Commit
•
9c9d41d
1
Parent(s):
89eacba
Update README.md
Browse files
README.md
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
---
|
2 |
-
license: mit
|
3 |
language:
|
4 |
- en
|
|
|
5 |
---
|
6 |
|
7 |
# TAT-LLM: A Specialized Language Model for Discrete Reasoning over Tabular and Textual Data
|
@@ -19,17 +19,16 @@ We present TAT-LLM, a specialized language model crafted through the innovative
|
|
19 |
| --- | --- | --- | --- | --- |
|
20 |
| GPT-3.5-Turbo | - | 58.00 | 59.47 | 52.74 |
|
21 |
| GPT-4 | - | 63.91 | 71.92 | 64.46 |
|
22 |
-
| TAT-LLM-7B-LORA | 7B | 65.13 | 76.49 | 71.38 |
|
23 |
-
| TAT-LLM-7B-FFT | 7B | 69.75 | 76.91 | 72.64 |
|
24 |
-
| TAT-LLM-13B-LORA | 13B | 71.93 | 77.51 | 72.22 |
|
25 |
-
| TAT-LLM-13B-FFT | 13B | 72.97 | 78.41 | 73.18 |
|
26 |
-
| TAT-LLM-70B-LORA | 70B | **76.81** | 81.42 | 76.55 |
|
27 |
-
| TAT-LLM-70B-FFT | 70B | 76.11 | **82.20** | **76.97** |
|
28 |
-
|
29 |
|
30 |
## Training
|
31 |
|
32 |
-
We train our TAT-LLM model in various sizes, including 7B, 13B, and 70B, using different methods such as parameter-efficient fine-tuning and full-parameter fine-tuning of LLaMA 2 on a combination of financial data from the FinQA, TAT-QA, and TAT-DQA datasets. To refine accuracy, we introduce an External Executor, enhancing the model by processing intermediate outputs to derive conclusive answers. Please refer to the [paper](https://arxiv.org/abs/2401.13223) for more details.
|
33 |
|
34 |
## Inference & Evaluation
|
35 |
|
@@ -48,4 +47,4 @@ If you find this model helpful, please consider citing our paper:
|
|
48 |
archivePrefix={arXiv},
|
49 |
primaryClass={cs.CL}
|
50 |
}
|
51 |
-
```
|
|
|
1 |
---
|
|
|
2 |
language:
|
3 |
- en
|
4 |
+
license: llama2
|
5 |
---
|
6 |
|
7 |
# TAT-LLM: A Specialized Language Model for Discrete Reasoning over Tabular and Textual Data
|
|
|
19 |
| --- | --- | --- | --- | --- |
|
20 |
| GPT-3.5-Turbo | - | 58.00 | 59.47 | 52.74 |
|
21 |
| GPT-4 | - | 63.91 | 71.92 | 64.46 |
|
22 |
+
| [TAT-LLM-7B-LORA](https://huggingface.co/next-tat/tat-llm-7b-lora) | 7B | 65.13 | 76.49 | 71.38 |
|
23 |
+
| [TAT-LLM-7B-FFT](https://huggingface.co/next-tat/tat-llm-7b-fft) | 7B | 69.75 | 76.91 | 72.64 |
|
24 |
+
| [TAT-LLM-13B-LORA](https://huggingface.co/next-tat/tat-llm-13b-lora) | 13B | 71.93 | 77.51 | 72.22 |
|
25 |
+
| [TAT-LLM-13B-FFT](https://huggingface.co/next-tat/tat-llm-13b-fft) | 13B | 72.97 | 78.41 | 73.18 |
|
26 |
+
| [TAT-LLM-70B-LORA](https://huggingface.co/next-tat/tat-llm-70b-lora) | 70B | **76.81** | 81.42 | 76.55 |
|
27 |
+
| [TAT-LLM-70B-FFT](https://huggingface.co/next-tat/tat-llm-70b-fft) | 70B | 76.11 | **82.20** | **76.97** |
|
|
|
28 |
|
29 |
## Training
|
30 |
|
31 |
+
We train our TAT-LLM model in various sizes, including 7B, 13B, and 70B, using different methods such as parameter-efficient fine-tuning and full-parameter fine-tuning of LLaMA 2 on a combination of financial data from the FinQA, TAT-QA, and TAT-DQA datasets([🤗HuggingFace Repo](https://huggingface.co/datasets/next-tat/tat-llm-instructions)). To refine accuracy, we introduce an External Executor, enhancing the model by processing intermediate outputs to derive conclusive answers. Please refer to the [paper](https://arxiv.org/abs/2401.13223) for more details.
|
32 |
|
33 |
## Inference & Evaluation
|
34 |
|
|
|
47 |
archivePrefix={arXiv},
|
48 |
primaryClass={cs.CL}
|
49 |
}
|
50 |
+
```
|