File size: 6,478 Bytes
71e55e7
eac2c48
014e235
71e55e7
cd24c9d
71e55e7
014e235
 
 
 
 
 
 
 
 
eac2c48
014e235
 
 
 
 
 
 
71e55e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
021c2c9
 
afa7e4f
021c2c9
afa7e4f
2ddfac0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71e55e7
2ddfac0
 
 
 
 
021c2c9
 
 
afa7e4f
021c2c9
 
 
 
 
 
 
afa7e4f
021c2c9
 
afa7e4f
021c2c9
 
 
 
 
 
 
 
 
afa7e4f
021c2c9
 
afa7e4f
021c2c9
 
 
 
 
 
 
 
afa7e4f
021c2c9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
afa7e4f
021c2c9
 
 
 
 
afa7e4f
021c2c9
afa7e4f
021c2c9
 
 
afa7e4f
021c2c9
 
 
 
afa7e4f
 
 
2ddfac0
 
 
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
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
from transformers import pipeline
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
from peft import PeftModel, PeftConfig
import gradio as gr
import torch


base_model = "unsloth/Llama-3.2-3B-Instruct"  # Replace with the correct base model
peft_model_path = "ivwhy/lora_model"

config = PeftConfig.from_pretrained(peft_model_path)
model = AutoModelForCausalLM.from_pretrained(base_model, torch_dtype=torch.bfloat16)
model = PeftModel.from_pretrained(model, peft_model_path)
tokenizer = AutoTokenizer.from_pretrained(base_model)

pipeline = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device=-1,  # CPU
)

chatbot = pipeline

message_list = []
response_list = []

def chat_function(message, history, system_prompt, max_new_tokens, temperature):
    messages = [{"role":"system","content":system_prompt},
                {"role":"user","content":message}]
    prompt = pipeline.tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,)
    terminators = [
        pipeline.tokenizer.eos_token_id,
        pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")]
    outputs = pipeline(
        prompt,
        max_new_tokens = max_new_tokens,
        eos_token_id = terminators,
        do_sample = True,
        temperature = 0.1,
        top_p = 0.9,)
    return outputs[0]["generated_text"][len(prompt):]

demo_chatbot = gr.ChatInterface(
    chat_function,
    textbox=gr.Textbox(placeholder="Enter message here", container=False, scale=7),
    chatbot=gr.Chatbot(height=400),
    additional_inputs=[
        gr.Textbox("You are helpful AI", label="System Prompt"),
        gr.Slider(500,4000, label="Max New Tokens"),
        gr.Slider(0,1,label="Temperature")
    ])

demo_chatbot.launch()

''' =================================== OLD VERSION ==============================================
import torch
import transformers
import gradio as gr
from unsloth import FastLanguageModel

# Load the fine-tuned Unsloth model
max_seq_length = 2048  # Adjust based on your training
dtype = None  # Auto-detect is fine for CPU

def load_model():
    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name="ivwhy/lora_model",  # Your fine-tuned model path
        max_seq_length=max_seq_length,
        dtype=dtype,
        load_in_4bit=True,  # Keep 4-bit loading enabled
    )

    # Optional: Add special tokens for chat if needed
    tokenizer.pad_token = tokenizer.eos_token

    # Create the pipeline for CPU
    pipeline = transformers.pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        device=-1  # Force CPU usage
    )
    
    return pipeline, tokenizer

# Load model globally
generation_pipeline, tokenizer = load_model()

def chat_function(message, history, system_prompt, max_new_tokens, temperature):
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": message}
    ]
    
    # Apply chat template
    prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
    )
    
    # Define terminators
    terminators = [
        tokenizer.eos_token_id,
        tokenizer.convert_tokens_to_ids("<|eot_id|>")
    ]
    
    # Generate response
    outputs = generation_pipeline(
        prompt,
        max_new_tokens=max_new_tokens,
        eos_token_id=terminators,
        do_sample=True,
        temperature=temperature,
        top_p=0.9,
    )
    
    # Extract and return just the generated text
    return outputs[0]["generated_text"][len(prompt):]

# Create Gradio interface
demo = gr.ChatInterface(
    chat_function,
    textbox=gr.Textbox(placeholder="Enter message here", container=False, scale=7),
    chatbot=gr.Chatbot(height=400),
    additional_inputs=[
        gr.Textbox("You are helpful AI", label="System Prompt"),
        gr.Slider(minimum=1, maximum=4000, value=500, label="Max New Tokens"),
        gr.Slider(minimum=0, maximum=1, value=0.7, label="Temperature")
    ]
)

if __name__ == "__main__":
    demo.launch()

================================== OLD VER ==============================
import torch
import transformers
import gradio as gr
from unsloth import FastLanguageModel

# Load the fine-tuned Unsloth model
max_seq_length = 2048  # Adjust based on your training
dtype = None  # None for auto detection

def load_model():
    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name="ivwhy/lora_model",  # Your fine-tuned model path
        max_seq_length=max_seq_length,
        dtype=dtype,
        load_in_4bit=True  # Optional: load in 4-bit for efficiency
    )

    # Optional: Add special tokens for chat if needed
    tokenizer.pad_token = tokenizer.eos_token

    # Create the pipeline
    pipeline = transformers.pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        device=0 if torch.cuda.is_available() else -1  # Use GPU if available
    )
    
    return pipeline, tokenizer

# Load model globally
generation_pipeline, tokenizer = load_model()

def chat_function(message, history, system_prompt, max_new_tokens, temperature):
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": message}
    ]
    
    # Apply chat template
    prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
    )
    
    # Define terminators
    terminators = [
        tokenizer.eos_token_id,
        tokenizer.convert_tokens_to_ids("<|eot_id|>")
    ]
    
    # Generate response
    outputs = generation_pipeline(
        prompt,
        max_new_tokens=max_new_tokens,
        eos_token_id=terminators,
        do_sample=True,
        temperature=temperature,
        top_p=0.9,
    )
    
    # Extract and return just the generated text
    return outputs[0]["generated_text"][len(prompt):]

# Create Gradio interface
demo = gr.ChatInterface(
    chat_function,
    textbox=gr.Textbox(placeholder="Enter message here", container=False, scale=7),
    chatbot=gr.Chatbot(height=400),
    additional_inputs=[
        gr.Textbox("You are helpful AI", label="System Prompt"),
        gr.Slider(minimum=1, maximum=4000, value=500, label="Max New Tokens"),
        gr.Slider(minimum=0, maximum=1, value=0.7, label="Temperature")
    ]
)

if __name__ == "__main__":
    demo.launch()

'''