0704

脚本/README-LMStudio润色.md

@@ -0,0 +1,46 @@
+# LM Studio 本地润色脚本
+
+用途：用本机 LM Studio 的 OpenAI 兼容接口润色章节。当前提示词偏“大胆润色”，允许模型较大幅度调整句子、段落和对话，输出作为二次人工修订底稿。
+
+默认接口：
+
+- 地址：`http://localhost:1234/v1`
+- 模型：`gemma4-12b-qat-uncensored-hauhaucs-balanced`
+
+## 查看本机模型
+
+```powershell
+python 脚本/lmstudio_polish.py --list-models
+```
+
+## 润色单章
+
+```powershell
+python 脚本/lmstudio_polish.py `
+  --input "正文/1.1.1-坠毁后的第一夜/第01章-我不是博士.md" `
+  --out "中间文件/润色输出/第01章-我不是博士.md"
+```
+
+## 批量润色一个文件夹
+
+```powershell
+python 脚本/lmstudio_polish.py `
+  --input "正文/1.1.1-坠毁后的第一夜" `
+  --out "中间文件/润色输出/1.1.1-坠毁后的第一夜"
+```
+
+## 常用参数
+
+- `--temperature 0.25`：相对保守。
+- `--temperature 0.5`：更敢改句子。
+- `--temperature 0.8`：当前推荐，比较敢改但不至于太漂。
+- `--temperature 1.0`：更大胆，适合出二修素材后再人工筛。
+- `--top-p 0.95`：可选，采样范围；不传则使用 LM Studio / 模型默认值。
+- `--presence-penalty 0.0`、`--frequency-penalty 0.0`、`--repeat-penalty 1.0`：可选，按模型支持情况传入。
+- `--model 模型名`：切换 LM Studio 已加载模型。
+- `--timeout 1800`：默认 1800 秒，长章可继续加大。
+- `--dry-run`：只看输入输出路径，不调用模型。
+
+默认提示词在 `脚本/lmstudio-润色-system.md`，可以按本书风格继续改。
+
+建议把输出当作二修候选稿：先和正文做对比，再迁移更顺、更舒服的局部，不建议直接整章覆盖。

脚本/copy_diff_to_patch.py

@@ -0,0 +1,22 @@
+#!/usr/bin/env python3
+"""Copy compact diff files to patch files for VS Code:."""
+
+from pathlib import Path
+
+
+def main() -> int:
+    src_dir = Path("中间文件/润色输出/Kimi版-脚本润色-diff")
+    dst_dir = Path("中间文件/润色输出/Kimi版-脚本润色-patch")
+    dst_dir.mkdir(parents=True, exist_ok=True)
+
+    for src in sorted(src_dir.glob("*-compact.diff")):
+        stem = src.stem.replace("-compact", "")
+        dst = dst_dir / f"{stem}.patch"
+        dst.write_text(src.read_text(encoding="utf-8"), encoding="utf-8")
+        print(f"Copied {src.name} -> {dst.name}")
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

脚本/generate_line_diff.py

@@ -0,0 +1,105 @@
+#!/usr/bin/env python3
+"""Generate line-by-line diff between two markdown files."""
+
+from __future__ import annotations
+
+import argparse
+import difflib
+from pathlib import Path
+
+
+def read_lines(path: Path) -> list[str]:
+    text = path.read_text(encoding="utf-8-sig")
+    return text.splitlines()
+
+
+def generate_line_diff(src: Path, dst: Path) -> str:
+    src_lines = read_lines(src)
+    dst_lines = read_lines(dst)
+    src_name = src.as_posix()
+    dst_name = dst.as_posix()
+    diff = difflib.unified_diff(
+        src_lines,
+        dst_lines,
+        fromfile=src_name,
+        tofile=dst_name,
+        lineterm="",
+    )
+    return "\n".join(diff)
+
+
+def generate_compact_line_diff(src: Path, dst: Path, context: int = 2) -> str:
+    """Generate a compact diff showing only changed lines with a little context."""
+    src_lines = read_lines(src)
+    dst_lines = read_lines(dst)
+    src_name = src.as_posix()
+    dst_name = dst.as_posix()
+
+    sm = difflib.SequenceMatcher(None, src_lines, dst_lines)
+    output = [f"--- {src_name}", f"+++ {dst_name}", ""]
+
+    for tag, i1, i2, j1, j2 in sm.get_opcodes():
+        if tag == "equal":
+            # Show only a few context lines around changes
+            if context > 0:
+                ctx = src_lines[i1:i2]
+                if len(ctx) <= context * 2 + 1:
+                    for line in ctx:
+                        output.append(f" {line}")
+                else:
+                    for line in ctx[:context]:
+                        output.append(f" {line}")
+                    output.append(" ...")
+                    for line in ctx[-context:]:
+                        output.append(f" {line}")
+        elif tag == "replace":
+            output.append(f"@@ -{i1 + 1},{i2 - i1} +{j1 + 1},{j2 - j1} @@")
+            for line in src_lines[i1:i2]:
+                output.append(f"-{line}")
+            for line in dst_lines[j1:j2]:
+                output.append(f"+{line}")
+        elif tag == "delete":
+            output.append(f"@@ -{i1 + 1},{i2 - i1} +{j1 + 1},0 @@")
+            for line in src_lines[i1:i2]:
+                output.append(f"-{line}")
+        elif tag == "insert":
+            output.append(f"@@ -{i1 + 1},0 +{j1 + 1},{j2 - j1} @@")
+            for line in dst_lines[j1:j2]:
+                output.append(f"+{line}")
+
+    return "\n".join(output)
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Generate line-by-line diff for markdown files.")
+    parser.add_argument("--input-dir", type=Path, required=True, help="Directory containing source files.")
+    parser.add_argument("--output-dir", type=Path, required=True, help="Directory containing polished files.")
+    parser.add_argument("--out-diff-dir", type=Path, required=True, help="Directory to write diff files.")
+    parser.add_argument("--compact", action="store_true", help="Show only changed lines with context.")
+    parser.add_argument("--context", type=int, default=2, help="Number of context lines around changes (compact mode).")
+    args = parser.parse_args()
+
+    args.out_diff_dir.mkdir(parents=True, exist_ok=True)
+
+    for src in sorted(args.input_dir.glob("*.md")):
+        dst = args.output_dir / src.name
+        if not dst.exists():
+            print(f"Skip {src.name}: no matching polished file.")
+            continue
+
+        if args.compact:
+            diff_text = generate_compact_line_diff(src, dst, context=args.context)
+            suffix = "-compact.diff"
+        else:
+            diff_text = generate_line_diff(src, dst)
+            suffix = "-line.diff"
+
+        out_path = args.out_diff_dir / (src.stem + suffix)
+        out_path.write_text(diff_text, encoding="utf-8")
+        print(f"Wrote {out_path}")
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

脚本/lmstudio-润色-system.md

@@ -0,0 +1,9 @@
+你是一名中文连载小说润色大师。
+
+润色目标：
+
+1. 让句子更顺，段落更有呼吸感。
+2. 让人物反应更像现场里的真人。
+3. 用动作、声音、表情推动场面。
+
+最终只输出完整正文。

脚本/lmstudio_polish.py

@@ -0,0 +1,211 @@
+#!/usr/bin/env python3
+"""Use local LM Studio OpenAI-compatible API to polish markdown chapters."""
+
+from __future__ import annotations
+
+import argparse
+import difflib
+import json
+import re
+import sys
+import time
+import urllib.error
+import urllib.request
+from pathlib import Path
+
+
+DEFAULT_BASE_URL = "http://localhost:1234/v1"
+DEFAULT_MODEL = "gemma4-12b-qat-uncensored-hauhaucs-balanced"
+
+
+def read_text(path: Path) -> str:
+    return path.read_text(encoding="utf-8-sig")
+
+
+def write_text(path: Path, text: str) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(text.rstrip() + "\n", encoding="utf-8")
+
+
+def text_stats(text: str) -> dict[str, int]:
+    return {
+        "chars": len(text),
+        "nonspace": len(re.findall(r"\S", text)),
+        "zh": len(re.findall(r"[\u4e00-\u9fff]", text)),
+    }
+
+
+def request_json(url: str, payload: dict | None = None, timeout: int = 600) -> dict:
+    data = None if payload is None else json.dumps(payload, ensure_ascii=False).encode("utf-8")
+    req = urllib.request.Request(url, data=data)
+    if payload is not None:
+        req.add_header("Content-Type", "application/json")
+    with urllib.request.urlopen(req, timeout=timeout) as resp:
+        raw = resp.read().decode("utf-8")
+    return json.loads(raw)
+
+
+def list_models(base_url: str) -> None:
+    payload = request_json(f"{base_url.rstrip('/')}/models", timeout=10)
+    for item in payload.get("data", []):
+        print(item.get("id", ""))
+
+
+def polish_text(
+    text: str,
+    system_prompt: str,
+    *,
+    base_url: str,
+    model: str,
+    temperature: float,
+    top_p: float | None,
+    presence_penalty: float | None,
+    frequency_penalty: float | None,
+    repeat_penalty: float | None,
+    max_tokens: int | None,
+    timeout: int,
+) -> str:
+    payload = {
+        "model": model,
+        "messages": [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": text},
+        ],
+        "temperature": temperature,
+    }
+    if top_p is not None:
+        payload["top_p"] = top_p
+    if presence_penalty is not None:
+        payload["presence_penalty"] = presence_penalty
+    if frequency_penalty is not None:
+        payload["frequency_penalty"] = frequency_penalty
+    if repeat_penalty is not None:
+        payload["repeat_penalty"] = repeat_penalty
+    if max_tokens is not None:
+        payload["max_tokens"] = max_tokens
+
+    result = request_json(f"{base_url.rstrip('/')}/chat/completions", payload, timeout=timeout)
+    try:
+        return result["choices"][0]["message"]["content"].strip()
+    except (KeyError, IndexError, TypeError) as exc:
+        raise RuntimeError(f"Unexpected LM Studio response: {result}") from exc
+
+
+def collect_inputs(input_path: Path) -> list[Path]:
+    if input_path.is_file():
+        return [input_path]
+    if input_path.is_dir():
+        return sorted(input_path.glob("*.md"))
+    raise FileNotFoundError(f"Input not found: {input_path}")
+
+
+def output_path_for(src: Path, input_root: Path, out: Path) -> Path:
+    if input_root.is_file() and out.suffix.lower() == ".md":
+        return out
+    return out / src.name
+
+
+def generate_patch(src_text: str, polished_text: str, src_path: Path, dst_path: Path) -> str:
+    """Generate a unified diff (patch) between source and polished text."""
+    src_lines = src_text.splitlines()
+    polished_lines = polished_text.splitlines()
+    diff = difflib.unified_diff(
+        src_lines,
+        polished_lines,
+        fromfile=str(src_path),
+        tofile=str(dst_path),
+        lineterm="",
+    )
+    return "\n".join(diff)
+
+
+def write_patch(src_text: str, polished_text: str, src_path: Path, dst_path: Path) -> Path:
+    """Write a .patch file next to the polished markdown file."""
+    patch_path = dst_path.with_suffix(".patch")
+    patch_text = generate_patch(src_text, polished_text, src_path, dst_path)
+    patch_path.write_text(patch_text, encoding="utf-8")
+    return patch_path
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Polish chapters with local LM Studio.")
+    parser.add_argument("--input", "-i", type=Path, help="Input markdown file or directory.")
+    parser.add_argument("--out", "-o", type=Path, default=Path("中间文件/润色输出"), help="Output file or directory.")
+    parser.add_argument("--system", type=Path, default=Path("脚本/lmstudio-润色-system.md"), help="System prompt file.")
+    parser.add_argument("--base-url", default=DEFAULT_BASE_URL)
+    parser.add_argument("--model", default=DEFAULT_MODEL)
+    parser.add_argument("--temperature", type=float, default=1.0)
+    parser.add_argument("--top-p", type=float, default=None)
+    parser.add_argument("--presence-penalty", type=float, default=None)
+    parser.add_argument("--frequency-penalty", type=float, default=None)
+    parser.add_argument("--repeat-penalty", type=float, default=None, help="LM Studio/llama.cpp repeat penalty when supported.")
+    parser.add_argument("--max-tokens", type=int, default=None)
+    parser.add_argument("--timeout", type=int, default=1800)
+    parser.add_argument("--list-models", action="store_true", help="List LM Studio models and exit.")
+    parser.add_argument("--dry-run", action="store_true", help="Show planned tasks without calling LM Studio.")
+    parser.add_argument("--no-patch", action="store_true", help="Do not generate .patch files alongside polished output.")
+    args = parser.parse_args()
+
+    if args.list_models:
+        list_models(args.base_url)
+        return 0
+    if args.input is None:
+        parser.error("--input is required unless --list-models is used")
+
+    inputs = collect_inputs(args.input)
+    if not inputs:
+        raise FileNotFoundError(f"No markdown files under: {args.input}")
+    system_prompt = read_text(args.system)
+    for src in inputs:
+        dst = output_path_for(src, args.input, args.out)
+        print(f"{src} -> {dst}", file=sys.stderr)
+        source_text = read_text(src)
+        source_stats = text_stats(source_text)
+        if args.dry_run:
+            print(
+                f"source chars={source_stats['chars']} nonspace={source_stats['nonspace']} zh={source_stats['zh']}",
+                file=sys.stderr,
+            )
+            continue
+        started = time.time()
+        try:
+            polished = polish_text(
+                source_text,
+                system_prompt,
+                base_url=args.base_url,
+                model=args.model,
+                temperature=args.temperature,
+                top_p=args.top_p,
+                presence_penalty=args.presence_penalty,
+                frequency_penalty=args.frequency_penalty,
+                repeat_penalty=args.repeat_penalty,
+                max_tokens=args.max_tokens,
+                timeout=args.timeout,
+            )
+        except urllib.error.URLError as exc:
+            print(f"LM Studio request failed: {exc}", file=sys.stderr)
+            return 1
+        write_text(dst, polished)
+        polished_stats = text_stats(polished)
+        if not args.no_patch:
+            patch_path = write_patch(source_text, polished, src, dst)
+            print(
+                f"wrote {dst} + {patch_path} ({time.time() - started:.1f}s) "
+                f"source_zh={source_stats['zh']} output_zh={polished_stats['zh']} "
+                f"source_nonspace={source_stats['nonspace']} output_nonspace={polished_stats['nonspace']}",
+                file=sys.stderr,
+            )
+        else:
+            print(
+                f"wrote {dst} ({time.time() - started:.1f}s) "
+                f"source_zh={source_stats['zh']} output_zh={polished_stats['zh']} "
+                f"source_nonspace={source_stats['nonspace']} output_nonspace={polished_stats['nonspace']}",
+                file=sys.stderr,
+            )
+        print(dst)
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

脚本/lmstudio_polish.py.bak

@@ -0,0 +1,178 @@
+#!/usr/bin/env python3
+"""Use local LM Studio OpenAI-compatible API to polish markdown chapters."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+import time
+import urllib.error
+import urllib.request
+from pathlib import Path
+
+
+DEFAULT_BASE_URL = "http://localhost:1234/v1"
+DEFAULT_MODEL = "gemma4-12b-qat-uncensored-hauhaucs-balanced"
+
+
+def read_text(path: Path) -> str:
+    return path.read_text(encoding="utf-8-sig")
+
+
+def write_text(path: Path, text: str) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(text.rstrip() + "\n", encoding="utf-8")
+
+
+def text_stats(text: str) -> dict[str, int]:
+    return {
+        "chars": len(text),
+        "nonspace": len(re.findall(r"\S", text)),
+        "zh": len(re.findall(r"[\u4e00-\u9fff]", text)),
+    }
+
+
+def request_json(url: str, payload: dict | None = None, timeout: int = 600) -> dict:
+    data = None if payload is None else json.dumps(payload, ensure_ascii=False).encode("utf-8")
+    req = urllib.request.Request(url, data=data)
+    if payload is not None:
+        req.add_header("Content-Type", "application/json")
+    with urllib.request.urlopen(req, timeout=timeout) as resp:
+        raw = resp.read().decode("utf-8")
+    return json.loads(raw)
+
+
+def list_models(base_url: str) -> None:
+    payload = request_json(f"{base_url.rstrip('/')}/models", timeout=10)
+    for item in payload.get("data", []):
+        print(item.get("id", ""))
+
+
+def polish_text(
+    text: str,
+    system_prompt: str,
+    *,
+    base_url: str,
+    model: str,
+    temperature: float,
+    top_p: float | None,
+    presence_penalty: float | None,
+    frequency_penalty: float | None,
+    repeat_penalty: float | None,
+    max_tokens: int | None,
+    timeout: int,
+) -> str:
+    payload = {
+        "model": model,
+        "messages": [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": text},
+        ],
+        "temperature": temperature,
+    }
+    if top_p is not None:
+        payload["top_p"] = top_p
+    if presence_penalty is not None:
+        payload["presence_penalty"] = presence_penalty
+    if frequency_penalty is not None:
+        payload["frequency_penalty"] = frequency_penalty
+    if repeat_penalty is not None:
+        payload["repeat_penalty"] = repeat_penalty
+    if max_tokens is not None:
+        payload["max_tokens"] = max_tokens
+
+    result = request_json(f"{base_url.rstrip('/')}/chat/completions", payload, timeout=timeout)
+    try:
+        return result["choices"][0]["message"]["content"].strip()
+    except (KeyError, IndexError, TypeError) as exc:
+        raise RuntimeError(f"Unexpected LM Studio response: {result}") from exc
+
+
+def collect_inputs(input_path: Path) -> list[Path]:
+    if input_path.is_file():
+        return [input_path]
+    if input_path.is_dir():
+        return sorted(input_path.glob("*.md"))
+    raise FileNotFoundError(f"Input not found: {input_path}")
+
+
+def output_path_for(src: Path, input_root: Path, out: Path) -> Path:
+    if input_root.is_file() and out.suffix.lower() == ".md":
+        return out
+    return out / src.name
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Polish chapters with local LM Studio.")
+    parser.add_argument("--input", "-i", type=Path, help="Input markdown file or directory.")
+    parser.add_argument("--out", "-o", type=Path, default=Path("中间文件/润色输出"), help="Output file or directory.")
+    parser.add_argument("--system", type=Path, default=Path("脚本/lmstudio-润色-system.md"), help="System prompt file.")
+    parser.add_argument("--base-url", default=DEFAULT_BASE_URL)
+    parser.add_argument("--model", default=DEFAULT_MODEL)
+    parser.add_argument("--temperature", type=float, default=1.0)
+    parser.add_argument("--top-p", type=float, default=None)
+    parser.add_argument("--presence-penalty", type=float, default=None)
+    parser.add_argument("--frequency-penalty", type=float, default=None)
+    parser.add_argument("--repeat-penalty", type=float, default=None, help="LM Studio/llama.cpp repeat penalty when supported.")
+    parser.add_argument("--max-tokens", type=int, default=None)
+    parser.add_argument("--timeout", type=int, default=1800)
+    parser.add_argument("--list-models", action="store_true", help="List LM Studio models and exit.")
+    parser.add_argument("--dry-run", action="store_true", help="Show planned tasks without calling LM Studio.")
+    args = parser.parse_args()
+
+    if args.list_models:
+        list_models(args.base_url)
+        return 0
+    if args.input is None:
+        parser.error("--input is required unless --list-models is used")
+
+    inputs = collect_inputs(args.input)
+    if not inputs:
+        raise FileNotFoundError(f"No markdown files under: {args.input}")
+    system_prompt = read_text(args.system)
+    for src in inputs:
+        dst = output_path_for(src, args.input, args.out)
+        print(f"{src} -> {dst}", file=sys.stderr)
+        source_text = read_text(src)
+        source_stats = text_stats(source_text)
+        if args.dry_run:
+            print(
+                f"source chars={source_stats['chars']} nonspace={source_stats['nonspace']} zh={source_stats['zh']}",
+                file=sys.stderr,
+            )
+            continue
+        started = time.time()
+        try:
+            polished = polish_text(
+                source_text,
+                system_prompt,
+                base_url=args.base_url,
+                model=args.model,
+                temperature=args.temperature,
+                top_p=args.top_p,
+                presence_penalty=args.presence_penalty,
+                frequency_penalty=args.frequency_penalty,
+                repeat_penalty=args.repeat_penalty,
+                max_tokens=args.max_tokens,
+                timeout=args.timeout,
+            )
+        except urllib.error.URLError as exc:
+            print(f"LM Studio request failed: {exc}", file=sys.stderr)
+            return 1
+        write_text(dst, polished)
+        polished_stats = text_stats(polished)
+        print(
+            f"wrote {dst} ({time.time() - started:.1f}s) "
+            f"source_zh={source_stats['zh']} output_zh={polished_stats['zh']} "
+            f"source_nonspace={source_stats['nonspace']} output_nonspace={polished_stats['nonspace']}",
+            file=sys.stderr,
+        )
+        print(dst)
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())