Fine-tuning

training workload

Curated

Adapting a pretrained model to a task or domain. Full fine-tuning approaches training cost; parameter-efficient methods (LoRA/QLoRA) fit far smaller.

Classification

Category: training
Compute profile: mixed
Scaling: single-node

Infrastructure considerations

Qualitative — no fabricated numbers

Interconnect sensitivity: medium
Network sensitivity: low

Curated · Memory · Full: weights + gradients + optimizer. PEFT/LoRA: weights + small adapter state.

Curated · Storage · Dataset staging plus adapter/checkpoint storage.

Typical frameworks

Curated — not a catalog relationship

PyTorchHugging Face PEFTDeepSpeed

Relevant models

Models that run this workload

deepseek-ai/deepseek-r1
meta-llama/llama-3.1-8b-instruct
meta-llama/llama-3.3-70b-instruct
qwen/qwen2.5-7b-instruct
qwen/qwen2.5-72b-instruct
mistralai/mistral-7b-instruct-v0.3

Data class

Curated · The workload taxonomy is authored qualitative reference, not an ingested source. Sensitivities and profiles are classifications, not measurements; a workload's concrete VRAM comes from the specific model it runs.