the brain splice commit !!

app.py

@@ -42,7 +42,7 @@ from torch import tensor as torch_tensor
 from datasets import load_dataset
 
 
-from greg_funcs import greg_search
+from greg_funcs import get_llm_response
 
 """# import models"""
 
@@ -491,6 +491,10 @@ class ChatWrapper:
     ):
         """Execute the chat functionality."""
         self.lock.acquire()
+
+
+
+        # import ipdb; ipdb.set_trace()
         try:
             print("\n==== date/time: " + str(datetime.datetime.now()) + " ====")
             print("inp: " + inp)
@@ -504,7 +508,10 @@ class ChatWrapper:
             output = AUTH_ERR_MSG
             hidden_text = output
 
-            # if chain and chain != "":
+
+            output = get_llm_response(inp)
+
+            """
             if chain:
                 # Set OpenAI key
                 import openai
@@ -515,12 +522,25 @@ class ChatWrapper:
                         chain, inp, capture_hidden_text=trace_chain)
                 else:
                     output, hidden_text = inp, None
+            """
+            print("original output", output)
+            print("using these knobs:",
+                  (
+                      formality, anticipation_level, joy_level,
+                      trust_level,
+                      fear_level, surprise_level, sadness_level, disgust_level, anger_level,
+                      lang_level, translate_to, literary_style
+                      )
+
+                  )
 
             output = transform_text(output, express_chain, num_words, formality, anticipation_level, joy_level,
                                     trust_level,
                                     fear_level, surprise_level, sadness_level, disgust_level, anger_level,
                                     lang_level, translate_to, literary_style)
 
+            print("transformed output", output)
+
             text_to_display = output
             if trace_chain:
                 text_to_display = hidden_text + "\n\n" + output

greg_funcs.py

@@ -4,6 +4,13 @@ from sentence_transformers import SentenceTransformer, CrossEncoder, util
 from torch import tensor as torch_tensor
 from datasets import load_dataset
 
+from langchain.llms import OpenAI
+from langchain.docstore.document import Document
+
+from langchain.chains.qa_with_sources import load_qa_with_sources_chain
+
+
+
 
 
 """# import models"""
@@ -25,7 +32,7 @@ dataset_embed = load_dataset("gfhayworth/hack_policy_embed", split='train')
 dataset_embed_pd = dataset_embed.to_pandas()
 mycorpus_embeddings = torch_tensor(dataset_embed_pd.values)
 
-def greg_search(query, passages = mypassages, doc_embedding = mycorpus_embeddings, top_k=20, top_n = 1):
+def search(query, passages = mypassages, doc_embedding = mycorpus_embeddings, top_k=20, top_n = 1):
     question_embedding = bi_encoder.encode(query, convert_to_tensor=True)
     question_embedding = question_embedding #.cuda()
     hits = util.semantic_search(question_embedding, doc_embedding, top_k=top_k)
@@ -45,3 +52,34 @@ def greg_search(query, passages = mypassages, doc_embedding = mycorpus_embedding
     # for hit in hits[0:3]:
     #         print("\t{:.3f}\t{}".format(hit['cross-score'], mypassages[hit['corpus_id']].replace("\n", " ")))
 
+
+
+def get_text_fmt(qry, passages = mypassages, doc_embedding=mycorpus_embeddings):
+    predictions = search(qry, passages = passages, doc_embedding = doc_embedding, top_n=5, )
+    prediction_text = []
+    for hit in predictions:
+        page_content = passages[hit['corpus_id']]
+        metadata = {"source": hit['corpus_id']}
+        result = Document(page_content=page_content, metadata=metadata)
+        prediction_text.append(result)
+    return prediction_text
+
+
+chain_qa = load_qa_with_sources_chain(OpenAI(temperature=0), chain_type="stuff")
+
+
+def get_llm_response(message):
+    mydocs = get_text_fmt(message)
+    response = chain_qa.run(input_documents=mydocs, question=message)
+    return response
+
+def chat(message, history):
+    history = history or []
+    message = message.lower()
+    
+    response = get_llm_response(message)
+    history.append((message, response))
+    return history, history
+
+
+