Alternatives
Products that do what How I topped the HuggingFace open LLM leaderboard on two gaming GPUs does
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…
- 1D3
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
- 2IV
The video demo runs a 7b Model on a normal gaming GPU. I think it already works quite well (accounting for the limited hardware power). :)
2024 · github.com
- 38F
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
- 4L3
Hi everyone, I'm kinda involved in some retrogaming and with some experiments I ran into the following question: "It would be possible to run transformer models bypassing the cpu/ram, connecting the gpu to the nvme?" This is the result of that question itself and some weekend vibecoding (it has the linked library repository in the readme as well), it seems to work, even on consumer gpus, it should work better on professional ones tho
Feb 2026 · github.com
- 5WM
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
- 6FT
Aug 2026 · github.com
- 7IL
I have been working in AI space for a while now, first at FAANG with ML since 2021, then with LLM in start-ups since early 2023. I think LLM Application development is extremely iterative, more so than any other types of development. This is because to improve an LLM application performance (accuracy, hallucinations, latency, cost), you need to try various combinations of LLM models, prompt templates (e.g., few-shot, chain-of-thought), prompt context with different RAG architecture, different agent architecture, and more. There are thousands of possible combinations and you need a process…
2024 · github.com
- 8RQ
Sep 2025 · github.com
- 9UD
Hey HN! I’m the founder of Unify, and we’ve just released our Model Hub, which provides a collection of LLM endpoints with live runtime benchmarks all plotted across time: https://unify.ai/hub A key finding is that static tabular runtime benchmarks for LLMs simply do not work. It’s necessary to take a time-series perspective, and plot the variations through time. We currently have 21 models provided by: Anyscale, Perplexity AI, Replicate, Together AI, OctoAI, Mistral AI and OpenAI, with more on the roadmap. We test across different regions (Asia, US, Europe), with varied…
2024
- 10ML
Aug 2026 · github.com
- 11LF
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
- 12CW
Hello HN! I was fed up switching between multiple UIs to ask GPT, Claude, etc… the same question and comparing the answers. So I built a way to ask multiple models the same question efficiently by having the LLM compare the responses and only show you new and valuable information from the 2nd model. This way you still get a fast response as normal from the 1st model, but also get any added value provided by the 2nd model. Initially I built my own UI to use this, but stumbled upon Open WebUI (formerly Ollama WebUI) which is fantastic, but is made more for local access to LLMs. So I talked to…
2025 · polychat.co
- 13ZA
This is a low-level opensource library I developed for my own use and decided to share, as it makes it possible to process large checkpoints of neural networks without renting high-RAM instances, on a regular PC. It replaces torch.load() with a custom function that produces a dictionary that materializes tensors on the fly. Compared to other solutions it doesn't require sharding or re-encoding checkpoints and uses them completely as-is. It is a foundation to make it possible to run inference and compress language models and other large models one layer at a time - in principle, even one…
2023 · github.com
- 145L
We've built InferX, a specialized runtime environment that fundamentally changes how LLMs are served. The core problem we solve is the latency bottleneck in AI inference, especially with large models. Current systems waste resources or suffer from painfully slow cold starts. InferX's AI-native architecture, with its "snapshot" technology, enables: * *Sub-2s cold starts:* Spin up models instantly. * *High density:* Serve more LLMs on the same GPUs. * *Optimal efficiency:* Maximize GPU utilization. This isn't just another API; it's a new execution layer designed from the ground up for the…
2025 · github.com
- 15HF
We have a massive GPU cluster and developed our own infrastructure to manage the cluster and train massive models. There's how it works: 1. You upload the dataset with preconfigured format into HuggingFaсe [1]. 2. Choose your LLM (e.g. LLaMa 70B, Mistral 7B) 3. Place your submission into the queue 4. Wait for it to get trained. 5. Then you get your trained model there on HuggingFace. Essentially, why would we want to do it? 1. We already have an experience with training big LLMs. 2. We could achieve near-perfect infrastructure performance for training. 3. Sometimes GPUs have just nothing to…
2023 · higgsfield.xyz
- 16LT
I’m excited to share a project I’ve been working on for over a year, which I believe will fundamentally change our approach to language models. We’ve designed a new architecture, which replaces the hidden state of an RNN with a machine learning model. This model compresses context through actual gradient descent on input tokens. We call our method “Test-Time-Training layers.” TTT layers directly replace attention, and unlock linear complexity architectures with expressive memory, allowing us to train LLMs with millions (someday billions) of tokens in context. Our instantiations, TTT-Linear…
2024
- 17AT
We kept shipping “simple” LLM features that were fluent-but-wrong. After too many postmortems we wrote down the failure patterns and added a small reasoning layer in front of the model. It’s model-agnostic, sits beside your existing stack, and you can implement it from a single PDF (MIT). What’s inside the PDF A problem map of 16 failure modes we kept hitting in real systems (OCR/layout drift, table-to-question mismatches, embedding≠meaning, pre-deploy collapse, etc.). Four lightweight gates you can add today: Knowledge-boundary canaries (empty/adversarial/known-fact probes).…
2025 · github.com
- 18WC
Nov 2025 · myclone.is
- 19BH
Hello HN, I recently posted a work-in-progress paper, along with code necessary for replicating all its results, at: https://github.com/glassroom/heinsen_routing Among other things, the code in this repo outperforms Hinton et al.'s recent state-of-the-art result in visual recognition[0] while requiring fewer parameters and an order-of-magnitude fewer training epochs. Most of the original research we do at work tends to be either proprietary in nature or tightly coupled to internal code, so we cannot share it with the world. In this case, however, I was able to remove all…
2019
- 20
- 21AT
May 2026 · github.com
- 22RG
I wanted to know how fast a 26B mixture-of-experts model could run on a desktop CPU with no GPU. Got ~40 tok/s single-stream (lossless) and ~124 batched. The surprising part was the byte budget: for this model you compress the output head (32% of per-token bytes), not the experts (16%). The writeup has the bandwidth roofline and the dead-ends; the repo has the reproducible recipe. Happy to answer questions. Repo: https://github.com/arun-prasath2005/gemma4-cpu-moe
Jun 2026 · apeg.dev
- 23TI
Series on reservoir computing, focusing on echo state networks. Covers GPU simulations, addressing variable training lengths and data processing. Analyzes Gaussian noise impact on network performance, with practical demonstrations and theoretical discussions. Relevant for those interested in neural network behavior in noisy environments.
2023 · ampolloreno.com
- 24IC
I spent the past week implementing a 1 Layer Neural Net and training it on MNIST within the visual scripting language provided by scratch.mit.edu. It was tedious, but ultimately not too difficult. The code runs incredibly slowly, so much so that 64 samples of MNIST takes 5+ hours to train on my machine. There were a lot of little mini challenges that were fun to overcome (implementing softmax was very tricky). If you're interested, I encourage you to try and improve on it! More details in the linked blog post.
2024 · bell-boy.github.io
Ranked by how close each launch is in meaning, then by votes. Refine with a description →