Update README.md
Browse files
README.md
CHANGED
|
@@ -3,29 +3,37 @@ license: mit
|
|
| 3 |
base_model: TheBloke/zephyr-7B-alpha-GPTQ
|
| 4 |
tags:
|
| 5 |
- generated_from_trainer
|
|
|
|
|
|
|
| 6 |
model-index:
|
| 7 |
-
- name:
|
| 8 |
results: []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
---
|
| 10 |
|
| 11 |
-
|
| 12 |
-
should probably proofread and complete it, then remove this comment. -->
|
| 13 |
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
This model is a fine-tuned version of [TheBloke/zephyr-7B-alpha-GPTQ](https://huggingface.co/TheBloke/zephyr-7B-alpha-GPTQ) on an unknown dataset.
|
| 17 |
|
| 18 |
## Model description
|
| 19 |
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
## Intended uses & limitations
|
| 23 |
|
| 24 |
-
|
| 25 |
|
| 26 |
-
## Training
|
| 27 |
|
| 28 |
-
|
|
|
|
| 29 |
|
| 30 |
## Training procedure
|
| 31 |
|
|
@@ -33,14 +41,20 @@ More information needed
|
|
| 33 |
|
| 34 |
The following hyperparameters were used during training:
|
| 35 |
- learning_rate: 0.0002
|
|
|
|
| 36 |
- train_batch_size: 8
|
| 37 |
- eval_batch_size: 8
|
| 38 |
-
-
|
|
|
|
| 39 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 40 |
- lr_scheduler_type: cosine
|
| 41 |
- lr_scheduler_warmup_ratio: 0.1
|
| 42 |
- num_epochs: 10
|
| 43 |
- mixed_precision_training: Native AMP
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
|
| 45 |
### Framework versions
|
| 46 |
|
|
@@ -48,3 +62,9 @@ The following hyperparameters were used during training:
|
|
| 48 |
- Pytorch 2.1.0+cu121
|
| 49 |
- Datasets 2.16.1
|
| 50 |
- Tokenizers 0.15.1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
base_model: TheBloke/zephyr-7B-alpha-GPTQ
|
| 4 |
tags:
|
| 5 |
- generated_from_trainer
|
| 6 |
+
- gptq
|
| 7 |
+
- peft
|
| 8 |
model-index:
|
| 9 |
+
- name: thesa
|
| 10 |
results: []
|
| 11 |
+
datasets:
|
| 12 |
+
- loaiabdalslam/counselchat
|
| 13 |
+
language:
|
| 14 |
+
- en
|
| 15 |
+
pipeline_tag: text-generation
|
| 16 |
---
|
| 17 |
|
| 18 |
+
# Thesa: A Therapy Chatbot 👩🏻⚕️
|
|
|
|
| 19 |
|
| 20 |
+
Thesa is an experimental project of a therapy chatbot trained on mental health data and fine-tuned with the Zephyr GPTQ model that uses quantization to decrease high computatinal and storage costs.
|
|
|
|
|
|
|
| 21 |
|
| 22 |
## Model description
|
| 23 |
|
| 24 |
+
- Model type: A fine-tuned version of Zephyr 7B Alpha - GPTQ on various mental health datasets
|
| 25 |
+
- Language(s): English
|
| 26 |
+
- License: MIT
|
| 27 |
+
- Fine-tuned from: [TheBloke/zephyr-7B-alpha-GPTQ](https://huggingface.co/TheBloke/zephyr-7B-alpha-GPTQ)
|
| 28 |
|
| 29 |
## Intended uses & limitations
|
| 30 |
|
| 31 |
+
This model is purely experimental and should not be used as substitute for a mental health professional.
|
| 32 |
|
| 33 |
+
## Training evaluation
|
| 34 |
|
| 35 |
+
Training loss:
|
| 36 |
+
<img src="imgs/loss_27.2.24.png" alt="loss" width="550"/>
|
| 37 |
|
| 38 |
## Training procedure
|
| 39 |
|
|
|
|
| 41 |
|
| 42 |
The following hyperparameters were used during training:
|
| 43 |
- learning_rate: 0.0002
|
| 44 |
+
- warmup_ratio: 0.1
|
| 45 |
- train_batch_size: 8
|
| 46 |
- eval_batch_size: 8
|
| 47 |
+
- gradient_accumulation_steps: 1
|
| 48 |
+
- seed: 35
|
| 49 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 50 |
- lr_scheduler_type: cosine
|
| 51 |
- lr_scheduler_warmup_ratio: 0.1
|
| 52 |
- num_epochs: 10
|
| 53 |
- mixed_precision_training: Native AMP
|
| 54 |
+
- fp16: True
|
| 55 |
+
|
| 56 |
+
Learning rate overtime (warm up ratio was used during training):
|
| 57 |
+
<img src="imgs/lr_27.2.24.png" alt="lr" width="550"/>
|
| 58 |
|
| 59 |
### Framework versions
|
| 60 |
|
|
|
|
| 62 |
- Pytorch 2.1.0+cu121
|
| 63 |
- Datasets 2.16.1
|
| 64 |
- Tokenizers 0.15.1
|
| 65 |
+
- Accelerate 0.27.2
|
| 66 |
+
- PEFT 0.8.2
|
| 67 |
+
- Auto-GPTQ 0.6.0
|
| 68 |
+
- TRL 0.7.11
|
| 69 |
+
- Optimum 1.17.1
|
| 70 |
+
- Bitsandbytes 0.42.0
|