What are the hardware requirements for CodeFuse DeepSeek 33B?

CodeFuse DeepSeek 33B requires approximately 66.5 GB of VRAM and supports a context window of 16K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed CodeFuse DeepSeek 33B and how large is it?

CodeFuse DeepSeek 33B is developed by codefuse-ai, a model with 33b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does CodeFuse DeepSeek 33B perform on standard benchmarks?

CodeFuse DeepSeek 33B has the following published scores: HumanEval (Python) 76.83. Compare against reference models on this page or on the LLM Explorer leaderboards.

CodeFuse DeepSeek 33B by codefuse-ai — VRAM 66.5GB, 16K context

Name: CodeFuse DeepSeek 33B
Author: codefuse-ai

CodeFuse DeepSeek 33B is an open-source language model by codefuse-ai. Features: 33b LLM, VRAM: 66.5GB, Context: 16K, License: other, LLM Explorer Score: 0.11, HumanEval: 76.8.

Code Conversational En Endpoints compatible Llama Pytorch Region:us Safetensors Sharded Tensorflow Zh

Model Card on HF 🤗: https://huggingface.co/codefuse-ai/CodeFuse-DeepSeek-33B

CodeFuse DeepSeek 33B Parameters and Internals

Model Type

code-generation

Training Details

Methodology:

QLoRA

Input Output

Input Format:

Concatenated string in training data format

Accepted Modalities:

text

Output Format:

Generated code

Performance Tips:

Ensure input string ends with '\ bot' for generating answers.

Release Notes

Date:

2024-01-12

Notes:

Released with pass@1 score of 78.65% on HumanEval.

LLM Name	CodeFuse DeepSeek 33B
Repository 🤗	https://huggingface.co/codefuse-ai/CodeFuse-DeepSeek-33B
Model Size	33b
Required VRAM	66.5 GB
Updated	2026-07-26
Maintainer	codefuse-ai
Model Type	llama
Model Files	9.7 GB: 1-of-7 9.9 GB: 2-of-7 9.9 GB: 3-of-7 9.8 GB: 4-of-7 9.9 GB: 5-of-7 9.9 GB: 6-of-7 7.4 GB: 7-of-7 9.7 GB: 1-of-7 9.9 GB: 2-of-7 9.9 GB: 3-of-7 9.8 GB: 4-of-7 9.9 GB: 5-of-7 9.9 GB: 6-of-7 7.4 GB: 7-of-7
Supported Languages	en zh
Model Architecture	LlamaForCausalLM
License	other
Context Length	16384
Model Max Length	16384
Transformers Version	4.34.1
Tokenizer Class	LlamaTokenizerFast
Beginning of Sentence Token	<｜begin▁of▁sentence｜>
End of Sentence Token	<｜end▁of▁sentence｜>
Vocabulary Size	32256
Torch Data Type	bfloat16

Rank the CodeFuse DeepSeek 33B 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

Quantized Models of the CodeFuse DeepSeek 33B

Model	Likes	Downloads	VRAM
CodeFuse DeepSeek 33B 4bits	10	7	18 GB

Best Alternatives to CodeFuse DeepSeek 33B

Best Alternatives	Context / RAM	Downloads	Likes
Llm Jp 4 33B Thinking	64K / 66.8 GB	192	24
Llm Jp 4 33B Base	64K / 66.8 GB	89	8
Llm Jp 4 33B Thinking NVFP4	64K / 21.6 GB	79	1
Llm Jp 4 33B Thinking Nvfp4	64K / 21.5 GB	38	2
...angled Llama 33M 32K Base V0.1	32K / 0.1 GB	22	1
ReflectionCoder DS 33B	16K / 67 GB	8293	4
Deepseek Wizard 33B Slerp	16K / 35.3 GB	7	0
ValidateAI 33B Slerp	16K / 35.4 GB	7	0
ValidateAI 3 33B Ties	16K / 66.5 GB	7	0
ValidateAI 2 33B AT	16K / 66.5 GB	6	0

Note: green Score (e.g. "73.2") means that the model is better than codefuse-ai/CodeFuse-DeepSeek-33B.