Footer

Your go-to destination for discovering trending open-source projects and uncovering the insights that matter

Legal

  • Terms and conditions
  • Privacy policy

More

  • Pricing
  • Advertise
  • onedump
  • gosnakego
  • Contact us

Social

  • X (Twitter)
  • GitHub
Trendshift owl
What they build speaks louder than words
© 2026 Trendshift, created by Julian Li with ❤️
SignalAdvertise
Trendshift
SignalAdvertise
Sign in
Trendshift owl

Discover trends that matter

SignalAdvertise
DailyWeeklyMonthlyYearly
Live mentionsTopics
RepositoriesDevelopers
InsightsStats
Websitebench
Websitebench
Busbar
Busbar
Omnigraph
Omnigraph
Bifrost
Bifrost

MakazhanAlpamys/Soup

Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.

#LLM training
Visit GitHub
Python
10.5k
1.8k
14 contributors
Apache License 2.0
website
AchievementsView all →
FeaturedAdvertise to 180K+ monthly visitors
Websitebench

Can AI agents reverse-engineer real websites?

Busbar

Control what AI can do before it acts

Omnigraph

Versioned graph database on object storage. Git-style workflows and multimodal retrieval

Bifrost

Fastest Enterprise AI Gateway for Routing, Governing, and Securing all your AI Traffic

Momentum

How fast MakazhanAlpamys/Soup is gaining stars, and where it lands on Trendshift and GitHub Trending

Activity and trending

Stars, forks and merged pull requests each day or month, lined up with the repository's rank on Trendshift Trending

Sep 15 – Nov 14, 2026

Stars +4,080 in 61 days

0100200300
239

Forks +795 in 61 days

0204060
43

Merged PRs 1,007 in 61 days

0204060
41

Trendshift Trendingwhere the repository ranked each day it made the top 25

Top 25 in Python

2115
Sep 15Sep 23Oct 1Oct 9Oct 17Oct 25Nov 2Nov 10
The left and right arrow keys step from day to day; Home and End go to the first and the last

Stars, forks, merged pull requests and Trendshift Trending rank, by day
DayStars gainedForks gainedMerged PRsPython
Tue, Sep 15, 2026+164+2824Not trending
Wed, Sep 16, 2026+85+106Not trending
Thu, Sep 17, 2026+53+1218Not trending
Fri, Sep 18, 2026+54+1020Not trending
Sat, Sep 19, 2026+66+1112Not trending
Sun, Sep 20, 2026+39+80Not trending
Mon, Sep 21, 2026+50+710Not trending
Tue, Sep 22, 2026+36+1217Not trending
Wed, Sep 23, 2026+55+145Not trending
Thu, Sep 24, 2026+69+1218Not trending
Fri, Sep 25, 2026+73+1813Not trending
Sat, Sep 26, 2026+94+189Not trending
Sun, Sep 27, 2026+111+2327Not trending
Mon, Sep 28, 2026+71+1931Not trending
Tue, Sep 29, 2026+199+2517#21
Wed, Sep 30, 2026+169+2427Not trending
Thu, Oct 1, 2026+120+2230Not trending
Fri, Oct 2, 2026+63+137Not trending
Sat, Oct 3, 2026+34+120Not trending
Sun, Oct 4, 2026+74+912Not trending
Mon, Oct 5, 2026+69+1226Not trending
Tue, Oct 6, 2026+74+916Not trending
Wed, Oct 7, 2026+48+1241Not trending
Thu, Oct 8, 2026+44+1132Not trending
Fri, Oct 9, 2026+65+1211Not trending
Sat, Oct 10, 2026+61+120Not trending
Sun, Oct 11, 2026+66+110Not trending
Mon, Oct 12, 2026+63+1117Not trending
Tue, Oct 13, 2026+239+4319#15
Wed, Oct 14, 2026+159+3525Not trending
Thu, Oct 15, 2026+83+1915Not trending
Fri, Oct 16, 2026+82+1519Not trending
Sat, Oct 17, 2026+73+140Not trending
Sun, Oct 18, 2026+80+1922Not trending
Mon, Oct 19, 2026+88+1725Not trending
Tue, Oct 20, 2026+51+1321Not trending
Wed, Oct 21, 2026+48+834Not trending
Thu, Oct 22, 2026+35+511Not trending
Fri, Oct 23, 2026+33+415Not trending
Sat, Oct 24, 2026+24+1217Not trending
Sun, Oct 25, 2026+37+60Not trending
Mon, Oct 26, 2026+37+1224Not trending
Tue, Oct 27, 2026+33+95Not trending
Wed, Oct 28, 2026+36+914Not trending
Thu, Oct 29, 2026+30+824Not trending
Fri, Oct 30, 2026+33+622Not trending
Sat, Oct 31, 2026+29+615Not trending
Sun, Nov 1, 2026+38+81Not trending
Mon, Nov 2, 2026+133+2219Not trending
Tue, Nov 3, 2026+64+1325Not trending
Wed, Nov 4, 2026+66+1322Not trending
Thu, Nov 5, 2026+51+719Not trending
Fri, Nov 6, 2026+49+1124Not trending
Sat, Nov 7, 2026+28+417Not trending
Sun, Nov 8, 2026+38+118Not trending
Mon, Nov 9, 2026+58+1510Not trending
Tue, Nov 10, 2026+66+89Not trending
Wed, Nov 11, 2026+27+816Not trending
Thu, Nov 12, 2026+29+1038Not trending
Fri, Nov 13, 2026+24+726Not trending
Sat, Nov 14, 2026+10+10Not trending
Total+4,080+7951,007
  • 1#1 on the list
  • 7Rank that day
  • Not trending
4

    • GitHub trending history

    Where the repository ranked on GitHub Trending, one dot for each day it was on a list

    Aug 15, 2026Best #7

    #1#5#10#15#20#25
    Top 10
    Sep 2026OctNov
    The left and right arrow keys step from day to day; Home and End go to the first and the last

    Mentions and achievements

    Where people are talking about MakazhanAlpamys/Soup, and every badge it has earned
    1. Nov 141 post

      • Your fine-tune is only as good as what you feed it. 3 things that quietly ruin a dataset: • one-word replies ("ok" × 2,000) • duplicates (50 copies = 50× the weight) • off-task data (memes, group…

        Read full on X, posted by @Souptesters

        @Souptesters · x.com

    2. Nov 131 post

      • Fine-tuning, kitchen version: The model = soup that's already cooked. Your data = the ingredients you add. Training = one more stir. You don't need a restaurant kitchen. We did it on a 4 GB laptop…

        Read full on X, posted by @Souptesters

        @Souptesters · x.com

    3. Nov 81 post

      • We fine-tuned an 8B LLM on a 4 GB laptop GPU. Peak VRAM: 3.32 GB. RTX 3050 Laptop. The trick: the frozen base stays in RAM and streams to the GPU one layer at a time. Open source. Apache-2.0. Free.

        Read full on X, posted by @Souptesters

        @Souptesters · x.com

    4. Nov 41 post

      • A detail from Soup's August NF4 write-up: a gradient check passed while comparing zero tensors. Parameter names didn't match the reference model. The broken control passed too. They fixed name…

        Read full on X, posted by @WhyFaust

        @WhyFaust · x.com

    5. Nov 31 post

      • Top AI repos on GitHub today | Wed, Nov 4 • vercel-labs/agent-desktop: A command line tool that lets an AI agent use your computer. It can click around and read what is on screen. •…

        Read full on X, posted by @VT_zero1

        @VT_zero1 · x.com

    6. 7 days earlier
    7. Oct 271 post

      • 昨晚装了 Soup：想在自己电脑上微调模型，不用再啃一堆配置地狱。 装好后 soup init --template chat，改 YAML 里的数据和模型，直接 soup train。还有个 layer streaming，把底座按层从内存/硬盘喂进显存，官方测过 4GB 笔记本显卡也能训 8B。

        Read full on X, posted by @run4want

        @run4want · x.com

    8. 13 days earlier
    9. Oct 141 post

      • 别再花钱去云端租 A100 了！ 你手里那台 4GB 显存的旧笔记本，现在能直接微调 Llama-3.1-8B 了。 传统方式一跑就是 CUDA OOM 显存暴击。这个开源新神器用「分层流式」直接破壁：显存死死咬在 3.32 GB 顶峰，真机跑出 208.6 tok/s，一行命令生成配置，断网就能炼丹。

        Read full on X, posted by @zhaozhaozhao914

        @zhaozhaozhao914 · x.com

    10. 7 days earlier
    11. Oct 72 posts

      • for anyone with a 3090 that's been sitting there since their last game → hyperqwen: 24gb of vram is enough for qwen 3.8 27b at 150k context

        Read full on X, posted by @starmexxx

        @starmexxx · x.com

      • 有人想微调大模型，一看显存要求直接放弃；有人用 Soup，4 GB 显存笔记本显卡照样调 80 亿参数——差别就在它把基座模型放内存，训练时一层层轮流搬进显卡，显存里只留当前那层。 实测显存峰值 3.32 GB，训练结果和正常跑出来完全一致。 GitHub： 用起来一份配置文件加一条命令，批大小、模型压缩方式全自动配好，也不用远程登录服务器。 作者还放了免费 Colab…

        Read full on X, posted by @Gas1688

        @Gas1688 · x.com

    12. 10 days earlier
    13. Sep 271 post

      • 本期 6 个 GitHub 热门项目： ① Ponytail：约束 Agent 少写代码 ② TSP：自托管 A 股量化 ③ Soup：YAML 微调大模型 ④ Strands：生产级 Agent SDK ⑤ CLI-Anything：让 Agent 操作 GUI 软件 ⑥ PostHog：从产品数据发现问题 #GitHub精选 #开源项目 #AI工具 #人工智能 #Agent

        Read full on X, posted by @HongyeMain

        @HongyeMain · x.com

    You might also like

    Repositories close to MakazhanAlpamys/Soup
    FeaturedAdvertise to 180K+ monthly visitors
    Omnigraph

    Versioned graph database on object storage. Git-style workflows and multimodal retrieval

    Bifrost

    Fastest Enterprise AI Gateway for Routing, Governing, and Securing all your AI Traffic

    Websitebench

    Can AI agents reverse-engineer real websites?

    Busbar

    Control what AI can do before it acts

    Goekdeniz-Guelmez/mlx-lm-lora

    Python

    Train Large Language Models on MLX.

    47166

    axolotl-ai-cloud/axolotl

    Python

    Go ahead and axolotl questions

    12.7k1.7k

    AI-Hypercomputer/maxtext

    Python

    A simple, performant, and scalable Jax LLM!

    2.5k648

    DaoyuanLi2816/can-i-finetune-this

    Python

    Single-GPU LLM fine-tuning preflight: memory estimates, runnable LoRA/QLoRA recipes, and local measurements.

    795107

    Goekdeniz-Guelmez/MLX-LoRA-Studio

    Swift

    A native Mac App for LLM fine-tuning on Apple Silicon — fully on-device, fully open source.

    29031

    Mcourtyard/m-courtyard

    TypeScript

    M-Courtyard: Local AI Model Fine-tuning Assistant for Apple Silicon. Zero-code, zero-cloud, privacy-first desktop app powered by Tauri + React + mlx-lm.

    18214

    vukrosic/llm-research-kit

    Python

    LLM that can be trained on 1 or more GPUs for research.

    5917

    wenxin0319/llm-forge

    TypeScript

    Open-source AI platform for training customized, lightweight LLMs with GPU acceleration and parallel training

    6220

    Liquid4All/liquid-finetune

    Python

    A full model customization repo for Liquid Foundation Models

    24644