AI · alternatives · 2026

24 alternatives to Minimal LLM Post-Training Experiments on an 8GB GPU (SFT, DPO, GRPO)

Minimal, readable LLM post-training experiments on one 8GB GPU. Measures forgetting, seed variance, and RL emergence. - pochenai/nano-llm-posttraining

This project provides code and guidance for fine-tuning large language models using supervised fine-tuning (SFT), direct preference optimization (DPO), and GRPO techniques on consumer-grade hardware with 8GB of GPU… Below are 24 products that do a similar job, ranked by how close each is in meaning and then by launch-day votes.

1
Fine-tune an 8B model on a 4 GB laptop GPU▲139
Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU. - MakazhanAlpamys/Soup
Aug 2026 · github.com · its alternatives →
2
How I topped the HuggingFace open LLM leaderboard on two gaming GPUs▲495
I found that duplicating a specific block of 7 middle layers in Qwen2-72B, without modifying any weights, improved performance across all Open LLM Leaderboard benchmarks and took #1. As of 2026, the top 4 models on that leaderboard are still descendants. The weird finding: single-layer duplication does nothing. Too few layers, nothing. Too many, it gets worse. Only circuit-sized blocks of ~7 layers work. This suggests pretraining carves out discrete functional circuits in the layer stack that only work when preserved whole. The whole thing was developed on 2x RTX 4090s in my basement. I'm…
Mar 2026 · dnhkng.github.io · its alternatives →
3
80% faster, 50% less memory, 0% loss of accuracy Llama finetuning▲385
Hi HN! I'm just sharing a project I've been working on during the LLM Efficiency Challenge - you can now finetune Llama with QLoRA 5x faster than Huggingface's original implementation on your own local GPU. Some highlights: 1. Manual autograd engine - hand derived backprop steps. 2. QLoRA / LoRA 80% faster, 50% less memory. 3. All kernels written in OpenAI's Triton language. 4. 0% loss in accuracy - no approximation methods - all exact. 5. No change of hardware necessary. Supports NVIDIA GPUs since 2018+. CUDA 7.5+. 6. Flash Attention support via Xformers. 7. Supports 4bit and 16bit…
2023 · github.com · its alternatives →
4
I built a tiny LLM to demystify how language models work▲915
Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food. Fork it and swap the personality for your own character.
Apr 2026 · github.com · its alternatives →
5
Tiny-vLLM – high performance LLM inference engine in C++ and CUDA▲205
Build your own high performance LLM inference engine in C++ and CUDA - a smaller version of vLLM - jmaczan/tiny-vllm
May 2026 · github.com · its alternatives →
6
Finetune LLaMA-7B on commodity GPUs using your own text▲449
I've been playing around with https://github.com/zphang/minimal-llama/ and https://github.com/tloen/alpaca-lora/blob/main/finetune.py, and wanted to create a simple UI where you can just paste text, tweak the parameters, and finetune the model quickly using a modern GPU. To prepare the data, simply separate your text with two blank lines. There's an inference tab, so you can test how the tuned model behaves. This is my first foray into the world of LLM finetuning, Python, Torch, Transformers, LoRA, PEFT, and Gradio. Enjoy!
2023 · github.com · its alternatives →
7
Duplicate 3 layers in a 24B LLM, logical deduction .22→.76. No training▲265
I replicated David Ng's RYS method (https://dnhkng.github.io/posts/rys/) on consumer AMD GPUs (RX 7900 XT + RX 6950 XT) and found something I didn't expect. Transformers appear to have discrete "reasoning circuits" — contiguous blocks of 3-4 layers that act as indivisible cognitive units. Duplicate the right block and the model runs its reasoning pipeline twice. No weights change. No training. The model just thinks longer. The results on standard benchmarks (lm-evaluation-harness, n=50): Devstral-24B, layers 12-14 duplicated once: - BBH Logical Deduction: 0.22 → 0.76…
Mar 2026 · github.com · its alternatives →
8
Speeding up LLM inference 2x times (possibly)▲419
Here's a project I've been working on for the last few months. It's a new (I think) algorithm, that allows to adjust smoothly - and in real time - how many calculations you'd like to do during inference of an LLM model. It seems that it's possible to do just 20-25% of weight multiplications instead of all of them, and still get good inference results. I implemented it to run on M1/M2/M3 GPU. The mmul approximation itself can be pushed to run 2x fast before the quality of output collapses. The inference speed is just a bit faster than Llama.cpp's, because the rest of implementation…
2024 · asciinema.org · its alternatives →
9WM
We made glhf.chat – run almost any open-source LLM, including 405B▲161
Try it out! https://glhf.chat/ Hey HN! We’ve been working for the past few months on a website to let you easily run (almost) any open-source LLM on autoscaling GPU clusters. It’s free for now while we figure out how to price it, but we expect to be cheaper than most GPU offerings since we can run the models multi-tenant. Unlike Together AI, Fireworks, etc, we’ll run any model that the open-source vLLM project supports: we don’t have a hardcoded list. If you want a specific model or finetune, you don’t have to ask us for it: you can just paste the Hugging Face link in and…
2024 · glhf.chat · its alternatives →
10
TensorZero▲307
Hi HN! We're Gabriel & Viraj, and we're excited to open source TensorZero. To be a little cheeky, TensorZero is an open-source platform that helps LLM applications graduate from API wrappers into defensible AI products. 1. Integrate our model gateway 2. Send metrics or feedback 3. Unlock compounding improvements in quality, cost, and latency It enables a data & learning flywheel for LLMs by unifying: • Inference: one API for all LLMs, with <1ms P99 overhead • Observability: inference & feedback → your database • Optimization: better prompts, models, inference strategies • Experimentation:…
2025 · github.com · its alternatives →
11
Predibase Reinforcement Fine-Tuning▲175
LLM reinforcement fine-tuning platform to improve LLM output
2025 · rubrik.com · its alternatives →
12
Local fine tuning for Mistral and SDXL, GPU mem/latency optimization▲36
100% bootstrapped new startup. It lets you fine tune Mistral-7B and SDXL. In particular, for the LLM fine tuning we implemented a dataprep pipeline that turns websites/pdfs/doc files into question-answer pairs for training the small LLM using an big LLM. It includes a GPU scheduler that can do finegrained GPU memory scheduling (Kubernetes can only do whole-GPU, we do it per-GB of GPU memory to pack both inference and fine tuning jobs into the same fleet) to fit model instances into GPU memory to optimally trade off user facing latency with GPU memory utilization It's a pretty…
2023 · docs.helix.ml · its alternatives →
13
Multi-GPU Reinforcement Learning in Tensorflow for OpenAI Gym▲58
Multi-GPU reinforcement learning using Deep Q-Network in TensorFlow for OpenAI Gym - vishar0/dist-dqn
2016 · github.com · its alternatives →
14
Next-Gen AI Training: LLM-RLHF-Tuning with PPO and DPO▲30
Comprehensive toolkit for Reinforcement Learning from Human Feedback (RLHF) training, featuring instruction fine-tuning, reward model training, and support for PPO and DPO algorithms with various configurations for the Transformer 5 models including Qwen 3.0 - raghavc/LLM-RLHF-Tuning-with-PPO-and-DP
2024 · github.com · its alternatives →
15
Explore large language models with 512MB of RAM▲138
Explore large language models in 512MB of RAM. Contribute to jncraton/languagemodels development by creating an account on GitHub.
2023 · github.com · its alternatives →
16
Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it▲170
We recently used DeepSeek V4 Flash as a teacher for finance tasks with GPT-OSS-120B. Distillation works well on this problem. At a constrained 8k token budget, our self-distilled 120B scores 83.61% on FinanceReasoning, above Kimi K3 (81.93%) and Inkling (65.13%). We released the 20B open weights. With V4 as the teacher though, we realized it would be timely to measure if the censorship characteristic of it transferred to the distilled version of the base model. tl;dr it didn't, the teacher answered politically sensitive questions 7 SDs differently than expected, but the distilled model's…
Jul 2026 · ctgt.ai · its alternatives →
17AT
A tiny LLM running at 21,000 tok/s on a $250 FPGA (Live Demo)▲79
A 3.16M-parameter INT4 transformer running entirely in the on-chip memory of a Xilinx Kria KV260. Zero DRAM in the token loop, 59,965 tok/s on the fabric, bit-exact. Chat with it live.
Aug 2026 · mikeayles.com · its alternatives →
18
Dolly▲113
Democratizing the magic of ChatGPT with open models
2023 · github.com · its alternatives →
19
Fine-grained stylistic control of LLMs using model arithmetic▲85
We developed a new framework that enables flexible control of generated text in language models. By combining several models and/or system prompts in one mathematical formula, it lets you tweak your style and combine model outputs with ease. A handy tool for those working with LLMs, looking for more fine-grained control of stylistic output. More details in our paper: https://arxiv.org/abs/2311.14479. Feedback and potential applications are welcome.
2023 · github.com · its alternatives →
20
TuneTrain.ai▲79
Do you know that Small Language Models (SLM) can outperform LLMs if trained (fine-tuned) on domain specific datasets, like your companies knowledge base? The issue: fine-tuning a SLM usually requires a dataset of 10k - 100k records, which is huge. I created a platform for applying "human in the loop" augmentation techniques on your small dataset, so that you can start with maybe 100 records and build-up quickly huge datasets and launch a training without prior knowledge. I implemented 2 techniques, based on LLM distillation, however: more to come. HN, what Do you think? Do you see value in…
2025 · tunetrain.ai · its alternatives →
21
Free Inference Engineer and Model Training Roadmap▲16
Free, gamified roadmaps for LLM engineering: an Inference Engineering path (KV caches, CUDA kernels, production vLLM serving) and a Model Training path (pretraining on a budget, scaling laws, SFT/DPO/GRPO) — 185 tasks with auto-verified milestones instead of a paper certificate.
Aug 2026 · inferquest.org · its alternatives →
22
Monostate AItraining▲59
Fine-tuning, RL, and inference in one CLI
Dec 2025 · github.com · its alternatives →
23
I ran a language model on a PS2▲46
The Emotion Engine has 32 MB of RAM total, so the trick is streaming weights from CD-ROM one matrix at a time during the forward pass — only activations, KV cache and embeddings live in RAM. This means models bigger than the RAM can still run, they just read more from disc. Had to build a custom quantized format (PSNT), hack endianness, write a tokenizer pipeline, and most of the PS2 SDK from scratch (releasing that separately). The model itself is also custom — a 10M param Llama-style architecture I trained specifically for this. And it works. On real hardware.
Mar 2026 · github.com · its alternatives →
24TT
Trained Tiny Tales GPT(30M model)from scratch and deployed in $15▲18
For the last few weeks, I have been working on training an LLM from scratch and deploying it in production on Google Cloud Platform. Finally, I trained a 30 million parameter model on 1 billion tokens and deployed it as a web service. You can access the LLM using this site - https://kunalmishra.info The following steps were taken to build Tiny Tales GPT 1. Downloaded and preprocessed 8GB of dataset using multiprocessing library. 2. Tokenized the data using byte pair encoding to create 1 billion tokens sharded in different bin files. 3. Defined a training setup and trained the model…
2024 · kunalmishra.info · its alternatives →

Also compare

Ranked by how close each launch is in meaning, then by votes. Prices were read from each product’s own site when checked and can change. Refine with your own description →