What are the hardware requirements for Ctrl?

Ctrl requires approximately 6.5 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Ctrl and how large is it?

Ctrl is developed by Salesforce. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate Ctrl?

Ctrl is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Ctrl by Salesforce — VRAM 6.5GB

Name: Ctrl
Author: Salesforce

Ctrl is an open-source language model by Salesforce. Features: LLM, VRAM: 6.5GB, License: bsd-3-clause, LLM Explorer Score: 0.06.

Arxiv:1909.05858 Arxiv:1910.09700 Ctrl En Endpoints compatible Pytorch Region:us Tf

Model Card on HF 🤗: https://huggingface.co/Salesforce/ctrl

Ctrl Parameters and Internals

Model Type

Transformer-based language model

Supported Languages

primary_language (English), additional_languages (Some German, Spanish, French)

Training Details

Data Sources:

Wikipedia (English, German, Spanish, French), Project Gutenberg, 45 Subreddits, OpenWebText, news data, Amazon Reviews, Europarl and UN data from WMT, ELI5, MRQA shared tasks

Data Volume:

140 GB

Methodology:

Pre-trained using language modeling with control codes as first token

Training Time:

2 weeks

Hardware Used:

Cloud TPU v3 Pod

Model Architecture:

CTRL has model dimension d = 1280, inner dimension f = 8192, 48 layers, and 16 heads per layer. Dropout with probability 0.1 follows the residual connections in each layer.

LLM Name	Ctrl
Repository 🤗	https://huggingface.co/Salesforce/ctrl
Required VRAM	6.5 GB
Updated	2026-07-04
Maintainer	Salesforce
Model Type	ctrl
Model Files	6.5 GB
Supported Languages	en
Model Architecture	AutoModel
License	bsd-3-clause
Model Max Length	256
Vocabulary Size	246534

Rank the Ctrl capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Ctrl

Best Alternatives	Context / RAM	Downloads	Likes
Distil Longformer Base 4096	4K / 0.4 GB	6	0
Daedalus 1	1K / GB	9	1
Opengpt2 Pytorch Backward	1K / 6 GB	325	1
Tiny Random Detr	1K / 0.2 GB	15	0
Opengpt2 Pytorch Forward	1K / 6 GB	2	1
Finsent Transformer	0.5K / 0.4 GB	1	1
Bert Chinese L 12 H 768 A 12	0.5K / 0.4 GB	9	1
Simbert Chinese Base	0.5K / 0.4 GB	11	0
Simbert Chinese Tiny	0.5K / 0 GB	9	0
Bert Tiny	0.5K / 0 GB	1360936	147

Note: green Score (e.g. "73.2") means that the model is better than Salesforce/ctrl.