Added files

app.py

@@ -1,161 +1,29 @@
 import time
+import sys
 import streamlit as st
 import string
 from io import StringIO 
 import pdb
 import json
-from twc_embeddings import HFModel,SimCSEModel,SGPTModel
+from twc_embeddings import HFModel,SimCSEModel,SGPTModel,CausalLMModel,SGPTQnAModel
 import torch
 
 
 MAX_INPUT = 100
 
+SEM_SIMILARITY="1"
+DOC_RETRIEVAL="2"
+CLUSTERING="3"
+
+
+use_case = {"1":"Finding similar phrases/sentences","2":"Retrieving semantically matching information to a query. It may not be a factual match","3":"Clustering"}
+
+
+
 
 from transformers import BertTokenizer, BertForMaskedLM
 
-model_names = [
-
-            {   "name":"sentence-transformers/all-MiniLM-L6-v2", 
-                "model":"sentence-transformers/all-MiniLM-L6-v2",
-                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
-                "orig_author_url":"https://github.com/UKPLab",
-                "orig_author":"Ubiquitous Knowledge Processing Lab",
-                "sota_info": {   
-                                 "task":"Over 3.8  million downloads from huggingface",
-                                 "sota_link":"https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2"
-                            },
-                "paper_url":"https://arxiv.org/abs/1908.10084",
-                "mark":True,
-                "class":"HFModel"},
-            {   "name":"sentence-transformers/paraphrase-MiniLM-L6-v2", 
-                "model":"sentence-transformers/paraphrase-MiniLM-L6-v2",
-                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
-                "orig_author_url":"https://github.com/UKPLab",
-                "orig_author":"Ubiquitous Knowledge Processing Lab",
-                "sota_info": {   
-                                 "task":"Over 2 million downloads from huggingface",
-                                 "sota_link":"https://huggingface.co/sentence-transformers/paraphrase-MiniLM-L6-v2"
-                            },
-                "paper_url":"https://arxiv.org/abs/1908.10084",
-                "mark":True,
-                "class":"HFModel"},
-            {   "name":"sentence-transformers/bert-base-nli-mean-tokens", 
-                "model":"sentence-transformers/bert-base-nli-mean-tokens",
-                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
-                "orig_author_url":"https://github.com/UKPLab",
-                "orig_author":"Ubiquitous Knowledge Processing Lab",
-                "sota_info": {   
-                                 "task":"Over 700,000 downloads from huggingface",
-                                 "sota_link":"https://huggingface.co/sentence-transformers/bert-base-nli-mean-tokens"
-                            },
-                "paper_url":"https://arxiv.org/abs/1908.10084",
-                "mark":True,
-                "class":"HFModel"},
-            {   "name":"sentence-transformers/all-mpnet-base-v2", 
-                "model":"sentence-transformers/all-mpnet-base-v2",
-                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
-                "orig_author_url":"https://github.com/UKPLab",
-                "orig_author":"Ubiquitous Knowledge Processing Lab",
-                "sota_info": {   
-                                 "task":"Over 500,000 downloads from huggingface",
-                                 "sota_link":"https://huggingface.co/sentence-transformers/all-mpnet-base-v2"
-                            },
-                "paper_url":"https://arxiv.org/abs/1908.10084",
-                "mark":True,
-                "class":"HFModel"},
-            {   "name":"sentence-transformers/all-MiniLM-L12-v2",
-                "model":"sentence-transformers/all-MiniLM-L12-v2",
-                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
-                "orig_author_url":"https://github.com/UKPLab",
-                "orig_author":"Ubiquitous Knowledge Processing Lab",
-                "sota_info": {   
-                                 "task":"Over 500,000 downloads from huggingface",
-                                 "sota_link":"https://huggingface.co/sentence-transformers/all-MiniLM-L12-v2"
-                            },
-                "paper_url":"https://arxiv.org/abs/1908.10084",
-                "mark":True,
-                "class":"HFModel"},
-
-            {   "name":"SGPT-125M", 
-                "model":"Muennighoff/SGPT-125M-weightedmean-nli-bitfit",
-                "fork_url":"https://github.com/taskswithcode/sgpt",
-                "orig_author_url":"https://github.com/Muennighoff",
-                "orig_author":"Niklas Muennighoff",
-                "sota_info": {   
-                                 "task":"#1 in multiple information retrieval & search tasks(smaller variant)",
-                                 "sota_link":"https://paperswithcode.com/paper/sgpt-gpt-sentence-embeddings-for-semantic",
-                            },
-                "paper_url":"https://arxiv.org/abs/2202.08904v5",
-                "mark":True,
-                "class":"SGPTModel"},
-            {   "name":"SGPT-1.3B",
-                "model": "Muennighoff/SGPT-1.3B-weightedmean-msmarco-specb-bitfit",
-                "fork_url":"https://github.com/taskswithcode/sgpt",
-                "orig_author_url":"https://github.com/Muennighoff",
-                "orig_author":"Niklas Muennighoff",
-                "sota_info": {   
-                                 "task":"#1 in multiple information retrieval & search tasks(smaller variant)",
-                                 "sota_link":"https://paperswithcode.com/paper/sgpt-gpt-sentence-embeddings-for-semantic",
-                            },
-                "paper_url":"https://arxiv.org/abs/2202.08904v5",
-                "Note":"If this large model takes too long or fails to load , try this ",
-                "alt_url":"http://www.taskswithcode.com/sentence_similarity/",
-                "mark":True,
-                "class":"SGPTModel"},
-            {   "name":"SGPT-5.8B",
-                "model": "Muennighoff/SGPT-5.8B-weightedmean-msmarco-specb-bitfit" ,
-                "fork_url":"https://github.com/taskswithcode/sgpt",
-                "orig_author_url":"https://github.com/Muennighoff",
-                "orig_author":"Niklas Muennighoff",
-                "Note":"If this large model takes too long or fails to load , try this ",
-                "alt_url":"http://www.taskswithcode.com/sentence_similarity/",
-                "sota_info": {   
-                                 "task":"#1 in multiple information retrieval & search tasks",
-                                 "sota_link":"https://paperswithcode.com/paper/sgpt-gpt-sentence-embeddings-for-semantic",
-                            },
-                "paper_url":"https://arxiv.org/abs/2202.08904v5",
-                "mark":True,
-                "class":"SGPTModel"},
-
-            {   "name":"SIMCSE-large" ,
-                "model":"princeton-nlp/sup-simcse-roberta-large",
-                "fork_url":"https://github.com/taskswithcode/SimCSE",
-                "orig_author_url":"https://github.com/princeton-nlp",
-                "orig_author":"Princeton Natural Language Processing",
-                "Note":"If this large model takes too long or fails to load , try this ",
-                "alt_url":"http://www.taskswithcode.com/sentence_similarity/",
-                "sota_info": {   
-                                 "task":"Within top 10 in multiple semantic textual similarity tasks",
-                                 "sota_link":"https://paperswithcode.com/paper/simcse-simple-contrastive-learning-of"
-                            },
-                "paper_url":"https://arxiv.org/abs/2104.08821v4",
-                "mark":True,
-                "class":"SimCSEModel","sota_link":"https://paperswithcode.com/sota/semantic-textual-similarity-on-sick"},
-
-            {  "name":"SIMCSE-base" ,
-                "model":"princeton-nlp/sup-simcse-roberta-base",
-                "fork_url":"https://github.com/taskswithcode/SimCSE",
-                "orig_author_url":"https://github.com/princeton-nlp",
-                "orig_author":"Princeton Natural Language Processing",
-                "sota_info": {   
-                                 "task":"Within top 10 in multiple semantic textual similarity tasks(smaller variant)",
-                                 "sota_link":"https://paperswithcode.com/paper/simcse-simple-contrastive-learning-of"
-                            },
-                "paper_url":"https://arxiv.org/abs/2104.08821v4",
-                "mark":True,
-                "class":"SimCSEModel","sota_link":"https://paperswithcode.com/sota/semantic-textual-similarity-on-sick"},
-
-
-            ]
-
-
-
-
-
-example_file_names = {
-"Machine learning terms (30+ phrases)": "small_test.txt",
-"Customer feedback mixed with noise (50+ sentences)":"larger_test.txt"
-}
+
 
 view_count_file = "view_count.txt"
 
@@ -177,12 +45,12 @@ def get_views():
         
 
 
-def construct_model_info_for_display():
+def construct_model_info_for_display(model_names):
     options_arr  = []
     markdown_str = f"<div style=\"font-size:16px; color: #2f2f2f; text-align: left\"><br/><b>Models evaluated ({len(model_names)})</b></div>"
     for node in model_names:
         options_arr .append(node["name"])
-        if (node["mark"] == True):
+        if (node["mark"] == "True"):
             markdown_str += f"<div style=\"font-size:16px; color: #5f5f5f; text-align: left\">&nbsp;•&nbsp;Model:&nbsp;<a href=\'{node['paper_url']}\' target='_blank'>{node['name']}</a><br/>&nbsp;&nbsp;&nbsp;&nbsp;Code released by:&nbsp;<a href=\'{node['orig_author_url']}\' target='_blank'>{node['orig_author']}</a><br/>&nbsp;&nbsp;&nbsp;&nbsp;Model info:&nbsp;<a href=\'{node['sota_info']['sota_link']}\' target='_blank'>{node['sota_info']['task']}</a></div>"
             if ("Note" in node):
                 markdown_str += f"<div style=\"font-size:16px; color: #a91212; text-align: left\">&nbsp;&nbsp;&nbsp;&nbsp;{node['Note']}<a href=\'{node['alt_url']}\' target='_blank'>link</a></div>"
@@ -194,7 +62,7 @@ def construct_model_info_for_display():
     return options_arr,markdown_str
 
 
-st.set_page_config(page_title='TWC - Compare popular/state-of-the-art models for Sentence Similarity task', page_icon="logo.jpg", layout='centered', initial_sidebar_state='auto',
+st.set_page_config(page_title='TWC - Compare popular/state-of-the-art models for tasks using sentence embeddings', page_icon="logo.jpg", layout='centered', initial_sidebar_state='auto',
             menu_items={
              'About': 'This app was created by taskswithcode. http://taskswithcode.com'
              
@@ -206,7 +74,7 @@ with col:
 
 
 @st.experimental_memo
-def load_model(model_name):
+def load_model(model_name,model_names):
     try:
         ret_model = None
         for node in model_names:
@@ -235,18 +103,18 @@ def uncached_compute_similarity(sentences,_model,model_name,main_index):
     #st.success("Similarity computation complete")
     return results
 
-def get_model_info(model_name):
+def get_model_info(model_names,model_name):
     for node in model_names:
         if (model_name == node["name"]):
             return node
 
-def run_test(model_name,sentences,display_area,main_index,user_uploaded):
+def run_test(model_names,model_name,sentences,display_area,main_index,user_uploaded):
     display_area.text("Loading model:" + model_name)
-    model_info = get_model_info(model_name)
+    model_info = get_model_info(model_names,model_name)
     if ("Note" in model_info):
         fail_link = f"{model_info['Note']} [link]({model_info['alt_url']})"
         display_area.write(fail_link)
-    model = load_model(model_name)
+    model = load_model(model_name,model_names)
     display_area.text("Model " + model_name  + " load complete")
     try:
             if (user_uploaded):
@@ -266,13 +134,20 @@ def run_test(model_name,sentences,display_area,main_index,user_uploaded):
 
     
 
-def display_results(orig_sentences,main_index,results,response_info):
+def display_results(orig_sentences,main_index,results,response_info,app_mode):
     main_sent = f"<div style=\"font-size:14px; color: #2f2f2f; text-align: left\">{response_info}<br/><br/></div>"
-    main_sent += "<div style=\"font-size:14px; color: #6f6f6f; text-align: left\">Results sorted by cosine distance. Closest(1) to furthest(-1) away from main sentence</div>"
-    main_sent += f"<div style=\"font-size:16px; color: #2f2f2f; text-align: left\"><b>Main sentence:</b>&nbsp;&nbsp;{orig_sentences[main_index]}</div>"
+    score_text = "cosine_distance" if app_mode == "similarity" else "cosine_distance/score"
+    pivot_name = "main sentence" if app_mode == "similarity" else "query"
+    main_sent += f"<div style=\"font-size:14px; color: #6f6f6f; text-align: left\">Results sorted by {score_text}. Closest to furthest away from {pivot_name}</div>"
+    pivot_name = pivot_name[0].upper() + pivot_name[1:]
+    main_sent += f"<div style=\"font-size:16px; color: #2f2f2f; text-align: left\"><b>{pivot_name}:</b>&nbsp;&nbsp;{orig_sentences[main_index]}</div>"
     body_sent = []
     download_data = {}
+    first = True
     for key in results:
+        if (app_mode == DOC_RETRIEVAL and first):
+            first = False
+            continue
         index = orig_sentences.index(key) + 1
         body_sent.append(f"<div style=\"font-size:16px; color: #2f2f2f; text-align: left\">{index}]&nbsp;{key}&nbsp;&nbsp;&nbsp;<b>{results[key]:.2f}</b></div>")
         download_data[key] =  f"{results[key]:.2f}" 
@@ -287,9 +162,15 @@ def init_session():
     st.session_state["main_index"] = 1
     st.session_state["file_name"] = "default"
  
-def main():
+def app_main(app_mode,example_files,model_name_files):
   init_session()
-  st.markdown("<h5 style='text-align: center;'>Compare popular/state-of-the-art models for Sentence Similarity task</h5>", unsafe_allow_html=True)
+  with open(example_files) as fp:
+        example_file_names = json.load(fp) 
+  with open(model_name_files) as fp:
+        model_names = json.load(fp)
+  curr_use_case = use_case[app_mode].split(".")[0]
+  st.markdown("<h5 style='text-align: center;'>Compare popular/state-of-the-art models for tasks using sentence embeddings</h5>", unsafe_allow_html=True)
+  st.markdown(f"<div style='color: #4f4f4f; text-align: left'>Use cases for sentence embeddings<br/>&nbsp;&nbsp;&nbsp;•&nbsp;&nbsp;{use_case['1']}<br/>&nbsp;&nbsp;&nbsp;•&nbsp;&nbsp;{use_case['2']}<br/>&nbsp;&nbsp;&nbsp;•&nbsp;&nbsp;{use_case['3']}<br/><i>This app illustrates <b>'{curr_use_case}'</b> use case</i></div>", unsafe_allow_html=True)
   st.markdown(f"<div style='color: #9f9f9f; text-align: right'>views:&nbsp;{get_views()}</div>", unsafe_allow_html=True)
 
 
@@ -298,17 +179,23 @@ def main():
       
       with st.form('twc_form'):
 
-        uploaded_file = st.file_uploader("Step 1. Upload text file(one sentence in a line) or choose an example text file below", type=".txt")
+        step1_line = "Step 1. Upload text file(one sentence in a line) or choose an example text file below"
+        if (app_mode ==  DOC_RETRIEVAL):
+            step1_line += ". The first line is treated as the query"
+        uploaded_file = st.file_uploader(step1_line, type=".txt")
 
-        selected_file_index = st.selectbox(label='Example files ',  
+        selected_file_index = st.selectbox(label=f'Example files ({len(example_file_names)})',  
                     options = list(dict.keys(example_file_names)), index=0,  key = "twc_file")
         st.write("")
-        options_arr,markdown_str = construct_model_info_for_display()
+        options_arr,markdown_str = construct_model_info_for_display(model_names)
         selection_label = 'Step 2. Select Model'
         selected_model = st.selectbox(label=selection_label,  
                     options = options_arr, index=0,  key = "twc_model")
         st.write("")
-        main_index = st.number_input('Step 3. Enter index of sentence in file to make it the main sentence',value=1,min_value = 1)
+        if (app_mode == "similarity"):
+            main_index = st.number_input('Step 3. Enter index of sentence in file to make it the main sentence',value=1,min_value = 1)
+        else:
+            main_index = 1
         st.write("")
         submit_button = st.form_submit_button('Run')
 
@@ -321,8 +208,8 @@ def main():
                 st.session_state["file_name"]  = uploaded_file.name
                 sentences = StringIO(uploaded_file.getvalue().decode("utf-8")).read()
             else:
-                st.session_state["file_name"]  = example_file_names[selected_file_index]
-                sentences = open(example_file_names[selected_file_index]).read()
+                st.session_state["file_name"]  = example_file_names[selected_file_index]["name"]
+                sentences = open(example_file_names[selected_file_index]["name"]).read()
             sentences = sentences.split("\n")[:-1]
             if (len(sentences) < main_index):
                 main_index = len(sentences)
@@ -332,12 +219,12 @@ def main():
                 sentences = sentences[:MAX_INPUT]
             st.session_state["model_name"] = selected_model
             st.session_state["main_index"] = main_index
-            results = run_test(selected_model,sentences,display_area,main_index - 1,(uploaded_file is not None))
+            results = run_test(model_names,selected_model,sentences,display_area,main_index - 1,(uploaded_file is not None))
             display_area.empty()
             with display_area.container():
                 device = 'GPU' if torch.cuda.is_available() else 'CPU'
                 response_info = f"Computation time on {device}: {time.time() - start:.2f} secs for {len(sentences)} sentences"
-                display_results(sentences,main_index - 1,results,response_info)
+                display_results(sentences,main_index - 1,results,response_info,app_mode)
                 #st.json(results)
       st.download_button(
          label="Download results as json",
@@ -359,5 +246,8 @@ def main():
  
 
 if __name__ == "__main__":
-   main()
+   #print("comand line input:",len(sys.argv),str(sys.argv))
+   #app_main(sys.argv[1],sys.argv[2],sys.argv[3])
+   app_main("1","sim_app_examples.json","sim_app_models.json")
+   #app_main("2","doc_app_examples.json","doc_app_models.json")
 

sim_app_examples.json

@@ -0,0 +1,5 @@
+{
+"Machine learning terms (phrases test)": {"name":"small_test.txt"},
+"Customer feedback mixed with noise":{"name":"larger_test.txt"},
+"Movie reviews": {"name":"imdb_sent.txt"}
+}

sim_app_models.json

@@ -0,0 +1,134 @@
+[
+
+            {   "name":"sentence-transformers/all-MiniLM-L6-v2", 
+                "model":"sentence-transformers/all-MiniLM-L6-v2",
+                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
+                "orig_author_url":"https://github.com/UKPLab",
+                "orig_author":"Ubiquitous Knowledge Processing Lab",
+                "sota_info": {   
+                                 "task":"Over 3.8  million downloads from huggingface",
+                                 "sota_link":"https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2"
+                            },
+                "paper_url":"https://arxiv.org/abs/1908.10084",
+                "mark":"True",
+                "class":"HFModel"},
+            {   "name":"sentence-transformers/paraphrase-MiniLM-L6-v2", 
+                "model":"sentence-transformers/paraphrase-MiniLM-L6-v2",
+                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
+                "orig_author_url":"https://github.com/UKPLab",
+                "orig_author":"Ubiquitous Knowledge Processing Lab",
+                "sota_info": {   
+                                 "task":"Over 2 million downloads from huggingface",
+                                 "sota_link":"https://huggingface.co/sentence-transformers/paraphrase-MiniLM-L6-v2"
+                            },
+                "paper_url":"https://arxiv.org/abs/1908.10084",
+                "mark":"True",
+                "class":"HFModel"},
+            {   "name":"sentence-transformers/bert-base-nli-mean-tokens", 
+                "model":"sentence-transformers/bert-base-nli-mean-tokens",
+                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
+                "orig_author_url":"https://github.com/UKPLab",
+                "orig_author":"Ubiquitous Knowledge Processing Lab",
+                "sota_info": {   
+                                 "task":"Over 700,000 downloads from huggingface",
+                                 "sota_link":"https://huggingface.co/sentence-transformers/bert-base-nli-mean-tokens"
+                            },
+                "paper_url":"https://arxiv.org/abs/1908.10084",
+                "mark":"True",
+                "class":"HFModel"},
+            {   "name":"sentence-transformers/all-mpnet-base-v2", 
+                "model":"sentence-transformers/all-mpnet-base-v2",
+                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
+                "orig_author_url":"https://github.com/UKPLab",
+                "orig_author":"Ubiquitous Knowledge Processing Lab",
+                "sota_info": {   
+                                 "task":"Over 500,000 downloads from huggingface",
+                                 "sota_link":"https://huggingface.co/sentence-transformers/all-mpnet-base-v2"
+                            },
+                "paper_url":"https://arxiv.org/abs/1908.10084",
+                "mark":"True",
+                "class":"HFModel"},
+            {   "name":"sentence-transformers/all-MiniLM-L12-v2",
+                "model":"sentence-transformers/all-MiniLM-L12-v2",
+                "fork_url":"https://github.com/taskswithcode/sentence_similarity_hf_model",
+                "orig_author_url":"https://github.com/UKPLab",
+                "orig_author":"Ubiquitous Knowledge Processing Lab",
+                "sota_info": {   
+                                 "task":"Over 500,000 downloads from huggingface",
+                                 "sota_link":"https://huggingface.co/sentence-transformers/all-MiniLM-L12-v2"
+                            },
+                "paper_url":"https://arxiv.org/abs/1908.10084",
+                "mark":"True",
+                "class":"HFModel"},
+
+            {   "name":"SGPT-125M", 
+                "model":"Muennighoff/SGPT-125M-weightedmean-nli-bitfit",
+                "fork_url":"https://github.com/taskswithcode/sgpt",
+                "orig_author_url":"https://github.com/Muennighoff",
+                "orig_author":"Niklas Muennighoff",
+                "sota_info": {   
+                                 "task":"#1 in multiple information retrieval & search tasks(smaller variant)",
+                                 "sota_link":"https://paperswithcode.com/paper/sgpt-gpt-sentence-embeddings-for-semantic"
+                            },
+                "paper_url":"https://arxiv.org/abs/2202.08904v5",
+                "mark":"True",
+                "class":"SGPTModel"},
+            {   "name":"SGPT-1.3B",
+                "model": "Muennighoff/SGPT-1.3B-weightedmean-msmarco-specb-bitfit",
+                "fork_url":"https://github.com/taskswithcode/sgpt",
+                "orig_author_url":"https://github.com/Muennighoff",
+                "orig_author":"Niklas Muennighoff",
+                "sota_info": {   
+                                 "task":"#1 in multiple information retrieval & search tasks(smaller variant)",
+                                 "sota_link":"https://paperswithcode.com/paper/sgpt-gpt-sentence-embeddings-for-semantic"
+                            },
+                "paper_url":"https://arxiv.org/abs/2202.08904v5",
+                "Note":"If this large model takes too long or fails to load , try this ",
+                "alt_url":"http://www.taskswithcode.com/sentence_similarity/",
+                "mark":"True",
+                "class":"SGPTModel"},
+            {   "name":"SGPT-5.8B",
+                "model": "Muennighoff/SGPT-5.8B-weightedmean-msmarco-specb-bitfit" ,
+                "fork_url":"https://github.com/taskswithcode/sgpt",
+                "orig_author_url":"https://github.com/Muennighoff",
+                "orig_author":"Niklas Muennighoff",
+                "Note":"If this large model takes too long or fails to load , try this ",
+                "alt_url":"http://www.taskswithcode.com/sentence_similarity/",
+                "sota_info": {   
+                                 "task":"#1 in multiple information retrieval & search tasks",
+                                 "sota_link":"https://paperswithcode.com/paper/sgpt-gpt-sentence-embeddings-for-semantic"
+                            },
+                "paper_url":"https://arxiv.org/abs/2202.08904v5",
+                "mark":"True",
+                "class":"SGPTModel"},
+
+            {   "name":"SIMCSE-large" ,
+                "model":"princeton-nlp/sup-simcse-roberta-large",
+                "fork_url":"https://github.com/taskswithcode/SimCSE",
+                "orig_author_url":"https://github.com/princeton-nlp",
+                "orig_author":"Princeton Natural Language Processing",
+                "Note":"If this large model takes too long or fails to load , try this ",
+                "alt_url":"http://www.taskswithcode.com/sentence_similarity/",
+                "sota_info": {   
+                                 "task":"Within top 10 in multiple semantic textual similarity tasks",
+                                 "sota_link":"https://paperswithcode.com/paper/simcse-simple-contrastive-learning-of"
+                            },
+                "paper_url":"https://arxiv.org/abs/2104.08821v4",
+                "mark":"True",
+                "class":"SimCSEModel","sota_link":"https://paperswithcode.com/sota/semantic-textual-similarity-on-sick"},
+
+            {  "name":"SIMCSE-base" ,
+                "model":"princeton-nlp/sup-simcse-roberta-base",
+                "fork_url":"https://github.com/taskswithcode/SimCSE",
+                "orig_author_url":"https://github.com/princeton-nlp",
+                "orig_author":"Princeton Natural Language Processing",
+                "sota_info": {   
+                                 "task":"Within top 10 in multiple semantic textual similarity tasks(smaller variant)",
+                                 "sota_link":"https://paperswithcode.com/paper/simcse-simple-contrastive-learning-of"
+                            },
+                "paper_url":"https://arxiv.org/abs/2104.08821v4",
+                "mark":"True",
+                "class":"SimCSEModel","sota_link":"https://paperswithcode.com/sota/semantic-textual-similarity-on-sick"}
+
+
+            ]