Update README.md
Browse files
README.md
CHANGED
|
@@ -40,7 +40,7 @@ widget:
|
|
| 40 |
---
|
| 41 |
# Model Card for MedGENIE-fid-flan-t5-base-medqa
|
| 42 |
|
| 43 |
-
MedGENIE comprises a collection of language models designed to utilize generated contexts, rather than retrieved ones, for addressing multiple-choice open-domain questions in the medical domain. Specifically, MedGENIE-fid-flan-t5-base-medqa is a fusion-in-decoder model based on flan-t5-base architecture, trained on the [MedQA-USMLE](https://huggingface.co/datasets/disi-unibo-nlp/medqa-5-opt-MedGENIE) dataset augmented with artificially generated contexts from PMC-LLaMA-13B. This model achieves a new state-of-the-art performance over the corresponding test set.
|
| 44 |
|
| 45 |
## Model description
|
| 46 |
|
|
@@ -56,47 +56,47 @@ At the time of release, MedGENIE-fid-flan-t5-base-medqa is a new lightweight SOT
|
|
| 56 |
| Model | Ground (Source) | Learning | Params | Accuracy (↓) |
|
| 57 |
|----------------------------------|--------------------|---------------------------|-----------------|-------------------------------|
|
| 58 |
| **MedGENIE-FID-Flan-T5** | G (PMC-LLaMA) | Fine-tuned | 250M | **53.1** |
|
| 59 |
-
| Codex <small>(Liévin et al.
|
| 60 |
-
| Codex <small>(Liévin et al.
|
| 61 |
-
| GPT-3.5-Turbo <small>(Yang et al.)</small> | R (Wikipedia) | k-shot | -- | 52.3 |
|
| 62 |
-
| MEDITRON <small>(Chen et al.)</small> | ∅ | Fine-tuned | 7B | 52.0 |
|
| 63 |
| Zephyr-β | R (MedWiki) | 2-shot | 7B | 50.4 |
|
| 64 |
-
| BioMedGPT <small>(Luo et al.)</small> | ∅ | k-shot | 10B | 50.4 |
|
| 65 |
-
| BioMedLM <small>(Singhal et al.)</small> | ∅ | Fine-tuned | 2.7B | 50.3 |
|
| 66 |
| PMC-LLaMA (AWQ) | ∅ | Fine-tuned | 13B | 50.2 |
|
| 67 |
-
| LLaMA-2 <small>(Chen et al.)</small> | ∅ | Fine-tuned | 7B | 49.6 |
|
| 68 |
| Zephyr-β | ∅ | 2-shot | 7B | 49.6 |
|
| 69 |
-
| Zephyr-β <small>(Chen et al.)</small> | ∅ | 3-shot | 7B | 49.2 |
|
| 70 |
-
| PMC-LLaMA <small>(Chen et al.)</small> | ∅ | Fine-tuned | 7B | 49.2 |
|
| 71 |
-
| DRAGON <small>(Yasunaga et al.)</small> | R (UMLS) | Fine-tuned | 360M | 47.5 |
|
| 72 |
-
| InstructGPT <small>(Liévin et al.)</small> | R (Wikipedia) | 0-shot | 175B | 47.3 |
|
| 73 |
-
| Flan-PaLM <small>(Singhal et al.)</small> | ∅ | 5-shot | 62B | 46.1 |
|
| 74 |
-
| InstructGPT <small>(Liévin et al.)</small> | ∅ | 0-shot | 175B | 46.0 |
|
| 75 |
-
| VOD <small>(Liévin et al. 2023)</small> | R (MedWiki) | Fine-tuned | 220M | 45.8 |
|
| 76 |
-
| Vicuna 1.3 <small>(Liévin et al.)</small> | ∅ | 0-shot | 33B | 45.2 |
|
| 77 |
-
| BioLinkBERT <small>(Singhal et al.)</small> | ∅ | Fine-tuned | 340M | 45.1 |
|
| 78 |
| Mistral-Instruct | R (MedWiki) | 2-shot | 7B | 45.1 |
|
| 79 |
| Galactica | ∅ | 0-shot | 120B | 44.4 |
|
| 80 |
-
| LLaMA-2 <small>(Liévin et al.)</small> | ∅ | 0-shot | 70B | 43.4 |
|
| 81 |
-
| BioReader <small>(
|
| 82 |
-
| Guanaco <small>(Liévin et al.)</small> | ∅ | 0-shot | 33B | 42.9 |
|
| 83 |
-
| LLaMA-2-chat <small>(Liévin et al.)</small> | ∅ | 0-shot | 70B | 42.3 |
|
| 84 |
-
| Vicuna 1.5 <small>(Liévin et al.)</small> | ∅ | 0-shot | 65B | 41.6 |
|
| 85 |
-
| Mistral-Instruct <small>(Chen et al.)</small> | ∅ | 3-shot | 7B | 41.1 |
|
| 86 |
-
| PaLM <small>(Singhal et al.)</small> | ∅ | 5-shot | 62B | 40.9 |
|
| 87 |
-
| Guanaco <small>(Liévin et al.)</small> | ∅ | 0-shot | 65B | 40.8 |
|
| 88 |
-
| Falcon-Instruct <small>(Liévin et al.)</small> | ∅ | 0-shot | 40B | 39.0 |
|
| 89 |
-
| Vicuna 1.3 <small>(Liévin et al.)</small> | ∅ | 0-shot | 13B | 38.7 |
|
| 90 |
| GreaseLM <small>(Zhang et al.)</small> | R (UMLS) | Fine-tuned | 359M | 38.5 |
|
| 91 |
-
| PubMedBERT <small>(Singhal et al.)</small> | ∅ | Fine-tuned | 110M | 38.1 |
|
| 92 |
| QA-GNN <small>(Yasunaga et al.)</small> | R (UMLS) | Fine-tuned | 360M | 38.0 |
|
| 93 |
-
| LLaMA-2 <small>(Yang et al.)</small> | R (Wikipedia) | k-shot | 13B | 37.6 |
|
| 94 |
| LLaMA-2-chat | R (MedWiki) | 2-shot | 7B | 37.2 |
|
| 95 |
| LLaMA-2-chat | ∅ | 2-shot | 7B | 37.2 |
|
| 96 |
| BioBERT <small>(Lee et al.)</small> | ∅ | Fine-tuned | 110M | 36.7 |
|
| 97 |
-
| MTP-Instruct <small>(Liévin et al.)</small> | ∅ | 0-shot | 30B | 35.1 |
|
| 98 |
-
| GPT-Neo <small>(Singhal et al.)</small> | ∅ | Fine-tuned | 2.5B | 33.3 |
|
| 99 |
-
| LLaMa-2-chat <small>(Liévin et al.)</small> | ∅ | 0-shot | 13B | 32.2 |
|
| 100 |
-
| LLaMa-2 <small>(Liévin et al.)</small> | ∅ | 0-shot | 13B | 31.1 |
|
| 101 |
-
| GPT-NeoX <small>(Liévin et al.)
|
| 102 |
|
|
|
|
| 40 |
---
|
| 41 |
# Model Card for MedGENIE-fid-flan-t5-base-medqa
|
| 42 |
|
| 43 |
+
MedGENIE comprises a collection of language models designed to utilize generated contexts, rather than retrieved ones, for addressing multiple-choice open-domain questions in the medical domain. Specifically, MedGENIE-fid-flan-t5-base-medqa is a fusion-in-decoder model based on flan-t5-base architecture, trained on the [MedQA-USMLE](https://huggingface.co/datasets/disi-unibo-nlp/medqa-5-opt-MedGENIE) dataset augmented with artificially generated contexts from PMC-LLaMA-13B. This model achieves a new state-of-the-art (SOTA) performance over the corresponding test set.
|
| 44 |
|
| 45 |
## Model description
|
| 46 |
|
|
|
|
| 56 |
| Model | Ground (Source) | Learning | Params | Accuracy (↓) |
|
| 57 |
|----------------------------------|--------------------|---------------------------|-----------------|-------------------------------|
|
| 58 |
| **MedGENIE-FID-Flan-T5** | G (PMC-LLaMA) | Fine-tuned | 250M | **53.1** |
|
| 59 |
+
| Codex <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-zhot | 175B | 52.5 |
|
| 60 |
+
| Codex <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | R (Wikipedia) | 0-shot | 175B | 52.5 |
|
| 61 |
+
| GPT-3.5-Turbo <small>([Yang et al.](https://arxiv.org/abs/2309.02233))</small> | R (Wikipedia) | k-shot | -- | 52.3 |
|
| 62 |
+
| MEDITRON <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small> | ∅ | Fine-tuned | 7B | 52.0 |
|
| 63 |
| Zephyr-β | R (MedWiki) | 2-shot | 7B | 50.4 |
|
| 64 |
+
| BioMedGPT <small>([Luo et al.](https://arxiv.org/abs/2308.09442v2))</small> | ∅ | k-shot | 10B | 50.4 |
|
| 65 |
+
| BioMedLM <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small> | ∅ | Fine-tuned | 2.7B | 50.3 |
|
| 66 |
| PMC-LLaMA (AWQ) | ∅ | Fine-tuned | 13B | 50.2 |
|
| 67 |
+
| LLaMA-2 <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small> | ∅ | Fine-tuned | 7B | 49.6 |
|
| 68 |
| Zephyr-β | ∅ | 2-shot | 7B | 49.6 |
|
| 69 |
+
| Zephyr-β <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small> | ∅ | 3-shot | 7B | 49.2 |
|
| 70 |
+
| PMC-LLaMA <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small> | ∅ | Fine-tuned | 7B | 49.2 |
|
| 71 |
+
| DRAGON <small>([Yasunaga et al.](https://arxiv.org/abs/2210.09338))</small> | R (UMLS) | Fine-tuned | 360M | 47.5 |
|
| 72 |
+
| InstructGPT <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | R (Wikipedia) | 0-shot | 175B | 47.3 |
|
| 73 |
+
| Flan-PaLM <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small> | ∅ | 5-shot | 62B | 46.1 |
|
| 74 |
+
| InstructGPT <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 175B | 46.0 |
|
| 75 |
+
| VOD <small>([Liévin et al. 2023](https://arxiv.org/abs/2210.06345))</small> | R (MedWiki) | Fine-tuned | 220M | 45.8 |
|
| 76 |
+
| Vicuna 1.3 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 33B | 45.2 |
|
| 77 |
+
| BioLinkBERT <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small> | ∅ | Fine-tuned | 340M | 45.1 |
|
| 78 |
| Mistral-Instruct | R (MedWiki) | 2-shot | 7B | 45.1 |
|
| 79 |
| Galactica | ∅ | 0-shot | 120B | 44.4 |
|
| 80 |
+
| LLaMA-2 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 70B | 43.4 |
|
| 81 |
+
| BioReader <small>(Frisoni et al.)</small> | R (PubMed-RCT) | Fine-tuned | 230M | 43.0 |
|
| 82 |
+
| Guanaco <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 33B | 42.9 |
|
| 83 |
+
| LLaMA-2-chat <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 70B | 42.3 |
|
| 84 |
+
| Vicuna 1.5 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 65B | 41.6 |
|
| 85 |
+
| Mistral-Instruct <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small> | ∅ | 3-shot | 7B | 41.1 |
|
| 86 |
+
| PaLM <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small> | ∅ | 5-shot | 62B | 40.9 |
|
| 87 |
+
| Guanaco <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 65B | 40.8 |
|
| 88 |
+
| Falcon-Instruct <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 40B | 39.0 |
|
| 89 |
+
| Vicuna 1.3 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 13B | 38.7 |
|
| 90 |
| GreaseLM <small>(Zhang et al.)</small> | R (UMLS) | Fine-tuned | 359M | 38.5 |
|
| 91 |
+
| PubMedBERT <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small> | ∅ | Fine-tuned | 110M | 38.1 |
|
| 92 |
| QA-GNN <small>(Yasunaga et al.)</small> | R (UMLS) | Fine-tuned | 360M | 38.0 |
|
| 93 |
+
| LLaMA-2 <small>([Yang et al.](https://arxiv.org/abs/2309.02233))</small> | R (Wikipedia) | k-shot | 13B | 37.6 |
|
| 94 |
| LLaMA-2-chat | R (MedWiki) | 2-shot | 7B | 37.2 |
|
| 95 |
| LLaMA-2-chat | ∅ | 2-shot | 7B | 37.2 |
|
| 96 |
| BioBERT <small>(Lee et al.)</small> | ∅ | Fine-tuned | 110M | 36.7 |
|
| 97 |
+
| MTP-Instruct <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 30B | 35.1 |
|
| 98 |
+
| GPT-Neo <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small> | ∅ | Fine-tuned | 2.5B | 33.3 |
|
| 99 |
+
| LLaMa-2-chat <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 13B | 32.2 |
|
| 100 |
+
| LLaMa-2 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 13B | 31.1 |
|
| 101 |
+
| GPT-NeoX <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small> | ∅ | 0-shot | 20B | 26.9 |
|
| 102 |
|