mini-swe-agent 1.13.4__py3-none-any.whl → 1.14.0__py3-none-any.whl

{mini_swe_agent-1.13.4.dist-info → mini_swe_agent-1.14.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mini-swe-agent
-Version: 1.13.4
+Version: 1.14.0
 Summary: Nano SWE Agent - A simple AI software engineering agent
 Author-email: Kilian Lieret <kilian.lieret@posteo.de>, "Carlos E. Jimenez" <carlosej@princeton.edu>
 License: MIT License

{mini_swe_agent-1.13.4.dist-info → mini_swe_agent-1.14.0.dist-info}/RECORD

@@ -1,5 +1,5 @@
-mini_swe_agent-1.13.4.dist-info/licenses/LICENSE.md,sha256=D3luWPkdHAe7LBsdD4vzqDAXw6Xewb3G-uczss0uh1s,1094
-minisweagent/__init__.py,sha256=BrTgbkr33E2jWEByqcdU-bYDBm8jgXYtL17_H44Yb0A,2016
+mini_swe_agent-1.14.0.dist-info/licenses/LICENSE.md,sha256=D3luWPkdHAe7LBsdD4vzqDAXw6Xewb3G-uczss0uh1s,1094
+minisweagent/__init__.py,sha256=x76NxcXdWcajgoVsrPVO4EaR79r1Bnk3bF6wifAmcYo,2016
 minisweagent/__main__.py,sha256=FIyAOiw--c3FQ2g240FOM1FdL0lk_PxSpixu0pQ7WFo,194
 minisweagent/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 minisweagent/agents/__init__.py,sha256=cpjJLzg1IGxLM-tZpoMJV9S33ye13XtdBO0x7DU_Lrk,48
@@ -23,17 +23,17 @@ minisweagent/environments/singularity.py,sha256=HSwRTWef7cMCgBiGAh5DIrxW8HkZ9C9Z
 minisweagent/environments/extra/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 minisweagent/environments/extra/bubblewrap.py,sha256=G12Dm63N30qByfLb1SKNsI4G4gLyKBfomnOIsPqRNZk,3662
 minisweagent/environments/extra/swerex_docker.py,sha256=WPYbohT_vqTHkde9cxpbV6chLXCpLl0PDAcgMbZsV0M,1707
-minisweagent/models/__init__.py,sha256=hMRTK5yfn2vymiXecZsSAGusxsTvrnQQVm8ih0y_PNM,4081
-minisweagent/models/anthropic.py,sha256=D8nHvvbgzPjla0He8p0O9kaXASPWg1Sai0pHsAj_Yn8,855
-minisweagent/models/litellm_model.py,sha256=RSvNzK6ksOOcTEY-kM36KJql9-iSH29kq0cKYXj2mxE,3034
-minisweagent/models/openrouter_model.py,sha256=ugI3322pyUQ8wBOuMysPoKlp0oRAmR38a6GcdjE6qHI,3748
-minisweagent/models/portkey_model.py,sha256=IcFNz33AYeYgj0N9PGWV7QCn1fT8MfMWZx_OEzLaPsc,4548
+minisweagent/models/__init__.py,sha256=Ol95N3oEDlojz7IBw10A7xf5rKfxtjESyJuEWOwZyzo,4171
+minisweagent/models/anthropic.py,sha256=4p-LxQ_RYQUX1rBsffAj3T1bBb2uMRhA4IyKfDcMpgo,1517
+minisweagent/models/litellm_model.py,sha256=qAYsFBzZGPRGGQfbNCJXKH9eRpzmJKZ9ORTJHsOdhL0,3419
+minisweagent/models/openrouter_model.py,sha256=7SwS4Raw-i_gAM9AGJ8EWoTrnCP8MfHqWf7UZ0sb4ds,4095
+minisweagent/models/portkey_model.py,sha256=lARromtKwiXJtxxXScOkp9PCl3oS76QMQh_6DeehoiI,4895
 minisweagent/models/test_models.py,sha256=ItCA6ddntzkYA7dzSuUEaLMV-AE8TBuXBFP8CzpiO3U,1351
 minisweagent/models/extra/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 minisweagent/models/extra/roulette.py,sha256=SqLj_wz9Vkbxou7i9Ef4Uzmg_eheDouNySkkV7pm2Ys,2093
 minisweagent/models/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-minisweagent/models/utils/cache_control.py,sha256=mG9cE56HQaUwXfoqvXoH6LcbMV_G1vlEE1aBBpikXYg,1608
-minisweagent/models/utils/key_per_thread.py,sha256=Vlxt--rapNNYCgIHrMCu1WVAkuiVIhC_awbarkbnkZQ,644
+minisweagent/models/utils/cache_control.py,sha256=rAjW_FPWxG7bnuNe94CARAq9LKaTgFtSSp9zMQvssRs,2535
+minisweagent/models/utils/key_per_thread.py,sha256=4YZXATIw-Fozi7M-1i-wyjIBf-GtQM71kkOHxSPkwrE,748
 minisweagent/run/__init__.py,sha256=WIoYgHVl7iZF2YncrfV3IttupG6P5KogroKHKECka3A,38
 minisweagent/run/github_issue.py,sha256=35mZoPLc4JV6XXJKRv55lnuKbXf5lDftd51N89-x9J0,3192
 minisweagent/run/hello_world.py,sha256=erLnEwNmPFLxq3-8zyv66Vy1kIqMqQf97vISX7LrQXg,959
@@ -50,8 +50,8 @@ minisweagent/run/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3
 minisweagent/run/utils/save.py,sha256=2xd-UnUzI7Fr_AUZ5KEJ53Aa4kpuuGYxkLwyUcvqyMM,2503
 minisweagent/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 minisweagent/utils/log.py,sha256=ruDMNKMrVC9NPvCeHwO3QYz5jsVNUGQB2dRAEAPAWp8,996
-mini_swe_agent-1.13.4.dist-info/METADATA,sha256=H4fwYy_JfcF7-Q6Z1_GRTFEheiqfSxnA65cDA-A3WG4,14151
-mini_swe_agent-1.13.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-mini_swe_agent-1.13.4.dist-info/entry_points.txt,sha256=d1_yRbTaGjs1UXHa6JQK0sKDGBIVGm8oeW0k2kfbJgQ,182
-mini_swe_agent-1.13.4.dist-info/top_level.txt,sha256=zKF4t8bFpV87fdVABZt2Da-vnb4Vkh_CxkwQx5YT4Ew,13
-mini_swe_agent-1.13.4.dist-info/RECORD,,
+mini_swe_agent-1.14.0.dist-info/METADATA,sha256=_quEJiibMIZ98LVaarGvBumSvBXWihTaeelQMGF7Wf8,14151
+mini_swe_agent-1.14.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+mini_swe_agent-1.14.0.dist-info/entry_points.txt,sha256=d1_yRbTaGjs1UXHa6JQK0sKDGBIVGm8oeW0k2kfbJgQ,182
+mini_swe_agent-1.14.0.dist-info/top_level.txt,sha256=zKF4t8bFpV87fdVABZt2Da-vnb4Vkh_CxkwQx5YT4Ew,13
+mini_swe_agent-1.14.0.dist-info/RECORD,,

minisweagent/__init__.py

@@ -8,7 +8,7 @@ This file provides:
   unless you want the static type checking.
 """
 
-__version__ = "1.13.4"
+__version__ = "1.14.0"
 
 import os
 from pathlib import Path

minisweagent/models/__init__.py

@@ -55,6 +55,13 @@ def get_model(input_model_name: str | None = None, config: dict | None = None) -
     if (from_env := os.getenv("MSWEA_MODEL_API_KEY")) and not str(type(model_class)).endswith("DeterministicModel"):
         config.setdefault("model_kwargs", {})["api_key"] = from_env
 
+    if (
+        any(s in resolved_model_name.lower() for s in ["anthropic", "sonnet", "opus", "claude"])
+        and "set_cache_control" not in config
+    ):
+        # Select cache control for Anthropic models by default
+        config["set_cache_control"] = "default_end"
+
     return model_class(**config)
 
 
@@ -98,11 +105,6 @@ def get_model_class(model_name: str, model_class: str = "") -> type:
             msg = f"Unknown model class: {model_class} (resolved to {full_path}, available: {_MODEL_CLASS_MAPPING})"
             raise ValueError(msg)
 
-    if any(s in model_name.lower() for s in ["anthropic", "sonnet", "opus", "claude"]):
-        from minisweagent.models.anthropic import AnthropicModel
-
-        return AnthropicModel
-
     # Default to LitellmModel
     from minisweagent.models.litellm_model import LitellmModel
 

minisweagent/models/anthropic.py

@@ -1,19 +1,35 @@
 import os
+import warnings
+from typing import Literal
 
-from minisweagent.models.litellm_model import LitellmModel
+from minisweagent.models.litellm_model import LitellmModel, LitellmModelConfig
 from minisweagent.models.utils.cache_control import set_cache_control
 from minisweagent.models.utils.key_per_thread import get_key_per_thread
 
 
+class AnthropicModelConfig(LitellmModelConfig):
+    set_cache_control: Literal["default_end"] | None = "default_end"
+    """Set explicit cache control markers, for example for Anthropic models"""
+
+
 class AnthropicModel(LitellmModel):
-    """For the use of anthropic models, we need to add explicit cache control marks
-    to the messages or we lose out on the benefits of the cache.
-    Because break points are limited per key, we also need to rotate between different keys
-    if running with multiple agents in parallel threads.
+    """This class is now only a thin wrapper around the LitellmModel class.
+    It is largely kept for backwards compatibility.
+    It will not be selected by `get_model` and `get_model_class` unless explicitly specified.
     """
 
+    def __init__(self, *, config_class: type = AnthropicModelConfig, **kwargs):
+        super().__init__(config_class=config_class, **kwargs)
+
     def query(self, messages: list[dict], **kwargs) -> dict:
         api_key = None
+        # Legacy only
         if rotating_keys := os.getenv("ANTHROPIC_API_KEYS"):
+            warnings.warn(
+                "ANTHROPIC_API_KEYS is deprecated and will be removed in the future. "
+                "Simply use the ANTHROPIC_API_KEY environment variable instead. "
+                "Key rotation is no longer required."
+            )
             api_key = get_key_per_thread(rotating_keys.split("::"))
-        return super().query(set_cache_control(messages), api_key=api_key, **kwargs)
+        messages = set_cache_control(messages, mode="default_end")
+        return super().query(messages, api_key=api_key, **kwargs)

minisweagent/models/litellm_model.py

@@ -3,7 +3,7 @@ import logging
 import os
 from dataclasses import asdict, dataclass, field
 from pathlib import Path
-from typing import Any
+from typing import Any, Literal
 
 import litellm
 from tenacity import (
@@ -15,6 +15,7 @@ from tenacity import (
 )
 
 from minisweagent.models import GLOBAL_MODEL_STATS
+from minisweagent.models.utils.cache_control import set_cache_control
 
 logger = logging.getLogger("litellm_model")
 
@@ -24,11 +25,13 @@ class LitellmModelConfig:
     model_name: str
     model_kwargs: dict[str, Any] = field(default_factory=dict)
     litellm_model_registry: Path | str | None = os.getenv("LITELLM_MODEL_REGISTRY_PATH")
+    set_cache_control: Literal["default_end"] | None = None
+    """Set explicit cache control markers, for example for Anthropic models"""
 
 
 class LitellmModel:
-    def __init__(self, **kwargs):
-        self.config = LitellmModelConfig(**kwargs)
+    def __init__(self, *, config_class: type = LitellmModelConfig, **kwargs):
+        self.config = config_class(**kwargs)
         self.cost = 0.0
         self.n_calls = 0
         if self.config.litellm_model_registry and Path(self.config.litellm_model_registry).is_file():
@@ -60,6 +63,8 @@ class LitellmModel:
             raise e
 
     def query(self, messages: list[dict[str, str]], **kwargs) -> dict:
+        if self.config.set_cache_control:
+            messages = set_cache_control(messages, mode=self.config.set_cache_control)
         response = self._query(messages, **kwargs)
         try:
             cost = litellm.cost_calculator.completion_cost(response)

minisweagent/models/openrouter_model.py

@@ -2,7 +2,7 @@ import json
 import logging
 import os
 from dataclasses import asdict, dataclass, field
-from typing import Any
+from typing import Any, Literal
 
 import requests
 from tenacity import (
@@ -14,6 +14,7 @@ from tenacity import (
 )
 
 from minisweagent.models import GLOBAL_MODEL_STATS
+from minisweagent.models.utils.cache_control import set_cache_control
 
 logger = logging.getLogger("openrouter_model")
 
@@ -22,6 +23,8 @@ logger = logging.getLogger("openrouter_model")
 class OpenRouterModelConfig:
     model_name: str
     model_kwargs: dict[str, Any] = field(default_factory=dict)
+    set_cache_control: Literal["default_end"] | None = None
+    """Set explicit cache control markers, for example for Anthropic models"""
 
 
 class OpenRouterAPIError(Exception):
@@ -90,6 +93,8 @@ class OpenRouterModel:
             raise OpenRouterAPIError(f"Request failed: {e}") from e
 
     def query(self, messages: list[dict[str, str]], **kwargs) -> dict:
+        if self.config.set_cache_control:
+            messages = set_cache_control(messages, mode=self.config.set_cache_control)
         response = self._query(messages, **kwargs)
 
         # Extract cost from usage information

minisweagent/models/portkey_model.py

@@ -3,7 +3,7 @@ import logging
 import os
 from dataclasses import asdict, dataclass, field
 from pathlib import Path
-from typing import Any
+from typing import Any, Literal
 
 import litellm
 from tenacity import (
@@ -15,6 +15,7 @@ from tenacity import (
 )
 
 from minisweagent.models import GLOBAL_MODEL_STATS
+from minisweagent.models.utils.cache_control import set_cache_control
 
 logger = logging.getLogger("portkey_model")
 
@@ -37,6 +38,8 @@ class PortkeyModelConfig:
     doesn't match the Portkey model name.
     Note that this might change if we get better support for Portkey and change how we calculate costs.
     """
+    set_cache_control: Literal["default_end"] | None = None
+    """Set explicit cache control markers, for example for Anthropic models"""
 
 
 class PortkeyModel:
@@ -85,6 +88,8 @@ class PortkeyModel:
         )
 
     def query(self, messages: list[dict[str, str]], **kwargs) -> dict:
+        if self.config.set_cache_control:
+            messages = set_cache_control(messages, mode=self.config.set_cache_control)
         response = self._query(messages, **kwargs)
         response_for_cost_calc = response.model_copy()
         if self.config.litellm_model_name_override:

minisweagent/models/utils/cache_control.py

@@ -1,3 +1,8 @@
+import copy
+import warnings
+from typing import Literal
+
+
 def _get_content_text(entry: dict) -> str:
     if isinstance(entry["content"], str):
         return entry["content"]
@@ -29,14 +34,39 @@ def _set_cache_control(entry: dict) -> None:
         entry["cache_control"] = {"type": "ephemeral"}
 
 
-def set_cache_control(messages: list[dict], last_n_messages_offset: int = 0) -> list[dict]:
+def set_cache_control(
+    messages: list[dict], *, mode: Literal["default_end"] | None = "default_end", last_n_messages_offset: int = 0
+) -> list[dict]:
     """This messages processor adds manual cache control marks to the messages."""
-    new_messages = []
-    n_tagged = 0
-    for i_entry, entry in enumerate(reversed(messages)):
+    if mode != "default_end":
+        raise ValueError(f"Invalid mode: {mode}")
+    if last_n_messages_offset:
+        warnings.warn("last_n_messages_offset is deprecated and will be removed in the future")
+    messages = copy.deepcopy(messages)
+
+    # Find all user messages
+    user_message_indices = []
+    for i, msg in enumerate(messages):
+        if msg["role"] == "user":
+            user_message_indices.append(i)
+
+    # Clear cache control from all messages first
+    for entry in messages:
         _clear_cache_control(entry)
-        if n_tagged < 2 and entry["role"] in ["user"] and i_entry >= last_n_messages_offset:
-            _set_cache_control(entry)
-            n_tagged += 1
-        new_messages.append(entry)
-    return list(reversed(new_messages))
+
+    # Add cache control to user messages, respecting offset
+    if user_message_indices:
+        # If offset is specified, exclude the last N messages from getting cache control
+        messages_to_cache = user_message_indices
+        if last_n_messages_offset > 0:
+            messages_to_cache = (
+                user_message_indices[:-last_n_messages_offset]
+                if last_n_messages_offset < len(user_message_indices)
+                else []
+            )
+
+        # Add cache control to the selected user messages
+        for idx in messages_to_cache:
+            _set_cache_control(messages[idx])
+
+    return messages

minisweagent/models/utils/key_per_thread.py

@@ -3,6 +3,7 @@ agents to not mess up prompt caching.
 """
 
 import threading
+import warnings
 from typing import Any
 
 _THREADS_THAT_USED_API_KEYS: list[Any] = []
@@ -10,6 +11,7 @@ _THREADS_THAT_USED_API_KEYS: list[Any] = []
 
 def get_key_per_thread(api_keys: list[Any]) -> Any:
     """Choose key based on thread name. Returns None if no keys are available."""
+    warnings.warn("get_key_per_thread is deprecated and will be removed in the future")
     thread_name = threading.current_thread().name
     if thread_name not in _THREADS_THAT_USED_API_KEYS:
         _THREADS_THAT_USED_API_KEYS.append(thread_name)