File size: 5,388 Bytes
39e35d5 56ac5bf 39e35d5 2d8d3e5 39e35d5 a83e4c7 39e35d5 2d8d3e5 39e35d5 a83e4c7 39e35d5 a83e4c7 39e35d5 2d8d3e5 a83e4c7 2d8d3e5 39e35d5 a83e4c7 2d8d3e5 39e35d5 a83e4c7 39e35d5 a83e4c7 39e35d5 2d8d3e5 39e35d5 2d8d3e5 39e35d5 2d8d3e5 39e35d5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 |
---
library_name: transformers
license: apache-2.0
---
# Model Card for mAInframer-1
LLM for the COBOL programming language.
## Model Details
### Model Description
mAInframer is a series of models with different parameter counts: 7b, 13b, 34b.
These models are pretrained on code (CodeLlama base models) and fine-tuned on COBOL code.
## Uses
This model can be used for code completion and fill-in-the middle (COBOL).
## How to Get Started with the Model
Model loading:
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("bloopai/mAInframer-7b", device_map="cuda", torch_dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained("codellama/CodeLlama-7b-hf")
```
### Completion:
```python
prompt = ''' IDENTIFICATION DIVISION.
PROGRAM-ID. SUM-OF-CUBES.
ENVIRONMENT DIVISION.
INPUT-OUTPUT SECTION.
DATA DIVISION.
WORKING-STORAGE SECTION.
01 STEP PIC S9(10).
01 CUBE PIC 9(7).
01 CUBE-SUM PIC 9(7) VALUE 0.
LINKAGE SECTION.
01 LINKED-ITEMS.
05 L-MAX-STEP PIC S9(10).
05 RESULT PIC S9(10).
*
* Given an integer number, return the sum of the of all the integers below it.
*
* Example:
*
* sum_of_cubes(3) == 1**3 + 2**3 == 9
* sum_of_cubes(5) == 100
*
* Store the result in the RESULT variable and mark the end of your program with END PROGRAM
'''
inputs = tokenizer.encode(prompt, return_tensors="pt", add_special_tokens=False).to("cuda")
outputs = model.generate(inputs, max_new_tokens=250, use_cache=True, do_sample=False, repetition_penalty=1.1)
print(self.tokenizer.decode(outputs[0]))
```
Result:
```cobol
IDENTIFICATION DIVISION.
PROGRAM-ID. SUM-OF-CUBES.
ENVIRONMENT DIVISION.
INPUT-OUTPUT SECTION.
DATA DIVISION.
WORKING-STORAGE SECTION.
01 STEP PIC S9(10).
01 CUBE PIC 9(7).
01 CUBE-SUM PIC 9(7) VALUE 0.
LINKAGE SECTION.
01 LINKED-ITEMS.
05 L-MAX-STEP PIC S9(10).
05 RESULT PIC S9(10).
*
* Given an integer number, return the sum of the of all the integers below it.
*
* Example:
*
* sum_of_cubes(3) == 1**3 + 2**3 == 9
* sum_of_cubes(5) == 100
*
* Store the result in the RESULT variable and mark the end of your program with END PROGRAM
PROCEDURE DIVISION USING LINKED-ITEMS.
MOVE L-MAX-STEP TO STEP.
PERFORM VARYING STEP FROM 1 BY 1 UNTIL STEP > L-MAX-STEP
COMPUTE CUBE = STEP ** 3
ADD CUBE TO CUBE-SUM
END-PERFORM.
DISPLAY CUBE-SUM.
MOVE CUBE-SUM TO RESULT.
GOBACK.
END PROGRAM SUM-OF-CUBES.
```
### Infilling
Follow the format: `<PRE>prefix<SUF>suffix<MID>`
To complete `PROCEDURE DIVISION` and infill `WORKING STORAGE SECTION.` to solve [COBOLEval](https://github.com/BloopAI/COBOLEval) problems:
```python
prompt = '''<PRE> IDENTIFICATION DIVISION.
PROGRAM-ID. SUM-OF-CUBES.
ENVIRONMENT DIVISION.
INPUT-OUTPUT SECTION.
DATA DIVISION.<SUF>
LINKAGE SECTION.
01 LINKED-ITEMS.
05 L-MAX-STEP PIC S9(10).
05 RESULT PIC S9(10).
*
* Given an integer number, return the sum of the of all the integers below it.
*
* Example:
*
* sum_of_cubes(3) == 1**3 + 2**3 == 9
* sum_of_cubes(5) == 100
*
* Store the result in the RESULT variable and mark the end of your program with END PROGRAM'''
```
Result:
```cobol
<PRE> IDENTIFICATION DIVISION.
PROGRAM-ID. MAX-ELEMENT.
ENVIRONMENT DIVISION.
INPUT-OUTPUT SECTION.
DATA DIVISION.<SUF>
LINKAGE SECTION.
01 LINKED-ITEMS.
05 L-L OCCURS 100 TIMES INDEXED BY NI PIC S9(10).
05 RESULT PIC S9(10).
* Return maximum element in the list.
* >>> max_element([1, 2, 3])
* 3
* >>> max_element([5, 3, -5, 2, -3, 3, 9, 0, 123, 1, -10])
* 123
*
* Store the result in the RESULT variable and mark the end of your program with END PROGRAM
PROCEDURE DIVISION USING LINKED-ITEMS.
MOVE ZERO TO WS-MAX-VALUE.
PERFORM VARYING NI FROM 1 BY 1 UNTIL NI > 100
IF L-L (NI) > WS-MAX-VALUE THEN
MOVE L-L (NI) TO WS-MAX-VALUE
END-IF
END-PERFORM.
DISPLAY 'THE MAXIMUM ELEMENT IS: ' WS-MAX-VALUE.
MOVE WS-MAX-VALUE TO RESULT.
GOBACK.
END PROGRAM MAX-ELEMENT.
<MID>
WORKING-STORAGE SECTION.
01 WS-MAX-VALUE PIC S9(10) VALUE ZERO.
```
## Training Details
Base model: CodeLlama
Finetuning type: LoRA
### Metrics
[COBOLEval](https://github.com/BloopAI/COBOLEval) is an adaptation of HumanEval where the problems are translated to COBOL.
| **Model** | CobolEval (pass@1) |
|----------------------|--------------------|
| **mAInframer-7b** | 6.16 |
| **mAInframer-13b** | 8.90 |
| **mAInframer-34b** | 10.27 |
## Citation
[Blog post]()
## Model Card Contact
[More Information Needed] |