Edit model card

Uploaded model

  • Compute sponsored by: Nvidia and Arrow ECS Denmark through Danish Data Science Community
  • Developed by: ThatsGroes
  • License: apache-2.0
  • Finetuned from model : meta-llama/Llama-3.1-70B-Instruct

LoRA adapter on Llama-3.1-70b loaded in 4-bit. Trained for 1 epoch with rank=lora_alpha=8

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

We ended up using 62.52 GB GPU memory (79.00%), of which 23.83 GB (30.12%) was used for LoRa.

[codecarbon INFO @ 11:07:59] Energy consumed for RAM : 2.574882 kWh. RAM Power : 188.78840446472168 W [codecarbon INFO @ 11:07:59] Energy consumed for all GPUs : 4.045188 kWh. Total GPU Power : 270.22211938762564 W [codecarbon INFO @ 11:07:59] Energy consumed for all CPUs : 0.579916 kWh. Total CPU Power : 42.5 W [codecarbon INFO @ 11:07:59] 7.199986 kWh of electricity used since the beginning.

Downloads last month
34
Safetensors
Model size
70.6B params
Tensor type
BF16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for ThatsGroes/Llama-3.1-70B-Instruct-SkoleGPT

Finetuned
(40)
this model
Quantizations
1 model

Dataset used to train ThatsGroes/Llama-3.1-70B-Instruct-SkoleGPT