粉丝与涨粉
快照 11月3日 · 7 日榜 272 个账号可排本期浏览最高的帖子
- 1
LFM3 by @liquidai is the first local AI model that consistently gets the question "when is Manchester City's next game" right on my Macbook :) Try it here: https://huggingface.co/LiquidAI/LFM3-1.7B-GGUF
- 2
Interesting, Grok @bot is currently a money-losing business, so they cut down usage credits "Elon Musk acknowledged on Sunday that the cost of running SpaceX’s AI agent Grok Bot was “unsustainably high” relative to the revenue it generated, while responding to subscriber complaints about usage limits less than two weeks after the billionaire touted the service as the fastest-growing product he had ever seen." - @Forbes Well, yes, I can see it lol. Can't use it anymore
- 3
Ugh I can’t have my pizza thanks to someone running GLM-5.3 or an abliterated DeepSeek on my local supermarket’s IT system
- 4
This extraordinary paper is now also available on Papers with Code! Find it here: https://paperswithcode.co/paper/131739
- 5
Multi-Token Prediction Done Right is currently #2 trending on @paperswithcodee Find it here: https://paperswithcode.co/pdf/2610.38331
同期站内报道里提到它
近 7 天的公开报道里没有按名字提到它的。
本周涨粉观察
全部观察每周一生成；中文区、7 日涨粉超过 1000 的账号会有条目。
发布节奏
最近 12 周每周中位浏览
最近 12 周发布时间
颜色 = 中位浏览爆款 vs 日常
爆款 = 浏览前 10%| 指标 | 爆款 | 日常 |
|---|---|---|
| 帖子 | 9 | 75 |
| 带媒体 | 66.7% | 65.3% |
| 回复 | 0% | 0% |
| 平均字数 | 177 | 205 |
| 中位浏览 | 2,023 | 291 |
打法建议
只陈述相关性- 互动最好的时段是 15:00 前后（北京时间），中位浏览 865。
- 一周里 周四 的中位浏览最高。
- 形式上 带图/视频 的中位浏览更高（350 vs 279）。
- 爆款里带媒体占 66.7%、回复占 0%；日常分别是 65.3%、0%。
- 当前节奏约 2.2 条/天，日均浏览 1,802。
内容结构
按本站分析的分类（近 120 天）还没有分类分析。
形式与语言
帖子占比- 图片44
- 纯文字29
- 视频11
回复占比 0% · 引用 53.6% · 带媒体 65.5%
互动最高的帖子
按浏览排序- 1
MiniCPM-V 4.7 running locally at 100 tokens/second! New multimodal models from @OpenBMB using only ~3 GB of memory. Below, I showcase how to run it with @llama_cpp's web UI 🧵
- 2
My prediction is that, despite Gemini 4 beating all models on most benchmarks, it won’t transfer to real world usage The last time I used Gemini for coding was Gemini-2.5-Pro-exp-03-25, released > 1.5 years ago I really hope I’m wrong though
- 3
Hi folks. Thanks for reporting https://paperswithcode.co is currently down I’m travelling South Africa for the moment and touching grass. Bear with me to make it available again. Might take some days. Planning to open-source the code base once I’m back
- 4
The Hugging Face for rl envs is literally just Hugging Face Find the new filter here: https://huggingface.co/datasets?other=rl-environment&sort=trending
- 5
I literally cannot choose between each of these tools
- 6
If you haven’t tried Jobs on @huggingface, I highly recommend it! It’s serverless compute (CPU or GPU) for anything you want to run on the cloud It’s agent-first thanks to the hf CLI and Skill. I use it to power all CRON jobs of Papers with Code, among other things Check it out at https://huggingface.co/docs/hub/jobs
- 7
Elite tweet
- 8
Papers with Code is live again! However, I hit my 5 hour limit after just 2 prompts, not sure what happened It’s temporarily fixed, better fix will be coming once I’m back
- 9
Made some improvements to the paper PDF reading experience on PwC! - faster PDF loading - jump to specific chapters Try it for example on @XiaomiMiMo's new work: https://paperswithcode.co/pdf/2610.11959
- 10
Anyone else going?
- 11
Interesting to see that the best American open-weight model is tied to the best open-weight European model Both are GPT-6 Luna-level
- 12
I’m so proud that when you ask “What’s the SOTA for small coding agents” on Papers with Code, it includes FrogNano in its answer 😍🤗 Congrats Microsoft on this release! https://staging.paperswithcode.co/share/f614f810-c938-42ce-940d-abb3d8516302
点赞、转发、浏览是采集那一刻的快照，不是帖子最终数据；时间为北京时间。KPI 卡片里的同比只在账号有 8 周以上数据时显示， 否则只看近 4 周。“打法建议”只说明数据里的相关性，不代表因果。