Update README.md
Browse files
README.md
CHANGED
|
@@ -1,35 +1,48 @@
|
|
| 1 |
---
|
| 2 |
license: llama3
|
| 3 |
library_name: peft
|
|
|
|
|
|
|
| 4 |
tags:
|
| 5 |
- trl
|
| 6 |
- sft
|
| 7 |
- unsloth
|
| 8 |
- generated_from_trainer
|
|
|
|
| 9 |
base_model: gradientai/Llama-3-8B-Instruct-262k
|
| 10 |
model-index:
|
| 11 |
- name: llama3-biotokenpretrain-kaniwa
|
| 12 |
results: []
|
| 13 |
---
|
| 14 |
|
| 15 |
-
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 16 |
-
should probably proofread and complete it, then remove this comment. -->
|
| 17 |
-
|
| 18 |
# llama3-biotokenpretrain-kaniwa
|
| 19 |
|
| 20 |
-
This model is a fine-tuned version of [gradientai/Llama-3-8B-Instruct-262k](https://huggingface.co/gradientai/Llama-3-8B-Instruct-262k) on the None dataset.
|
| 21 |
|
| 22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
-
|
|
|
|
|
|
|
| 25 |
|
| 26 |
-
##
|
|
|
|
| 27 |
|
| 28 |
-
|
|
|
|
|
|
|
| 29 |
|
| 30 |
-
|
| 31 |
|
| 32 |
-
More information needed
|
| 33 |
|
| 34 |
## Training procedure
|
| 35 |
|
|
@@ -47,14 +60,15 @@ The following hyperparameters were used during training:
|
|
| 47 |
- lr_scheduler_warmup_steps: 5
|
| 48 |
- training_steps: 280
|
| 49 |
|
| 50 |
-
### Training results
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
### Framework versions
|
| 55 |
|
| 56 |
- PEFT 0.10.0
|
| 57 |
- Transformers 4.40.2
|
| 58 |
- Pytorch 2.2.1+cu121
|
| 59 |
- Datasets 2.19.1
|
| 60 |
-
- Tokenizers 0.19.1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: llama3
|
| 3 |
library_name: peft
|
| 4 |
+
language:
|
| 5 |
+
- en
|
| 6 |
tags:
|
| 7 |
- trl
|
| 8 |
- sft
|
| 9 |
- unsloth
|
| 10 |
- generated_from_trainer
|
| 11 |
+
- dna
|
| 12 |
base_model: gradientai/Llama-3-8B-Instruct-262k
|
| 13 |
model-index:
|
| 14 |
- name: llama3-biotokenpretrain-kaniwa
|
| 15 |
results: []
|
| 16 |
---
|
| 17 |
|
|
|
|
|
|
|
|
|
|
| 18 |
# llama3-biotokenpretrain-kaniwa
|
| 19 |
|
|
|
|
| 20 |
|
| 21 |
+
This is a LoRA adapter.
|
| 22 |
+
|
| 23 |
+
The base model is the longer-context LLaMA-3-8b-Instruct developed by Gradient and Crusoe: `gradientai/Llama-3-8B-Instruct-262k`
|
| 24 |
+
|
| 25 |
+
The tokenizer has added "biotokens" ∎A, ∎C, ∎G, and ∎T.
|
| 26 |
+
|
| 27 |
+
The dataset was 0.5% of BYU's 2019 kaniwa (*Chenopodium pallidicaule*) genome, from https://genomevolution.org/coge/GenomeInfo.pl?gid=53872
|
| 28 |
+
|
| 29 |
+
The adapter was finetuned for 3 hours on an L4 GPU. The data was split into ~7k nucleotide snippets with an Alpaca like message format.
|
| 30 |
+
|
| 31 |
+
Training Notebook: https://colab.research.google.com/drive/1FKA3p_jnfRHYd-hqJdYmKn8MQpxec0t5?usp=sharing
|
| 32 |
|
| 33 |
+
Sample message:
|
| 34 |
+
```
|
| 35 |
+
Write information about the nucleotide sequence.
|
| 36 |
|
| 37 |
+
### Sequence:
|
| 38 |
+
∎G∎C∎C∎T∎A∎T∎A∎G∎T∎G∎T∎G∎T∎A∎G...
|
| 39 |
|
| 40 |
+
### Annotation:
|
| 41 |
+
Information about location in the kaniwa chromosome: >lcl|Cp5
|
| 42 |
+
```
|
| 43 |
|
| 44 |
+
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 45 |
|
|
|
|
| 46 |
|
| 47 |
## Training procedure
|
| 48 |
|
|
|
|
| 60 |
- lr_scheduler_warmup_steps: 5
|
| 61 |
- training_steps: 280
|
| 62 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
### Framework versions
|
| 64 |
|
| 65 |
- PEFT 0.10.0
|
| 66 |
- Transformers 4.40.2
|
| 67 |
- Pytorch 2.2.1+cu121
|
| 68 |
- Datasets 2.19.1
|
| 69 |
+
- Tokenizers 0.19.1
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
### Genome Citation
|
| 73 |
+
|
| 74 |
+
Mangelson H, et al. The genome of *Chenopodium pallidicaule*: an emerging Andean super grain. Appl. Plant Sci. 2019;7:e11300. doi: 10.1002/aps3.11300
|