Yehor/kulyk-uk-en
TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Jul 18, 2025License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold
The kulyk-uk-en model by Yehor is a lightweight 0.35 billion parameter machine translation model, fine-tuned from the LFM2 architecture, specifically designed for translating Ukrainian to English. It was trained on 40 million quality-filtered samples and achieves a BLEU score of 36.27 on the FLORES-200 benchmark. This model is optimized for efficient deployment, requiring only 1 GB of RAM with bf16, making it suitable for resource-constrained environments.
Loading preview...
Yehor/kulyk-uk-en: Lightweight Ukrainian to English Machine Translation
This model, developed by Yehor, is a compact 0.35 billion parameter machine translation system focused on translating text from Ukrainian to English. It is built upon the LFM2 architecture and has been fine-tuned for high performance in this specific language pair.
Key Capabilities & Features
- Dedicated Translation: Optimized for Ukrainian to English machine translation.
- Efficient Performance: Requires only 1 GB of RAM when running with bf16, making it suitable for deployment on systems with limited resources.
- Competitive Benchmarking: Achieves a BLEU score of 36.27 on the FLORES-200 dataset, indicating solid translation quality for its size.
- High Throughput: Capable of processing 229.93 tokens per second (batch size 1) and up to 8392.48 tokens per second (batch size 64).
- Training Details: Fine-tuned with 40 million quality-filtered samples over 1.4 epochs, utilizing Flash Attention 2 and DeepSpeed for efficient training on NVIDIA RTX 3090 Ti GPUs.
Use Cases
- Ukrainian-English Translation: Ideal for applications requiring fast and efficient translation from Ukrainian into English.
- Resource-Constrained Environments: Its low memory footprint makes it suitable for edge devices or servers with limited RAM.
- Integration into Services: Can be easily integrated into web services or applications for real-time translation tasks.