← Back to Hugging Face models & datasets

Hugging Face models & datasets

The top 9 hugging face models & datasets measured by RiseFinder over the 7d window ending Week 39 · 21 September 2026.

Period: Weekly
Window: 7d
Rows: 9
Data through: Week 39 · 21 September 2026

← All current risers All hugging face models & datasets editions Newer: Week 40 · 28 September 2026 Older: Week 38 · 16 September 2026

01/

Weekly box

9 entries

m-a-p/SheetSage2

SheetSage2 is a music transcription model on Hugging Face that turns music recordings into lead sheets and timed musical annotations, covering melody, chords, beats, key and structure. It transcribes complete songs into editable ABC or MIDI, renders piano previews, and can export embeddings and token predictions. The model is built on MERT-v2-FullSong, accepts file paths, encoded audio bytes, binary streams or raw waveforms, and handles long songs with overlapping windows.

Hugging Face model

Open source ↗

+375.3%

2,412 → 11,465 downloads

m-a-p/WildSongBench

WildSongBench is a benchmark dataset of 192 song-generation prompts, 94 in Chinese and 98 in English, used in the YuE2 benchmarks. The repository provides the prompts, exact inputs and seeds, reference scores, and a minimal evaluator for SongBench, Q3O, MuQ-MuLan, AllMusicCaps, and Qwen3-ASR PER. Audio is not included.

Hugging Face dataset

Open source ↗

+263.2%

467 → 1,696 downloads

openbmb/MiniCPM5-2B

MiniCPM5-2B is a dense 2B Transformer model from OpenBMB, the second model in the MiniCPM5 series. It is built for on-device, local deployment and resource-constrained scenarios. The release includes multiple formats such as BF16 checkpoints, GGUF for llama.cpp, Ollama and LM Studio, MLX for Apple Silicon, and GPTQ 4-bit quantized versions, and the training datasets are released alongside the model as part of the UltraData family.

Hugging Face model

Open source ↗

+180.2%

150,110 → 420,622 downloads

serdarcaglar/kiraat

serdarcaglar/kiraat is a Turkish read-speech dataset hosted on Hugging Face. It is a sentence-aligned corpus built from publicly available recordings on Turkish audiobook YouTube channels, with each clip labeled by its source channel. The dataset contains 1,840,404 clips totaling 3,105.7 hours of audio, drawn from 27 channels and 90 clustered speakers, with a recommended subset of 1,547,494 clips covering 2,575.2 hours.

Hugging Face dataset

Open source ↗

+151.4%

667 → 1,677 downloads

venvoo/china-a-share-l2-level2-limit-order-book-tick-data

venvoo/china-a-share-l2-level2-limit-order-book-tick-data is a Hugging Face dataset containing a historical archive of Chinese exchange Level-2 market data from 2017 to 2026, supplied through a vendor export and stored in Parquet format. It includes ten-level quote snapshots, individual order messages, and trade-stream records, covering A-share stocks as well as non-stock instruments such as ETFs and bonds.

Hugging Face dataset

Open source ↗

+143.5%

4,878 → 11,878 downloads

openbmb/UltraData-SFT-Agent-2609

UltraData-SFT-Agent-2609 is a machine learning dataset of executable Agent trajectories used for instruction-tuning. It is the L3 refined data for Agent instruction-tuning within UltraData's L0-L4 tiered data management framework, built for the post-training of MiniCPM5-2B, and complements UltraData-SFT-2605, which covers core-domain SFT.

Hugging Face dataset

Open source ↗

+123.6%

8,165 → 18,261 downloads

agentionai/Signal-3.8-27B-GGUF

Signal is a fine-tuned version of Qwen3.8-27B from AgentionAI, designed for lower generation latency and better token efficiency. Its creators report it produces 57% fewer answer tokens and 52% fewer thinking tokens on their held-out general-prompt evaluation while matching or improving measured answer quality.

Hugging Face model

Open source ↗

+92.3%

8,152 → 15,675 downloads

openbmb/MiniCPM5-2B-GGUF

MiniCPM5-2B-GGUF is the GGUF format release of MiniCPM5-2B, a dense 2B parameter Transformer language model from the MiniCPM5 series. The GGUF format is intended for use with llama.cpp, Ollama, and LM Studio. The underlying model is built for on-device, local deployment and resource-constrained scenarios.

Hugging Face model

Open source ↗

+81.1%

99,716 → 180,593 downloads

XHToken/Spark-X2.5-4B

Spark-X2.5-4B is a compact, general-purpose language model intended for everyday tasks such as conversation, writing, translation, reasoning, coding, tool use, and agentic workflows. It uses a hybrid attention architecture combining full-attention and sliding-window layers, natively supports context windows of up to 1M tokens, and supports more than 200 languages. The model is compatible with hardware platforms including NVIDIA and Huawei and with inference frameworks such as vLLM, SGLang, and llama.cpp.

Hugging Face model

Open source ↗

+44.2%

21,336 → 30,772 downloads

This dated edition is generated from the reviewed RiseFinder export. Movement figures retain their original source, period and direction.

Get early signals on the websites, apps, repositories and startups gaining momentum. Join the list for Risefinder email briefings.

By subscribing, you ask us to store your email and send Risefinder updates. You can unsubscribe at any time. Privacy.

Want a sample first? Read past daily emails.

Hugging Face models & datasets

Weekly box

Get email updates