refactor: rename default_model->model, add thinking_type support, fix pet urlopen error handling
This commit is contained in:
1 parent
0551967a7e
commit
74abb77a0b
5 files changed
+41
-27
No files matched your search
@@ -774,4 +774,4 @@ if __name__ == '__main__':
|
|||||||
pet = MacPet()
|
pet = MacPet()
|
||||||
pet.run()
|
pet.run()
|
||||||
else:
|
else:
|
||||||
pet = WinPet()
|
pet = WinPet('vita')
|
||||||
+3
-3
@@ -1222,7 +1222,7 @@ class ChatPanel(QWidget):
|
|||||||
|
|
||||||
for idx, tc in enumerate(self.agent.llmclients):
|
for idx, tc in enumerate(self.agent.llmclients):
|
||||||
b = tc.backend
|
b = tc.backend
|
||||||
name = f"{type(b).__name__}/{b.default_model}"
|
name = f"{type(b).__name__}/{b.model}"
|
||||||
is_current = idx == self.agent.llm_no
|
is_current = idx == self.agent.llm_no
|
||||||
|
|
||||||
row = QWidget()
|
row = QWidget()
|
||||||
@@ -1296,9 +1296,9 @@ class ChatPanel(QWidget):
|
|||||||
reply = backend.ask("你好", stream=False)
|
reply = backend.ask("你好", stream=False)
|
||||||
text = str(reply).strip() if reply else ""
|
text = str(reply).strip() if reply else ""
|
||||||
ok = len(text) > 0 and not text.startswith("Error") and not text.startswith("[")
|
ok = len(text) > 0 and not text.startswith("Error") and not text.startswith("[")
|
||||||
print(f"[HealthCheck] Backend #{idx} {type(backend).__name__}/{backend.default_model}: {'OK' if ok else 'FAIL'} -> {text[:60]}")
|
print(f"[HealthCheck] Backend #{idx} {type(backend).__name__}/{backend.model}: {'OK' if ok else 'FAIL'} -> {text[:60]}")
|
||||||
except Exception as e:
|
except Exception as e:
|
||||||
print(f"[HealthCheck] Backend #{idx} {type(backend).__name__}/{backend.default_model}: ERROR -> {e}")
|
print(f"[HealthCheck] Backend #{idx} {type(backend).__name__}/{backend.model}: ERROR -> {e}")
|
||||||
ok = False
|
ok = False
|
||||||
if hasattr(backend, 'raw_msgs') and backend.raw_msgs:
|
if hasattr(backend, 'raw_msgs') and backend.raw_msgs:
|
||||||
backend.raw_msgs = [m for m in backend.raw_msgs if m.get("prompt") != "你好"]
|
backend.raw_msgs = [m for m in backend.raw_msgs if m.get("prompt") != "你好"]
|
||||||
|
|||||||
+5
-1
@@ -48,7 +48,11 @@ def render_sidebar():
|
|||||||
pet_script = os.path.join(os.path.dirname(__file__), 'desktop_pet_v2.pyw')
|
pet_script = os.path.join(os.path.dirname(__file__), 'desktop_pet_v2.pyw')
|
||||||
if not os.path.exists(pet_script): pet_script = os.path.join(os.path.dirname(__file__), 'desktop_pet.pyw')
|
if not os.path.exists(pet_script): pet_script = os.path.join(os.path.dirname(__file__), 'desktop_pet.pyw')
|
||||||
subprocess.Popen([sys.executable, pet_script], **kwargs)
|
subprocess.Popen([sys.executable, pet_script], **kwargs)
|
||||||
def _pet_req(q): threading.Thread(target=lambda: urlopen(f'http://127.0.0.1:51983/?{q}', timeout=2), daemon=True).start()
|
def _pet_req(q):
|
||||||
|
def _do():
|
||||||
|
try: urlopen(f'http://127.0.0.1:51983/?{q}', timeout=2)
|
||||||
|
except Exception: pass
|
||||||
|
threading.Thread(target=_do, daemon=True).start()
|
||||||
agent._pet_req = _pet_req
|
agent._pet_req = _pet_req
|
||||||
if not hasattr(agent, '_turn_end_hooks'): agent._turn_end_hooks = {}
|
if not hasattr(agent, '_turn_end_hooks'): agent._turn_end_hooks = {}
|
||||||
def _pet_hook(ctx):
|
def _pet_hook(ctx):
|
||||||
|
|||||||
+30
-20
@@ -94,9 +94,9 @@ class SiderLLMSession:
|
|||||||
def __init__(self, cfg):
|
def __init__(self, cfg):
|
||||||
from sider_ai_api import Session # 不使用sider的话没必要安装这个包
|
from sider_ai_api import Session # 不使用sider的话没必要安装这个包
|
||||||
self._core = Session(cookie=cfg['apikey'], proxies=proxies)
|
self._core = Session(cookie=cfg['apikey'], proxies=proxies)
|
||||||
self.default_model = cfg.get('model', 'gemini-3.0-flash')
|
self.model = cfg.get('model', 'gemini-3.0-flash')
|
||||||
def ask(self, prompt, stream=False):
|
def ask(self, prompt, stream=False):
|
||||||
model = self.default_model
|
model = self.model
|
||||||
if len(prompt) > 28000:
|
if len(prompt) > 28000:
|
||||||
print(f"[Warn] Prompt too long ({len(prompt)} chars), truncating.")
|
print(f"[Warn] Prompt too long ({len(prompt)} chars), truncating.")
|
||||||
prompt = prompt[-28000:]
|
prompt = prompt[-28000:]
|
||||||
@@ -425,33 +425,46 @@ class BaseSession:
|
|||||||
def __init__(self, cfg):
|
def __init__(self, cfg):
|
||||||
self.api_key = cfg['apikey']
|
self.api_key = cfg['apikey']
|
||||||
self.api_base = cfg['apibase'].rstrip('/')
|
self.api_base = cfg['apibase'].rstrip('/')
|
||||||
self.default_model = cfg.get('model', '')
|
self.model = cfg.get('model', '')
|
||||||
self.context_win = cfg.get('context_win', 24000)
|
self.context_win = cfg.get('context_win', 24000)
|
||||||
self.history = []
|
self.history = []
|
||||||
self.lock = threading.Lock()
|
self.lock = threading.Lock()
|
||||||
self.system = ""
|
self.system = ""
|
||||||
self.name = cfg.get('name', self.default_model)
|
self.name = cfg.get('name', self.model)
|
||||||
proxy = cfg.get('proxy')
|
proxy = cfg.get('proxy')
|
||||||
self.proxies = {"http": proxy, "https": proxy} if proxy else None
|
self.proxies = {"http": proxy, "https": proxy} if proxy else None
|
||||||
self.max_retries = max(0, int(cfg.get('max_retries', 1)))
|
self.max_retries = max(0, int(cfg.get('max_retries', 1)))
|
||||||
self.connect_timeout = max(1, int(cfg.get('timeout', 5)))
|
self.connect_timeout = max(1, int(cfg.get('timeout', 5)))
|
||||||
self.read_timeout = max(5, int(cfg.get('read_timeout', 30)))
|
self.read_timeout = max(5, int(cfg.get('read_timeout', 30)))
|
||||||
effort = cfg.get('reasoning_effort')
|
def _enum(key, valid):
|
||||||
effort = None if effort is None else str(effort).strip().lower()
|
v = cfg.get(key); v = None if v is None else str(v).strip().lower()
|
||||||
self.reasoning_effort = effort if effort in ('none', 'minimal', 'low', 'medium', 'high', 'xhigh') else None
|
return v if not v or v in valid else print(f"[WARN] Invalid {key} {v!r}, ignored.")
|
||||||
if effort and not self.reasoning_effort: print(f"[WARN] Invalid reasoning_effort {effort!r}, ignored.")
|
self.reasoning_effort = _enum('reasoning_effort', {'none', 'minimal', 'low', 'medium', 'high', 'xhigh'})
|
||||||
|
self.thinking_type = _enum('thinking_type', {'adaptive', 'enabled', 'disabled'})
|
||||||
|
self.thinking_budget_tokens = cfg.get('thinking_budget_tokens')
|
||||||
mode = str(cfg.get('api_mode', 'chat_completions')).strip().lower().replace('-', '_')
|
mode = str(cfg.get('api_mode', 'chat_completions')).strip().lower().replace('-', '_')
|
||||||
self.api_mode = 'responses' if mode in ('responses', 'response') else 'chat_completions'
|
self.api_mode = 'responses' if mode in ('responses', 'response') else 'chat_completions'
|
||||||
self.temperature = cfg.get('temperature', 1.0)
|
self.temperature = cfg.get('temperature', 1.0)
|
||||||
self.max_tokens = cfg.get('max_tokens', 8192)
|
self.max_tokens = cfg.get('max_tokens', 8192)
|
||||||
|
def _apply_claude_thinking(self, payload):
|
||||||
|
if self.thinking_type:
|
||||||
|
thinking = {"type": self.thinking_type}
|
||||||
|
if self.thinking_type == 'enabled':
|
||||||
|
if self.thinking_budget_tokens is None: print("[WARN] thinking_type='enabled' requires thinking_budget_tokens, ignored.")
|
||||||
|
else:
|
||||||
|
thinking["budget_tokens"] = self.thinking_budget_tokens; payload["thinking"] = thinking
|
||||||
|
else: payload["thinking"] = thinking
|
||||||
|
if self.reasoning_effort:
|
||||||
|
effort = {'low': 'low', 'medium': 'medium', 'high': 'high', 'xhigh': 'max'}.get(self.reasoning_effort)
|
||||||
|
if effort: payload["output_config"] = {"effort": effort}
|
||||||
|
else: print(f"[WARN] reasoning_effort {self.reasoning_effort!r} is unsupported for Claude output_config.effort, ignored.")
|
||||||
def ask(self, prompt, stream=False):
|
def ask(self, prompt, stream=False):
|
||||||
def _ask_gen():
|
def _ask_gen():
|
||||||
content = ''
|
|
||||||
with self.lock:
|
with self.lock:
|
||||||
self.history.append({"role": "user", "content": [{"type": "text", "text": prompt}]})
|
self.history.append({"role": "user", "content": [{"type": "text", "text": prompt}]})
|
||||||
trim_messages_history(self.history, self.context_win)
|
trim_messages_history(self.history, self.context_win)
|
||||||
messages = self.make_messages(self.history)
|
messages = self.make_messages(self.history)
|
||||||
content_blocks = None
|
content_blocks = None; content = ''
|
||||||
gen = self.raw_ask(messages)
|
gen = self.raw_ask(messages)
|
||||||
try:
|
try:
|
||||||
while True: chunk = next(gen); content += chunk; yield chunk
|
while True: chunk = next(gen); content += chunk; yield chunk
|
||||||
@@ -466,10 +479,9 @@ class BaseSession:
|
|||||||
|
|
||||||
class ClaudeSession(BaseSession):
|
class ClaudeSession(BaseSession):
|
||||||
def raw_ask(self, messages):
|
def raw_ask(self, messages):
|
||||||
model = self.default_model
|
|
||||||
headers = {"x-api-key": self.api_key, "Content-Type": "application/json", "anthropic-version": "2023-06-01", "anthropic-beta": "prompt-caching-2024-07-31"}
|
headers = {"x-api-key": self.api_key, "Content-Type": "application/json", "anthropic-version": "2023-06-01", "anthropic-beta": "prompt-caching-2024-07-31"}
|
||||||
payload = {"model": model, "messages": messages, "temperature": self.temperature, "max_tokens": self.max_tokens, "stream": True}
|
payload = {"model": self.model, "messages": messages, "temperature": self.temperature, "max_tokens": self.max_tokens, "stream": True}
|
||||||
if self.reasoning_effort: payload["reasoning_effort"] = self.reasoning_effort
|
self._apply_claude_thinking(payload)
|
||||||
if self.system: payload["system"] = [{"type": "text", "text": self.system, "cache_control": {"type": "persistent"}}]
|
if self.system: payload["system"] = [{"type": "text", "text": self.system, "cache_control": {"type": "persistent"}}]
|
||||||
try:
|
try:
|
||||||
with requests.post(auto_make_url(self.api_base, "messages"), headers=headers, json=payload, stream=True, timeout=(self.connect_timeout, self.read_timeout)) as r:
|
with requests.post(auto_make_url(self.api_base, "messages"), headers=headers, json=payload, stream=True, timeout=(self.connect_timeout, self.read_timeout)) as r:
|
||||||
@@ -487,7 +499,7 @@ class ClaudeSession(BaseSession):
|
|||||||
|
|
||||||
class LLMSession(BaseSession):
|
class LLMSession(BaseSession):
|
||||||
def raw_ask(self, messages):
|
def raw_ask(self, messages):
|
||||||
return (yield from _openai_stream(self.api_base, self.api_key, messages, self.default_model, self.api_mode,
|
return (yield from _openai_stream(self.api_base, self.api_key, messages, self.model, self.api_mode,
|
||||||
temperature=self.temperature, reasoning_effort=self.reasoning_effort,
|
temperature=self.temperature, reasoning_effort=self.reasoning_effort,
|
||||||
max_tokens=self.max_tokens, max_retries=self.max_retries,
|
max_tokens=self.max_tokens, max_retries=self.max_retries,
|
||||||
connect_timeout=self.connect_timeout, read_timeout=self.read_timeout, proxies=self.proxies))
|
connect_timeout=self.connect_timeout, read_timeout=self.read_timeout, proxies=self.proxies))
|
||||||
@@ -521,7 +533,7 @@ class NativeClaudeSession(BaseSession):
|
|||||||
self.tools = None
|
self.tools = None
|
||||||
def raw_ask(self, messages):
|
def raw_ask(self, messages):
|
||||||
messages = _fix_messages(messages)
|
messages = _fix_messages(messages)
|
||||||
model = self.default_model
|
model = self.model
|
||||||
beta_parts = ["claude-code-20250219", "interleaved-thinking-2025-05-14", "redact-thinking-2026-02-12", "prompt-caching-scope-2026-01-05"]
|
beta_parts = ["claude-code-20250219", "interleaved-thinking-2025-05-14", "redact-thinking-2026-02-12", "prompt-caching-scope-2026-01-05"]
|
||||||
if "[1m]" in model.lower():
|
if "[1m]" in model.lower():
|
||||||
beta_parts.insert(1, "context-1m-2025-08-07"); model = model.replace("[1m]", "").replace("[1M]", "")
|
beta_parts.insert(1, "context-1m-2025-08-07"); model = model.replace("[1m]", "").replace("[1M]", "")
|
||||||
@@ -531,7 +543,7 @@ class NativeClaudeSession(BaseSession):
|
|||||||
if self.api_key.startswith("sk-ant-"): headers["x-api-key"] = self.api_key
|
if self.api_key.startswith("sk-ant-"): headers["x-api-key"] = self.api_key
|
||||||
else: headers["authorization"] = f"Bearer {self.api_key}"
|
else: headers["authorization"] = f"Bearer {self.api_key}"
|
||||||
payload = {"model": model, "messages": messages, "temperature": self.temperature, "max_tokens": self.max_tokens, "stream": True}
|
payload = {"model": model, "messages": messages, "temperature": self.temperature, "max_tokens": self.max_tokens, "stream": True}
|
||||||
if self.reasoning_effort: payload["reasoning_effort"] = self.reasoning_effort
|
self._apply_claude_thinking(payload)
|
||||||
payload["metadata"] = {"user_id": json.dumps({"device_id": self._device_id, "account_uuid": self._account_uuid, "session_id": self._session_id}, separators=(',', ':'))}
|
payload["metadata"] = {"user_id": json.dumps({"device_id": self._device_id, "account_uuid": self._account_uuid, "session_id": self._session_id}, separators=(',', ':'))}
|
||||||
if self.tools:
|
if self.tools:
|
||||||
claude_tools = openai_tools_to_claude(self.tools)
|
claude_tools = openai_tools_to_claude(self.tools)
|
||||||
@@ -587,7 +599,7 @@ class NativeOAISession(NativeClaudeSession):
|
|||||||
def raw_ask(self, messages):
|
def raw_ask(self, messages):
|
||||||
"""OpenAI streaming. yields text chunks, generator return = list[content_block]"""
|
"""OpenAI streaming. yields text chunks, generator return = list[content_block]"""
|
||||||
msgs = ([{"role": "system", "content": self.system}] if self.system else []) + _msgs_claude2oai(messages)
|
msgs = ([{"role": "system", "content": self.system}] if self.system else []) + _msgs_claude2oai(messages)
|
||||||
return (yield from _openai_stream(self.api_base, self.api_key, msgs, self.default_model, self.api_mode,
|
return (yield from _openai_stream(self.api_base, self.api_key, msgs, self.model, self.api_mode,
|
||||||
temperature=self.temperature, max_tokens=self.max_tokens,
|
temperature=self.temperature, max_tokens=self.max_tokens,
|
||||||
tools=self.tools, reasoning_effort=self.reasoning_effort,
|
tools=self.tools, reasoning_effort=self.reasoning_effort,
|
||||||
max_retries=self.max_retries, connect_timeout=self.connect_timeout,
|
max_retries=self.max_retries, connect_timeout=self.connect_timeout,
|
||||||
@@ -643,8 +655,6 @@ class ToolClient:
|
|||||||
_write_llm_log('Response', raw_text)
|
_write_llm_log('Response', raw_text)
|
||||||
return self._parse_mixed_response(raw_text)
|
return self._parse_mixed_response(raw_text)
|
||||||
|
|
||||||
#def _should_use_structured_messages(self, messages): return isinstance(self.backend, LLMSession) and any(isinstance(m.get("content"), list) for m in messages)
|
|
||||||
|
|
||||||
def _estimate_content_len(self, content):
|
def _estimate_content_len(self, content):
|
||||||
if isinstance(content, str): return len(content)
|
if isinstance(content, str): return len(content)
|
||||||
if isinstance(content, list):
|
if isinstance(content, list):
|
||||||
@@ -802,7 +812,7 @@ class MixinSession:
|
|||||||
import copy; self._sessions[0] = copy.copy(self._sessions[0])
|
import copy; self._sessions[0] = copy.copy(self._sessions[0])
|
||||||
self._orig_raw_asks = [s.raw_ask for s in self._sessions]
|
self._orig_raw_asks = [s.raw_ask for s in self._sessions]
|
||||||
self._sessions[0].raw_ask = self._raw_ask
|
self._sessions[0].raw_ask = self._raw_ask
|
||||||
self.default_model = getattr(self._sessions[0], 'default_model', None)
|
self.model = getattr(self._sessions[0], 'model', None)
|
||||||
self._cur_idx, self._switched_at = 0, 0.0
|
self._cur_idx, self._switched_at = 0, 0.0
|
||||||
def __getattr__(self, name): return getattr(self._sessions[0], name)
|
def __getattr__(self, name): return getattr(self._sessions[0], name)
|
||||||
_BROADCAST_ATTRS = frozenset({'system', 'tools', 'temperature', 'max_tokens', 'reasoning_effort'})
|
_BROADCAST_ATTRS = frozenset({'system', 'tools', 'temperature', 'max_tokens', 'reasoning_effort'})
|
||||||
|
|||||||
@@ -231,7 +231,7 @@ class TestMiniMaxLLMSessionConfig(unittest.TestCase):
|
|||||||
'read_timeout': 120,
|
'read_timeout': 120,
|
||||||
}
|
}
|
||||||
session = LLMSession(cfg)
|
session = LLMSession(cfg)
|
||||||
self.assertEqual(session.default_model, 'MiniMax-M2.7')
|
self.assertEqual(session.model, 'MiniMax-M2.7')
|
||||||
self.assertEqual(session.api_base, 'https://api.minimax.io/v1')
|
self.assertEqual(session.api_base, 'https://api.minimax.io/v1')
|
||||||
self.assertEqual(session.context_win, 50000)
|
self.assertEqual(session.context_win, 50000)
|
||||||
self.assertEqual(session.max_retries, 2)
|
self.assertEqual(session.max_retries, 2)
|
||||||
@@ -246,7 +246,7 @@ class TestMiniMaxLLMSessionConfig(unittest.TestCase):
|
|||||||
'model': 'MiniMax-M2.7-highspeed',
|
'model': 'MiniMax-M2.7-highspeed',
|
||||||
}
|
}
|
||||||
session = LLMSession(cfg)
|
session = LLMSession(cfg)
|
||||||
self.assertEqual(session.default_model, 'MiniMax-M2.7-highspeed')
|
self.assertEqual(session.model, 'MiniMax-M2.7-highspeed')
|
||||||
|
|
||||||
|
|
||||||
class TestMiniMaxNativeToolClientThinkTag(unittest.TestCase):
|
class TestMiniMaxNativeToolClientThinkTag(unittest.TestCase):
|
||||||
|
|||||||
Reference in new issue
Block a user