File size: 4,905 Bytes
c5f49b9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import json
import os
import random
import re


SEED = 1337
VAL_RATIO = 0.08
MAX_GENERAL_TRAIN = 3000
MAX_GENERAL_VAL = 420
REFINE_BOOST_RATIO = 0.65

SRC_GENERAL_TRAIN = os.path.join("data", "jarvis_train.txt")
SRC_GENERAL_VAL = os.path.join("data", "jarvis_val.txt")
SRC_REFINE_TRAIN = os.path.join("data", "jarvis_refine_train.txt")
SRC_REFINE_VAL = os.path.join("data", "jarvis_refine_val.txt")

OUT_TRAIN = os.path.join("data", "jarvis_mix_train.txt")
OUT_VAL = os.path.join("data", "jarvis_mix_val.txt")
OUT_REPORT = os.path.join("data", "jarvis_mix_report.json")

PAIR_RE = re.compile(r"User:\s*(.*?)\nAssistant:\s*(.*?)(?=\n\nUser:|\Z)", flags=re.S)

LOW_QUALITY_PATTERNS = [
    "set a clear target, run one controlled test",
    "keep only measurable improvements",
    "steps: 1) prepare ingredients, 2) cook in short stages",
]


def normalize(text: str) -> str:
    return re.sub(r"\s+", " ", text).strip()


def read_pairs(path):
    if not os.path.exists(path):
        return []
    text = open(path, "r", encoding="utf-8", errors="ignore").read()
    pairs = []
    for user, assistant in PAIR_RE.findall(text):
        u = normalize(user)
        a = normalize(assistant)
        if len(u) < 4 or len(a) < 8:
            continue
        if len(u) > 420 or len(a) > 840:
            continue
        pairs.append((u, a))
    return pairs


def dedupe(pairs):
    seen = set()
    out = []
    for u, a in pairs:
        key = (u.lower(), a.lower())
        if key in seen:
            continue
        seen.add(key)
        out.append((u, a))
    return out


def filter_low_quality(pairs):
    out = []
    for u, a in pairs:
        u_low = u.lower()
        a_low = a.lower()
        if any(p in a_low for p in LOW_QUALITY_PATTERNS):
            continue
        if "context:" in u_low:
            continue
        if "\n" in u:
            continue
        if "[" in a or "]" in a:
            continue
        if "â" in a:
            continue
        a_words = len(re.findall(r"[A-Za-z0-9']+", a))
        if a_words < 7 or a_words > 140:
            continue
        if not re.search(r"[.!?]$", a):
            continue
        out.append((u, a))
    return out


def sample_without_replacement(rng, items, count):
    if count <= 0:
        return []
    if count >= len(items):
        return list(items)
    return rng.sample(items, count)


def sample_with_replacement(rng, items, count):
    if count <= 0 or not items:
        return []
    return [rng.choice(items) for _ in range(count)]


def write_pairs(path, pairs):
    with open(path, "w", encoding="utf-8") as f:
        for u, a in pairs:
            f.write(f"User: {u}\nAssistant: {a}\n\n")


def main():
    rng = random.Random(SEED)

    general_train = dedupe(filter_low_quality(read_pairs(SRC_GENERAL_TRAIN)))
    general_val = dedupe(filter_low_quality(read_pairs(SRC_GENERAL_VAL)))
    refine_train = dedupe(filter_low_quality(read_pairs(SRC_REFINE_TRAIN)))
    refine_val = dedupe(filter_low_quality(read_pairs(SRC_REFINE_VAL)))

    rng.shuffle(general_train)
    rng.shuffle(general_val)

    general_train = general_train[:MAX_GENERAL_TRAIN]
    general_val = general_val[:MAX_GENERAL_VAL]

    refine_key = {(u.lower(), a.lower()) for u, a in refine_train}
    general_only_train = [p for p in general_train if (p[0].lower(), p[1].lower()) not in refine_key]

    train_core = refine_train + general_only_train
    rng.shuffle(train_core)

    boost_count = int(len(refine_train) * REFINE_BOOST_RATIO)
    boosted_refine = sample_with_replacement(rng, refine_train, boost_count)
    mix_train = train_core + boosted_refine
    rng.shuffle(mix_train)

    # Validation set from dedicated val files + small holdout from non-boosted train core.
    val_holdout_n = max(120, int(len(train_core) * VAL_RATIO))
    holdout = sample_without_replacement(rng, train_core, val_holdout_n)
    mix_val = dedupe(refine_val + general_val + holdout)

    # Remove validation leakage from training.
    val_key = {(u.lower(), a.lower()) for u, a in mix_val}
    mix_train = [(u, a) for u, a in mix_train if (u.lower(), a.lower()) not in val_key]
    rng.shuffle(mix_train)

    os.makedirs("data", exist_ok=True)
    write_pairs(OUT_TRAIN, mix_train)
    write_pairs(OUT_VAL, mix_val)

    report = {
        "seed": SEED,
        "val_ratio": VAL_RATIO,
        "general_train_used": len(general_train),
        "general_val_used": len(general_val),
        "refine_train_used": len(refine_train),
        "refine_val_used": len(refine_val),
        "boosted_refine_rows": len(boosted_refine),
        "mix_train_rows": len(mix_train),
        "mix_val_rows": len(mix_val),
        "out_train": OUT_TRAIN,
        "out_val": OUT_VAL,
    }
    with open(OUT_REPORT, "w", encoding="utf-8") as f:
        json.dump(report, f, indent=2)

    print(json.dumps(report, indent=2))


if __name__ == "__main__":
    main()