# IFM releases K2-Horizon-7B-Uno diffusion-augmented LLM, claims 5200 tps speedup

Digest AI · Generative AI & Models · published 2026-09-11T00:34:34Z

Canonical: https://digestai.news/story/ifm-releases-k2-horizon-7b-uno-diffusion-augmented-llm-claims-5200-tps

## Summary

The Institute of Foundation Models (IFM) has posted a new 7‑billion‑parameter language model, **K2‑Horizon‑7B‑Uno**, on Hugging Face. The model adds a diffusion pathway to the existing K2‑Horizon‑7B architecture via LoRA‑based adapters, promising the same generation quality while running at roughly **5,200 tokens per second** on benchmark tests. The repository contains only the adapters; the base‑model weights are hosted separately under the IFM/K2‑Horizon‑7B name.

Developers can load the model with PEFT by wrapping the base model in a `PeftModel` object, and the project provides example notebooks for Google Colab and Kaggle. Selected evaluation scripts are available, with a full benchmark suite slated for release later. The model has already attracted attention, recording **46,547 downloads in the past month**. The authors ask users to cite the accompanying paper and project page when using the model.

If the performance claims hold, K2‑Horizon‑7B‑Uno could offer a practical route to faster inference for applications that need high‑throughput text generation without sacrificing quality, especially for developers already using PEFT‑based fine‑tuning pipelines.

## Key points

- K2‑Horizon‑7B‑Uno adds LoRA diffusion adapters to a 7B LLM, claiming 5,200 tps with unchanged quality.
- Model and adapters are hosted on Hugging Face; base weights are separate under IFM/K2‑Horizon‑7B.
- 46,557 downloads in the last month indicate strong early community interest.

## Why it matters

A faster, quality‑preserving LLM can lower inference costs for SaaS products and research workloads, accelerating adoption of diffusion‑augmented architectures.

## Sources

1. [IFM/K2-Horizon-7B-Uno · Hugging Face - 5200tps with no quality loss](https://huggingface.co/IFM/K2-Horizon-7B-Uno) (huggingface.co, 2026-09-11, primary source)

Part of the developing story: [Rapid Release of Advanced AI Language Models](https://digestai.news/thread/intern-s2-397b-hugging-face-s-new-multimodal-foundation-model) (2 stories)

## Cite

Digest AI, "IFM releases K2-Horizon-7B-Uno diffusion-augmented LLM, claims 5200 tps speedup", 11 September 2026, https://digestai.news/story/ifm-releases-k2-horizon-7b-uno-diffusion-augmented-llm-claims-5200-tps

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/ifm-releases-k2-horizon-7b-uno-diffusion-augmented-llm-claims-5200-tps.json