Tasks:
- * Analyze convergence and overfitting
- * Analyze results and prepare experiment reports
- * Build evaluation datasets and automated evaluation scripts
- * Build supervised fine-tuning and preference datasets
- * Clean, synthesize, convert, and quality-check multimodal training data
- * Compare data, hyperparameter, and training strategy performance
- * Document experiments and technical findings
- * Experiment with edge inference
- * Explore model distillation and quantization
- * Monitor training loss and metrics
- * Run model fine-tuning experiments using SFT, LoRA, QLoRA, and DPO
- * Write training and evaluation scripts
Perks/Benefits:
Skills/Tech stack required:
[Attention] [AWQ] [C++] [Data Processing] [Deep learning] [DeepSpeed] [Direct Preference Optimization] [Direct Preference Optimization (DPO)] [Distributed Training] [Docker] [Embeddings] [Fully Sharded Data Parallel] [Fully Sharded Data Parallel (FSDP)] [Git] [GPTQ] [Hugging Face] [Hugging Face Transformers] [Language Models (LLMs)] [Large Language Models] [Large Language Models (LLMs)] [Linux] [Llama.cpp] [LoRA] [Loss Functions] [Model Distillation] [Model Evaluation] [Model Quantization] [Multimodal Data] [Multimodal Data Processing] [ONNX Runtime] [Optimization Algorithms] [Preference Datasets] [Preference Optimization (DPO)] [Python] [PyTorch] [QLoRA] [Reinforcement Learning] [Reinforcement Learning from Human Feedback] [Reinforcement Learning from Human Feedback (RLHF)] [Supervised Fine Tuning] [Supervised Fine-Tuning (SFT)] [TensorRT] [Tokenization] [Transformer] [Vision-language] [Vision-language-action] [Vision-Language-Action Models] [Vision-language-action models (VLAs)] [Vision Language Models] [Vision-language models (VLMs)]
Educational requirements:
N/A
Role(s):
[Intern] [Large Language Model Training Intern] [Learning Intern] [Machine Learning Intern] [Training Intern]