lm-deluge 0.0.71__py3-none-any.whl → 0.0.73__py3-none-any.whl

lm_deluge/api_requests/base.py

@@ -90,9 +90,32 @@ class APIRequestBase(ABC):
         start -> poll -> result style of request.
         """
         assert self.context.status_tracker, "no status tracker"
-        start_time = time.time()
+        poll_interval = 5.0
+        attempt_start = time.monotonic()
+        deadline = attempt_start + self.context.request_timeout
+        response_id: str | None = None
+        last_status: str | None = None
+
         async with aiohttp.ClientSession() as session:
-            last_status: str | None = None
+
+            async def cancel_response(reason: str) -> None:
+                nonlocal response_id
+                if not response_id:
+                    return
+                cancel_url = f"{self.url}/{response_id}/cancel"
+                try:
+                    async with session.post(
+                        url=cancel_url,
+                        headers=self.request_header,
+                    ) as cancel_response:
+                        cancel_response.raise_for_status()
+                    print(f"Background req {response_id} cancelled: {reason}")
+                except (
+                    Exception
+                ) as cancel_err:  # pragma: no cover - best effort logging
+                    print(
+                        f"Failed to cancel background req {response_id}: {cancel_err}"
+                    )
 
             try:
                 self.context.status_tracker.total_requests += 1
@@ -109,14 +132,11 @@ class APIRequestBase(ABC):
                     last_status = data["status"]
 
                 while True:
-                    if time.time() - start_time > self.context.request_timeout:
-                        # cancel the response
-                        async with session.post(
-                            url=f"{self.url}/{response_id}/cancel",
-                            headers=self.request_header,
-                        ) as http_response:
-                            http_response.raise_for_status()
-
+                    now = time.monotonic()
+                    remaining = deadline - now
+                    if remaining <= 0:
+                        elapsed = now - attempt_start
+                        await cancel_response(f"timed out after {elapsed:.1f}s")
                         return APIResponse(
                             id=self.context.task_id,
                             model_internal=self.context.model_name,
@@ -128,8 +148,9 @@ class APIRequestBase(ABC):
                             content=None,
                             usage=None,
                         )
+
                     # poll for the response
-                    await asyncio.sleep(5.0)
+                    await asyncio.sleep(min(poll_interval, max(remaining, 0)))
                     async with session.get(
                         url=f"{self.url}/{response_id}",
                         headers=self.request_header,
@@ -146,6 +167,8 @@ class APIRequestBase(ABC):
                             return await self.handle_response(http_response)
 
             except Exception as e:
+                if response_id:
+                    await cancel_response(f"errored: {type(e).__name__}")
                 raise_if_modal_exception(e)
                 tb = traceback.format_exc()
                 print(tb)

lm_deluge/api_requests/openai.py

@@ -67,10 +67,12 @@ async def _build_oa_chat_request(
                 effort = "minimal"
             else:
                 effort = "low"
-        if effort == "minimal" and "gpt-5" not in model.id:
-            print(
-                "WARNING: 'minimal' reasoning effort only allowed for gpt-5. setting to 'low'."
-            )
+        # GPT-5.1 models don't support 'minimal', they support 'none' instead
+        if effort == "minimal" and "gpt-5.1" in model.id:
+            maybe_warn("WARN_MINIMAL_TO_NONE", model_name=context.model_name)
+            effort = "none"
+        elif effort == "minimal" and "gpt-5" not in model.id:
+            maybe_warn("WARN_MINIMAL_TO_LOW", model_name=context.model_name)
             effort = "low"
         request_json["reasoning_effort"] = effort
     else:
@@ -271,16 +273,24 @@ async def _build_oa_responses_request(
         request_json["max_output_tokens"] = sampling_params.max_new_tokens
 
     if model.reasoning_model:
-        if sampling_params.reasoning_effort in [None, "none"]:
+        effort = sampling_params.reasoning_effort
+        if effort in [None, "none"]:
             # gemini models can switch reasoning off
             if "gemini" in model.id:
-                sampling_params.reasoning_effort = "none"
+                effort = "none"
             else:
-                sampling_params.reasoning_effort = "low"
+                effort = "low"
+        # GPT-5.1 models don't support 'minimal', they support 'none' instead
+        if effort == "minimal" and "gpt-5.1" in model.id:
+            maybe_warn("WARN_MINIMAL_TO_NONE", model_name=context.model_name)
+            effort = "none"
+        elif effort == "minimal" and "gpt-5" not in model.id:
+            maybe_warn("WARN_MINIMAL_TO_LOW", model_name=context.model_name)
+            effort = "low"
         request_json["temperature"] = 1.0
         request_json["top_p"] = 1.0
         request_json["reasoning"] = {
-            "effort": sampling_params.reasoning_effort,
+            "effort": effort,
             "summary": "auto",
         }
     else:

lm_deluge/client.py

@@ -44,10 +44,14 @@ class _LLMClient(BaseModel):
     Keeps all validation, serialization, and existing functionality.
     """
 
-    _REASONING_SUFFIXES: ClassVar[dict[str, Literal["low", "medium", "high"]]] = {
+    _REASONING_SUFFIXES: ClassVar[
+        dict[str, Literal["low", "medium", "high", "minimal", "none"]]
+    ] = {
         "-low": "low",
         "-medium": "medium",
         "-high": "high",
+        "-minimal": "minimal",
+        "-none": "none",
     }
 
     model_names: str | list[str] = ["gpt-4.1-mini"]
@@ -149,9 +153,11 @@ class _LLMClient(BaseModel):
 
     def _normalize_model_names(
         self, models: list[str]
-    ) -> tuple[list[str], list[Literal["low", "medium", "high"] | None]]:
+    ) -> tuple[
+        list[str], list[Literal["low", "medium", "high", "minimal", "none"] | None]
+    ]:
         normalized: list[str] = []
-        efforts: list[Literal["low", "medium", "high"] | None] = []
+        efforts: list[Literal["low", "medium", "high", "minimal", "none"] | None] = []
 
         for name in models:
             base_name = self._preprocess_openrouter_model(name)
@@ -164,7 +170,10 @@ class _LLMClient(BaseModel):
         return normalized, efforts
 
     def _align_sampling_params(
-        self, per_model_efforts: list[Literal["low", "medium", "high"] | None]
+        self,
+        per_model_efforts: list[
+            Literal["low", "medium", "high", "minimal", "none"] | None
+        ],
     ) -> None:
         if len(per_model_efforts) < len(self.model_names):
             per_model_efforts = per_model_efforts + [None] * (
@@ -332,7 +341,7 @@ class _LLMClient(BaseModel):
     @classmethod
     def _strip_reasoning_suffix_if_registered(
         cls, model_name: str
-    ) -> tuple[str, Literal["low", "medium", "high"] | None]:
+    ) -> tuple[str, Literal["low", "medium", "high", "minimal", "none"] | None]:
         """Remove reasoning suffix only when the trimmed model already exists."""
         for suffix, effort in cls._REASONING_SUFFIXES.items():
             if model_name.endswith(suffix) and len(model_name) > len(suffix):
@@ -364,6 +373,15 @@ class _LLMClient(BaseModel):
             assert (
                 self.use_responses_api
             ), "background mode only allowed for responses api"
+
+        # codex models require responses api
+        for model_name in self.model_names:
+            if "codex" in model_name.lower() and not self.use_responses_api:
+                raise ValueError(
+                    f"Model '{model_name}' requires use_responses_api=True. "
+                    "Codex models are only available via the Responses API."
+                )
+
         # Auto-generate name if not provided
         if self.name is None:
             if len(self.model_names) == 1:

lm_deluge/models/openai.py

@@ -10,6 +10,48 @@ OPENAI_MODELS = {
     #                ░███
     #                █████
     #               ░░░░░
+    "gpt-5.1": {
+        "id": "gpt-5.1",
+        "name": "gpt-5.1",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": False,
+        "supports_logprobs": True,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 1.25,
+        "cached_input_cost": 0.125,
+        "output_cost": 10.0,
+        "reasoning_model": True,
+    },
+    "gpt-5.1-codex": {
+        "id": "gpt-5.1-codex",
+        "name": "gpt-5.1-codex",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": False,
+        "supports_logprobs": True,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 1.25,
+        "cached_input_cost": 0.125,
+        "output_cost": 10.0,
+        "reasoning_model": True,
+    },
+    "gpt-5.1-codex-mini": {
+        "id": "gpt-5.1-codex-mini",
+        "name": "gpt-5.1-codex-mini",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": False,
+        "supports_logprobs": True,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 0.25,
+        "cached_input_cost": 0.025,
+        "output_cost": 2.0,
+        "reasoning_model": True,
+    },
     "gpt-5-codex": {
         "id": "gpt-5-codex",
         "name": "gpt-5-codex",

lm_deluge/warnings.py

@@ -7,6 +7,8 @@ WARNINGS: dict[str, str] = {
     "WARN_REASONING_UNSUPPORTED": "Ignoring reasoning_effort param for non-reasoning model: {model_name}.",
     "WARN_CACHING_UNSUPPORTED": "Cache parameter '{cache_param}' is not supported, ignoring for {model_name}.",
     "WARN_LOGPROBS_UNSUPPORTED": "Ignoring logprobs param for non-logprobs model: {model_name}",
+    "WARN_MINIMAL_TO_LOW": "'minimal' reasoning effort only allowed for gpt-5 models. Setting to 'low' for {model_name}.",
+    "WARN_MINIMAL_TO_NONE": "GPT-5.1 models don't support 'minimal' reasoning effort. Converting to 'none' for {model_name}.",
 }
 
 

{lm_deluge-0.0.71.dist-info → lm_deluge-0.0.73.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.71
+Version: 0.0.73
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10

{lm_deluge-0.0.71.dist-info → lm_deluge-0.0.73.dist-info}/RECORD

@@ -2,7 +2,7 @@ lm_deluge/__init__.py,sha256=zF5lAitfgJ8A28IXJ5BE9OUCqGOqSnGOWn3ZIlizNyY,822
 lm_deluge/batches.py,sha256=Km6QM5_7BlF2qEyo4WPlhkaZkpzrLqf50AaveHXQOoY,25127
 lm_deluge/cache.py,sha256=xO2AIYvP3tUpTMKQjwQQYfGRJSRi6e7sMlRhLjsS-u4,4873
 lm_deluge/cli.py,sha256=Ilww5gOw3J5v0NReq_Ra4hhxU4BCIJBl1oTGxJZKedc,12065
-lm_deluge/client.py,sha256=nBKuP6buwQYNMCP9f2SOuPkfowKRijJv4-bI-STg7Iw,40824
+lm_deluge/client.py,sha256=WOYYSJopBqN3SPlRorkkgapYnSUvo6CveDoPMfZz8QQ,41409
 lm_deluge/config.py,sha256=s3wFBRD6pi0wtXMJRmQDT2vdiqSvhjUPmLehbkv41i0,943
 lm_deluge/embed.py,sha256=CO-TOlC5kOTAM8lcnicoG4u4K664vCBwHF1vHa-nAGg,13382
 lm_deluge/errors.py,sha256=oHjt7YnxWbh-eXMScIzov4NvpJMo0-2r5J6Wh5DQ1tk,209
@@ -15,16 +15,16 @@ lm_deluge/rerank.py,sha256=-NBAJdHz9OB-SWWJnHzkFmeVO4wR6lFV7Vw-SxG7aVo,11457
 lm_deluge/tool.py,sha256=Kp2O5lDq_WVo_ASxjLQSHzVRbaxZkS6J0JIIskBjux0,28909
 lm_deluge/tracker.py,sha256=aeS9GUJpgOSQRVXAnGDvlMO8qYpSxpTNLYj2hrMg0m8,14757
 lm_deluge/usage.py,sha256=xz9tAw2hqaJvv9aAVhnQ6N1Arn7fS8Shb28VwCW26wI,5136
-lm_deluge/warnings.py,sha256=nlDJMCw30VhDEFxqLO2-bfXH_Tv5qmlglzUSbokCSw8,1498
+lm_deluge/warnings.py,sha256=xXXYXEfaaSVr__16BKOEEWLdfZi1L-2ylzTrXTRyO18,1748
 lm_deluge/api_requests/__init__.py,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
 lm_deluge/api_requests/anthropic.py,sha256=QGq3G5jJIGcoM2HdRt73GgkvZs4GOViyjYexWex05Vk,8927
-lm_deluge/api_requests/base.py,sha256=GCcydwBRx4_xAuYLvasXlyj-TgqvKAVhVvxRfJkvPbY,9471
+lm_deluge/api_requests/base.py,sha256=mXEM85mcU_5LD-ugELpCl28tv-tpHKcaxerTIVLQZVo,10436
 lm_deluge/api_requests/bedrock.py,sha256=Uppne03GcIEk1tVYzoGu7GXK2Sg94a_xvFTLDRN_phY,15412
 lm_deluge/api_requests/chat_reasoning.py,sha256=sJvstvKFqsSBUjYcwxzGt2_FH4cEp3Z6gKcBPyPjGwk,236
 lm_deluge/api_requests/common.py,sha256=BZ3vRO5TB669_UsNKugkkuFSzoLHOYJIKt4nV4sf4vc,422
 lm_deluge/api_requests/gemini.py,sha256=4uD7fQl0yWyAvYkPNi3oO1InBnvYfo5_QR6k-va-2GI,7838
 lm_deluge/api_requests/mistral.py,sha256=8JZP2CDf1XZfaPcTk0WS4q-VfYYj58ptpoH8LD3MQG4,4528
-lm_deluge/api_requests/openai.py,sha256=ezlGYNGHFvQGgs-xuxhDDeiEembHhVh_KqJBdRBqSlM,26038
+lm_deluge/api_requests/openai.py,sha256=ZqzQxs8CNUk757Q-1AfpKODkg5yPFHZjsqLm8bwjYDs,26584
 lm_deluge/api_requests/response.py,sha256=vG194gAH5p7ulpNy4qy5Pryfb1p3ZV21-YGoj__ru3E,7436
 lm_deluge/api_requests/deprecated/bedrock.py,sha256=WrcIShCoO8JCUSlFOCHxg6KQCNTZfw3TpYTvSpYk4mA,11320
 lm_deluge/api_requests/deprecated/cohere.py,sha256=KgDScD6_bWhAzOY5BHZQKSA3kurt4KGENqC4wLsGmcU,5142
@@ -58,7 +58,7 @@ lm_deluge/models/kimi.py,sha256=1voigLdNO2CxpWv0KDpQPP3Wolx5WrqgAlYL9ObJFuQ,1117
 lm_deluge/models/meta.py,sha256=BBgnscL1gMcIdPbRqrlDl_q9YAYGSrkw9JkAIabXtLs,1883
 lm_deluge/models/minimax.py,sha256=rwW9gNotAYfDVtMlqmSYegN6GoZM_9DSNNZU2yPOmaU,275
 lm_deluge/models/mistral.py,sha256=x67o5gckBGmPcIGdVbS26XZAYFKBYM4tsxEAahGp8bk,4323
-lm_deluge/models/openai.py,sha256=6J4eAt6Iu5RopokyldUQzRlviFBXBqhLqpVP5tztzqI,11074
+lm_deluge/models/openai.py,sha256=t6fcXo0YXgPQ6YiftZJP8gPw8FOBqoVapSavMVmtaOw,12411
 lm_deluge/models/openrouter.py,sha256=O-Po4tmHjAqFIVU96TUL0QnK01R4e2yDN7Z4sYJ-CuE,2120
 lm_deluge/models/together.py,sha256=AjKhPsazqBgqyLwHkNQW07COM1n_oSrYQRp2BFVvn9o,4381
 lm_deluge/presets/cerebras.py,sha256=MDkqj15qQRrj8wxSCDNNe_Cs7h1WN1UjV6lTmSY1olQ,479
@@ -69,8 +69,8 @@ lm_deluge/util/logprobs.py,sha256=UkBZakOxWluaLqHrjARu7xnJ0uCHVfLGHJdnYlEcutk,11
 lm_deluge/util/spatial.py,sha256=BsF_UKhE-x0xBirc-bV1xSKZRTUhsOBdGqsMKme20C8,4099
 lm_deluge/util/validation.py,sha256=hz5dDb3ebvZrZhnaWxOxbNSVMI6nmaOODBkk0htAUhs,1575
 lm_deluge/util/xml.py,sha256=Ft4zajoYBJR3HHCt2oHwGfymGLdvp_gegVmJ-Wqk4Ck,10547
-lm_deluge-0.0.71.dist-info/licenses/LICENSE,sha256=uNNXGXPCw2TC7CUs7SEBkA-Mz6QBQFWUUEWDMgEs1dU,1058
-lm_deluge-0.0.71.dist-info/METADATA,sha256=kgq3xiS7tMIbXpx5UkhCEA_yJAJvgGOPaie_ZlScTxQ,13514
-lm_deluge-0.0.71.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lm_deluge-0.0.71.dist-info/top_level.txt,sha256=hqU-TJX93yBwpgkDtYcXyLr3t7TLSCCZ_reytJjwBaE,10
-lm_deluge-0.0.71.dist-info/RECORD,,
+lm_deluge-0.0.73.dist-info/licenses/LICENSE,sha256=uNNXGXPCw2TC7CUs7SEBkA-Mz6QBQFWUUEWDMgEs1dU,1058
+lm_deluge-0.0.73.dist-info/METADATA,sha256=zr-cq_RyQNDDplrt6CUbVfGNpsVQEbACExRg5w6mE5M,13514
+lm_deluge-0.0.73.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lm_deluge-0.0.73.dist-info/top_level.txt,sha256=hqU-TJX93yBwpgkDtYcXyLr3t7TLSCCZ_reytJjwBaE,10
+lm_deluge-0.0.73.dist-info/RECORD,,