Delete README.md
Browse files
README.md
DELETED
@@ -1,45 +0,0 @@
|
|
1 |
-
---
|
2 |
-
language:
|
3 |
-
- en
|
4 |
-
tags:
|
5 |
-
- pytorch
|
6 |
-
- causal-lm
|
7 |
-
license: apache-2.0
|
8 |
-
datasets:
|
9 |
-
- EleutherAI/pile
|
10 |
-
---
|
11 |
-
This is a d-Matrix functional reference of the [EleutherAI/gpt-j-6b](https://huggingface.co/EleutherAI/gpt-j-6b) model.
|
12 |
-
|
13 |
-
The reference provides the following functional *configurations*:
|
14 |
-
Configuration | Explanation
|
15 |
-
:-- | :--
|
16 |
-
**`BASELINE`** | a reference functionally equivalent to the original model
|
17 |
-
**`BASIC`** | all linear algebraic operands quantized to `BFP16-64`, and all other operations transformed to approximated kernel simulations
|
18 |
-
|
19 |
-
### Usage
|
20 |
-
|
21 |
-
Install d-Matrix [ML Tools](https://github.com/d-matrix-ai/dmx-mltools) first.
|
22 |
-
|
23 |
-
```sh
|
24 |
-
pip install dmx-mltools
|
25 |
-
```
|
26 |
-
|
27 |
-
The following is an example model and its evaluation.
|
28 |
-
|
29 |
-
```python
|
30 |
-
from mltools.dmx import pipeline
|
31 |
-
|
32 |
-
pipe = pipeline(
|
33 |
-
task="text-generation",
|
34 |
-
model="d-matrix/gpt-j-6b",
|
35 |
-
dmx_config="BASELINE", # see above for other variants
|
36 |
-
)
|
37 |
-
|
38 |
-
results = pipe.evaluate(
|
39 |
-
metric="d-matrix/dmx_perplexity",
|
40 |
-
dataset="wikitext",
|
41 |
-
dataset_version="wikitext-2-raw-v1",
|
42 |
-
)
|
43 |
-
```
|
44 |
-
|
45 |
-
### Evaluation results
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|