A coding model learns to make watercolor-style images
An open training recipe links generated drawing code to a visual reward loop.
Article·4 minEXPLORE THE ECOSYSTEM
Training
Fine-tuning, evaluation, and reproducible learning recipes.
12 articlesNewest announcements first
Article·4 min
A small GRPO experiment targeted structured-output reliability
A public recipe measured whether light post-training could improve a compact model’s schema compliance.
Article·4 min
Teaching a retriever the vocabulary and document lengths of your domain
Multi-vector fine-tuning offers a route to domain-specific retrieval, but good relevance data and evaluation matter as much as the model.
Article·3 min
Parameter-efficient fine-tuning had more to offer than LoRA alone
A comparative guide encouraged choosing an adaptation method instead of accepting the default.
Article·3 min
TRL 1.0 balanced post-training experimentation with stability
The release framed a rapidly changing research area as infrastructure people increasingly rely on.
Article·4 min
Transformers v5 made tokenizer internals easier to separate
The redesign distinguished tokenizer architecture from the learned vocabulary attached to it.
Article·4 min
Transformers v5 rethought the model-definition layer
The initial v5 release candidate focused on a library serving a much wider AI ecosystem.
Article·4 min
Trackio gives experiments a lightweight place to live
Local dashboards and optional Spaces sharing make training runs easier to compare.
Article·4 min
nanoVLM made vision-language training easier to read
A minimal educational toolkit exposed the moving parts of a vision-language model without a sprawling codebase.
Article·4 min
Open-R1 turned a reasoning-model release into an open research agenda
The project aimed to reproduce the missing pieces of a reasoning model’s training process.
Article·4 min