Update README.md
Browse files
README.md
CHANGED
@@ -56,6 +56,26 @@ This model was developed using English corpora, and thus may be considered Engli
|
|
56 |
|
57 |
Further, this model was developed in part using the [PMC-15M](https://aka.ms/biomedclip-paper) dataset. The figure-caption pairs that make up this dataset may contain biases reflecting the current practice of academic publication. For example, the corresponding papers may be enriched for positive findings, contain examples of extreme cases, and otherwise reflect distributions that are not representative of other sources of biomedical data.
|
58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
|
60 |
## Evaluation
|
61 |
|
@@ -138,29 +158,7 @@ torchrun --nnodes=1 --nproc_per_node=8 --master_port=25001 \
|
|
138 |
```
|
139 |
</details>
|
140 |
|
141 |
-
|
142 |
-
|
143 |
-
| Model Delta Weights | Size |
|
144 |
-
| --- | ---: |
|
145 |
-
| [llava_med_in_text_60k_delta.zip](https://hanoverprod.z21.web.core.windows.net/med_llava/models/llava_med_in_text_60k_delta.zip) | 11.06 GB |
|
146 |
-
|
147 |
-
The model weights above are *delta* weights. The usage of LLaVA-Med checkpoints should comply with the base LLM's model license: [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md).
|
148 |
-
|
149 |
-
Instructions:
|
150 |
-
|
151 |
-
1. Download the delta weights [llava_med_in_text_60k_delta.zip](https://hanoverprod.z21.web.core.windows.net/med_llava/models/llava_med_in_text_60k_delta.zip) and unzip.
|
152 |
-
1. Get the original LLaMA weights in the huggingface format by following the instructions [here](https://huggingface.co/docs/transformers/main/model_doc/llama).
|
153 |
-
1. Use the following scripts to get LLaVA-Med weights by applying our delta. In the script below, set the --delta argument to the path of the unzipped `llava_med_in_text_60k_delta` directory. It can be adapted for other delta weights by changing the `--delta` argument (and base/target accordingly).
|
154 |
-
|
155 |
-
```bash
|
156 |
-
python3 -m llava.model.apply_delta \
|
157 |
-
--base /path/to/llama-7b \
|
158 |
-
--target /output/path/to/llava_med_in_text_60k \
|
159 |
-
--delta path/to/llava_med_in_text_60k_delta
|
160 |
-
```
|
161 |
-
|
162 |
-
|
163 |
-
## - Evaluation
|
164 |
|
165 |
Depending on which checkpoint is employed in evaluation, zero-shot performance is reported on medical instruct tuned checkpoint (eg, [LLaVA-Med-7B](/path/to/checkpoint_llava_med_instruct_60k_inline_mention)), and fine-tuned performance is reported on checkpoint that has been further tuned on training set of the downstream datasets (eg, [LLaVA-Med-7B-VQA-Rad](/path/to/checkpoint_llava_med_instruct_60k_inline_mention/fine_tuned/vqa_rad) ).
|
166 |
|
|
|
56 |
|
57 |
Further, this model was developed in part using the [PMC-15M](https://aka.ms/biomedclip-paper) dataset. The figure-caption pairs that make up this dataset may contain biases reflecting the current practice of academic publication. For example, the corresponding papers may be enriched for positive findings, contain examples of extreme cases, and otherwise reflect distributions that are not representative of other sources of biomedical data.
|
58 |
|
59 |
+
## Serving
|
60 |
+
|
61 |
+
| Model Delta Weights | Size |
|
62 |
+
| --- | ---: |
|
63 |
+
| [llava_med_in_text_60k_delta.zip](https://hanoverprod.z21.web.core.windows.net/med_llava/models/llava_med_in_text_60k_delta.zip) | 11.06 GB |
|
64 |
+
|
65 |
+
The model weights above are *delta* weights. The usage of LLaVA-Med checkpoints should comply with the base LLM's model license: [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md).
|
66 |
+
|
67 |
+
Instructions:
|
68 |
+
|
69 |
+
1. Download the delta weights [llava_med_in_text_60k_delta.zip](https://hanoverprod.z21.web.core.windows.net/med_llava/models/llava_med_in_text_60k_delta.zip) and unzip.
|
70 |
+
1. Get the original LLaMA weights in the huggingface format by following the instructions [here](https://huggingface.co/docs/transformers/main/model_doc/llama).
|
71 |
+
1. Use the following scripts to get LLaVA-Med weights by applying our delta. In the script below, set the --delta argument to the path of the unzipped `llava_med_in_text_60k_delta` directory. It can be adapted for other delta weights by changing the `--delta` argument (and base/target accordingly).
|
72 |
+
|
73 |
+
```bash
|
74 |
+
python3 -m llava.model.apply_delta \
|
75 |
+
--base /path/to/llama-7b \
|
76 |
+
--target /output/path/to/llava_med_in_text_60k \
|
77 |
+
--delta path/to/llava_med_in_text_60k_delta
|
78 |
+
```
|
79 |
|
80 |
## Evaluation
|
81 |
|
|
|
158 |
```
|
159 |
</details>
|
160 |
|
161 |
+
#### - Evaluation
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
162 |
|
163 |
Depending on which checkpoint is employed in evaluation, zero-shot performance is reported on medical instruct tuned checkpoint (eg, [LLaVA-Med-7B](/path/to/checkpoint_llava_med_instruct_60k_inline_mention)), and fine-tuned performance is reported on checkpoint that has been further tuned on training set of the downstream datasets (eg, [LLaVA-Med-7B-VQA-Rad](/path/to/checkpoint_llava_med_instruct_60k_inline_mention/fine_tuned/vqa_rad) ).
|
164 |
|