Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,81 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
library_name: transformers
|
| 3 |
+
model_name: multiscreen_psi16_768
|
| 4 |
license: apache-2.0
|
| 5 |
+
datasets:
|
| 6 |
+
- roneneldan/TinyStories
|
| 7 |
+
tags:
|
| 8 |
+
- generated_from_trainer
|
| 9 |
+
- sft
|
| 10 |
+
- trl
|
| 11 |
+
- multiscreen
|
| 12 |
+
- tiny-stories
|
| 13 |
+
- small-language-model
|
| 14 |
+
- experimental
|
| 15 |
+
- research
|
| 16 |
+
- arxiv:2604.01178
|
| 17 |
+
- arxiv:2305.07759
|
| 18 |
---
|
| 19 |
+
|
| 20 |
+
# Model Card for multiscreen_psi16_768
|
| 21 |
+
|
| 22 |
+
This model is an **unofficial** experimental pre-traind model of multiscreen with [ TinyStories](https://huggingface.co/datasets/roneneldan/TinyStories) datasets.
|
| 23 |
+
It has been trained using [TRL](https://github.com/huggingface/trl).
|
| 24 |
+
|
| 25 |
+
## Quick start
|
| 26 |
+
|
| 27 |
+
```python
|
| 28 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
model_id = "kurogane/tinystorys_multiscreen_vocab768"
|
| 32 |
+
cache_dir = r"/media/kurogane/backup/cache"
|
| 33 |
+
|
| 34 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 35 |
+
model_id,
|
| 36 |
+
trust_remote_code=True,
|
| 37 |
+
cache_dir=cache_dir,
|
| 38 |
+
)
|
| 39 |
+
model.to("cuda:0")
|
| 40 |
+
|
| 41 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
| 42 |
+
model_id,
|
| 43 |
+
padding_side="left",
|
| 44 |
+
cache_dir=cache_dir,
|
| 45 |
+
)
|
| 46 |
+
|
| 47 |
+
model_inputs = tokenizer(["A list of colors: red, blue"], return_tensors="pt").to(model.device)
|
| 48 |
+
generated_ids = model.generate(**model_inputs)
|
| 49 |
+
|
| 50 |
+
s_output = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
| 51 |
+
print(s_output)
|
| 52 |
+
```
|
| 53 |
+
|
| 54 |
+
### result example
|
| 55 |
+
> A list of colors: red, blue, yellow, green, orange. All the people
|
| 56 |
+
|
| 57 |
+
## Training procedure
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
This model was trained with SFT.
|
| 63 |
+
|
| 64 |
+
### Framework versions
|
| 65 |
+
|
| 66 |
+
- TRL: 0.24.0
|
| 67 |
+
- Transformers: 5.8.0
|
| 68 |
+
- Pytorch: 2.11.0+cu129
|
| 69 |
+
- Datasets: 4.3.0
|
| 70 |
+
- Tokenizers: 0.22.2
|
| 71 |
+
|
| 72 |
+
## Used archtechture
|
| 73 |
+
This model is an experimental tiny language model trained on TinyStories using a Multiscreen-style architecture inspired by the paper *Screening Is Enough* by Ken M. Nakanishi.
|
| 74 |
+
This model implementation was developed as an experimental Hugging Face Transformers port, with reference to the unofficial PyTorch implementation `dieOD/multiscreen-pytorch`. This model is not an official implementation released by the author of the Multiscreen paper.
|
| 75 |
+
- Multiscreen paper: https://arxiv.org/abs/2604.01178
|
| 76 |
+
- Reference implementation: https://github.com/dieOD/multiscreen-pytorch
|
| 77 |
+
|
| 78 |
+
## Used dataset
|
| 79 |
+
The training data is based on the TinyStories dataset by Ronen Eldan and Yuanzhi Li.
|
| 80 |
+
- TinyStories paper: https://arxiv.org/abs/2305.07759
|
| 81 |
+
- TinyStories dataset: https://huggingface.co/datasets/roneneldan/TinyStories
|