Add support models which does not support structed output

gpt_shows_rename/__main__.py

@@ -1,12 +1,17 @@
 from .config import load_config
 from .file import gen_input_list, link_files
 from .gpt import get_response
+import asyncio
 
 
-cfg = load_config()
-files = gen_input_list(cfg.input)
-res = get_response(cfg, cfg.input, files, cfg.series_name, cfg.year, cfg.tmdb_id, cfg.tvdb_id)
-for f in res.files:
-    print(files[f.index], '->', f.name)
-input('Continue?')
-link_files(cfg.input, files, res, cfg.output, cfg.hardlink)
+async def main():
+    cfg = load_config()
+    files = gen_input_list(cfg.input)
+    res = await get_response(cfg, cfg.input, files, cfg.series_name, cfg.year, cfg.tmdb_id, cfg.tvdb_id)
+    for f in res.files:
+        print(files[f.index], '->', f.name)
+    input('Continue?')
+    link_files(cfg.input, files, res, cfg.output, cfg.hardlink)
+
+
+asyncio.run(main())

gpt_shows_rename/gpt.py

@@ -35,10 +35,39 @@ def gen_files_list(files: List[str]):
     return prompt
 
 
-def get_response(cfg: Config, inp: str, files: List[str],
-                 series_name: str = None, year: int = None, tmdb_id: int = None,
-                 tvdb_id: int = None) -> Files:
+SUPPORTED_MODELS = ["gpt-4o", "gpt-4o-mini", "gpt-4o-2024-08-06", "gpt-4o-2024-11-20", "gpt-4o-mini-2024-07-18"]
+
+
+def is_support_structed_output(model: str):
+    if model.startswith('ft:'):
+        models = model.split(':')
+        if len(models) < 2:
+            return False
+        model = models[1]
+    return model in SUPPORTED_MODELS
+
+
+def parse_result(result: str):
+    lines = result.splitlines()
+    files = []
+    for line in lines:
+        line = line.strip()
+        if not line:
+            continue
+        if line.startswith("```"):
+            continue
+        try:
+            files.append(File(**json.loads(line)))
+        except Exception:
+            raise ValueError(f"Failed to parse message: {line}")
+    return Files(files=files)
+
+
+async def get_response(cfg: Config, inp: str, files: List[str],
+                       series_name: str = None, year: int = None, tmdb_id: int = None,
+                       tvdb_id: int = None) -> Files:
     prompt = f'The input directory is `{inp}`.'
+    prompt += '\n' + gen_files_list(files)
     if series_name:
         prompt += f'\nThe series name is `{series_name}`.'
     if year:
@@ -47,18 +76,41 @@ def get_response(cfg: Config, inp: str, files: List[str],
         prompt += f'\nThe TMDB ID is `{tmdb_id}`.'
     if tvdb_id:
         prompt += f'\nThe TVDB ID is `{tvdb_id}`.'
-    prompt += '\n' + gen_files_list(files)
-    http_client = httpx.Client(proxy=cfg.proxy)
-    client = openai.Client(api_key=cfg.api_key, base_url=cfg.base_url, http_client=http_client)
-    res = client.beta.chat.completions.parse(
-        model=cfg.model,
-        messages=[
-            {"role": "system", "content": SYSTEM_PROMPT},
-            {"role": "user", "content": prompt},
-        ],
-        response_format=Files,
-    )
-    mes = res.choices[0].message
-    if mes.refusal:
-        raise ValueError(f"Model refused to answer: {mes.refusal}")
-    return mes.parsed
+    if is_support_structed_output(cfg.model):
+        http_client = httpx.Client(proxy=cfg.proxy)
+        client = openai.Client(api_key=cfg.api_key, base_url=cfg.base_url, http_client=http_client)
+        res = client.beta.chat.completions.parse(
+            model=cfg.model,
+            messages=[
+                {"role": "system", "content": SYSTEM_PROMPT},
+                {"role": "user", "content": prompt},
+            ],
+            response_format=Files,
+        )
+        mes = res.choices[0].message
+        if mes.refusal:
+            raise ValueError(f"Model refused to answer: {mes.refusal}")
+        if not mes.parsed:
+            raise ValueError(f"Unhandled error: {mes.to_json()}")
+        return mes.parsed
+    else:
+        http_client = httpx.AsyncClient(proxy=cfg.proxy)
+        client = openai.AsyncClient(api_key=cfg.api_key, base_url=cfg.base_url, http_client=http_client)
+        res = await client.chat.completions.create(
+            model=cfg.model,
+            messages=[
+                {"role": "system", "content": SYSTEM_PROMPT + "\nThe format of the returned result is consistent with the input file list."},
+                {"role": "user", "content": prompt},
+            ],
+            stream=True,
+        )
+        mes = ''
+        async for chunk in res:
+            if chunk.choices:
+                choice = chunk.choices[0]
+                if choice.delta and choice.delta.content:
+                    data = choice.delta.content
+                    mes += data
+                    print(data, end='', flush=True)
+        print(flush=True)
+        return parse_result(mes)