jiazhengli
commited on
Commit
•
19a512a
1
Parent(s):
b94494e
Update README.md
Browse files
README.md
CHANGED
@@ -9,31 +9,54 @@ base_model: meta-llama/Meta-Llama-3-8B
|
|
9 |
model-index:
|
10 |
- name: sft_trained_woaqa_llama3
|
11 |
results: []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
---
|
13 |
|
14 |
-
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
15 |
-
should probably proofread and complete it, then remove this comment. -->
|
16 |
|
17 |
-
#
|
18 |
|
19 |
-
|
20 |
-
It achieves the following results on the evaluation set:
|
21 |
-
- Loss: 0.8950
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
More information needed
|
26 |
|
27 |
## Intended uses & limitations
|
28 |
|
29 |
-
|
30 |
|
31 |
## Training and evaluation data
|
32 |
|
33 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
## Training procedure
|
36 |
|
|
|
|
|
37 |
### Training hyperparameters
|
38 |
|
39 |
The following hyperparameters were used during training:
|
@@ -59,7 +82,6 @@ The following hyperparameters were used during training:
|
|
59 |
| 0.8548 | 3.16 | 500 | 0.8958 |
|
60 |
| 0.8468 | 3.79 | 600 | 0.8952 |
|
61 |
|
62 |
-
|
63 |
### Framework versions
|
64 |
|
65 |
- PEFT 0.10.0
|
|
|
9 |
model-index:
|
10 |
- name: sft_trained_woaqa_llama3
|
11 |
results: []
|
12 |
+
datasets:
|
13 |
+
- jiazhengli/Synthetic_Rationale
|
14 |
+
- jiazhengli/Rationale_MCTS
|
15 |
+
language:
|
16 |
+
- en
|
17 |
+
metrics:
|
18 |
+
- accuracy
|
19 |
+
- f1
|
20 |
---
|
21 |
|
|
|
|
|
22 |
|
23 |
+
# Meta-Llama-3-8B-QLoRA-Assessment-Rationale-sft
|
24 |
|
25 |
+
The model trained with w/o private data from the EMNLP 2024 Paper: Calibrating LLMs with Preference Optimization on Thought Trees for Generating Rationale in Science Question Scoring.
|
|
|
|
|
26 |
|
27 |
+
- **Paper:** [Calibrating LLMs with Preference Optimization on Thought Trees for Generating Rationale in Science Question Scoring](https://arxiv.org/abs/2406.19949) (EMNLP 2024 Findings)
|
28 |
+
- **GitHub Repository:** [Thought Tree Assessment Repository](https://github.com/lijiazheng99/thought_tree_assessment)
|
|
|
29 |
|
30 |
## Intended uses & limitations
|
31 |
|
32 |
+
This model offers a valuable resource for research in explainable AI within educational technology. The model is trained with **noisy** response-level rationales. This makes them **unsuitable** for direct application in high-stakes assessments without additional verification.
|
33 |
|
34 |
## Training and evaluation data
|
35 |
|
36 |
+
We trained and evaluated the model on the [Synthetic Rationale data](https://huggingface.co/datasets/jiazhengli/Synthetic_Rationale), which was generated from the [Rationale MCTS data](https://huggingface.co/datasets/jiazhengli/Rationale_MCTS).
|
37 |
+
|
38 |
+
## Citation
|
39 |
+
|
40 |
+
Please cite the following work if you utilize this model:
|
41 |
+
|
42 |
+
**BibTeX:**
|
43 |
+
|
44 |
+
```bibtex
|
45 |
+
@misc{li2024calibratingllmspreferenceoptimization,
|
46 |
+
title={Calibrating LLMs with Preference Optimization on Thought Trees for Generating Rationale in Science Question Scoring},
|
47 |
+
author={Jiazheng Li and Hainiu Xu and Zhaoyue Sun and Yuxiang Zhou and David West and Cesare Aloisi and Yulan He},
|
48 |
+
year={2024},
|
49 |
+
eprint={2406.19949},
|
50 |
+
archivePrefix={arXiv},
|
51 |
+
primaryClass={cs.CL},
|
52 |
+
url={https://arxiv.org/abs/2406.19949},
|
53 |
+
}
|
54 |
+
```
|
55 |
|
56 |
## Training procedure
|
57 |
|
58 |
+
Please refer to our [paper](https://arxiv.org/abs/2406.19949).
|
59 |
+
|
60 |
### Training hyperparameters
|
61 |
|
62 |
The following hyperparameters were used during training:
|
|
|
82 |
| 0.8548 | 3.16 | 500 | 0.8958 |
|
83 |
| 0.8468 | 3.79 | 600 | 0.8952 |
|
84 |
|
|
|
85 |
### Framework versions
|
86 |
|
87 |
- PEFT 0.10.0
|