# Crowd-v1

> Crowd-v1 is the first official model architecture released for CrowdGPT, a community-driven distributed AI project.

- **Organisation:** Vxtzq
- **Category:** Chat & Text Generation
- **Parameters:** 1B
- **Licence:** MIT (Commercial use OK)
- **Released:** 2026-08-21
- **Hugging Face:** 91K downloads (30 days), 40 likes

## Estimated memory

| Quant | ≈ GB |
|---|---|
| Q2_K | 1.18 |
| Q3_K_M | 1.34 |
| Q4_K_M | 1.45 |
| Q5_K_M | 1.56 |
| Q6_K | 1.67 |
| Q8_0 | 1.9 |
| F16 | 2.82 |

## About

Crowd-v1 is the first official model architecture released for CrowdGPT, a community-driven distributed AI project.

Unlike a conventional pretrained model release, Crowd-v1 is distributed with partially trained weights. The purpose of this release is to provide a common model definition and weight format that CrowdGPT clients can download and collectively train.

The model is designed to be consumed by the CrowdGPT distributed training infrastructure, where individual participants contribute compute toward training a shared model.

Important: Crowd-v1 undergoing pretraining. Its initial weights contain partial useful language knowledge and the model should not be expected to generate very coherent text.

Crowd-v1 contains approximately 1 billion parameters.

Grouped-Query Attention (GQA) Crowd-v1 uses 16 query heads but only 4 key/value heads, reducing KV-cache and attention-related memory requirements.

Rotary Position Embeddings (RoPE) Rotary embeddings are applied to query and key representations.

SwiGLU The feed-forward network uses a SwiGLU-style gated MLP with a hidden dimension of 2,560.

Weight tying The token embedding matrix and language-model head share weights, reducing the total parameter count.

Causal attention Each token can only attend to preceding tokens, making the architecture suitable for autoregressive language modeling.

The released weights are generated deterministically using seed 42.

Non-normalization parameters are initialized from:

N(0, 0.02)

## Alternatives

- [openbmb/MiniCPM5-1B](https://selfhost.directory/models/openbmb/minicpm5-1b) (1.1B, Apache 2.0, ≈1.33 GB at Q4): We are releasing MiniCPM5-1B, the first model in the MiniCPM5 series. It is a dense 1B Transformer built for on-device, local deployment, and resource-constrained scenarios, reaching 1B-class open-source SOTA.
- [Qwen/Qwen3-0.6B](https://selfhost.directory/models/qwen/qwen3-0.6b) (752M, Apache 2.0, ≈1.82 GB at Q4): Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.
- [IFM/K2-Horizon-0.9B](https://selfhost.directory/models/ifm/k2-horizon-0.9b) (1.1B, Apache 2.0, ≈1.58 GB at Q4): K2-Horizon-0.9B is the compact dense member of the K2-Horizon family: a 0.9B-class decoder-only model with a 128K context window.
- [LiquidAI/LFM2.5-1.2B-Instruct](https://selfhost.directory/models/liquidai/lfm2.5-1.2b-instruct) (1.2B, lfm1.0, ≈1.44 GB at Q4): LFM2.5 is a new family of hybrid models designed for on-device deployment. It builds on the LFM2 architecture with extended pre-training and reinforcement learning.
- [google/gemma-3-1b-it](https://selfhost.directory/models/google/gemma-3-1b-it) (1B, Gemma Terms of Use, ≈1.3 GB at Q4): Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models.
- [paradigma-inc/limite-1b-violetto](https://selfhost.directory/models/paradigma-inc/limite-1b-violetto) (1B, Apache 2.0, ≈1.49 GB at Q4): Limite 1B - Violetto is Paradigma’s first model, designed for high-throughput solutions of difficult mathematical problems.

More: https://selfhost.directory/models/vxtzq/crowd-v1/alternatives

## Runs well on

- [GeForce RTX 4060](https://selfhost.directory/models/vxtzq/crowd-v1/on/rtx-4060): F16, est. 75 tok/s
- [Apple M4 (16 GB)](https://selfhost.directory/models/vxtzq/crowd-v1/on/apple-m4-16gb): F16, est. 33 tok/s
- [Radeon RX 6700 XT](https://selfhost.directory/models/vxtzq/crowd-v1/on/rx-6700-xt): F16, est. 106 tok/s
- [GeForce RTX 3060 12GB](https://selfhost.directory/models/vxtzq/crowd-v1/on/rtx-3060): F16, est. 99 tok/s
- [Arc B580](https://selfhost.directory/models/vxtzq/crowd-v1/on/arc-b580): F16, est. 125 tok/s
- [Tesla T4](https://selfhost.directory/models/vxtzq/crowd-v1/on/tesla-t4): F16, est. 88 tok/s

## Similar size

- [Qwen/Qwen3-0.6B](https://selfhost.directory/models/qwen/qwen3-0.6b) (752M)
- [google/gemma-3-1b-it](https://selfhost.directory/models/google/gemma-3-1b-it) (1B)
- [meta-llama/Llama-3.2-1B-Instruct](https://selfhost.directory/models/meta-llama/llama-3.2-1b-instruct) (1.2B)
- [TinyLlama/TinyLlama-1.1B-Chat-v1.0](https://selfhost.directory/models/tinyllama/tinyllama-1.1b-chat-v1.0) (1.1B)
- [LiquidAI/LFM2.5-1.2B-Instruct](https://selfhost.directory/models/liquidai/lfm2.5-1.2b-instruct) (1.2B)
- [openbmb/MiniCPM5-1B](https://selfhost.directory/models/openbmb/minicpm5-1b) (1.1B)

## FAQ

**How much VRAM does Crowd-v1 need?**
About 1.4 GB at Q4_K_M with an 8K context. Q8_0 needs about 1.9 GB and full precision about 2.8 GB. Longer contexts need more memory for the KV cache.

**Can Crowd-v1 run on the RTX 4060?**
Yes. It fits at FP16 / BF16 (about 2.8 GB), at an estimated 75 tokens/s. It fits comfortably on the Apple M4 (16 GB), Radeon RX 6700 XT, RTX 3060 12GB.

**Can I use Crowd-v1 commercially?**
Yes. It is released under MIT. Permissive: do almost anything as long as you keep the copyright notice.

## Links

- huggingface: https://huggingface.co/Vxtzq/Crowd-v1
- github: https://github.com/Vxtzq/CrowdGPT

---
Source: https://selfhost.directory/models/vxtzq/crowd-v1