What are the hardware requirements for Moss Moon 003 Sft?

Moss Moon 003 Sft requires approximately 33.5 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Moss Moon 003 Sft and how large is it?

Moss Moon 003 Sft is developed by fnlp. 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 Moss Moon 003 Sft?

Moss Moon 003 Sft is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Moss Moon 003 Sft by fnlp — VRAM 33.5GB

Name: Moss Moon 003 Sft
Author: fnlp

Moss Moon 003 Sft is an open-source language model by fnlp. Features: LLM, VRAM: 33.5GB, License: agpl-3.0, LLM Explorer Score: 0.07.

Arxiv:2203.13474 Autotrain compatible Custom code Dataset:fnlp/moss-002-sft-data En Moss Pytorch Region:us Sharded Zh

Model Card on HF 🤗: https://huggingface.co/fnlp/moss-moon-003-sft

Moss Moon 003 Sft Benchmarks

LLME Score: 0.07306

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

Moss Moon 003 Sft (fnlp/moss-moon-003-sft)

🌟 Advertise your project 🚀

Moss Moon 003 Sft Parameters and Internals

Model Type

conversational, plugin-augmented

Use Cases

Areas:

research, commercial applications

Applications:

conversational agents, AI assistance

Primary Use Cases:

instruction following, dialogue systems

Limitations:

Possible misleading, incorrect, or biased information generation

Considerations:

Users should verify model outputs for correctness and bias.

Additional Notes

Model is in continuous development with new features and improvements being implemented.

Supported Languages

languages_supported (en, zh), proficiency_level (fluent)

Training Details

Data Sources:

fnlp/moss-002-sft-data, fnlp/moss-003-sft-data, gpt-3.5-turbo

Data Volume:

100B Chinese tokens, 20B English tokens

Methodology:

Pre-trained on a large-scale dataset, followed by supervised fine-tuning and preference-aware training with multi-turn conversational data and plugin augmentation.

Context Length:

2048

Hardware Used:

A100 GPU, NVIDIA 3090 GPU

Model Architecture:

Pre-trained on multi-turn conversational data and plugin-augmented data with supervised and preference-aware fine-tuning.

Safety Evaluation

Methodologies:

red-teaming

Risk Categories:

misinformation, bias

Ethical Considerations:

Acknowledges possible biased, misleading, or incorrect information generation.

Responsible Ai Considerations

Fairness:

Training includes a diverse set of conversation data to address bias.

Transparency:

Model details and code are open-sourced.

Accountability:

Fudan University

Mitigation Strategies:

Fine-tuning on preference data to improve factuality and safety.

Input Output

Input Format:

Text input following conversational format.

Accepted Modalities:

text

Output Format:

Text output with capabilities to use plugins for specialized tasks.

Performance Tips:

Use INT-4/8 models for lower memory requirements; align text input for best context understanding.

Release Notes

Version:

moss-moon-003-base

Date:

N/A

Notes:

Base pre-trained model with 16B parameters, intended for conversational use.

Version:

moss-moon-003-sft

Date:

N/A

Notes:

Supervised fine-tuned version capable of handling multi-turn dialogues.

Version:

moss-moon-003-sft-plugin

Date:

N/A

Notes:

Plugin-augmented for enhanced functionality in specific tasks.

LLM Name	Moss Moon 003 Sft
Repository 🤗	https://huggingface.co/fnlp/moss-moon-003-sft
Required VRAM	33.5 GB
Updated	2025-09-23
Maintainer	fnlp
Model Type	moss
Model Files	9.8 GB: 1-of-4 9.9 GB: 2-of-4 9.8 GB: 3-of-4 4.0 GB: 4-of-4
Supported Languages	en zh
Model Architecture	MossForCausalLM
License	agpl-3.0
Model Max Length	2048
Transformers Version	4.25.1
Tokenizer Class	GPT2Tokenizer
Beginning of Sentence Token	<\|endoftext\|>
End of Sentence Token	<eom>
Unk Token	<\|endoftext\|>
Vocabulary Size	107008
Torch Data Type	float16
Activation Function	gelu_new
Errors	replace

Best Alternatives to Moss Moon 003 Sft

Best Alternatives	Context / RAM	Downloads	Likes
Moss Moon 003 Base	0K / 33.5 GB	1117	132
Moss Moon 003 Sft	0K / 33.5 GB	852	128
Moss Moon 003 Sft Plugin	0K / 33.5 GB	978	71
Moss Moon 003 Base	0K / 33.5 GB	247	131
...Moon 003 Sft Int4 Fix Autotune	0K / 10.8 GB	10	1
Moss Moon 003 Sft Int8	0K / 18.5 GB	209	14
Moss Moon 003 Sft Plugin	0K / 33.5 GB	7	69
Moss Moon 003 Sft Plugin Int4	0K / 10.8 GB	168	18
Moss Moon 003 Sft Int4	0K / 10.8 GB	41	41
Moss Moon 003 Sft Int4	0K / 10.8 GB	42	40

Note: green Score (e.g. "73.2") means that the model is better than fnlp/moss-moon-003-sft.

Rank the Moss Moon 003 Sft Capabilities

🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟

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

What open-source LLMs or SLMs are you in search of? 54028 in total.

Email us: info@extractum.io. Our Privacy Policy | Terms and Conditions | Suggest an improvement.

Our Social Media →

Original data from HuggingFace, OpenCompass and various public git repos.

Check out Ag3ntum — our secure, self-hosted AI agent for server management.

Release v20260328a

Support LLM Explorer

Moss Moon 003 Sft by fnlp

» All LLMs » fnlp » Moss Moon 003 Sft URL Share it on

Moss Moon 003 Sft Benchmarks

Moss Moon 003 Sft Parameters and Internals

Best Alternatives to Moss Moon 003 Sft

Rank the Moss Moon 003 Sft Capabilities

What open-source LLMs or SLMs are you in search of? 54028 in total.