Hugging Face's logo

ksg-dfci
/
MatchMiner-AI

Model card Files Community
MatchMiner-AI
File size: 5,121 Bytes 
b63318a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
 
{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f986b468-c428-4ca3-9101-1cbabe6ad73f",
   "metadata": {},
   "outputs": [],
   "source": [
    "from vllm import LLM, SamplingParams\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import torch.nn.functional as F\n",
    "import torch\n",
    "from transformers import AutoTokenizer\n",
    "from transformers import AutoModelForCausalLM\n",
    "import re\n",
    "import os\n",
    "#os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"1\"\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8e0537c8-85cc-4bae-97de-6dd6f70ea5a3",
   "metadata": {},
   "outputs": [],
   "source": [
    "llama = LLM(model='hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4', tensor_parallel_size = 2, download_dir = \"../meta_ai/\", gpu_memory_utilization=0.80, max_model_len=10000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6119a0b4-5a63-4f91-a637-484b5e9dc29c",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "71a24b19-5e1c-4c24-b13b-ac04c1e94bd2",
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_synthetic_clinical_notes(num_reports, llama_model):\n",
    "\n",
    "    tokenizer = llama_model.get_tokenizer()\n",
    "    prompts = []\n",
    "    cancer_types = np.random.choice(['breast', 'non-small cell lung', 'small cell lung', 'colorectal', 'pancreatic', 'urothelial', 'prostate', 'gastric', 'esophageal', 'thymoma', 'thymic carcinoma', 'adrenal', 'ovarian', 'endometrial', 'melanoma', 'renal cell', 'sarcoma', 'head and neck', 'Hodgkin lymphoma', 'Non-Hodgkin lymphoma', 'myeloma', 'acute myeloid leukemia', 'chronic myeloid leukemia', 'acute lymphoblastic leukemia', 'chronic lymphocytic leukemia/lymphoma', 'primary brain tumor'], size=num_reports) \n",
    "\n",
    "    for i in range(num_reports):\n",
    "        messages = [\n",
    "            {'role':'system', 'content': \"\"\"Your job is to generate synthetic oncologist clinical progress notes for hypothetical patients with cancer.\n",
    "            You know all there is to know about cancer and its treatments, so be detailed.           \n",
    "        \"\"\"},      \n",
    "              \n",
    "            {'role':'user', 'content': \"\"\"Imagine a patient with cancer. \n",
    "            The cancer type is\"\"\" + cancer_types[i] + \".\" + \"\"\"\n",
    "            The patient might have any stage of disease. Use everything you know about cancer, including biomarkers, epidemiology, and heterogeneity in disease presentations.\n",
    "            The note might correspond to any point along the disease trajectory, from initial diagnosis to curative intent treatment to palliative intent treatment.\n",
    "            The note should include a chief complaint, oncologic history including prior treatments, past medical history/comorbidities, current subjective clinical status and physical exam including vital signs and ECOG performance status, laboratory values, radiology excerpts, and an assessment and plan.\n",
    "            The note should should be approximately two pages long. It will not be used for clinical care, so do not include disclaimers.\"\"\"}\n",
    "        ]\n",
    "    \n",
    "        prompts.append(tokenizer.apply_chat_template(conversation=messages, add_generation_prompt=True, tokenize=False))\n",
    "    \n",
    "\n",
    "    \n",
    "    responses = llama_model.generate(\n",
    "        prompts,   \n",
    "        SamplingParams(\n",
    "        temperature=1.0,\n",
    "        top_p=0.9,\n",
    "        max_tokens=4000,\n",
    "        stop_token_ids=[tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids(\"<|eot_id|>\")],  # KEYPOINT HERE\n",
    "    ))\n",
    "\n",
    "    response_texts = [x.outputs[0].text for x in responses]\n",
    "\n",
    "\n",
    "    return pd.DataFrame({'cancer_type':cancer_types, 'synthetic_clinical_note':response_texts})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "63642559-3583-4771-88f1-5d4e8b6e033f",
   "metadata": {},
   "outputs": [],
   "source": [
    "results = generate_synthetic_clinical_notes(10000, llama)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6eeb2f9c-4ca8-424b-b30e-dcee57323fe2",
   "metadata": {},
   "outputs": [],
   "source": [
    "results.to_csv('synthetic_clinical_notes.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0470b150-dda1-4900-adb6-a90e5cf74e39",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.12.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}