What are the hardware requirements for Replit Code V1 3B?

Replit Code V1 3B requires approximately 10.4 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Replit Code V1 3B and how large is it?

Replit Code V1 3B is developed by replit, a model with 3b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Replit Code V1 3B perform on standard benchmarks?

Replit Code V1 3B has the following published scores: HumanEval (Python) 20.12. Compare against reference models on this page or on the LLM Explorer leaderboards.

Replit Code V1 3B by replit — VRAM 10.4GB

Name: Replit Code V1 3B
Author: replit

Replit Code V1 3B is an open-source language model by replit. Features: 3b LLM, VRAM: 10.4GB, License: cc-by-sa-4.0, LLM Explorer Score: 0.09, HumanEval: 20.1.

Arxiv:2108.12409 Arxiv:2205.14135 Arxiv:2211.15533 Arxiv:2302.06675 Code Custom code Dataset:bigcode/the-stack-dedu... Endpoints compatible Model-index Mpt Pytorch Region:us

Model Card on HF 🤗: https://huggingface.co/replit/replit-code-v1-3b

Replit Code V1 3B Parameters and Internals

Model Type

code-generation

Use Cases

Primary Use Cases:

Code completion across 20 different programming languages

Limitations:

Pre-training dataset may contain offensive or inappropriate content, Not suitable for applications that may cause harm or distress

Considerations:

Exercise reasonable caution when using in production systems.

Supported Languages

code programming_language (> Markdown, Java, JavaScript, Python, TypeScript, PHP, SQL, JSX, reStructuredText, Rust, C, CSS, Go, C++, HTML, Vue, Ruby, Jupyter Notebook, R, Shell)

Training Details

Data Sources:

bigcode/the-stack-dedup, Stack Dedup v1.2 dataset

Data Volume:

525B tokens

Methodology:

Trained for code completion using various techniques such as Flash Attention and AliBi positional embeddings.

Hardware Used:

256 x A100-40GB GPUs

Model Architecture:

Causal Language Model

Input Output

Accepted Modalities:

text input for code generation

Output Format:

generated code output compatible with programming languages

Performance Tips:

Experiment with different decoding methods and parameters. Post-processing generated code is important.

LLM Name	Replit Code V1 3B
Repository 🤗	https://huggingface.co/replit/replit-code-v1-3b
Model Size	3b
Required VRAM	10.4 GB
Updated	2026-08-08
Maintainer	replit
Model Type	mpt
Model Files	10.4 GB
Supported Languages	code
Model Architecture	MPTForCausalLM
License	cc-by-sa-4.0
Model Max Length	2048
Transformers Version	4.28.1
Tokenizer Class	ReplitLMTokenizer
Padding Token	<\|pad\|>
Vocabulary Size	32768
Torch Data Type	float32

Rank the Replit Code V1 3B 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 Replit Code V1 3B

Best Alternatives	Context / RAM	Downloads	Likes
Replit Code V1.5 3B	0K / 6.6 GB	544	316
Sea Lion 3B	0K / 6.4 GB	778	17
Code Millenials 3B	0K / 5.2 GB	5	1
SEA LION V1 3B	0K / 6.4 GB	186	19
Mpt 3B 8K Instruct	0K / 6.9 GB	5	3
Glaive Function Calling V1	0K / 10.4 GB	8	75
...aive Function Calling V2 Small	0K / 10.4 GB	8	15
Evol Replit V1	0K / 10.4 GB	9	8
Replit V2 CodeInstruct 3B	0K / 10.4 GB	17	71
Replit CodeInstruct V3	0K / 10.4 GB	7	2

Note: green Score (e.g. "73.2") means that the model is better than replit/replit-code-v1-3b.