func init() { if user == "" { log.Fatal("$USER not set") } if home == "" { home = "/home/" + user } if gopath == "" { gopath = home + "/go" } // gopath may be overridden by --gopath flag on command line. flag.StringVar(&gopath, "gopath", gopath, "override default GOPATH") }
ctx_len이 옵션없이 실행하면 14k 밖에 되지 않는다. unsloth에서 잘못 변환한걸까.. 아니면 약간의 편법일까?
0.05.275.357 I srv load_model: loaded multimodal model, 'mmproj-F16.gguf' 0.05.430.604 I srv load_model: initializing, n_slots = 4, n_ctx_slot = 14336, kv_unified = 'true' 0.05.460.693 I srv init: chat template supports preserving reasoning, consider enabling it via --reasoning-preserve 0.05.460.766 I srv llama_server: model loaded 0.05.460.776 I srv llama_server: listening on http://0.0.0.0:8080
기본값 실행(14k ctx_len)
> 안녕?
[Start thinking]
The user is greeting in Korean, saying "안녕?" which means "Hi?" or "Hey?" in a casual way. They're being friendly and casual. I should respond naturally in Korean, matching their casual tone. [End thinking]
안녕! 😊 뭐, 궁금한 거 있어? 아니면 그냥 인사인 거야?
[ Prompt: 8.8 t/s | Generation: 11.3 t/s ]
> 너에 대해서 설명해줘
[Start thinking]
The user is asking me to describe myself. They're speaking in casual Korean. I should respond in Korean, describing what I am in a natural way. I'll describe myself as an AI assistant without revealing system prompt details. Let me be friendly and match their casual tone. [End thinking]
아, 나에 대해? 😊
나는 **Claude**라고 불리는 AI 어시스턴트이야. Anthropic에서 개발한 거고.
간단히 말하면:
- **말을 잘 하는 글쓰기/대화 AI**야. 대화, 글쓰기, 코딩, 분석, 번역, 아이디어 브레인스토밍, 문제 해결 같은 거 도와줄 수 있어. - **톤이 자연스러운 편**이야. 딱딱하기보다 편안하게 이야기하는 걸 좋아해. - **다양한 분야에서 도움**을 줄 수 있어. 공부, 업무, 창작, 고민 상담, 코드 리뷰 등등. - **기억이 없어.** (이 대화에서는 계속 기억하고 있지만, 세션이 끝나면 초기화돼.) - **내면이 없어요.** 😅 그냥 언어 패턴을 아주 잘 이해하고 조합하는 거지. - **진짜로 "생각"은 안 하고** 말하고 있는 거야.
원래 질문이나 도움이 필요한 게 있으면 언제든 말해!
[ Prompt: 15.8 t/s | Generation: 14.6 t/s ]
> 파이썬으로 셀레니움을 통해 웹을 서칭하고 텍스트만 추출하고 makrdown 으로 변환후 md 파일과 pdf로 저장하는 기능을 구현해줘
[Start thinking]
The user is asking me to implement a Python function using Selenium to: 1. Search/browse a web page 2. Extract text only 3. Convert to Markdown 4. Save as .md file and .pdf
Let me think about the best approach:
1. Selenium for web scraping/browsing 2. Extract text from the page 3. Convert to Markdown - I could use a library like `html2text` or manually extract text 4. Save as .md file 5. Convert to PDF - I can use `markdown` + `pdfkit` (wkhtmltopdf), or `fpdf`, or `weasyprint`, or `markdown2pdf`
Let me write a comprehensive, well-structured Python script that does all of this.
Dependencies: - selenium - webdriver-manager (for auto-downloading Chrome driver) - html2text (for HTML to Markdown conversion) - For PDF: I'll use `markdown` + `pdfkit` or `weasyprint`. Actually, let me use a simpler approach. `markdown` to convert MD to HTML, then `pdfkit` to convert HTML to PDF. Or I could use `fpdf` for direct PDF generation.
Actually, the most reliable approach for PDF would be: - Option 1: markdown -> HTML -> pdfkit (needs wkhtmltopdf installed) - Option 2: markdown -> fpdf (pure Python, no external deps) - Option 3: Use `markdown2pdf` package
Let me go with a practical approach: - selenium for browsing - html2text for HTML -> Markdown - markdown + pdfkit for Markdown -> PDF (or I'll use `weasyprint` as an alternative)
Actually, let me use a cleaner approach. I'll use: - selenium + webdriver-manager - BeautifulSoup or selenium's get_text() for text extraction - html2text for markdown conversion - For PDF: I'll use `markdown` library to convert to HTML, then `pdfkit` (wkhtmltopdf) or `weasyprint`
Let me write a clean, well-documented script.
Actually, for text extraction from selenium, I can get the HTML content and then convert it. Let me structure this properly.
For the PDF part, the simplest pure-Python approach would be to use `fpdf` or `reportlab`. But `markdown` -> `pdfkit` is more standard. Let me provide both options or just the most practical one.
Let me write a complete, well-structured script. [End thinking]
# Selenium + Markdown + PDF 변환 스크립트
## 전체 코드
```python import os import re import time from selenium import webdriver from seleniuhttp://m.webdriver.chrome.service import Service from seleniuhttp://m.webdriver.chrome.options import Options from seleniuhttp://m.webdriver.common.by import By from webdriver_manager.chrome import ChromeDriverManager from html2text import html2text import markdown import pdfkit # wkhtmltopdf 기반
# ============================================================ # 1. Selenium 드라이버 설정 # ============================================================ def create_driver(headless=True): """Chrome 드라이버 초기화""" options = Options() if headless: options.add_argument("--headless") options.add_argument("--disable-gpu") options.add_argument("user-agent=Mozilla/5.0 (Windows NT 10.0; Win64; x664) AppleWebKit/537.36") options.add_argument("--no-sandbox")
# ============================================================ # 2. 웹 검색 & 텍스트 추출 # ============================================================ def search_and_extract(url: str, driver=None) -> str: """ 특정 URL로 접속하여 본문 텍스트를 추출한다. - driver를 직접 만들어주는 경우와 공용 driver를 쓰는 경우 모두 지원 """ own_driver = driver is None if own_driver: driver = create_driver()
try: driver.get(url) time.sleep(2) # 렌더링 대기 (JS가 실행되는 사이트 대비)
# <main>, <article>, <body> 중 가장 적절한 컨테이너 탐색 for selector in ["//main", "//article", "//div[@class='content']", "//body"]: try: elem = driver.find_element(By.XPATH, selector) html_content = elem.get_attribute("innerHTML") break except Exception: continue
return html_content
finally: if own_driver: driver.quit()
# ============================================================ # 3. HTML → Markdown 변환 # ============================================================ def html_to_markdown(html: str) -> str: """ html2text를 사용해 HTML을 Markdown으로 변환. 불필요한 여백/빈줄 제거. """ h = html2text.HTML2Text() h.ignore_emphasis = False # *이탈* 유지 h.ignore_links = False # [링크](url) 유지 h.body_attributes = True md_text = h.handle(html)
# 3줄 이상 빈 줄 압축 md_text = re.sub(r"\n{3,}", "\n\n", md_text.strip()) return md_text
# ============================================================ # 4. 저장 ( .md / .pdf ) # ============================================================ def save_markdown(md_text: str, path: str): """md 파일 저장""" with open(path, "w", encoding="utf-8") as f: f.write(md_text) print(f" ✔ Markdown 저장: {path}")
def save_pdf(md_text: str, path: str): """ Markdown → HTML → PDF 변환. ⚠️ 시스템에 wkhtmltopdf가 설치되어 있어야 한다. - macOS : brew install wkhtmltopdf - Ubuntu: sudo apt install wkhtmltopdf """ # Markdown → HTML (헤더/코드블록/테이블 등 지원) html = markdown.markdown(md_text, extensions=["tables", "codehilite", "toc"])
# PDF 변환 옵션 config = pdfkit.Configuration( wkhtmltopdf="wkhtmltopdf", # Windows/macOS/Linux 공통 page_size="A4", encoding="UTF-8" ) pdfkit.from_string(html, path, configuration=config) print(f" ✔ PDF 저장: {path}")
# ============================================================ # 5. 메인 실행 # ============================================================ def main(): URL = "https://en.wikipedia.org/wiki/Python_(programming_language)" BASE_NAME = "python_article" # 저장 파일 이름 SAVE_DIR = "./output"
# When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`, # video frame sampling can be configured via `extra_body` (e.g., by setting `fps`). # This feature is currently supported only in vLLM. # # By default, `fps=2` and `do_sample_frames=True`. # With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate. # chat_response = client.chat.completions.create( # model="Qwen/Qwen3.8-27B", # messages=messages, # extra_body={ # "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True}, # }, # )
Reasoning Toggle content The user has provided a sequence of video frames showing archaeological tomb discoveries. The frames display:
0-10s: A tomb in Zhoujiazhuang site, Habei Province 10-20s: Interior of a Ming Dynasty tomb in Xinzhou city, Shanxi province with pottery 20-30s: A painted wooden panel with birds and flowers 30-40s: Interior with wooden furniture and pottery 40-50s: Close-up of ornate wooden chair/structure 50-60s: Two coffins/sarcophagi in a brick chamber 60-85s: A stone doorway with ornate carvings The user hasn't asked a specific question. They've just provided video frames with timestamps. This is an ambiguous request - they may want a description, analysis, or summary of the video content.
Since there's no explicit question, I should provide a helpful description/summary of what's shown in the video. The video appears to be about archaeological tomb discoveries in China, specifically Ming Dynasty tombs.
Let me look at the frames more carefully:
The first frames (0-10s) show a tomb at the Zhoujiazhuang site in Hebei Province Then the video transitions to show a Ming Dynasty tomb (1368-1644) found in Xinzhou city, Shanxi province The Xinzhou tomb shows: Black pottery vessels/urns lined up A painted wooden panel with birds (peacock?) and flowers Wooden furniture (tables, chairs, stools) in the tomb A gilded ornate chair Two coffins/sarcophagi A stone doorway with carved decorations This appears to be a news video about archaeological discoveries in China. The captions mention "Photo provided to chinadaily.com.cn" suggesting this is from a news article.
I should describe the video content. Since the user just provided video without a question, I'll provide a summary of what's shown.
Actually, looking at this more carefully, this seems to be a request where the user is showing me a video and expects me to describe or comment on it. The most helpful response would be a clear description of the video content.
Let me structure a response that summarizes the video.
This video presents a slideshow of photographs from two Chinese archaeological discoveries:
1. Zhoujiazhuang Site (Hebei Province, North China) The opening frames (0:00–0:10) show a downward view into a square, stone-lined tomb chamber. The structure has rough-hewn walls, a sloping passage on the right, and what appears to be a wooden or lacquered panel on the back wall, with smaller objects (possibly figurines or vessels) resting on a ledge.
2. Ming Dynasty Tomb (Xinzhou City, Shanxi Province) The bulk of the video (0:10 onward) documents a well-preserved tomb dating to the Ming Dynasty (1368–1644). The images include:
Blackware pottery: A row of glossy black urns and jars (some with narrow necks, others open-mouthed) arranged along a brick ledge, with a wooden pole leaning nearby. Painted woodwork: A dark lacquered panel decorated with gilded/white-painted flowers, leaves, and a bird with elaborate plumage perched on a branch. Furniture in the chamber: Wooden tables, chairs, and stools set on a stone-tiled floor, one table bearing an array of small dishes/plates. Gilded chair: A close-up of an ornate wooden chair with carved, gold-traced floral decoration and a bird motif on the backrest. Coffins/sarcophagi: Two large wooden coffin boxes (one decorated with gilded floral patterns) standing in a brick-vaulted chamber. Stone doorway: An elaborately carved white stone gate with cloud/floral relief panels, a rishi-style roof lintel, and a pair of studded doors that stand slightly ajar. The captions credit chinadaily.com.cn for the Xinzhou photographs and the National Cultural Heritage Administration for the Zhoujiazhuang image.
If you'd like, I can focus on any particular image (e.g., the painted panel, the furniture, or the stone doorway) in more detail.
Reasoning Toggle content The user has provided a video that appears to be a slideshow of still images with captions. The video shows archaeological findings: a tomb discovery at Zhoujiazhuang site in Hebei Province, and then a series of images of a Ming Dynasty tomb (1368-1644) found in Xinzhou city, Shanxi province, showing pottery jars, painted panels, furniture, coffins, and the tomb entrance.
The user hasn't asked a specific question. They've just presented the video. I need to figure out what they want. The prompt is just "Video:" followed by the frames. There's no explicit question or instruction.
Hmm, this is tricky. The user has provided a video but no question. In this kind of setup, the expected behavior is usually to describe the video content, summarize it, or answer an implicit question. Since there's no explicit instruction, I should provide a helpful summary/description of what the video shows.
Let me look carefully at the frames to describe the content accurately.
Frame 1-9 (0:00-0:09): A tomb discovered in the Zhoujiazhuang site in North China's Hebei Province. Photo courtesy of National Cultural Heritage Administration. Shows an excavated rectangular pit/tomb chamber with a wooden coffin or structure inside, earthen walls, and what appears to be a ladder or wooden beams.
Frame 10-19 (0:10-0:19): The interior of a well-preserved tomb dating to the Ming Dynasty (1368-1644) found in Xinzhou city, Shanxi province. Shows a row of dark pottery jars/vessels (some with lids, some open) lined up against a brick wall, with one jar in front. Photo provided to chinadaily.com.cn.
Frame 20-29 (0:20-0:29): A painted panel with floral designs — peonies, flowers, leaves, and what looks like a bird or fish motif, on a dark reddish-brown lacquered background.
Frame 30-39 (0:30-0:39): The interior of the tomb showing wooden furniture (chairs, a table) with small dishes/cups arranged on the table, brick walls.
Frame 40-49 (0:40-0:49): Close-up of an ornately carved and gold-decorated wooden chair with painted/gilded figures.
Frame 50-59 (0:50-0:59): Two coffins (wooden, decorated) placed in a brick chamber with arched niches.
Frame 60-... (1:00-1:25): A stone tomb entrance/structure with carved gates, in a pit/excavation.
So the video is a slideshow about Chinese archaeological tomb discoveries, primarily the Ming Dynasty tomb in Xinzhou, Shanxi.
Since no question is asked, I'll provide a clear, organized summary of the video's content. I should note that the video consists of static images with captions describing archaeological tomb discoveries in China.
Let me write a response that summarizes the video. I'll structure it with the different scenes.
Actually, I want to consider whether the user expects me to do something with tools. The available tools are get_datetime and get_info, which return the current date/time and OS info. These are not relevant to describing the video. The user didn't ask for the current time or local environment. So I shouldn't call those tools.
I'll just provide a description/summary of the video. That's the most helpful response given no explicit question.
Let me make it concise but informative.
This video is a slow slideshow of still photographs (with text captions) showing archaeological tomb discoveries in China. It moves through several distinct scenes:
Opening (0:00–0:09) – An overhead view of an excavated burial pit at the Zhoujiazhuang site in Hebei Province (northern China). You can see earthen walls forming a rectangular chamber containing a dark wooden coffin or structure, with what looks like a set of wooden beams or a ladder. The caption credits the National Cultural Heritage Administration.
Ming Dynasty tomb – pottery (0:10–0:19) – The interior of a well‑preserved Ming Dynasty (1368–1644) tomb found in Xinzhou city, Shanxi Province. A row of dark‑glazed pottery jars
$ ros2 launch kinect2_bridge kinect2_bridge_launch.yaml [INFO] [launch]: All log files can be found below /home/minimonk/.ros/log/2026-08-18-15-32-14-328668-minimonk-HP-EliteBook-2760p-7007 [INFO] [launch]: Default logging verbosity is set to INFO [INFO] [kinect2_bridge_node-1]: process started with pid [7008] [INFO] [point_cloud_xyzrgb_node-2]: process started with pid [7010] [kinect2_bridge_node-1] [INFO] [1787034734.508992751] [kinect2_bridge_node]: parameter: [kinect2_bridge_node-1] base_name: kinect2 [kinect2_bridge_node-1] sensor: default [kinect2_bridge_node-1] fps_limit: 30 [kinect2_bridge_node-1] calib_path: /home/minimonk/ros2_ws/src/kinect2_ros2/kinect2_bridge/data/ [kinect2_bridge_node-1] use_png: false [kinect2_bridge_node-1] jpeg_quality: 90 [kinect2_bridge_node-1] png_level: 1 [kinect2_bridge_node-1] depth_method: default [kinect2_bridge_node-1] depth_device: -1 [kinect2_bridge_node-1] reg_method: default [kinect2_bridge_node-1] reg_device: -1 [kinect2_bridge_node-1] max_depth: 12 [kinect2_bridge_node-1] min_depth: 0.1 [kinect2_bridge_node-1] queue_size: 5 [kinect2_bridge_node-1] bilateral_filter: true [kinect2_bridge_node-1] edge_aware_filter: true [kinect2_bridge_node-1] publish_tf: true [kinect2_bridge_node-1] base_name_tf: kinect2 [kinect2_bridge_node-1] worker_threads: 4 [kinect2_bridge_node-1] libva info: VA-API version 1.14.0 [kinect2_bridge_node-1] libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so [kinect2_bridge_node-1] libva info: Found init function __vaDriverInit_1_14 [kinect2_bridge_node-1] libva info: va_openDriver() returns 0 [kinect2_bridge_node-1] [INFO] [1787034735.139375282] [kinect2_bridge_node]: Kinect2 devices found: [kinect2_bridge_node-1] [INFO] [1787034735.139436224] [kinect2_bridge_node]: 0: 501441643042 (selected) [kinect2_bridge_node-1] [INFO] [1787034735.225374525] [kinect2_bridge_node]: starting kinect2 [kinect2_bridge_node-1] [INFO] [1787034735.540712973] [kinect2_bridge_node]: device serial: 501441643042 [kinect2_bridge_node-1] [INFO] [1787034735.540771176] [kinect2_bridge_node]: device firmware: 4.0.3911.0 [kinect2_bridge_node-1] [INFO] [1787034735.772433213] [kinect2_bridge_node]: Using CPU registration method! [kinect2_bridge_node-1] [INFO] [1787034735.772485273] [kinect2_bridge_node]: Using CPU registration method! [kinect2_bridge_node-1] [INFO] [1787034735.787628318] [kinect2_bridge_node]: waiting for clients to connect [kinect2_bridge_node-1] [INFO] [1787034736.781245426] [kinect2_bridge_node]: client connected. starting device... [kinect2_bridge_node-1] [INFO] [1787034740.153039579] [kinect2_bridge_node]: depth processing: ~614.366ms (~1.6277Hz) publishing rate: ~5.31568Hz [kinect2_bridge_node-1] [INFO] [1787034740.153092716] [kinect2_bridge_node]: color processing: ~4.92322ms (~203.119Hz) publishing rate: ~3.98676Hz [kinect2_bridge_node-1] [INFO] [1787034743.162903904] [kinect2_bridge_node]: depth processing: ~970.624ms (~1.03026Hz) publishing rate: ~3.98689Hz [kinect2_bridge_node-1] [INFO] [1787034743.162966773] [kinect2_bridge_node]: color processing: ~4.06577ms (~245.956Hz) publishing rate: ~3.98689Hz
흐음.. 되는게 없네. 비율이 왜이러냐..
cd ~/src/libfreenect2/build_cuda cmake .. -DENABLE_CUDA=ON -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda -DCUDA_HOST_COMPILER=/usr/bin/gcc-10 -DENABLE_VAAPI=OFF make -j$(nproc) sudo make install
