File size: 2,490 Bytes
77f55a0
 
 
08a4a14
 
 
 
 
 
 
 
77f55a0
 
 
 
08a4a14
77f55a0
 
 
 
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
 
08a4a14
 
 
 
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
08a4a14
77f55a0
 
 
 
 
 
08a4a14
77f55a0
 
 
 
 
 
 
 
08a4a14
77f55a0
 
 
 
 
 
 
 
 
08a4a14
77f55a0
 
 
 
 
08a4a14
77f55a0
 
 
08a4a14
77f55a0
08a4a14
77f55a0
 
 
 
 
 
 
 
 
 
 
 
08a4a14
 
 
77f55a0
 
 
 
08a4a14
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
---
library_name: peft
base_model: core42/jais-13b
license: mit
datasets:
- arbml/Ashaar_dataset
language:
- ar
metrics:
- perplexity
- bertscore
---

# Model Card for Model ID

Fine-tuned using QLoRA for poem generation task.


### Model Description

We utilize Ashaar dataset and fine-tune the model to generate poems.

The input to the model is structred as follows:

'''

\#\#\# Instruction: Generate a poem based on the following title, and the given era:

\#\#\# Input: \{Title of a poem + poet era\}

\#\#\# Response: \{Poem verses\}

'''


- **Developed by:** Abdelrahman ’Boda’ Sadallah, Anastasiia Demidova, Daria Kotova
- **Model type:** Causal LM
- **Language(s) (NLP):** Arabic
- **Finetuned from model [optional]:** core42/jais-13b

### Model Sources

- **Repository:** https://github.com/BodaSadalla98/llm-optimized-fintuning

## Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

The model is the result of our AI project. If you intend to use it, please, refer to the repo.


### Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

For improving stories generation, you can play parameters: temeperature, top_p/top_k, repetition_penalty, etc.


## Training Details

### Training Data

<!-- This should link to a Data Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

Link to the dataset on huggungface: https://huggingface.co/datasets/arbml/ashaar.


## Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

### Testing Data, Factors & Metrics


Test split of the same dataset.

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

We are using perplexity and BERTScore.

### Results

Perplexity: 48.3125

BERTScore: 59.33

## Training procedure


The following `bitsandbytes` quantization config was used during training:
- quant_method: bitsandbytes
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: fp4
- bnb_4bit_use_double_quant: False
- bnb_4bit_compute_dtype: float32

### Framework versions


- PEFT 0.6.0.dev0