Commit
•
70dbc50
1
Parent(s):
7597694
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,131 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
base_model: amazingvince/Not-WizardLM-2-7B
|
4 |
+
pipeline_tag: text-generation
|
5 |
+
---
|
6 |
+
|
7 |
+
# QuantFactory/Not-WizardLM-2-7B-GGUF
|
8 |
+
This is quantized version of [amazingvince/Not-WizardLM-2-7B]() created using llama.cpp
|
9 |
+
|
10 |
+
# Model Description
|
11 |
+
|
12 |
+
<a href="https://colab.research.google.com/gist/pszemraj/d3d74ceab942722b49188606785e2bfd/not-wizardlm-2-7b-inference.ipynb">
|
13 |
+
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
|
14 |
+
</a>
|
15 |
+
|
16 |
+
Included is code ripped from fastchat with the expected chat templating.
|
17 |
+
|
18 |
+
Also wiz.pdf is a pdf of the github blog showing the apache 2 release.
|
19 |
+
Link to wayback machine included: https://web.archive.org/web/20240415221214/https://wizardlm.github.io/WizardLM2/
|
20 |
+
|
21 |
+
|
22 |
+
|
23 |
+
## example
|
24 |
+
|
25 |
+
```python
|
26 |
+
import dataclasses
|
27 |
+
from enum import auto, Enum
|
28 |
+
from typing import List, Tuple, Any
|
29 |
+
|
30 |
+
|
31 |
+
class SeparatorStyle(Enum):
|
32 |
+
"""Different separator style."""
|
33 |
+
SINGLE = auto()
|
34 |
+
TWO = auto()
|
35 |
+
|
36 |
+
|
37 |
+
@dataclasses.dataclass
|
38 |
+
class Conversation:
|
39 |
+
"""A class that keeps all conversation history."""
|
40 |
+
system: str
|
41 |
+
roles: List[str]
|
42 |
+
messages: List[List[str]]
|
43 |
+
offset: int
|
44 |
+
sep_style: SeparatorStyle = SeparatorStyle.SINGLE
|
45 |
+
sep: str = "###"
|
46 |
+
sep2: str = None
|
47 |
+
|
48 |
+
# Used for gradio server
|
49 |
+
skip_next: bool = False
|
50 |
+
conv_id: Any = None
|
51 |
+
|
52 |
+
def get_prompt(self):
|
53 |
+
if self.sep_style == SeparatorStyle.SINGLE:
|
54 |
+
ret = self.system
|
55 |
+
for role, message in self.messages:
|
56 |
+
if message:
|
57 |
+
ret += self.sep + " " + role + ": " + message
|
58 |
+
else:
|
59 |
+
ret += self.sep + " " + role + ":"
|
60 |
+
return ret
|
61 |
+
elif self.sep_style == SeparatorStyle.TWO:
|
62 |
+
seps = [self.sep, self.sep2]
|
63 |
+
ret = self.system + seps[0]
|
64 |
+
for i, (role, message) in enumerate(self.messages):
|
65 |
+
if message:
|
66 |
+
ret += role + ": " + message + seps[i % 2]
|
67 |
+
else:
|
68 |
+
ret += role + ":"
|
69 |
+
return ret
|
70 |
+
else:
|
71 |
+
raise ValueError(f"Invalid style: {self.sep_style}")
|
72 |
+
|
73 |
+
def append_message(self, role, message):
|
74 |
+
self.messages.append([role, message])
|
75 |
+
|
76 |
+
def to_gradio_chatbot(self):
|
77 |
+
ret = []
|
78 |
+
for i, (role, msg) in enumerate(self.messages[self.offset:]):
|
79 |
+
if i % 2 == 0:
|
80 |
+
ret.append([msg, None])
|
81 |
+
else:
|
82 |
+
ret[-1][-1] = msg
|
83 |
+
return ret
|
84 |
+
|
85 |
+
def copy(self):
|
86 |
+
return Conversation(
|
87 |
+
system=self.system,
|
88 |
+
roles=self.roles,
|
89 |
+
messages=[[x, y] for x, y in self.messages],
|
90 |
+
offset=self.offset,
|
91 |
+
sep_style=self.sep_style,
|
92 |
+
sep=self.sep,
|
93 |
+
sep2=self.sep2,
|
94 |
+
conv_id=self.conv_id)
|
95 |
+
|
96 |
+
def dict(self):
|
97 |
+
return {
|
98 |
+
"system": self.system,
|
99 |
+
"roles": self.roles,
|
100 |
+
"messages": self.messages,
|
101 |
+
"offset": self.offset,
|
102 |
+
"sep": self.sep,
|
103 |
+
"sep2": self.sep2,
|
104 |
+
"conv_id": self.conv_id,
|
105 |
+
}
|
106 |
+
|
107 |
+
|
108 |
+
|
109 |
+
conv = Conversation(
|
110 |
+
system="A chat between a curious user and an artificial intelligence assistant. "
|
111 |
+
"The assistant gives helpful, detailed, and polite answers to the user's questions.",
|
112 |
+
roles=("USER", "ASSISTANT"),
|
113 |
+
messages=[],
|
114 |
+
offset=0,
|
115 |
+
sep_style=SeparatorStyle.TWO,
|
116 |
+
sep=" ",
|
117 |
+
sep2="</s>",
|
118 |
+
)
|
119 |
+
|
120 |
+
conv.append_message(conv.roles[0], "Why would Microsoft take this down?")
|
121 |
+
conv.append_message(conv.roles[1], None)
|
122 |
+
prompt = conv.get_prompt()
|
123 |
+
|
124 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
125 |
+
|
126 |
+
result = model.generate(**inputs, max_new_tokens=1000)
|
127 |
+
generated_ids = result[0]
|
128 |
+
generated_text = tokenizer.decode(generated_ids, skip_special_tokens=True)
|
129 |
+
print(generated_text)
|
130 |
+
|
131 |
+
```
|