Switch to coze v2 api

Author: polyrabbit

hacker_news/llm/coze.py

@@ -58,7 +58,7 @@ def summarize_by_coze(content: str) -> str:
     content = sanitize_for_openai(content, overhead=1000)
 
     # For model: GPT-4 Turbo (128K), temperature: 0.5 - GPT-4 Turbo is an excellent rule follower.
-    prompt = (f'Use third person mood to summarize the main points of the following article delimited by triple backticks in 2 concise sentences. '
+    prompt = (f'Use third person mood to summarize the main points of the following article delimited by triple backticks in 2 concise English sentences. '
               f'Ensure the summary does not exceed 300 characters.\n'
               f'```{content}.```')
 
@@ -75,20 +75,7 @@ def summarize_by_coze(content: str) -> str:
                                 'stream': False,
                             })
         resp.raise_for_status()
-
-        for line in resp.iter_lines():
-            if line and line.startswith(b'data:'):
-                line = line[len(b'data:'):].strip()
-                try:
-                    resp_json = json.loads(line)
-                except json.JSONDecodeError as e:
-                    logger.warning(f'Failed to decode coze response, unexpected json {line}, error: {e}')
-                    return ''
-                break
-        else:
-            logger.warning(f'Unexpected coze response, no data line found')
-            return ''
-
+        resp_json = resp.json()
     except Exception as e:
         logger.warning(f'Failed to summarize using coze, {e}')
         return ''

hacker_news/llm/openai.py

@@ -8,7 +8,10 @@ def sanitize_for_openai(text, overhead):
 
     # one token generally corresponds to ~4 characters, from https://platform.openai.com/tokenizer
     if len(text) > 4096 * 2:
-        enc = tiktoken.encoding_for_model(config.openai_model)
+        try:
+            enc = tiktoken.encoding_for_model(config.openai_model)  # We have openai compatible apis now
+        except KeyError:
+            enc = tiktoken.encoding_for_model('gpt-3.5-turbo')
         tokens = enc.encode(text)
         if len(tokens) > 4096 - overhead:  # 4096: model's context limit
             text = enc.decode(tokens[:4096 - overhead])

hacker_news/news.py

@@ -147,12 +147,13 @@ def summarize_by_openai(self, content):
                  f'3 - Provide a Chinese translation of sentence: "{title}".\n' \
                  f'```{content.strip(".")}.```'
         try:
-            answer = self.openai_complete(prompt, True)
-            summary = self.parse_step_answer(answer).strip()
-            if not summary:  # If step parse failed, ignore the translation
-                summary = self.openai_complete(
-                    f'Summarize the article delimited by triple backticks in 2 sentences.\n'
-                    f'```{content.strip(".")}.```', False)
+            # Too many exceptions to support translation, give up...
+            # answer = self.openai_complete(prompt, True)
+            # summary = self.parse_step_answer(answer).strip().strip(' *-')
+            # if not summary:  # If step parse failed, ignore the translation
+            summary = self.openai_complete(
+                f'Use third person mood to summarize the main points of the following article delimited by triple backticks in 2 concise sentences. Ensure the summary does not exceed 100 characters.\n'
+                f'```{content.strip(".")}.```', False)
             return summary
         except Exception as e:
             logger.exception(f'Failed to summarize using openai, key #{config.openai_key_index}, {e}')  # Make this error explicit in the log
@@ -219,7 +220,7 @@ def openai_complete(self, prompt, need_json):
         return answer
 
     def parse_step_answer(self, answer):
-        if not answer:
+        if not answer or isinstance(answer, str):
             return answer
         db.translation.add(answer.get('summary', ''), answer.get('summary_zh', ''), 'zh')
         db.translation.add(self.title, self.parse_title_translation(answer.get('translation', '')), 'zh')

page_content_extractor/http.py

@@ -5,7 +5,6 @@
 import urllib3
 from fake_useragent import UserAgent
 from requests.adapters import HTTPAdapter
-from urllib3.exceptions import InsecureRequestWarning
 from urllib3.util import timeout
 from urllib3.util.ssl_ import create_urllib3_context