Update README.md
Browse files
README.md
CHANGED
|
@@ -1,52 +1,128 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
-
|
| 4 |
tags:
|
| 5 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
model-index:
|
| 7 |
-
- name:
|
| 8 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
---
|
| 10 |
|
| 11 |
-
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 12 |
-
should probably proofread and complete it, then remove this comment. -->
|
| 13 |
|
| 14 |
-
# jaleah-ai-model
|
| 15 |
|
| 16 |
-
|
| 17 |
|
| 18 |
-
## Model
|
|
|
|
| 19 |
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
-
|
| 23 |
|
| 24 |
-
|
|
|
|
| 25 |
|
| 26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
|
| 28 |
-
|
| 29 |
|
| 30 |
-
##
|
|
|
|
| 31 |
|
| 32 |
-
###
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
|
| 34 |
-
|
| 35 |
-
- learning_rate: 5e-05
|
| 36 |
-
- train_batch_size: 4
|
| 37 |
-
- eval_batch_size: 8
|
| 38 |
-
- seed: 42
|
| 39 |
-
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
| 40 |
-
- lr_scheduler_type: linear
|
| 41 |
-
- num_epochs: 5
|
| 42 |
|
| 43 |
-
###
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
|
|
|
|
| 46 |
|
| 47 |
-
###
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
|
| 49 |
-
|
| 50 |
-
-
|
| 51 |
-
-
|
| 52 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
language:
|
| 3 |
+
- code
|
| 4 |
tags:
|
| 5 |
+
- code-generation
|
| 6 |
+
- ai-assistant
|
| 7 |
+
- code-completion
|
| 8 |
+
- python
|
| 9 |
+
license: mit
|
| 10 |
+
datasets:
|
| 11 |
+
- github-code
|
| 12 |
+
- stackoverflow
|
| 13 |
model-index:
|
| 14 |
+
- name: Jaleah AI Code Generator
|
| 15 |
+
results:
|
| 16 |
+
- task:
|
| 17 |
+
type: text-generation
|
| 18 |
+
name: Code Generation
|
| 19 |
+
dataset:
|
| 20 |
+
name: Python Code Corpus
|
| 21 |
+
type: generated
|
| 22 |
+
metrics:
|
| 23 |
+
- type: BLEU
|
| 24 |
+
value: experimental
|
| 25 |
+
- type: CodeBLEU
|
| 26 |
+
value: experimental
|
| 27 |
+
- type: perplexity
|
| 28 |
+
value: experimental
|
| 29 |
---
|
| 30 |
|
|
|
|
|
|
|
| 31 |
|
|
|
|
| 32 |
|
| 33 |
+
# Jaleah AI Code Generation Model
|
| 34 |
|
| 35 |
+
## Model Description
|
| 36 |
+
Jaleah AI is a fine-tuned version of the Microsoft CodeGPT small Python model, specialized in generating high-quality Python code snippets across various domains.
|
| 37 |
|
| 38 |
+
### Model Details
|
| 39 |
+
- **Developed by:** TeckMill AI Research Team
|
| 40 |
+
- **Base Model:** microsoft/CodeGPT-small-py
|
| 41 |
+
- **Language:** Python
|
| 42 |
+
- **Version:** 1.0
|
| 43 |
|
| 44 |
+
# Jaleah AI Code Generation Model
|
| 45 |
|
| 46 |
+
## Model Description
|
| 47 |
+
Jaleah AI is a fine-tuned version of the Microsoft CodeGPT small Python model, specialized in generating high-quality Python code snippets across various domains.
|
| 48 |
|
| 49 |
+
### Model Details
|
| 50 |
+
- **Developed by:** TeckMill AI Research Team
|
| 51 |
+
- **Base Model:** microsoft/CodeGPT-small-py
|
| 52 |
+
- **Language:** Python
|
| 53 |
+
- **Version:** 1.0
|
| 54 |
|
| 55 |
+
# Jaleah AI Code Generation Model
|
| 56 |
|
| 57 |
+
## Model Description
|
| 58 |
+
Jaleah AI is a fine-tuned version of the Microsoft CodeGPT small Python model, specialized in generating high-quality Python code snippets across various domains.
|
| 59 |
|
| 60 |
+
### Model Details
|
| 61 |
+
- **Developed by:** TeckMill AI Research Team
|
| 62 |
+
- **Base Model:** microsoft/CodeGPT-small-py
|
| 63 |
+
- **Language:** Python
|
| 64 |
+
- **Version:** 1.0
|
| 65 |
|
| 66 |
+
## Intended Uses & Limitations
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
|
| 68 |
+
### Intended Uses
|
| 69 |
+
- Code snippet generation
|
| 70 |
+
- Assisting developers with Python programming
|
| 71 |
+
- Providing intelligent code suggestions
|
| 72 |
+
- Rapid prototyping of Python functions and classes
|
| 73 |
|
| 74 |
+
### Limitations
|
| 75 |
+
- May generate syntactically incorrect code
|
| 76 |
+
- Requires human review and validation
|
| 77 |
+
- Performance may vary across different coding domains
|
| 78 |
+
- Not suitable for complete project generation
|
| 79 |
|
| 80 |
+
## Training Data
|
| 81 |
|
| 82 |
+
### Data Sources
|
| 83 |
+
The model was trained on a diverse dataset including:
|
| 84 |
+
- GitHub trending repositories
|
| 85 |
+
- Stack Overflow top-rated code answers
|
| 86 |
+
- Open-source Python project codebases
|
| 87 |
+
- Synthetic code generation
|
| 88 |
+
- Complex algorithmic implementations
|
| 89 |
|
| 90 |
+
### Data Preprocessing
|
| 91 |
+
- Syntax validation
|
| 92 |
+
- Comment and docstring removal
|
| 93 |
+
- Length and complexity filtering
|
| 94 |
+
|
| 95 |
+
## Training Procedure
|
| 96 |
+
|
| 97 |
+
### Training Hyperparameters
|
| 98 |
+
- **Learning Rate:** 5e-05
|
| 99 |
+
- **Batch Size:** 4
|
| 100 |
+
- **Epochs:** 12
|
| 101 |
+
- **Optimizer:** AdamW
|
| 102 |
+
- **Learning Rate Scheduler:** Linear
|
| 103 |
+
- **Weight Decay:** 0.01
|
| 104 |
+
|
| 105 |
+
### Training Process
|
| 106 |
+
- Fine-tuning of pre-trained CodeGPT model
|
| 107 |
+
- Multi-source code collection
|
| 108 |
+
- Advanced synthetic code generation
|
| 109 |
+
- Rigorous code validation
|
| 110 |
+
|
| 111 |
+
## Evaluation
|
| 112 |
+
Detailed evaluation metrics to be added in future versions.
|
| 113 |
+
|
| 114 |
+
## Ethical Considerations
|
| 115 |
+
- Designed to assist, not replace, human developers
|
| 116 |
+
- Encourages learning and code understanding
|
| 117 |
+
|
| 118 |
+
## How to Use
|
| 119 |
+
```python
|
| 120 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 121 |
+
|
| 122 |
+
model = AutoModelForCausalLM.from_pretrained("teckmill/jaleah-ai-model")
|
| 123 |
+
tokenizer = AutoTokenizer.from_pretrained("teckmill/jaleah-ai-model")
|
| 124 |
+
|
| 125 |
+
def generate_code(prompt, max_length=200):
|
| 126 |
+
input_ids = tokenizer.encode(prompt, return_tensors="pt")
|
| 127 |
+
output = model.generate(input_ids, max_length=max_length, num_return_sequences=1)
|
| 128 |
+
return tokenizer.decode(output[0], skip_special_tokens=True)
|