What are the hardware requirements for Songcomposer Sft?

Songcomposer Sft requires approximately 16.7 GB of VRAM and supports a context window of 32K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Songcomposer Sft and how large is it?

Songcomposer Sft is developed by Mar2Ding. 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 Songcomposer Sft?

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

Songcomposer Sft by Mar2Ding — VRAM 16.7GB, 32K context

Name: Songcomposer Sft
Author: Mar2Ding

Songcomposer Sft is an open-source language model by Mar2Ding. Features: LLM, VRAM: 16.7GB, Context: 32K, License: apache-2.0, LLM Explorer Score: 0.12.

Arxiv:2402.17645 Custom code En Feature-extraction Internlm Pytorch Region:us Sharded Zh

Model Card on HF 🤗: https://huggingface.co/Mar2Ding/songcomposer_sft

Songcomposer Sft Parameters and Internals

Model Type

lyric and melody composition, instruction-following song generation

Use Cases

Applications:

lyric to melody, melody to lyric, song continuation, text to song

Supported Languages

languages_supported_en (English), languages_supported_zh (Chinese)

Training Details

Data Sources:

InternLM2 as the initialization of the LLM

Methodology:

Pretrained with InternLM2 and finetuned for instruction-following song generation

Input Output

Input Format:

Prompts in text format

Accepted Modalities:

text

Output Format:

Song lyrics and/or melody

LLM Name	Songcomposer Sft
Repository 🤗	https://huggingface.co/Mar2Ding/songcomposer_sft
Required VRAM	16.7 GB
Updated	2026-07-28
Maintainer	Mar2Ding
Model Type	internlm
Model Files	10.0 GB: 1-of-2 6.7 GB: 2-of-2
Supported Languages	en zh
Model Architecture	InternLM2ForCausalLM
License	apache-2.0
Context Length	32768
Model Max Length	32768
Transformers Version	4.31.0
Is Biased	0
Tokenizer Class	InternLMTokenizer
Padding Token	</s>
Vocabulary Size	93194
Torch Data Type	float16

Rank the Songcomposer 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 Songcomposer Sft

Best Alternatives	Context / RAM	Downloads	Likes
ShareCaptioner Video	32K / 17.2 GB	95	19
Songcomposer Pretrain	32K / 16.7 GB	26	5
MMedLM2	32K / 30.9 GB	12	12
Internlm2 5 Step Prover	8K / 15.4 GB	488	5
Internlm2 Step Prover	8K / 15.4 GB	36	23

Note: green Score (e.g. "73.2") means that the model is better than Mar2Ding/songcomposer_sft.