license: apache-2.0
language:
- en
datasets:
- appvoid/no-prompt-15k
pipeline_tag: text-generation
palmer
a better base model
palmer is a series of ~1b parameters language models fine-tuned to be used as base models instead of using custom prompts for tasks. This means that it can be further fine-tuned on more data with custom prompts as usual or be used for downstream tasks as any base model you can get. The model has the best of both worlds: some "bias" to act as an assistant, but also the abillity to predict the next-word from its internet knowledge base. It's a 1.3b llama 2 model so you can use it with your favorite tools/frameworks.
evaluation
| Model | ARC_C | HellaSwag | PIQA | Winogrande |
|---|---|---|---|---|
| tinyllama-2t | 0.2807 | 0.5463 | 0.7067 | 0.5683 |
| palmer-001 | 0.2807 | 0.5524 | 0.7106 | 0.5896 |
| sheared-1.3b | 0.2910 | 0.5935 | 0.7339 | 0.5809 |
| no-prompt-1.3b | 0.3157 | 0.6022 | 0.7334 | 0.5864 |
| falcon-rw-1b-instruct-openorca (sota) | 0.3362 | 0.5997 | 0.7394 | 0.6148 |
This model was trained on less than 25% of the dataset yet achieves competitive performance to current sota on open llm leaderboard. Wait for what it's coming!
training
Training took ~5 P100 gpu hours. It was trained on 15,000 gpt-4 shuffled samples. palmer was fine-tuned using lower learning rates ensuring it keeps as much general knowledge as possible.
prompt
no prompt
limitations
Hallucinations are frequent, just as any transformer model this size.
