File size: 3,546 Bytes
4f77518
 
 
3974b7a
4f77518
 
 
 
 
 
 
 
 
3974b7a
 
 
 
 
d23fc4b
3974b7a
 
 
f37371f
3974b7a
d23fc4b
 
3974b7a
4f77518
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
125b498
4f77518
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2f309cd
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
import gradio as gr
from huggingface_hub import InferenceClient
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
import spaces
import torch
from safetensors import safe_open
from jaxtyping import Float, Int
from typing import List, Callable
from torch import Tensor
from threading import Thread
import einops

model_id = "MaziyarPanahi/Meta-Llama-3-70B-Instruct-GPTQ"
tokenizer = AutoTokenizer.from_pretrained(model_id)
quantize_config = BaseQuantizeConfig(
        bits=4,
        group_size=128,
        desc_act=False
    )
model = AutoGPTQForCausalLM.from_quantized(
        model_id,
        device="cuda:0",
        use_safetensors=True,
        disable_exllamav2=True,
        use_marlin=True,
        quantize_config=quantize_config).eval()


@spaces.GPU
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    messages.append({"role": "user", "content": message})

    response = ""

    inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(device=torch.device("cuda"))
    streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True, skip_prompt=True)

    thread = Thread(
        target=model.generate,
        kwargs={
            "inputs": inputs,
            "max_new_tokens": max_tokens,
            "temperature": temperature,
            "top_p": top_p,
            "streamer": streamer,
        },
    )
    thread.start()

    for new_text in streamer:
        token = new_text.choices[0].delta.content

        response += token
        yield response
    
def get_orthogonalized_matrix(matrix: Float[Tensor, '... d_model'], vec: Float[Tensor, 'd_model']) -> Float[Tensor, '... d_model']:
    device = matrix.device
    vec = vec.to(device)
    proj = einops.einsum(matrix, vec.view(-1, 1), '... d_model, d_model single -> ... single') * vec
    return matrix - proj

"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)


if __name__ == "__main__":
    # get refusal_dir from refusal_dir.safetensors file.
    with safe_open("refusal_dir.safetensors", framework="pt", device="cpu") as f:
        refusal_dir = f.get_tensor("refusal_dir")
    refusal_dir = refusal_dir.cpu().float()

    model.model.embed_tokens.weight.data = get_orthogonalized_matrix(model.model.embed_tokens.weight, refusal_dir)

    for block in model.model.layers:
        block.self_attn.o_proj.weight.data = get_orthogonalized_matrix(block.self_attn.o_proj.weight, refusal_dir)
        block.mlp.down_proj.weight.data = get_orthogonalized_matrix(block.mlp.down_proj.weight.T, refusal_dir).T
    
    demo.launch()