Beschreibung
We are currently looking for a freelance Recipe Intelligence - Language Model Fine-Tuning Specialist (m/f/d) for our client.Client Details
Start: asap
Project duration: 2 months + Extension
Utilisation: 100%
Location: Remote / Wuppertal
Industry: Production
Project language: English necessary/German
Description
We are actively fine-tuning a large-scale language model on in-house data with the goal of translating natural language inputs into machine-readable JSON configurations. We already developed a promising end-to-end prototype, including a custom data loader, a fine-tuned LLM and model serving. The next step is to further enhance the model performance and to generalize the approach to other data sources.
Tasks:
- Adapt our custom ETL pipeline to an updated training data scheme, -Craft a series of LLM prompts to generate broader training data
- Develop custom metrics to track the domain specific performance of the model on a custom test dataset
- Experiment and improve the model performance by fine-tuning larger multi-gpu models, with state-of-the-art tools like DeepSpeed, LoRa-Peft, GaLore optimizers and grammar based generation
- Serve the model via Databricks model serving API
Profile
- Deep understanding of NLP, with hands-on experience in machine learning frameworks like PyTorch, and familiarity with NLP libraries such as Hugging Face Transformers.
- Experience with fine-tuning large language models like Flan-T5, and knowledge of state-of-the-art optimization techniques and tools, including DeepSpeed, LoRA-peft, and GaLore optimizers.
- Expertise in utilizing multi-GPU environments for distributed training, and familiarity with tools and methods to optimize machine learning models for high performance and efficiency.
- Knowledge of Databricks, including API model serving, and experience with deploying models in production environments at scale.
- Strong programming skills in Python, and the ability to write maintainable, efficient, and reliable code.
- English fluent
Job Offer
Does the project sound interesting?
I look forward to hearing from you with the following information:
Your current project availability (earliest start date)
Your maximum workload/week in total
Can you offer this workload on a regular basis?
Your hourly rate (on-site / remote)
Your current profile
A brief feedback on the fit (please address the points listed above)