What are the hardware requirements for ProSparse MiniCPM 1B Sft?

ProSparse MiniCPM 1B Sft requires approximately 2.7 GB of VRAM and supports a context window of 4K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed ProSparse MiniCPM 1B Sft and how large is it?

ProSparse MiniCPM 1B Sft is developed by SparseLLM, a model with 1b parameters. 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 ProSparse MiniCPM 1B Sft?

ProSparse MiniCPM 1B Sft is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

ProSparse MiniCPM 1B Sft by SparseLLM — VRAM 2.7GB, 4K context

Name: ProSparse MiniCPM 1B Sft
Author: SparseLLM

ProSparse MiniCPM 1B Sft is an open-source language model by SparseLLM. Features: 1b LLM, VRAM: 2.7GB, Context: 4K, License: apache-2.0, LLM Explorer Score: 0.13.

Arxiv:2310.04564 Arxiv:2312.12456 Arxiv:2402.03804 Arxiv:2402.13516 Arxiv:2404.06395 Custom code En Minicpm Modelbest Pytorch Region:us Thunlp Zh

Model Card on HF 🤗: https://huggingface.co/SparseLLM/ProSparse-MiniCPM-1B-sft

ProSparse MiniCPM 1B Sft Parameters and Internals

Model Type

language model, sparsity-enhanced

Use Cases

Areas:

research, commercial applications

Applications:

text generation, code generation, reading comprehension

Primary Use Cases:

Inference speed enhancement due to sparsity

Limitations:

Potential inaccuracies with activation predictor, Evaluation issues with LM-Eval

Considerations:

Use UltraEval for evaluation; handle limitations in inference accuracy carefully.

Supported Languages

en (proficient), zh (proficient)

Training Details

Data Sources:

ProSparse training methodology

Data Volume:

473.02 billion tokens

Methodology:

Activation Function Substitution, Progressive Sparsity Regularization, Activation Threshold Shifting

Model Architecture:

Modified version of MiniCPM using ProSparse method with ReLU activation substitution

Input Output

Input Format:

Tokenized text

Accepted Modalities:

text

Output Format:

Tokenized text

Performance Tips:

Utilize PowerInfer for acceleration.

LLM Name	ProSparse MiniCPM 1B Sft
Repository 🤗	https://huggingface.co/SparseLLM/ProSparse-MiniCPM-1B-sft
Model Size	1b
Required VRAM	2.7 GB
Updated	2026-06-19
Maintainer	SparseLLM
Model Files	2.7 GB
Supported Languages	en zh
Model Architecture	MiniCPMForCausalLM
License	apache-2.0
Context Length	4096
Model Max Length	4096
Transformers Version	4.36.0
Tokenizer Class	LlamaTokenizer
Vocabulary Size	73440
Torch Data Type	bfloat16

Rank the ProSparse MiniCPM 1B Sft 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 ProSparse MiniCPM 1B Sft

Best Alternatives	Context / RAM	Downloads	Likes
BitCPM4 1B	32K / 2.7 GB	75	26
MiniCPM 1B Sft Bf16	4K / 2.7 GB	15365	19
MiniCPM S 1B Sft	4K / 2.7 GB	666	12

Note: green Score (e.g. "73.2") means that the model is better than SparseLLM/ProSparse-MiniCPM-1B-sft.