Commit
•
005b47c
1
Parent(s):
0b50e35
Update README.md
Browse files
README.md
CHANGED
|
@@ -3,65 +3,73 @@ license: mit
|
|
| 3 |
base_model: microsoft/Phi-3-mini-4k-instruct
|
| 4 |
tags:
|
| 5 |
- alignment-handbook
|
| 6 |
-
- trl
|
| 7 |
- dpo
|
| 8 |
-
- generated_from_trainer
|
| 9 |
- trl
|
| 10 |
-
-
|
| 11 |
-
- generated_from_trainer
|
| 12 |
datasets:
|
| 13 |
-
-
|
| 14 |
-
- original
|
| 15 |
model-index:
|
| 16 |
-
- name:
|
| 17 |
results: []
|
| 18 |
---
|
| 19 |
|
|
|
|
|
|
|
| 20 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 21 |
should probably proofread and complete it, then remove this comment. -->
|
| 22 |
|
| 23 |
-
# 0.001_SELM-Phi-3-mini-4k-instruct_iter_1
|
| 24 |
|
| 25 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
## Model description
|
| 28 |
|
| 29 |
-
More information needed
|
| 30 |
|
| 31 |
-
## Intended uses & limitations
|
| 32 |
|
| 33 |
-
|
|
|
|
|
|
|
|
|
|
| 34 |
|
| 35 |
-
##
|
| 36 |
|
| 37 |
-
More information needed
|
| 38 |
|
| 39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
### Training hyperparameters
|
| 42 |
|
| 43 |
The following hyperparameters were used during training:
|
| 44 |
-
-
|
|
|
|
| 45 |
- train_batch_size: 4
|
| 46 |
-
- eval_batch_size: 4
|
| 47 |
- seed: 42
|
| 48 |
- distributed_type: multi-GPU
|
| 49 |
- num_devices: 8
|
| 50 |
- gradient_accumulation_steps: 4
|
| 51 |
- total_train_batch_size: 128
|
| 52 |
-
- total_eval_batch_size: 32
|
| 53 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 54 |
-
- lr_scheduler_type: cosine
|
| 55 |
-
- lr_scheduler_warmup_ratio: 0.1
|
| 56 |
- num_epochs: 1
|
| 57 |
|
| 58 |
-
### Training results
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
### Framework versions
|
| 63 |
|
| 64 |
- Transformers 4.40.2
|
| 65 |
-
- Pytorch 2.
|
| 66 |
- Datasets 2.14.6
|
| 67 |
- Tokenizers 0.19.1
|
|
|
|
| 3 |
base_model: microsoft/Phi-3-mini-4k-instruct
|
| 4 |
tags:
|
| 5 |
- alignment-handbook
|
|
|
|
| 6 |
- dpo
|
|
|
|
| 7 |
- trl
|
| 8 |
+
- selm
|
|
|
|
| 9 |
datasets:
|
| 10 |
+
- HuggingFaceH4/ultrafeedback_binarized
|
|
|
|
| 11 |
model-index:
|
| 12 |
+
- name: SELM-Phi-3-mini-4k-instruct-iter-1
|
| 13 |
results: []
|
| 14 |
---
|
| 15 |
|
| 16 |
+
|
| 17 |
+
|
| 18 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 19 |
should probably proofread and complete it, then remove this comment. -->
|
| 20 |
|
|
|
|
| 21 |
|
| 22 |
+
|
| 23 |
+
Self-Exploring Language Models: Active Preference Elicitation for Online Alignment.
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
# SELM-Phi-3-mini-4k-instruct-iter-1
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
This model is a fine-tuned version of [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) using synthetic data based on on the HuggingFaceH4/ultrafeedback_binarized dataset.
|
| 32 |
+
|
| 33 |
+
|
| 34 |
|
| 35 |
## Model description
|
| 36 |
|
|
|
|
| 37 |
|
|
|
|
| 38 |
|
| 39 |
+
- Model type: A 3.8B parameter Phi3-instruct-based Self-Exploring Language Models (SELM).
|
| 40 |
+
- License: MIT
|
| 41 |
+
|
| 42 |
+
|
| 43 |
|
| 44 |
+
## Results
|
| 45 |
|
|
|
|
| 46 |
|
| 47 |
+
|
| 48 |
+
| | AlpacaEval 2.0 (LC WR) | MT-Bench (Average) |
|
| 49 |
+
|----------------------------------------|------------------------|--------------------|
|
| 50 |
+
| [SELM-Phi-3-mini-4k-instruct-iter-3](https://huggingface.co/ZhangShenao/SELM-Phi-3-mini-4k-instruct-iter-3) |        27.98 |       8.32 |
|
| 51 |
+
| [SELM-Phi-3-mini-4k-instruct-iter-2](https://huggingface.co/ZhangShenao/SELM-Phi-3-mini-4k-instruct-iter-2) |        26.79 |       8.44 |
|
| 52 |
+
| [SELM-Phi-3-mini-4k-instruct-iter-1](https://huggingface.co/ZhangShenao/SELM-Phi-3-mini-4k-instruct-iter-1) |        27.33 |       8.37 |
|
| 53 |
+
| [Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) |        23.05 |       8.12 |
|
| 54 |
+
|
| 55 |
|
| 56 |
### Training hyperparameters
|
| 57 |
|
| 58 |
The following hyperparameters were used during training:
|
| 59 |
+
- alpha: 0.001
|
| 60 |
+
- beta: 0.01
|
| 61 |
- train_batch_size: 4
|
|
|
|
| 62 |
- seed: 42
|
| 63 |
- distributed_type: multi-GPU
|
| 64 |
- num_devices: 8
|
| 65 |
- gradient_accumulation_steps: 4
|
| 66 |
- total_train_batch_size: 128
|
|
|
|
| 67 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
|
|
|
|
|
|
| 68 |
- num_epochs: 1
|
| 69 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
### Framework versions
|
| 71 |
|
| 72 |
- Transformers 4.40.2
|
| 73 |
+
- Pytorch 2.1.2+cu121
|
| 74 |
- Datasets 2.14.6
|
| 75 |
- Tokenizers 0.19.1
|