Update README.md
Browse files
README.md
CHANGED
|
@@ -11,8 +11,6 @@ base_model:
|
|
| 11 |
|
| 12 |
**Transform English instructions with data context into executable pandas code with AI**
|
| 13 |
|
| 14 |
-
## Model Details
|
| 15 |
-
|
| 16 |
### Model Description
|
| 17 |
|
| 18 |
Csv-AI-Cleaner converts **natural language instructions** into **pandas code** for data cleaning, filtering, grouping, sorting, merges, and more.
|
|
@@ -28,7 +26,6 @@ Fine-tuned on synthetic + real-world datasets using CodeT5 with LoRA for efficie
|
|
| 28 |
|
| 29 |
- **Repository:** https://huggingface.co/arhansd1/Csv-AI-Cleaner-V3
|
| 30 |
|
| 31 |
-
## Uses
|
| 32 |
|
| 33 |
### Direct Use
|
| 34 |
- Input: Context (sample dataset) + instruction in natural language
|
|
@@ -95,14 +92,10 @@ Instruction: Show IT department employees earning over 45000
|
|
| 95 |
print(generate_code(input_example))
|
| 96 |
```
|
| 97 |
|
| 98 |
-
## Training Details
|
| 99 |
-
|
| 100 |
### Training Data
|
| 101 |
- Combination of synthetic data cleaning instructions + public dataset column contexts
|
| 102 |
- Augmented with filtered StackOverflow code snippets for pandas tasks
|
| 103 |
|
| 104 |
-
### Training Procedure
|
| 105 |
-
|
| 106 |
#### Preprocessing
|
| 107 |
- Normalized table format for context section
|
| 108 |
- Instruction phrasing normalized to imperative form
|
|
@@ -114,7 +107,6 @@ print(generate_code(input_example))
|
|
| 114 |
- Batch size: 8
|
| 115 |
- Precision: fp16 mixed precision
|
| 116 |
|
| 117 |
-
## Evaluation
|
| 118 |
|
| 119 |
### Testing Data
|
| 120 |
- Held-out set of 500 natural language → pandas task pairs
|
|
@@ -136,8 +128,6 @@ High syntax accuracy, good partial match rate, slightly lower exact match on mul
|
|
| 136 |
- **Compute Region:** US-East
|
| 137 |
- **Carbon Emitted:** ~1.2 kg CO2eq (estimate)
|
| 138 |
|
| 139 |
-
## Technical Specifications
|
| 140 |
-
|
| 141 |
### Model Architecture and Objective
|
| 142 |
- CodeT5-base (encoder-decoder)
|
| 143 |
- Objective: Seq2Seq code generation from natural language + data context
|
|
@@ -149,4 +139,4 @@ High syntax accuracy, good partial match rate, slightly lower exact match on mul
|
|
| 149 |
## Model Card Contact
|
| 150 |
- **Author:** ArhanSD1
|
| 151 |
- **Hugging Face:** https://huggingface.co/arhansd1
|
| 152 |
-
- **Email:**
|
|
|
|
| 11 |
|
| 12 |
**Transform English instructions with data context into executable pandas code with AI**
|
| 13 |
|
|
|
|
|
|
|
| 14 |
### Model Description
|
| 15 |
|
| 16 |
Csv-AI-Cleaner converts **natural language instructions** into **pandas code** for data cleaning, filtering, grouping, sorting, merges, and more.
|
|
|
|
| 26 |
|
| 27 |
- **Repository:** https://huggingface.co/arhansd1/Csv-AI-Cleaner-V3
|
| 28 |
|
|
|
|
| 29 |
|
| 30 |
### Direct Use
|
| 31 |
- Input: Context (sample dataset) + instruction in natural language
|
|
|
|
| 92 |
print(generate_code(input_example))
|
| 93 |
```
|
| 94 |
|
|
|
|
|
|
|
| 95 |
### Training Data
|
| 96 |
- Combination of synthetic data cleaning instructions + public dataset column contexts
|
| 97 |
- Augmented with filtered StackOverflow code snippets for pandas tasks
|
| 98 |
|
|
|
|
|
|
|
| 99 |
#### Preprocessing
|
| 100 |
- Normalized table format for context section
|
| 101 |
- Instruction phrasing normalized to imperative form
|
|
|
|
| 107 |
- Batch size: 8
|
| 108 |
- Precision: fp16 mixed precision
|
| 109 |
|
|
|
|
| 110 |
|
| 111 |
### Testing Data
|
| 112 |
- Held-out set of 500 natural language → pandas task pairs
|
|
|
|
| 128 |
- **Compute Region:** US-East
|
| 129 |
- **Carbon Emitted:** ~1.2 kg CO2eq (estimate)
|
| 130 |
|
|
|
|
|
|
|
| 131 |
### Model Architecture and Objective
|
| 132 |
- CodeT5-base (encoder-decoder)
|
| 133 |
- Objective: Seq2Seq code generation from natural language + data context
|
|
|
|
| 139 |
## Model Card Contact
|
| 140 |
- **Author:** ArhanSD1
|
| 141 |
- **Hugging Face:** https://huggingface.co/arhansd1
|
| 142 |
+
- **Email:** N/A
|