mini-swe-agent 1.13.4__tar.gz → 1.14.0__tar.gz

{mini_swe_agent-1.13.4/src/mini_swe_agent.egg-info → mini_swe_agent-1.14.0}/PKG-INFO

@@ -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 → mini_swe_agent-1.14.0/src/mini_swe_agent.egg-info}/PKG-INFO

@@ -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 → mini_swe_agent-1.14.0}/src/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

{mini_swe_agent-1.13.4 → mini_swe_agent-1.14.0}/src/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
 

mini_swe_agent-1.14.0/src/minisweagent/models/anthropic.py

@@ -0,0 +1,35 @@
+import os
+import warnings
+from typing import Literal
+
+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):
+    """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("::"))
+        messages = set_cache_control(messages, mode="default_end")
+        return super().query(messages, api_key=api_key, **kwargs)

{mini_swe_agent-1.13.4 → mini_swe_agent-1.14.0}/src/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)

{mini_swe_agent-1.13.4 → mini_swe_agent-1.14.0}/src/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

{mini_swe_agent-1.13.4 → mini_swe_agent-1.14.0}/src/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:

mini_swe_agent-1.14.0/src/minisweagent/models/utils/cache_control.py

@@ -0,0 +1,72 @@
+import copy
+import warnings
+from typing import Literal
+
+
+def _get_content_text(entry: dict) -> str:
+    if isinstance(entry["content"], str):
+        return entry["content"]
+    assert len(entry["content"]) == 1, "Expected single message in content"
+    return entry["content"][0]["text"]
+
+
+def _clear_cache_control(entry: dict) -> None:
+    if isinstance(entry["content"], list):
+        assert len(entry["content"]) == 1, "Expected single message in content"
+        entry["content"][0].pop("cache_control", None)
+    entry.pop("cache_control", None)
+
+
+def _set_cache_control(entry: dict) -> None:
+    if not isinstance(entry["content"], list):
+        entry["content"] = [  # type: ignore
+            {
+                "type": "text",
+                "text": _get_content_text(entry),
+                "cache_control": {"type": "ephemeral"},
+            }
+        ]
+    else:
+        entry["content"][0]["cache_control"] = {"type": "ephemeral"}
+    if entry["role"] == "tool":
+        # Workaround for weird bug
+        entry["content"][0].pop("cache_control", None)
+        entry["cache_control"] = {"type": "ephemeral"}
+
+
+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."""
+    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)
+
+    # 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

{mini_swe_agent-1.13.4 → mini_swe_agent-1.14.0}/src/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)

mini_swe_agent-1.13.4/src/minisweagent/models/anthropic.py

@@ -1,19 +0,0 @@
-import os
-
-from minisweagent.models.litellm_model import LitellmModel
-from minisweagent.models.utils.cache_control import set_cache_control
-from minisweagent.models.utils.key_per_thread import get_key_per_thread
-
-
-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.
-    """
-
-    def query(self, messages: list[dict], **kwargs) -> dict:
-        api_key = None
-        if rotating_keys := os.getenv("ANTHROPIC_API_KEYS"):
-            api_key = get_key_per_thread(rotating_keys.split("::"))
-        return super().query(set_cache_control(messages), api_key=api_key, **kwargs)

mini_swe_agent-1.13.4/src/minisweagent/models/utils/cache_control.py

@@ -1,42 +0,0 @@
-def _get_content_text(entry: dict) -> str:
-    if isinstance(entry["content"], str):
-        return entry["content"]
-    assert len(entry["content"]) == 1, "Expected single message in content"
-    return entry["content"][0]["text"]
-
-
-def _clear_cache_control(entry: dict) -> None:
-    if isinstance(entry["content"], list):
-        assert len(entry["content"]) == 1, "Expected single message in content"
-        entry["content"][0].pop("cache_control", None)
-    entry.pop("cache_control", None)
-
-
-def _set_cache_control(entry: dict) -> None:
-    if not isinstance(entry["content"], list):
-        entry["content"] = [  # type: ignore
-            {
-                "type": "text",
-                "text": _get_content_text(entry),
-                "cache_control": {"type": "ephemeral"},
-            }
-        ]
-    else:
-        entry["content"][0]["cache_control"] = {"type": "ephemeral"}
-    if entry["role"] == "tool":
-        # Workaround for weird bug
-        entry["content"][0].pop("cache_control", None)
-        entry["cache_control"] = {"type": "ephemeral"}
-
-
-def set_cache_control(messages: list[dict], 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)):
-        _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))