Scivity/parameter-golf-armenia
A submission to the "Parameter Golf Armenia" competition, focusing on hyperparameter tuning for a GPT language model training script. The author modified optimizer settings (Muon + AdamW), learning rates, sequence length, and dataset sharding to achieve improved validation bits-per-byte (bpb) on the FineWeb dataset compared to a baseline configuration.
A capstone project evaluating GPT-3.5-Turbo's performance on Armenian language tasks, including extractive QA, multip...
A competition support repository for Armenian participants in the OpenAI Parameter Golf challenge. It outlines rules,...
A project exploring Armenian language autocomplete using multiple NLP approaches including Word2Vec, LSTM, BERT trans...
ArmBench-LLM is a specialized evaluation framework for benchmarking large language models on Armenian language tasks....
ArmBench-TextEmbed is a specialized Python benchmark for evaluating the performance of text embedding models on the A...
A research project exploring how unsupervised pre-training on untranscribed audio from different language families af...