|
--- |
|
license: apache-2.0 |
|
datasets: |
|
- kobprof/skolegpt-instruct |
|
language: |
|
- da |
|
--- |
|
|
|
|
|
# Munin-7b-alpha instruction fined tuned |
|
[Munin-7b-alpha](https://huggingface.co/danish-foundation-models/munin-7b-alpha) from [Danish Foundation Models](https://www.foundationmodels.dk/) fine-tuned by [yours truly](https://www.linkedin.com/in/kaspergroesludvigsen/) for 1 epoch on [kobprof/skolegpt-instruct](https://huggingface.co/datasets/kobprof/skolegpt-instruct) using the code from [this notebook](https://github.com/alexandrainst/d3a-llm-workshop) by The Alexandra Institute |
|
|
|
Trained on a single Nvidia RTX A4000 GPU using 13.82 GB GPU memory (87.84%), of which 8.71 GB (55.39%) was used for LoRa. |
|
|
|
The model trained for just shy of 4 hours consuming a total of 0.694 KWh (as per estimates produced with CodeCarbon) and emitting approximately 57 gCO2e (average CO2e emissions per KWh during training was 82.5 g as per https://www.energidataservice.dk/tso-electricity/CO2Emis) |
