Fine-Tuning xLSTM-7B on a budget: An Experimental Journey with Chat Templates
• 1
ethicalabs.ai
non-profit
Verified
AI & ML interests
Building practical, ethical and sustainable AI/ML projects.
Recent Activity
View all activity
Articles
Benchmarks, Resources Footprints and Tasks
#1 opened 23 days ago
by 
mrs83
Post
106
Currently HF Papers only ingests arXiv IDs. While arXiv is standard, it imposes endorsement requirements and legacy moderation bottlenecks that exclude a growing segment of independent researchers, decentralized collectives, and open-source practitioners who publish peer-traceable preprints through CERN/Zenodo, OpenAIRE or OSF.
Zenodo records provide immutable DOIs, versioned artifact linking, and standardized metadata via open APIs.
Integrating Zenodo/DOI ingestion alongside arXiv would drastically broaden paper discovery for the open-source community.
If the engineering bandwidth on the HF team is focused on other roadmaps, I’d be glad to collaborate, help spec the ingestion pipeline, or contribute to an initial integration PR to map Zenodo metadata into the Papers schema cleanly.
Is extending support to external DOI providers currently on the radar, or open to community contributions?
Forum discussion: https://discuss.huggingface.co/t/feature-request-question-non-arxiv-preprints-zenodo-on-hf-papers/179555
Zenodo records provide immutable DOIs, versioned artifact linking, and standardized metadata via open APIs.
Integrating Zenodo/DOI ingestion alongside arXiv would drastically broaden paper discovery for the open-source community.
If the engineering bandwidth on the HF team is focused on other roadmaps, I’d be glad to collaborate, help spec the ingestion pipeline, or contribute to an initial integration PR to map Zenodo metadata into the Papers schema cleanly.
Is extending support to external DOI providers currently on the radar, or open to community contributions?
Forum discussion: https://discuss.huggingface.co/t/feature-request-question-non-arxiv-preprints-zenodo-on-hf-papers/179555
mrs83
updated a
model about 1 month ago
Post
1584
Hi Hugging Face community! 👋 we just submitted our project, OpenAIRE Graph Evaluator, to the OpenAIRE AI Hackathon, and community voting is open until September 11th!
We developed a new classifier model and an LLM-as-judge pipeline to automatically classify millions of unlabelled research papers within the OpenAIRE Graph.
The goal is to save researchers hundreds of hours spent manually sorting datasets, methodologies, and reviews, while creating a self-improving pipeline to generate datasets for training future classifiers!
We would love your support: please vote for "02-OpenAIRE Graph Evaluator" (Theme B) here: https://innovation.openaire.eu/component/content/article/hackathon-community-vote.html?catid=8
We built a custom model, Echo-DSRN, a hybrid recurrent architecture designed for resource-constrained tasks. It features an internal surprise gate and decouples computation into a fast GRU, sliding-window attention, and a gated slow state.
It comes in 4 variants:
- Hybrid RNN: Lightweight generation and speculative drafting.
- Transformers Hybrid: DSRN memory blocks injected into backbones like Qwen 2.5.
- Embedding: Dense sentence embeddings compatible with sentence-transformers.
- Classifier: The specialized engine powering our hackathon submission!
Thanks for your support! Let us know if you have any questions about the architecture!
---
Collab Hub: https://openaire-2026.ethicalabs.ai/collab
GitHub: https://github.com/ethicalabs-ai/Echo-DSRN
HF Hackathon Collection: https://huggingface.co/collections/ethicalabs/openaire-ai-hackathon-2026
We developed a new classifier model and an LLM-as-judge pipeline to automatically classify millions of unlabelled research papers within the OpenAIRE Graph.
The goal is to save researchers hundreds of hours spent manually sorting datasets, methodologies, and reviews, while creating a self-improving pipeline to generate datasets for training future classifiers!
We would love your support: please vote for "02-OpenAIRE Graph Evaluator" (Theme B) here: https://innovation.openaire.eu/component/content/article/hackathon-community-vote.html?catid=8
We built a custom model, Echo-DSRN, a hybrid recurrent architecture designed for resource-constrained tasks. It features an internal surprise gate and decouples computation into a fast GRU, sliding-window attention, and a gated slow state.
It comes in 4 variants:
- Hybrid RNN: Lightweight generation and speculative drafting.
- Transformers Hybrid: DSRN memory blocks injected into backbones like Qwen 2.5.
- Embedding: Dense sentence embeddings compatible with sentence-transformers.
- Classifier: The specialized engine powering our hackathon submission!
Thanks for your support! Let us know if you have any questions about the architecture!
---
Collab Hub: https://openaire-2026.ethicalabs.ai/collab
GitHub: https://github.com/ethicalabs-ai/Echo-DSRN
HF Hackathon Collection: https://huggingface.co/collections/ethicalabs/openaire-ai-hackathon-2026
- 1 reply
mrs83
updated 3
datasets about 1 month ago
mrs83
updated 8
models about 1 month ago
ethicalabs/Echo-DSRN-v0.1.3-Research-Intent-CLF
Text Classification • 98.3M • Updated • 48
ethicalabs/Echo-DSRN-v0.1.4-Embed-Intent-CLF
Text Classification • 98.4M • Updated • 247 • 1
ethicalabs/Echo-DSRN-v0.1.3-Embed-Intent
Sentence Similarity • 98.3M • Updated • 171
ethicalabs/Echo-DSRN-v0.1.3-Embed-Exp
Sentence Similarity • 98.3M • Updated • 1.23k • 1
ethicalabs/Echo-DSRN-v0.1.3-Intent-CLF
Text Classification • 98.3M • Updated • 55 • 1
ethicalabs/Echo-SmolTools-114M-NSFW-CLF
Text Classification • 98.3M • Updated • 36
ethicalabs/Echo-SmolTools-114M-Intent-CLF-Gen
Text Classification • 0.1B • Updated • 39
ethicalabs/Echo-DSRN-114M-v0.1.2
Text Generation • 0.1B • Updated • 202
mrs83
updated a
Space about 2 months ago
mrs83
updated a
collection about 2 months ago
mrs83
published a
model about 2 months ago
mrs83
updated a
collection about 2 months ago