File size: 10,130 Bytes
301dd5b
 
3bf2541
301dd5b
3bf2541
301dd5b
 
 
 
 
3bf2541
301dd5b
 
 
 
 
 
3bf2541
 
301dd5b
 
3bf2541
 
301dd5b
 
3bf2541
 
301dd5b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3bf2541
301dd5b
 
 
3bf2541
301dd5b
 
 
 
 
3bf2541
301dd5b
 
 
 
 
 
3bf2541
 
301dd5b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3bf2541
 
301dd5b
 
 
3bf2541
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
301dd5b
 
3bf2541
 
 
 
301dd5b
 
 
 
 
3bf2541
301dd5b
 
 
 
 
3bf2541
 
 
 
 
301dd5b
 
3bf2541
 
301dd5b
 
 
 
3bf2541
301dd5b
 
 
 
 
 
 
 
3bf2541
 
 
301dd5b
 
 
 
 
 
3bf2541
 
301dd5b
 
 
 
3bf2541
301dd5b
 
 
 
 
 
 
 
 
 
 
 
 
3bf2541
 
 
 
301dd5b
 
 
 
3bf2541
 
 
 
 
 
301dd5b
 
 
 
 
 
 
 
 
 
 
 
3bf2541
 
301dd5b
 
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
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
from __future__ import annotations

import re
import logging
from enum import Enum
from typing import Optional

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel, Field
import tiktoken

logger = logging.getLogger("promptzip")

# ── App ───────────────────────────────────────────────────────────────────────
app = FastAPI(
    title="PromptZip API",
    description="Compress large text, code, and logs to save LLM context window space.",
    version="0.1.0",
)

origins = ["*"]

app.add_middleware(
    CORSMiddleware,
    allow_origins=origins,
    allow_credentials=False,
    allow_methods=["*"],
    allow_headers=["*"],
)

# ── Tokenizer (loaded once at startup) ───────────────────────────────────────
_encoder = tiktoken.get_encoding("cl100k_base")
_COST_PER_MILLION: float = 5.00  # USD β€” GPT-4o standard input rate


def count_tokens(text: str) -> int:
    """Exact token count via cl100k_base (GPT-4 / GPT-4o)."""
    return len(_encoder.encode(text))


def estimate_cost(token_count: int) -> float:
    """USD cost rounded to 6 dp at $5.00 / 1 M tokens."""
    return round((token_count / 1_000_000) * _COST_PER_MILLION, 6)


# ── LLMlingua (optional β€” lazy-loaded so startup is never blocked) ────────────
_llmlingua_compressor = None
_llmlingua_error: Optional[str] = None


def _get_llmlingua():
    """Return a cached PromptCompressor, or raise HTTPException if unavailable."""
    global _llmlingua_compressor, _llmlingua_error
    if _llmlingua_compressor is not None:
        return _llmlingua_compressor
    if _llmlingua_error is not None:
        raise HTTPException(
            status_code=503,
            detail=f"LLMlingua failed to load: {_llmlingua_error}. "
                   "Use mode='code' or mode='logs' for regex-based compression.",
        )
    try:
        from llmlingua import PromptCompressor
        _llmlingua_compressor = PromptCompressor(
            model_name="microsoft/llmlingua-2-bert-base-multilingual-cased-meetingbank",
            use_llmlingua2=True,
            device_map="cpu",
        )
        logger.info("LLMlingua initialised successfully.")
        return _llmlingua_compressor
    except Exception as exc:
        _llmlingua_error = str(exc)
        logger.error("LLMlingua init failed: %s", exc)
        raise HTTPException(
            status_code=503,
            detail=f"LLMlingua failed to load: {exc}. "
                   "Use mode='code' or mode='logs' for regex-based compression.",
        )


# ── Compression logic ─────────────────────────────────────────────────────────

# Aggression β†’ target retention ratio for LLMlingua
_TEXT_RATIOS = {1: 0.8, 2: 0.6, 3: 0.4}


def compress_logs(text: str, aggression: int) -> str:
    """
    Regex-based log compression:
      1. Strip common timestamp patterns.
      2. Optionally strip IPv4 addresses (aggression >= 2).
      3. Collapse consecutive duplicate lines (repeating errors).
      4. Collapse runs of blank lines to a single blank.
    """
    # --- Timestamps ---
    # ISO-8601 / syslog / common log format variants
    timestamp_patterns = [
        # [2023-10-12 14:00:00.123] or 2023-10-12T14:00:00Z
        r"\d{4}-\d{2}-\d{2}[T ]\d{2}:\d{2}:\d{2}(?:[.,]\d+)?(?:Z|[+-]\d{2}:\d{2})?\s*",
        # [12/Oct/2023:14:00:00 +0000]
        r"\[\d{2}/\w+/\d{4}:\d{2}:\d{2}:\d{2} [+-]\d{4}\]\s*",
        # Jan 12 14:00:00
        r"\b(?:Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec)\s+\d{1,2}\s+\d{2}:\d{2}:\d{2}\s*",
        # [14:00:00]
        r"\[\d{2}:\d{2}:\d{2}(?:\.\d+)?\]\s*",
    ]
    for pat in timestamp_patterns:
        text = re.sub(pat, "", text)

    # --- IP addresses (aggression >= 2) ---
    if aggression >= 2:
        text = re.sub(
            r"\b(?:\d{1,3}\.){3}\d{1,3}(?::\d+)?\b",
            "<ip>",
            text,
        )

    # --- Collapse consecutive duplicate lines ---
    lines = text.splitlines()
    deduped: list[str] = []
    prev = None
    repeat_count = 0
    for line in lines:
        stripped = line.strip()
        if stripped == prev and stripped:  # skip blank dedup here
            repeat_count += 1
            if repeat_count == 1:
                deduped.append(f"  [repeated {repeat_count + 1}x ↑]")
            else:
                deduped[-1] = f"  [repeated {repeat_count + 1}x ↑]"
        else:
            repeat_count = 0
            deduped.append(line)
            prev = stripped

    # --- Collapse blank lines ---
    text = "\n".join(deduped)
    text = re.sub(r"\n{3,}", "\n\n", text)

    # --- Strip leading/trailing whitespace per line (aggression 3) ---
    if aggression == 3:
        text = "\n".join(l.strip() for l in text.splitlines())
        text = re.sub(r"\n{2,}", "\n", text)

    return text.strip()


def compress_code(text: str, aggression: int) -> str:
    """
    Regex-based code comment & whitespace stripping:
      - Remove /* ... */ block comments (including docblock variants /** */)
      - Remove Python/Ruby # single-line comments
      - Remove C++/JS // single-line comments
      - Remove Python/Java triple-quoted docstrings (aggression >= 2)
      - Remove blank / whitespace-only lines (aggression >= 2)
      - Strip trailing whitespace and over-indent (aggression 3)
    """
    # --- Block comments: /* ... */ (non-greedy, dotall) ---
    text = re.sub(r"/\*.*?\*/", "", text, flags=re.DOTALL)

    # --- Triple-quoted Python docstrings (aggression >= 2) ---
    if aggression >= 2:
        text = re.sub(r'""".*?"""', "", text, flags=re.DOTALL)
        text = re.sub(r"'''.*?'''", "", text, flags=re.DOTALL)

    # --- Single-line comments ---
    # // comments (not inside strings β€” best-effort with regex)
    text = re.sub(r"(?m)(?<!:)//.*$", "", text)
    # # comments β€” skip shebang on line 1
    text = re.sub(r"(?m)(?<!^#!)#.*$", "", text)

    # --- Trailing whitespace ---
    text = re.sub(r"(?m)[ \t]+$", "", text)

    # --- Blank lines (aggression >= 2) ---
    if aggression >= 2:
        text = re.sub(r"\n{2,}", "\n", text)

    # --- Aggressive: remove all indentation & collapse to single lines per block ---
    if aggression == 3:
        text = re.sub(r"(?m)^[ \t]+", "", text)

    return text.strip()


def compress_text(text: str, aggression: int) -> str:
    """
    Semantic compression via LLMlingua PromptCompressor.
    Falls back gracefully if the model cannot be loaded.
    """
    compressor = _get_llmlingua()
    ratio = _TEXT_RATIOS[aggression]
    result = compressor.compress_prompt(
        text,
        rate=ratio,
        force_tokens=["\n"],     # preserve newline structure
    )
    return result.get("compressed_prompt", text)


# ── Schemas ───────────────────────────────────────────────────────────────────
class Mode(str, Enum):
    text = "text"
    code = "code"
    logs = "logs"


class CompressRequest(BaseModel):
    text: str = Field(..., description="The raw text, code, or log to compress.")
    mode: Mode = Field(Mode.text, description="Compression strategy: text | code | logs.")
    aggression_level: int = Field(
        2,
        ge=1,
        le=3,
        description="1 = gentle, 2 = balanced, 3 = aggressive.",
    )


class CompressResponse(BaseModel):
    compressed_text: str
    original_tokens: int
    new_tokens: int
    tokens_saved: int
    percent_saved: float = Field(..., description="Percentage of tokens removed.")
    dollars_saved: float = Field(..., description="Estimated API cost delta in USD.")
    mode: Mode
    aggression_level: int


# ── Endpoints ─────────────────────────────────────────────────────────────────
@app.get("/health", tags=["Health"])
async def health_check():
    """Check that the API is alive and responding."""
    return {"status": "ok", "service": "promptzip-api"}


@app.post("/api/tokenize", tags=["Tokenizer"])
async def tokenize(body: dict):
    """Count exact tokens and return estimated cost."""
    text = body.get("text", "")
    tokens = count_tokens(text)
    return {
        "token_count": tokens,
        "estimated_cost_usd": estimate_cost(tokens),
        "encoding": "cl100k_base",
        "rate_per_million_usd": _COST_PER_MILLION,
    }


@app.post("/api/compress", response_model=CompressResponse, tags=["Compress"])
async def compress(body: CompressRequest):
    """
    Compress *text* using the chosen strategy:
    - **logs** β€” regex strips timestamps, IPs, and repeating lines
    - **code** β€” regex strips comments, docstrings, blank lines
    - **text** β€” semantic compression via LLMlingua PromptCompressor
    """
    if not body.text.strip():
        raise HTTPException(status_code=400, detail="text must not be empty.")

    dispatch = {
        Mode.logs: compress_logs,
        Mode.code: compress_code,
        Mode.text: compress_text,
    }
    compressed = dispatch[body.mode](body.text, body.aggression_level)

    original_tokens = count_tokens(body.text)
    new_tokens = count_tokens(compressed)
    saved = original_tokens - new_tokens
    pct = round((saved / original_tokens) * 100, 2) if original_tokens else 0.0

    return CompressResponse(
        compressed_text=compressed,
        original_tokens=original_tokens,
        new_tokens=new_tokens,
        tokens_saved=saved,
        percent_saved=pct,
        dollars_saved=round(estimate_cost(original_tokens) - estimate_cost(new_tokens), 6),
        mode=body.mode,
        aggression_level=body.aggression_level,
    )