Update README.md
Browse files
README.md
CHANGED
|
@@ -32,13 +32,8 @@ Misal 7B, a pretrained and instruction tuned large language model based on Meta
|
|
| 32 |
|
| 33 |
Detailed blog [here](https://smallstep.ai/making-misal).
|
| 34 |
|
| 35 |
-
##
|
| 36 |
-
|
| 37 |
-
We did a manual round of evaluations using internet data (we have released the evaluation data here). This is a fairly small dataset with 100 questions taken from the internet. We understand that a better evaluation method is needed to benchmark our model, this being the first iteration we decided to proceed with manual evaluation.
|
| 38 |
-
|
| 39 |
-
Our main aim was to see if the model understands basic instructions, if so how well is it able to understand it, hence we have limited our evaluation to Reading comprehension, Translation, Sentiment Analysis, Paraphrasing like tasks.
|
| 40 |
-
|
| 41 |
-
[Manual Evaluation Set ](https://huggingface.co/datasets/smallstepai/Misal-Evaluation-v0.1)
|
| 42 |
|
| 43 |
| | Misal | ChatGPT3.5 | Krutrim | MahaMarathi |
|
| 44 |
| --------------------- | ----- | ---------- | ------- | ----------- |
|
|
@@ -48,10 +43,12 @@ Our main aim was to see if the model understands basic instructions, if so how w
|
|
| 48 |
| translation | 76 | 96 | 80 | 0 |
|
| 49 |
| average | 81 | 85 | 67 | 0 |
|
| 50 |
|
|
|
|
|
|
|
|
|
|
| 51 |
## Summary :
|
| 52 |
|
| 53 |
Our model beats ChatGPT 3.5 at reading comprehension.
|
| 54 |
-
|
| 55 |
While we are not able to beat ChatGPT 3.5 on remaining tasks like sentiment analysis, paraphrasing, translation, our model beats Ola Krutrim at all the tasks except translation.
|
| 56 |
|
| 57 |
|
|
|
|
| 32 |
|
| 33 |
Detailed blog [here](https://smallstep.ai/making-misal).
|
| 34 |
|
| 35 |
+
## Evaluation :
|
| 36 |
+
We did a manual round of evaluations using internet data. This is a fairly small dataset with 100 questions taken from the internet. We understand that a better evaluation method is needed to benchmark our model, this being the first iteration we decided to proceed with manual evaluation. Our main aim was to see if the model understands basic instructions, if so how well is it able to understand it, hence we have limited our evaluation to Reading comprehension, Translation, Sentiment Analysis, Paraphrasing like tasks.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
| | Misal | ChatGPT3.5 | Krutrim | MahaMarathi |
|
| 39 |
| --------------------- | ----- | ---------- | ------- | ----------- |
|
|
|
|
| 43 |
| translation | 76 | 96 | 80 | 0 |
|
| 44 |
| average | 81 | 85 | 67 | 0 |
|
| 45 |
|
| 46 |
+
We have released the evaluation data here:
|
| 47 |
+
- [Manual Evaluation Set](https://huggingface.co/datasets/smallstepai/Misal-Evaluation-v0.1)
|
| 48 |
+
|
| 49 |
## Summary :
|
| 50 |
|
| 51 |
Our model beats ChatGPT 3.5 at reading comprehension.
|
|
|
|
| 52 |
While we are not able to beat ChatGPT 3.5 on remaining tasks like sentiment analysis, paraphrasing, translation, our model beats Ola Krutrim at all the tasks except translation.
|
| 53 |
|
| 54 |
|