File size: 10,695 Bytes
8b6d1b6
 
d82cd6f
 
2b66ced
 
 
08d76ce
 
2b66ced
 
 
 
 
 
 
 
d82cd6f
 
2b66ced
 
 
 
 
 
 
 
 
 
 
 
7c57fe0
 
 
d82cd6f
08d76ce
2b66ced
 
 
 
 
 
08d76ce
 
2b66ced
08d76ce
2b66ced
 
 
449ac0a
2b66ced
 
 
 
 
 
 
 
449ac0a
 
 
 
 
 
 
 
 
2b66ced
 
 
08d76ce
 
06bb18a
08d76ce
06bb18a
 
 
 
 
cd7e21b
 
 
06bb18a
08d76ce
2b66ced
d82cd6f
 
 
cd7e21b
 
 
08d76ce
8b6d1b6
 
3a616f6
 
 
 
 
 
 
 
 
 
8b6d1b6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2574e16
 
645be2c
 
 
 
2574e16
8b6d1b6
 
2574e16
 
 
 
 
 
 
 
 
 
645be2c
2574e16
 
 
 
645be2c
2574e16
645be2c
2574e16
645be2c
8b6d1b6
 
 
44f7b3c
8b6d1b6
 
 
2574e16
8b6d1b6
 
 
 
 
 
 
2574e16
 
 
8b6d1b6
2574e16
8b6d1b6
2574e16
 
 
8b6d1b6
2574e16
08d76ce
8b6d1b6
2574e16
645be2c
 
 
 
 
 
 
 
 
 
 
 
 
 
8b6d1b6
 
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
import json

SYSTEM_PROMPT = "You are a helpful assistant that provide concise and accurate answers."

def set_cora_preset():
    return (
        "gsarti/cora_mgen",  # model_name_or_path
        "<Q>: {current} <P>: {context}",  # input_template
        "<Q>: {current}",  # input_current_text_template
    )


def set_default_preset():
    return (
        "gpt2",  # model_name_or_path
        "{current} {context}",  # input_template
        "{current}",  # output_template
        "{current}",  # contextless_input_template
        "{current}",  # contextless_output_template
        [],  # special_tokens_to_keep
        "",  # decoder_input_output_separator
        "{}",  # model_kwargs
        "{}",  # tokenizer_kwargs
        "{}",  # generation_kwargs
        "{}",  # attribution_kwargs
    )


def set_zephyr_preset():
    return (
        "stabilityai/stablelm-2-zephyr-1_6b",  # model_name_or_path
        "<|system|>{system_prompt}<|endoftext|>\n<|user|>\n{context}\n\n{current}<|endoftext|>\n<|assistant|>".replace("{system_prompt}", SYSTEM_PROMPT),  # input_template
        "\n",  # decoder_input_output_separator
        "<|system|>{system_prompt}<|endoftext|>\n<|user|>\n{current}<|endoftext|>\n<|assistant|>".replace("{system_prompt}", SYSTEM_PROMPT),  # input_current_text_template
        ["<|im_start|>", "<|im_end|>", "<|endoftext|>"],  # special_tokens_to_keep
        '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
    )


def set_chatml_preset():
    return (
        "Qwen/Qwen1.5-0.5B-Chat",  # model_name_or_path
        "<|im_start|>system\n{system_prompt}<|im_end|>\n<|im_start|>user\n{context}\n\n{current}<|im_end|>\n<|im_start|>assistant\n".replace("{system_prompt}", SYSTEM_PROMPT),  # input_template
        "<|im_start|>system\n{system_prompt}<|im_end|>\n<|im_start|>user\n{current}<|im_end|>\n<|im_start|>assistant\n".replace("{system_prompt}", SYSTEM_PROMPT),  # input_current_text_template
        ["<|im_start|>", "<|im_end|>"],  # special_tokens_to_keep
        '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
    )


def set_mbart_mmt_preset():
    return (
        "facebook/mbart-large-50-one-to-many-mmt",  # model_name_or_path
        "{context} {current}",  # input_template
        "{context} {current}",  # output_template
        '{\n\t"src_lang": "en_XX",\n\t"tgt_lang": "fr_XX"\n}',  # tokenizer_kwargs
    )


def set_nllb_mmt_preset():
    return (
        "facebook/nllb-200-distilled-600M",  # model_name_or_path
        "{context} {current}",  # input_template
        "{context} {current}",  # output_template
        '{\n\t"src_lang": "eng_Latn",\n\t"tgt_lang": "fra_Latn"\n}',  # tokenizer_kwargs
    )


def set_towerinstruct_preset():
    return (
        "Unbabel/TowerInstruct-7B-v0.1",  # model_name_or_path
        "<|im_start|>user\nSource: {current}\nContext: {context}\nTranslate the above text into French. Use the context to guide your answer.\nTarget:<|im_end|>\n<|im_start|>assistant\n",  # input_template
        "<|im_start|>user\nSource: {current}\nTranslate the above text into French.\nTarget:<|im_end|>\n<|im_start|>assistant\n",  # input_current_text_template
        ["<|im_start|>", "<|im_end|>"],  # special_tokens_to_keep
        '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
    )

def set_gemma_preset():
    return (
        "google/gemma-2b-it", # model_name_or_path
        "<start_of_turn>user\n{context}\n{current}<end_of_turn>\n<start_of_turn>model", # input_template
        "\n",  # decoder_input_output_separator
        "<start_of_turn>user\n{current}<end_of_turn>\n<start_of_turn>model", # input_current_text_template
        ["<start_of_turn>", "<end_of_turn>"], # special_tokens_to_keep
        '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
    )

def set_mistral_instruct_preset():
    return (
        "mistralai/Mistral-7B-Instruct-v0.2", # model_name_or_path
        "[INST]{context}\n{current}[/INST]", # input_template
        "[INST]{current}[/INST]", # input_current_text_template
        '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
    )

def set_phi3_preset():
    return (
        "microsoft/Phi-3-mini-4k-instruct", # model_name_or_path
        "<|system|>\n{system_prompt}<|end|>\n<|user|>\n{context}\n\n{current}<|end|>\n<|assistant|>".replace("{system_prompt}", SYSTEM_PROMPT), # input_template
        "\n",  # decoder_input_output_separator
        "<|system|>\n{system_prompt}<|end|>\n<|user|>\n{current}<|end|>\n<|assistant|>".replace("{system_prompt}", SYSTEM_PROMPT), # input_current_text_template
        ["<|system|>", "<|end|>", "<|assistant|>", "<|user|>"], # special_tokens_to_keep
        '{\n\t"max_new_tokens": 50\n}',  # generation_kwargs
    )

def update_code_snippets_fn(
    input_current_text: str,
    input_context_text: str,
    output_current_text: str,
    output_context_text: str,
    model_name_or_path: str,
    attribution_method: str,
    attributed_fn: str | None,
    context_sensitivity_metric: str,
    context_sensitivity_std_threshold: float,
    context_sensitivity_topk: int,
    attribution_std_threshold: float,
    attribution_topk: int,
    input_template: str,
    output_template: str,
    contextless_input_template: str,
    contextless_output_template: str,
    special_tokens_to_keep: str | list[str] | None,
    decoder_input_output_separator: str,
    model_kwargs: str,
    tokenizer_kwargs: str,
    generation_kwargs: str,
    attribution_kwargs: str,
) -> tuple[str, str]:
    if not input_current_text:
        input_current_text = "<MISSING INPUT CURRENT TEXT, REQUIRED>"
    nl = "\n"
    tq = "\"\"\""
    def escape_quotes(s: str) -> str:
        return s.replace('"', '\\"')
    def py_get_kwargs_str(kwargs: str, name: str, pad: str = " " * 4) -> str:
        kwargs_dict = json.loads(kwargs)
        return nl + pad + name + '=' + str(kwargs_dict) + ',' if kwargs_dict else ''
    def py_get_if_specified(arg: str | int | float | list | None, name: str, pad: str = " " * 4) -> str:
        if arg is None or (isinstance(arg, (str, list)) and not arg) or (isinstance(arg, (int, float)) and arg <= 0):
            return ""
        elif isinstance(arg, str):
            return nl + pad + name + "=" + tq + arg + tq + ","
        elif isinstance(arg, list):
            return nl + pad + name + "=" + str(arg) + ","
        else:
            return nl + pad + name + "=" + str(arg) + ","
    def sh_get_kwargs_str(kwargs: str, name: str, pad: str = " " * 4) -> str:
        return nl + pad + f"--{name} " + '"' + escape_quotes("".join(x.strip() for x in str(kwargs).split("\n"))) + '"' + " \\" if json.loads(kwargs) else ''
    def sh_get_if_specified(arg: str | int | float | list | None, name: str, pad: str = " " * 4) -> str:
        if arg is None or (isinstance(arg, (str, list)) and not arg) or (isinstance(arg, (int, float)) and arg <= 0):
            return ""
        elif isinstance(arg, str):
            return nl + pad + f"--{name} " + '"' + escape_quotes(arg) + '"' + " \\" 
        elif isinstance(arg, list):
            return nl + pad + f"--{name} " + " ".join(str(arg)) + " \\"
        else:
            return nl + pad + f"--{name} " + str(arg) + " \\"
    # Python
    python = f"""#!pip install inseq
import inseq
from inseq.commands.attribute_context.attribute_context import attribute_context_with_model, AttributeContextArgs

inseq_model = inseq.load_model(
    "{model_name_or_path}",
    "{attribution_method}",{py_get_kwargs_str(model_kwargs, "model_kwargs")}{py_get_kwargs_str(tokenizer_kwargs, "tokenizer_kwargs")}
)

pecore_args = AttributeContextArgs(
    model_name_or_path="{model_name_or_path}",
    attribution_method="{attribution_method}",
    attributed_fn="{attributed_fn}",
    context_sensitivity_metric="{context_sensitivity_metric}",
    context_sensitivity_std_threshold={context_sensitivity_std_threshold},{py_get_if_specified(context_sensitivity_topk, "context_sensitivity_topk")}
    attribution_std_threshold={attribution_std_threshold},{py_get_if_specified(attribution_topk, "attribution_topk")}
    input_current_text=\"\"\"{input_current_text}\"\"\",{py_get_if_specified(input_context_text, "input_context_text")}
    contextless_input_current_text=\"\"\"{contextless_input_template}\"\"\",
    input_template=\"\"\"{input_template}\"\"\",{py_get_if_specified(output_current_text, "output_current_text")}{py_get_if_specified(output_context_text, "output_context_text")}
    contextless_output_current_text=\"\"\"{contextless_output_template}\"\"\",
    output_template="{output_template}",{py_get_if_specified(special_tokens_to_keep, "special_tokens_to_keep")}{py_get_if_specified(decoder_input_output_separator, "decoder_input_output_separator")}
    save_path="pecore_output.json",
    viz_path="pecore_output.html",{py_get_kwargs_str(model_kwargs, "model_kwargs")}{py_get_kwargs_str(tokenizer_kwargs, "tokenizer_kwargs")}{py_get_kwargs_str(generation_kwargs, "generation_kwargs")}{py_get_kwargs_str(attribution_kwargs, "attribution_kwargs")}
)

out = attribute_context_with_model(pecore_args, inseq_model)"""
    # Bash
    bash = f"""# pip install inseq
inseq attribute-context \\
    --model_name_or_path "{model_name_or_path}" \\
    --attribution_method "{attribution_method}" \\
    --attributed_fn "{attributed_fn}" \\
    --context_sensitivity_metric "{context_sensitivity_metric}" \\
    --context_sensitivity_std_threshold {context_sensitivity_std_threshold} \\{sh_get_if_specified(context_sensitivity_topk, "context_sensitivity_topk")}
    --attribution_std_threshold {attribution_std_threshold} \\{sh_get_if_specified(attribution_topk, "attribution_topk")}
    --input_current_text "{escape_quotes(input_current_text)}" \\{sh_get_if_specified(input_context_text, "input_context_text")}
    --contextless_input_current_text "{escape_quotes(contextless_input_template)}" \\
    --input_template "{escape_quotes(input_template)}" \\{sh_get_if_specified(output_current_text, "output_current_text")}{sh_get_if_specified(output_context_text, "output_context_text")}
    --contextless_output_current_text "{escape_quotes(contextless_output_template)}" \\
    --output_template "{escape_quotes(output_template)}" \\{sh_get_if_specified(special_tokens_to_keep, "special_tokens_to_keep")}{sh_get_if_specified(decoder_input_output_separator, "decoder_input_output_separator")}
    --save_path "pecore_output.json" \\
    --viz_path "pecore_output.html" \\{sh_get_kwargs_str(model_kwargs, "model_kwargs")}{sh_get_kwargs_str(tokenizer_kwargs, "tokenizer_kwargs")}{sh_get_kwargs_str(generation_kwargs, "generation_kwargs")}{sh_get_kwargs_str(attribution_kwargs, "attribution_kwargs")}""".strip("\\")
    return python, bash