Evals show that our SFT (LFM-1.3B-Distill) model performs slightly better with a 32k token budget. It's actually competitive with models based on DeepSeek-R1-Distill-Qwen-1.5B while being 15% smaller.
OPEN SOURCE
-
SFT Fine-tuning Strategy for 1.3B Chat Model
By
–
The first step is SFTing our general-purpose 1.3B chat model on the right mixture. We iterated over many open-source datasets to find the ideal mix. This is great to raise performance, but also makes the model extremely verbose (>10k tokens on average).
-
Google VideoPrism tool for automated video scene segmentation
By
–
Aprovechando la salida del nuevo modelo opensource de Google VideoPrism, para análisis de vídeo, me he programado una herramienta que a partir de analizar la similitud de los embeddings obtenidos corta el vídeo en sus diferentes escenas.
— Carlos Santana (@DotCSV) 24 juin 2025
Idealmente en un futuro usaré la… pic.twitter.com/LTbgrM2RXqAprovechando la salida del nuevo modelo opensource de Google VideoPrism, para análisis de vídeo, me he programado una herramienta que a partir de analizar la similitud de los embeddings obtenidos corta el vídeo en sus diferentes escenas. Idealmente en un futuro usaré la
-
Father’s Open Source Project Generates Music Videos from JSON
By
–
my dads new open source project lets you generate music videos from a JSON file
— Yohei (@yoheinakajima) 24 juin 2025
(the JSON file included in thread basically has prompts for the music and each image/video) https://t.co/w47uoRqWr5my dads new open source project lets you generate music videos from a JSON file (the JSON file included in thread basically has prompts for the music and each image/video)
-
Living Notes System: Open Source Knowledge Management Approach
By
–
Here @Dries details his inspiration and thought process behind his system of "living notes". For him, this method aligns with the open source philosophy of continuously learning and sharing in the open https://
bit.ly/44snCop #drupal #opensource #foss -
DeepSeek’s Open Source Win Overshadows Meta’s Llama Leadership
By
–
Y algo que también dije y no me equivoqué: La gran perjudicada con la salida de DeepSeek fue Meta, a la que le rompieron el liderazgo de sacar modelos open source competentes. ¿Alguien se acuerda de Llama 4?
-
Warp 2.0 introduces Agentic Developer Environment
By
–
BREAKING 🚨: Warp is getting a Warp 2.0 upgrade and introduces ADE!
— 🚨 AI News | TestingCatalog (@testingcatalog) 24 juin 2025
ADE is an Agentic Developer Environment, running right in your terminal. It supports MCPs as well!
Testing time 👀 https://t.co/SbWnAtEqn1 pic.twitter.com/UjVp2vnvIGBREAKING : Warp is getting a Warp 2.0 upgrade and introduces ADE! ADE is an Agentic Developer Environment, running right in your terminal. It supports MCPs as well! Testing time
-
NeoBERT Model Released on HuggingFace
By
–
our model is already available on HuggingFace. the implementation is optimized for efficiency so it's not bloated like a lot of other models (wont name names here…) star NeoBERT! download NeoBERT! finetune NeoBERT! http://
huggingface.co/chandar-lab/Ne
oBERT
… -
NeoBERT: New State-of-the-Art Small Encoder Model
By
–
In the beginning, there was BERT. Eventually BERT gave rise to RoBERTa. Then, DeBERTa. Later, ModernBERT. And now, NeoBERT. The new state-of-the-art small-sized encoder: