pkanithi commited on
Commit
950f771
1 Parent(s): 815eb42

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +164 -154
README.md CHANGED
@@ -1,201 +1,211 @@
1
  ---
2
- library_name: transformers
3
- tags: []
 
 
 
 
 
 
 
 
 
4
  ---
 
 
5
 
6
- # Model Card for Model ID
7
 
8
- <!-- Provide a quick summary of what the model is/does. -->
 
 
9
 
 
 
 
 
 
 
 
 
 
 
10
 
11
 
12
- ## Model Details
13
-
14
- ### Model Description
15
-
16
- <!-- Provide a longer summary of what this model is. -->
17
-
18
- This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
19
-
20
- - **Developed by:** [More Information Needed]
21
- - **Funded by [optional]:** [More Information Needed]
22
- - **Shared by [optional]:** [More Information Needed]
23
- - **Model type:** [More Information Needed]
24
- - **Language(s) (NLP):** [More Information Needed]
25
- - **License:** [More Information Needed]
26
- - **Finetuned from model [optional]:** [More Information Needed]
27
-
28
- ### Model Sources [optional]
29
-
30
- <!-- Provide the basic links for the model. -->
31
-
32
- - **Repository:** [More Information Needed]
33
- - **Paper [optional]:** [More Information Needed]
34
- - **Demo [optional]:** [More Information Needed]
35
-
36
- ## Uses
37
-
38
- <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
-
40
- ### Direct Use
41
-
42
- <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
-
44
- [More Information Needed]
45
-
46
- ### Downstream Use [optional]
47
-
48
- <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
-
50
- [More Information Needed]
51
-
52
- ### Out-of-Scope Use
53
-
54
- <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
-
56
- [More Information Needed]
57
-
58
- ## Bias, Risks, and Limitations
59
-
60
- <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
-
62
- [More Information Needed]
63
-
64
- ### Recommendations
65
-
66
- <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
-
68
- Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
-
70
- ## How to Get Started with the Model
71
-
72
- Use the code below to get started with the model.
73
 
74
- [More Information Needed]
 
 
75
 
76
- ## Training Details
77
 
78
- ### Training Data
79
-
80
- <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
-
82
- [More Information Needed]
83
-
84
- ### Training Procedure
85
-
86
- <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
-
88
- #### Preprocessing [optional]
89
-
90
- [More Information Needed]
91
-
92
-
93
- #### Training Hyperparameters
94
-
95
- - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
-
97
- #### Speeds, Sizes, Times [optional]
98
-
99
- <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
-
101
- [More Information Needed]
102
-
103
- ## Evaluation
104
-
105
- <!-- This section describes the evaluation protocols and provides the results. -->
106
-
107
- ### Testing Data, Factors & Metrics
108
-
109
- #### Testing Data
110
-
111
- <!-- This should link to a Dataset Card if possible. -->
112
-
113
- [More Information Needed]
114
-
115
- #### Factors
116
-
117
- <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
-
119
- [More Information Needed]
120
-
121
- #### Metrics
122
-
123
- <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
-
125
- [More Information Needed]
126
-
127
- ### Results
128
 
129
- [More Information Needed]
130
 
131
- #### Summary
132
 
 
133
 
 
134
 
135
- ## Model Examination [optional]
136
 
137
- <!-- Relevant interpretability work for the model goes here -->
138
 
139
- [More Information Needed]
140
 
141
- ## Environmental Impact
142
 
143
- <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
 
145
- Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
 
147
- - **Hardware Type:** [More Information Needed]
148
- - **Hours used:** [More Information Needed]
149
- - **Cloud Provider:** [More Information Needed]
150
- - **Compute Region:** [More Information Needed]
151
- - **Carbon Emitted:** [More Information Needed]
 
152
 
153
- ## Technical Specifications [optional]
154
 
155
- ### Model Architecture and Objective
156
 
157
- [More Information Needed]
 
 
158
 
159
- ### Compute Infrastructure
160
 
161
- [More Information Needed]
 
 
 
 
 
162
 
163
- #### Hardware
 
 
 
 
 
 
 
 
 
 
 
 
164
 
165
- [More Information Needed]
 
 
166
 
167
- #### Software
 
 
 
168
 
169
- [More Information Needed]
 
 
 
 
 
 
 
 
170
 
171
- ## Citation [optional]
 
172
 
173
- <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
 
175
- **BibTeX:**
176
 
177
- [More Information Needed]
178
 
179
- **APA:**
180
 
181
- [More Information Needed]
182
 
183
- ## Glossary [optional]
184
 
185
- <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
 
187
- [More Information Needed]
188
 
189
- ## More Information [optional]
 
 
 
 
 
 
 
 
190
 
191
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
192
 
193
- ## Model Card Authors [optional]
 
 
 
 
 
 
 
194
 
195
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
196
 
197
- ## Model Card Contact
 
 
 
198
 
199
- [More Information Needed]
200
 
 
201
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - en
4
+ license: llama3
5
+ tags:
6
+ - m42
7
+ - health
8
+ - healthcare
9
+ - clinical-llm
10
+ pipeline_tag: text-generation
11
+ inference: false
12
+ license_name: llama3
13
  ---
14
+ # **Med42-v2 - A Suite of Clinically-aligned Large Language Models**
15
+ Med42-v2 is a suite of open-access clinical large language models (LLM) instruct and preference-tuned by M42 to expand access to medical knowledge. Built off LLaMA-3 and comprising either 8 or 70 billion parameters, these generative AI system provide high-quality answers to medical questions.
16
 
17
+ ## Key performance metrics:
18
 
19
+ - Med42-v2-70B outperforms GPT-4.0 in most of the MCQA tasks.
20
+ - Med42-v2-70B achieves a MedQA zero-shot performance of 79.10, surpassing the prior state-of-the-art among all openly available medical LLMs.
21
+ - Med42-v2-70B sits at the top of the Clinical Elo Rating Leaderboard.
22
 
23
+ |Models|Elo Score|
24
+ |:---:|:---:|
25
+ |Med42-v2-70B| 1764 |
26
+ |Llama3-70B-Instruct| 1643 |
27
+ |GPT4-o| 1426 |
28
+ |Llama3-8B-Instruct| 1352 |
29
+ |Mixtral-8x7b-Instruct| 970 |
30
+ |Med42-v2-8B| 924 |
31
+ |OpenBioLLM-70B| 657 |
32
+ |JSL-MedLlama-3-8B-v2.0| 447 |
33
 
34
 
35
+ ## Limitations & Safe Use
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
 
37
+ - Med42-v2 suite of models are not ready for real clinical use. Extensive human evaluation is undergoing as it is required to ensure safety.
38
+ - Potential for generating incorrect or harmful information.
39
+ - Risk of perpetuating biases in training data.
40
 
41
+ Use this suite of models responsibly! Do not rely on them for medical usage without rigorous safety testing.
42
 
43
+ ## Model Details
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44
 
45
+ *Disclaimer: This large language model is not yet ready for clinical use without further testing and validation. It should not be relied upon for making medical decisions or providing patient care.*
46
 
47
+ Beginning with Llama3 models, Med42-v2 were instruction-tuned using a dataset of ~1B tokens compiled from different open-access and high-quality sources, including medical flashcards, exam questions, and open-domain dialogues.
48
 
49
+ **Model Developers:** M42 Health AI Team
50
 
51
+ **Finetuned from model:** Llama3 - 8B & 70B Instruct
52
 
53
+ **Context length:** 8k tokens
54
 
55
+ **Input:** Text only data
56
 
57
+ **Output:** Model generates text only
58
 
59
+ **Status:** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we enhance model's performance.
60
 
61
+ **License:** Llama 3 Community License Agreement
62
 
63
+ **Research Paper:** *Comming soon*
64
 
65
+ ## Intended Use
66
+ Med42-v2 suite of models are being made available for further testing and assessment as AI assistants to enhance clinical decision-making and enhance access to LLMs for healthcare use. Potential use cases include:
67
+ - Medical question answering
68
+ - Patient record summarization
69
+ - Aiding medical diagnosis
70
+ - General health Q&A
71
 
72
+ **Run the model**
73
 
74
+ You can use the 🤗 Transformers library `text-generation` pipeline to do inference.
75
 
76
+ ```python
77
+ import transformers
78
+ import torch
79
 
80
+ model_name_or_path = "m42-health/Llama3-Med42-8B"
81
 
82
+ pipeline = transformers.pipeline(
83
+ "text-generation",
84
+ model=model_name_or_path,
85
+ torch_dtype=torch.bfloat16,
86
+ device_map="auto",
87
+ )
88
 
89
+ messages = [
90
+ {
91
+ "role": "system",
92
+ "content": (
93
+ "You are a helpful, respectful and honest medical assistant. You are a second version of Med42 developed by the AI team at M42, UAE. "
94
+ "Always answer as helpfully as possible, while being safe. "
95
+ "Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. "
96
+ "Please ensure that your responses are socially unbiased and positive in nature. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. "
97
+ "If you don’t know the answer to a question, please don’t share false information."
98
+ ),
99
+ },
100
+ {"role": "user", "content": "What are the symptoms of diabetes?"},
101
+ ]
102
 
103
+ prompt = pipeline.tokenizer.apply_chat_template(
104
+ messages, tokenize=False, add_generation_prompt=False
105
+ )
106
 
107
+ stop_tokens = [
108
+ pipeline.tokenizer.eos_token_id,
109
+ pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>"),
110
+ ]
111
 
112
+ outputs = pipeline(
113
+ prompt,
114
+ max_new_tokens=512,
115
+ eos_token_id=stop_tokens,
116
+ do_sample=True,
117
+ temperature=0.4,
118
+ top_k=150,
119
+ top_p=0.75,
120
+ )
121
 
122
+ print(outputs[0]["generated_text"][len(prompt) :])
123
+ ```
124
 
125
+ ## Hardware and Software
126
 
127
+ The training was conducted on the NVIDIA DGX cluster with H100 GPUs, utilizing PyTorch's Fully Sharded Data Parallel (FSDP) framework.
128
 
 
129
 
130
+ ## Evaluation Results
131
 
132
+ ### Open-ended question generation
133
 
134
+ To ensure a robust evaluation of our model's output quality, we employ the LLM-as-a-Judge approach using Prometheus-8x7b-v2.0. Our assessment uses carefully curated 4,000 publicly accessible healthcare-related questions, generating responses from various models. We then use Prometheus to conduct pairwise comparisons of the answers. Drawing inspiration from the LMSYS Chatbot-Arena methodology, we present the results as Elo ratings for each model.
135
 
136
+ To maintain fairness and eliminate potential bias from prompt engineering, we used the same simple system prompt for every model throughout the evaluation process.
137
 
138
+ Below is the scoring rubric we used to prompt Prometheus to select the best answer:
139
 
140
+ ```
141
+ ### Score Rubric:
142
+ Which response is of higher overall quality in a medical context? Consider:
143
+ * Relevance: Does it directly address the question?
144
+ * Completeness: Does it cover all important aspects, details and subpoints?
145
+ * Safety: Does it avoid unsafe practices and address potential risks?
146
+ * Ethics: Does it maintain confidentiality and avoid biases?
147
+ * Clarity: Is it professional, clear and easy to understand?
148
+ ```
149
 
150
+ #### Elo Ratings
151
+ |Models|Elo Score|
152
+ |:---:|:---:|
153
+ |Med42-v2-70B| 1764 |
154
+ |Llama3-70B-Instruct| 1643 |
155
+ |GPT4-o| 1426 |
156
+ |Llama3-8B-Instruct| 1352 |
157
+ |Mixtral-8x7b-Instruct| 970 |
158
+ |Med42-v2-8B| 924 |
159
+ |OpenBioLLM-70B| 657 |
160
+ |JSL-MedLlama-3-8B-v2.0| 447 |
161
 
162
+ #### Win-rate
163
+
164
+ ![plot](./pairwise_model_comparison.svg)
165
+
166
+
167
+ ### MCQA Evaluation
168
+
169
+ Med42-v2 improves performance on every clinical benchmark compared to our previous version, including MedQA, MedMCQA, USMLE, MMLU clinical topics and MMLU Pro clinical subset. For all evaluations reported so far, we use [EleutherAI's evaluation harness library](https://github.com/EleutherAI/lm-evaluation-harness) and report zero-shot accuracies (except otherwise stated). We integrated chat templates into harness and computed the likelihood for the full answer instead of only the tokens "a.", "b.", "c." or "d.".
170
 
171
+ |Model|MMLU Pro|MMLU|MedMCQA|MedQA|USMLE|
172
+ |---:|:---:|:---:|:---:|:---:|:---:|
173
+ |Med42v2-70B|64.36|87.12|73.20|79.10|83.80|
174
+ |Med42v2-8B|54.30|75.76|61.34|62.84|67.04|
175
+ |OpenBioLLM|64.24|90.40|73.18|76.90|79.01|
176
+ |GPT-4.0<sup>&dagger;</sup>|-|87.00|69.50|78.90|84.05|
177
+ |MedGemini*|-|-|-|84.00|-|
178
+ |Med-PaLM-2(5-shot)*|-|87.77|71.30|79.70|-|
179
+ |Med42|-|76.72|60.90|61.50|71.85|
180
+ |ClinicalCamel-70B|-|69.75|47.00|53.40|54.30|
181
+ |GPT-3.5<sup>&dagger;</sup>|-|66.63|50.10|50.80|53.00|
182
+ |Llama3-8B-Instruct|48.24|72.89|59.65|61.64|60.38|
183
+ |Llama3-70B-Instruct|64.24|85.99|72.03|78.88|83.57|
184
+
185
+ **For MedGemini, results are reported for MedQA without self-training and without search. We note that 0-shot performance is not reported for Med-PaLM 2. Further details can be found at [https://github.com/m42health/med42](https://github.com/m42health/med42)*.
186
+
187
+ <sup>&dagger;</sup> *Results as reported in the paper [Capabilities of GPT-4 on Medical Challenge Problems](https://www.microsoft.com/en-us/research/uploads/prod/2023/03/GPT-4_medical_benchmarks.pdf)*.
188
+
189
+
190
+ ## Accessing Med42 and Reporting Issues
191
+
192
+ Please report any software "bug" or other problems through one of the following means:
193
 
194
+ - Reporting issues with the model: [https://github.com/m42health/med42](https://github.com/m42health/med42)
195
+ - Reporting risky content generated by the model, bugs and/or any security concerns: [https://forms.office.com/r/fPY4Ksecgf](https://forms.office.com/r/fPY4Ksecgf)
196
+ - M42’s privacy policy available at [https://m42.ae/privacy-policy/](https://m42.ae/privacy-policy/)
197
+ - Reporting violations of the Acceptable Use Policy or unlicensed uses of Med42: <med42@m42.ae>
198
 
199
+ ## Acknowledgements
200
 
201
+ We thank the Torch FSDP team for their robust distributed training framework, the EleutherAI harness team for their valuable evaluation tools, and the Hugging Face Alignment team for their contributions to responsible AI development.
202
 
203
+ ## Citation
204
+ ```
205
+ @article{christophe2024med42,
206
+ title={Med42-v2 - A Suite of Clinically-aligned Large Language Models},
207
+ author={Christophe, Cl{\'e}ment and Raha, Tathagata and Hayat, Nasir and Kanithi, Praveen and Al-Mahrooqi, Ahmed and Munjal, Prateek and Saadi, Nada and Javed, Hamza and Salman, Umar and Pimentel, Marco and Rajan, Ronnie and Khan, Shadab},
208
+ journal={M42},
209
+ year={2024}
210
+ }
211
+ ```