File size: 5,501 Bytes
11df34d
 
 
 
8091023
 
11df34d
 
 
 
b299ae3
fba1123
11df34d
 
b299ae3
11df34d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b299ae3
11df34d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b299ae3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11df34d
 
 
 
3d15ec6
fba1123
 
11df34d
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
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
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
##### `🦙_Alpaca.py`
##### Alpaca Model
##### https://github.com/seemanne/llamacpypy
##### https://github.com/shaunabanana/llama.py
##### https://github.com/thomasantony/llamacpp-python
##### https://github.com/abetlen/llama-cpp-python
##### Please reach out to [email protected] for any questions
#### Loading needed Python libraries
import streamlit as st
#from llamacpypy import Llama
#import llamacpp
from llama_cpp import Llama
import os
import subprocess
import sys




#### Streamlit initial setup
st.set_page_config(
  page_title = "🦙 Alpaca",
  page_icon = "images/Logo.png",
  layout = "centered",
  initial_sidebar_state = "expanded"
)




#### Functions of the Python Wrapper
def llama_stream(
        prompt = '',
        skip_prompt = True,
        trim_prompt = 0,
        executable = 'pages/llama.cpp/main',
        model = 'models/7B/ggml-model-q4_0.bin',
        threads = 4,
        temperature = 0.7,
        top_k = 40,
        top_p = 0.5,
        repeat_last_n = 256,
        repeat_penalty = 1.17647,
        n = 4096,
        interactive = False,
        reverse_prompt = "User:"
):
    command = [
        executable,
        '-m', model,
        '-t', str(threads),
        '--temp', str(temperature),
        '--top_k', str(top_k),
        '--top_p', str(top_p),
        '--repeat_last_n', str(repeat_last_n),
        '--repeat_penalty', str(repeat_penalty),
        '-n', str(n),
        '-p', prompt
    ]
    if interactive:
        command += ['-i', '-r', reverse_prompt]
    
    process = subprocess.Popen(
        command,
        stdin = subprocess.PIPE,
        stdout = subprocess.PIPE,
        stderr = subprocess.PIPE,
    )
    
    token = b''
    generated = ''
    while True:
        token += process.stdout.read(1)
        if token:  # neither empty string nor None
            try:
                decoded = token.decode('utf-8')
                
                trimmed_prompt = prompt
                if trim_prompt > 0:
                    trimmed_prompt = prompt[:-trim_prompt]
                prompt_finished = generated.startswith(trimmed_prompt)
                reverse_prompt_encountered = generated.endswith(reverse_prompt)
                if not skip_prompt or prompt_finished:
                    yield decoded
                if interactive and prompt_finished and reverse_prompt_encountered:
                    user_input = input()
                    process.stdin.write(user_input.encode('utf-8') + b'\n')
                    process.stdin.flush()
                
                generated += decoded
                token = b''
            except UnicodeDecodeError:
                continue
        elif process.poll() is not None:
            return


def llama(
        prompt = '',
        stream = False,
        skip_prompt = False,
        trim_prompt = 0,
        executable = 'pages/llama.cpp/main',
        model = 'models/7B/ggml-model-q4_0.bin',
        threads = 4,
        temperature = 0.7,
        top_k = 40,
        top_p = 0.5,
        repeat_last_n = 256,
        repeat_penalty = 1.17647,
        n = 4096,
        interactive = False,
        reverse_prompt = "User:"
):
    streamer = llama_stream(
        prompt = prompt,
        skip_prompt = skip_prompt,
        trim_prompt = trim_prompt,
        executable = executable,
        model = model,
        threads = threads,
        temperature = temperature,
        top_k = top_k,
        top_p = top_p,
        repeat_last_n = repeat_last_n,
        repeat_penalty = repeat_penalty,
        n = n,
        interactive = interactive,
        reverse_prompt = reverse_prompt
    )
    if stream:
        return streamer
    else:
        return ''.join(list(streamer))



### Python Wrapper (functions above)
#text = []
#for token in llama(prompt = 'What is your purpose?', repeat_penalty = 1.05, skip_prompt = False, interactive = False):
#    print(token, end = '', flush = True)
#    text.append(token)
#st.subheader('Debug')
#st.experimental_show(text[0])
#st.experimental_show(text[1])
#st.subheader('Answer')
#st.write(''.join(text))



### llamacpypy
#llama = Llama(model_name = 'models/7B/ggml-model-q4_0.bin', warm_start = True)
#llama.load_model()
#var = llama.generate("This is the weather report, we are reporting a clown fiesta happening at backer street. The clowns ")
#st.write(var)



### llamacpp
#model_path = "./models/7B/ggml-model-q4_0.bin"
#params = llamacpp.gpt_params(model_path, 4096, 40, 0.1, 0.7, 2.0)
#model = llamacpp.PyLLAMA(model_path, params)
#text = model.predict("Hello, I'm a llama.", 10)
#st.write(text)
#params = llamacpp.gpt_params('./models/7B/ggml-model-q4_0.bin',  # model,
#    512,  # ctx_size
#    100,  # n_predict
#    40,  # top_k
#    0.95,  # top_p
#    0.85,  # temp
#    1.30,  # repeat_penalty
#    -1,  # seed
#    8,  # threads
#    64,  # repeat_last_n
#    8,  # batch_size
#)
#model = llamacpp.PyLLAMA(params)
#model.add_bos()     # Adds "beginning of string" token
#model.update_input("A llama is a")
#model.print_startup_stats()
#model.prepare_context()

#model.ingest_all_pending_input(True)
#while not model.is_finished():
#    text, is_finished = model.infer_text()
#    print(text, end="")

#    if is_finished:
#        break

# Flush stdout
#sys.stdout.flush()
#model.print_end_stats()



### Llama cpp
llm = Llama(model_path = "xfh/alpaca.cpp_65b_ggml")
output = llm("Q: Name the planets in the solar system? A: ", max_tokens = 32, stop = ["Q:", "\n"], echo = True)
st.write(output)