Planck-OpenLAiNN-10M-GGUF 🤗

Hey there fellow researchers, developers, and AI enthusiasts! Today I'm releasing a new family of Models, Planck LAiNN, These are probably some of the smallest LLMs that are on HF. They aren't super useful but it was a fun expierment!~

These are the GGUF quants of the models. For the original models, you can find them here.

Models Overview

• Panck-OpenLAiNN-10M: A Truely Tiny model with just 10 Million parameters, this is probably boarderline useless, but it IS functional.
• Panck-OpenLAiNN-25M: The second smallest model, 25 million parameters, it's not that much better.
• Panck-OpenLAiNN-50M: Surprisingly smart, it's 50 Million parameters and could potentially maybe, Possibly even be useful ;)
• Panck-OpenLAiNN-75M: The current ""heavy"" weight of the Plank-OpenLAiNN Models.

Pretraining Details

Plank-OpenLAiNN was trained on 32B tokens of the Fineweb dataset, it's the same one that was used for the Pico-LAiNN family of models. The model was pretrained with a context length of 1024 tokens.

Other information:

• Compatibility: Built to be compatible with existing projects that use LLAMA 2's tokenizer and architecture.
• Ease of Use: No need to reinvent the wheel. These models are ready to be plugged into your applications.
• Open Source: Fully open source, so you can tweak, tune, and twist them to your heart's content.

Benchy

| Tasks | Value | |Stderr|

|--------------|------:|---|-----:|

|arc_challenge | 0.1766|± |0.0111|

|arc_easy | 0.3144|± |0.0095|

|boolq | 0.5847|± |0.0086|

|hellaswag | 0.2622|± |0.0044|

|lambada_openai| 0.0047|± |0.0009| # Yes, really

|piqa | 0.5718|± |0.0115|

|winogrande | 0.4957|± |0.0141|

Future Plans

• More Models: I'm currenetly training the bigger siblings of Pico-OpenLAiNN, including a 1B parameter version and beyond. 2-4 Billion parameter versions are planned. These will be Released as OpenLAiNN.
• New architecture: This is still up in the air and I'm still developing it, things are going well and I'll post updates.
• Paper: A detailed paper or training data will be posted at some point.

Credit Where Credit's Due

If you find these models useful and decide to use these models, a link to this repository would be highly appreciated. I am a one man show running this and I'm doing this for free, Thanks 🤗