuned on Nous Research's OpenHermes 2.5 chat dataset.
|
9 |
|
10 |
+
The chat format is identical to the default HuggingFace format, **EXCEPT** that we replace the `<|im_start|>` and `<|im_end|>` special tokens with the unused ASCII codes `"\x16"` and `"\x17"` respectively. This was done to ensure complete compatibility withe standard RWKV World tokenizer.
|
11 |
|
12 |
+
An example interaction:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
```
|
14 |
+
\x16user
|
15 |
+
Write a poem about the beauty of mathematics in the style of Emily Dickinson.\x17
|
16 |
+
\x16assistant
|
17 |
+
In the heart of Numbers, so precise,
|
18 |
+
A Beauty dwells - both cool and nice.
|
19 |
+
An elegance that sings in sums,
|
20 |
+
And dances to algebraic drums.
|
21 |
+
|
22 |
+
Addition's grace, subtraction's might,
|
23 |
+
Multiplication's sheer delight,
|
24 |
+
Division's firm yet gentle hand,
|
25 |
+
All part of this enchanted land.
|
26 |
+
|
27 |
+
Infinity, a lover bold,
|
28 |
+
To whom the stars their secrets told,
|
29 |
+
While Zero, cloaked in mystery,
|
30 |
+
Unlocks each mathematical decree.
|
31 |
+
|
32 |
+
Fractals bloom like flowers wild,
|
33 |
+
Geometry, Nature's favored child,
|
34 |
+
Calculus, with steady gaze,
|
35 |
+
Maps out life's intricate maze.
|
36 |
+
|
37 |
+
Pi, an endless serenade,
|
38 |
+
In circles' arms forever laid,
|
39 |
+
And prime numbers,
|
40 |
+
standing lone,
|
41 |
+
Make the universe their own.
|
42 |
+
|
43 |
+
Each equation, theorem scribed,
|
44 |
+
In
|
45 |
+
parchment old or computer imbibed,
|
46 |
+
Is but a whisper from the skies,
|
47 |
+
A cosmic truth in disguise.\x17
|
|
|
48 |
```
|
49 |
+
In the above, newlines are printed without escaping, while ASCII codes are escaped for readability.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|