Hugging Face's logo
Activity Feed Request to join this org

AI & ML interests

None defined yet.

Recent Activity
osansevieroย  authored a paper 3 days ago
EmbeddingGemma 2: Unified and Efficient Multimodal Representations
gimmy256ย  authored a paper 17 days ago
CulturalRiddles: A Multicultural Benchmark of Riddles
jefequienย  authored a paper about 1 month ago
S2PD: Serial-to-Parallel Diffusion for Physically and Logically Consistent Video Generation
View all activity
danielhanchenย 
posted an update about 10 hours ago
view post
Post
382
Qwen3.8-Flash-Next GGUFs now run faster with less thinking and less token use.

Unsloth Dynamic V3.1 GGUFs bring thinking & token use to match 95% of full precision Qwen3.8-Flash.

Results: 50% fewer tokens + 15% higher accuracy vs same size quants.

GGUF: unsloth/Qwen3.8-Flash-Next-GGUF
Analysis/Guide: https://unsloth.ai/docs/models/qwen3.8-next
danielhanchenย 
posted an update 4 days ago
view post
Post
7,068
Weโ€™re releasing new Qwen3.8-Flash-Next GGUFs with 15% higher accuracy.

Unsloth Dynamic V3.1 outperforms others by >15% on Terminal Bench, DeepSWE & more benchmarks (256K context).

Run on 32GB+ RAM setups.

GGUF: unsloth/Qwen3.8-Flash-Next-GGUF
Blog: https://unsloth.ai/docs/models/qwen3.8-next
danielhanchenย 
posted an update 9 days ago
danielhanchenย 
posted an update 10 days ago
danielhanchenย 
posted an update 20 days ago
danielhanchenย 
posted an update about 1 month ago
view post
Post
5,177
You can now train your own Decision model like Jev locally!

We increased Qwen3.5 0.8Bโ€™s aggregate accuracy from 20.7% to 74.3% across 3 decision benchmarks - on just 4GB VRAM.

Turn any LLM like Qwen3.8, Gemma 4 into decision models with our open-source Unsloth repo.

GitHub: https://github.com/unslothai/unsloth
Guide: https://unsloth.ai/docs/basics/train-your-own-decision-model-with-unsloth
  • 5 replies
ยท
danielhanchenย 
posted an update about 1 month ago
view post
Post
4,156
Google releases EmbeddingGemma 2, a new open embedding model that runs locally on 0.5GB RAM.

The 740M parameter Apache 2.0 model combines a 270M text model with vision (170M) + audio (300M).

Run & train the model via Unsloth.

GGUF: unsloth/embeddinggemma-2-GGUF
Guide: https://unsloth.ai/docs/models/embeddinggemma-2
losercheersย 
authored 2 papers about 1 month ago
losercheersย 
posted an update about 1 month ago
view post
Post
109
Sharing two recent explorations in attention design from our team.

We started with two straightforward questions: Does every attention head need to repeatedly attend to the entire causal history? Once attention scores have been computed, do regions with very little contribution still need the full subsequent computation?

We explored two approaches:

CoWindow Attention (CoWA): Let heads share the work of accessing history. Heads share local context and divide distant context into complementary windows. Each head attends sparsely, while the heads collectively cover the full causal history.
CoWindow Attention: Full Causal Coverage Is a Collective Property (2609.32704)

MassAlloc Attention (MALA): Let attention allocate its own compute. MALA preserves full causal QK scoring, then uses attentionโ€™s own softmax statistics to reduce subsequent computation in low-contribution regions.
MassAlloc Attention: Let Attention Allocate Its Own Compute (2609.32712)

Both approaches support training forward and backward passes, as well as inference prefill and decoding. In attention-operator benchmarks at 128K tokens on 8ร—H100 with TP=8, compared with FullAttn:

CoWA: 7.4ร— forward, 8.6ร— backward, and 3.0ร— decoding speedups.
MALA: 2.2ร— forward, 3.0ร— backward, and 1.6ร— decoding speedups.

We also conducted scaling experiments from 0.6B to 14B, alongside separate continued-training experiments at 32B. During 14B training with 32K context, CoWA and MALA reduced total training FLOPs by 28.5% and 23.1%, respectively, while maintaining performance comparable to FullAttn on the evaluated model capabilities.

From method design to kernel implementation to model training, our goal was to explore which attention computations can be eliminated, and how to turn those savings into practical gains in ML infrastructure.
losercheersย 
submitted 2 papers to Daily Papers about 1 month ago
danielhanchenย 
posted an update about 1 month ago
view post
Post
9,206
Unsloth has surpassed 500M model downloads on Hugging Face! ๐Ÿฆฅ๐Ÿค—

Qwen3.8-27B GGUF is already Unslothโ€™s #1 most-downloaded model ever.

Thanks for all your support! Follow us:
unsloth
  • 1 reply
ยท
danielhanchenย 
posted an update about 2 months ago
danielhanchenย 
posted an update 2 months ago
Aurelien-Morganย 
posted an update 2 months ago
view post
Post
2,473
@retrain-pipelines execution engine is in perpetual evolution, with the aim to establish itself as SOTA, and for the long run.

However, we neglect no aspect of ML-Eng centricity.

If notebooks is where you like to do dev most,
we support you there 100% too.

Build crazy combos of inline tasks, deep parallel sub-DAG branches, nested asynchronous groups...

... the DAG renderer is undergoing an incremental upgrade

until the next one.

* starring toy tasks here. No ML has been hurt in this video ๐Ÿ™‚
  • 1 reply
ยท
danielhanchenย 
posted an update 3 months ago
view post
Post
7,591
Introducing Unsloth Desktop ๐Ÿฆฅ
The first desktop app to run and train models locally.

โ€ข Open-source. Runs on Mac, Windows and Linux
โ€ข Supports MLX, diffusion image/video, audio, GGUF
โ€ข Connect Claude Code and Codex to local LLMs
โ€ข 50% more accurate, self-healing tool calls + sandboxed code exec
โ€ข Works for CPU + multiGPU setups - NVIDIA, AMD, Intel, Mac
โ€ข Train models 2ร— faster with 70% less VRAM
โ€ข Private web search, deep research, RAG, MCP and exports (NVFP4, GGUF)
โ€ข Use Unslothโ€™s OpenAI-compatible API and cloud models
โ€ข Securely deploy LLMs remotely and access anywhere

Unsloth Desktop is now available on http://unsloth.ai
and GitHub.

GitHub: https://github.com/unslothai/unsloth
Blog and Guide: https://unsloth.ai/docs/desktop
  • 6 replies
ยท
danielhanchenย 
posted an update 3 months ago
danielhanchenย 
posted an update 3 months ago
view post
Post
3,589
We compared 1-bit Kimi K3 to Claude Opus 5 and GPT 5.6. ๐Ÿคฏ

We gave 4 models the same prompt: Create a glass aquarium whose side panel develops a visible crack and then bursts...

1-bit Kimi K3 GGUF ran locally on 4x B200s at 36 tok/s.

GGUF: unsloth/Kimi-K3-GGUF
GitHub repo: https://github.com/unslothai/unsloth
  • 1 reply
ยท
danielhanchenย 
posted an update 3 months ago
view post
Post
5,791
Kimi K3 can now be run locally! โœจ

The 1-bit model retains ~78.9% accuracy after we shrunk it from 1.56TB to 594GB (-62% size).

Run on a Mac Studio connected with 128GB RAM device. Kimi K3 is the strongest open model to date.

GGUF: unsloth/Kimi-K3-GGUF
Guide: https://unsloth.ai/docs/models/kimi-k3
  • 5 replies
ยท