Fix Openai API for embeddings

config.py

@@ -4,9 +4,12 @@ import os
 
 class Settings(BaseSettings):
     huggingface_key: str = os.environ.get("huggingface_key")
+    OPENAI_API_KEY: str = os.environ.get("OPENAI_API_KEY")
     cache_dir: str = "cache"
     embedding_api: str = "sbert"
     embedding_model: str = "mano-wii/BAAI_bge-base-en-v1.5-tunned-for-blender-issues"
+    # embedding_api: str = "openai"
+    # embedding_model: str = "text-embedding-ada-002"
 
 
 settings = Settings()

requirements-fastapi.txt

@@ -3,4 +3,5 @@ uvicorn[standard]
 python-multipart
 pydantic-settings
 huggingface_hub
-sentence_transformers
+sentence_transformers
+openai

routers/tool_find_related.py

@@ -62,6 +62,7 @@ class EmbeddingContext:
     # Set when creating the object
     lock = None
     model = None
+    openai_client = None
     model_name = ''
     config_type = ''
 
@@ -86,7 +87,11 @@ class EmbeddingContext:
                 self.model = self.model.to('cuda')
 
         elif config_type == 'openai':
-            # openai.api_base = settings.openai.api_base
+            from openai import OpenAI
+            self.openai_client = OpenAI(
+                # base_url = settings.openai_api_base
+                api_key=settings.OPENAI_API_KEY,
+            )
             self.encode = self.encode_openai
 
         self.model_name = model_name
@@ -100,7 +105,6 @@ class EmbeddingContext:
 
     def encode_openai(self, texts_to_embed):
         import math
-        import openai
         import time
 
         tokens_count = 0
@@ -116,8 +120,11 @@ class EmbeddingContext:
                 end = start + chunk_size
                 chunk = texts_to_embed[start:end]
 
-                embeddings_tmp = openai.Embedding.create(
-                    input=chunk, model=self.model_name)['data']
+                embeddings_tmp = self.openai_client.embeddings.create(
+                    model=self.model_name,
+                    input=chunk,
+                ).data
+
                 if embeddings_tmp is None:
                     break
 
@@ -126,7 +133,7 @@ class EmbeddingContext:
                 if i < chunks_num - 1:
                     time.sleep(60)  # Wait 1 minute before the next call
 
-            return torch.stack([torch.tensor(embedding['embedding'], dtype=torch.float32) for embedding in embeddings])
+            return torch.stack([torch.tensor(embedding.embedding, dtype=torch.float32) for embedding in embeddings])
 
     def get_tokens(self, text):
         if self.model:
@@ -204,18 +211,20 @@ class EmbeddingContext:
             issues = gitea_fetch_issues(
                 owner, repo, since=date_old, issue_attr_filter=self.issue_attr_filter, exclude=black_list)
 
-            # WORKAROUND:
-            # Consider that if the time hasn't changed, it's the same issue.
-            issues = [
-                issue for issue in issues if issue['updated_at'] != date_old]
+            # Get the most recent date
+            date_new = _find_latest_date(issues, date_old)
 
-            if len(issues) == 0:
+            if date_new == date_old:
+                # Nothing changed
                 return data
 
-            # Get the most recent date
-            date_new = _find_latest_date(issues, date_old)
+            data['updated_at'] = date_new
 
 # autopep8: off
+            # WORKAROUND:
+            # Consider that if the time hasn't changed, it's the same issue.
+            issues = [issue for issue in issues if issue['updated_at'] != date_old]
+
             numbers_old = data['numbers']
             titles_old = data['titles']
             embeddings_old = data['embeddings']
@@ -239,6 +248,9 @@ class EmbeddingContext:
                         old_closed.append(i_old)
                         break
 
+            if not old_closed and not issues_open:
+                return data
+
             mask_open = torch.ones(len(numbers_open), dtype=torch.bool)
             need_sort = False
             change_map = []
@@ -286,10 +298,13 @@ class EmbeddingContext:
                     i_new += 1
 
                 assert i_new == total
-            else:
+            elif mask_open.any():
                 titles_new = titles_old + [issue['title'] for i, issue in enumerate(issues_open) if mask_open[i]]
                 numbers_new = numbers_old + [number for i, number in enumerate(numbers_open) if mask_open[i]]
                 embeddings_new = torch.cat([embeddings_old, embeddings[mask_open]])
+            else:
+                # Only Updated Data changed
+                return data
 
             if need_sort:
                 sorted_indices = sorted(range(len(numbers_new)), key=lambda k: numbers_new[k])
@@ -297,7 +312,6 @@ class EmbeddingContext:
                 numbers_new = [numbers_new[i] for i in sorted_indices]
                 embeddings_new = embeddings_new[sorted_indices]
 
-            data['updated_at'] = date_new
             data['titles'] = titles_new
             data['numbers'] = numbers_new
             data['embeddings'] = embeddings_new