lm-deluge 0.0.81__py3-none-any.whl → 0.0.83__py3-none-any.whl

lm_deluge/api_requests/anthropic.py

@@ -101,11 +101,14 @@ def _build_anthropic_request(
             request_json["max_tokens"] += budget
         else:
             request_json["thinking"] = {"type": "disabled"}
+            if "kimi" in model.id and "thinking" in model.id:
+                maybe_warn("WARN_KIMI_THINKING_NO_REASONING")
 
     else:
         request_json["thinking"] = {"type": "disabled"}
         if sampling_params.reasoning_effort:
             print("ignoring reasoning_effort for non-reasoning model")
+
     if system_message is not None:
         request_json["system"] = system_message
 
@@ -231,6 +234,9 @@ class AnthropicRequest(APIRequestBase):
                 data = await http_response.json()
                 response_content = data["content"]
 
+                # print("=== CONTENT ===")
+                # print(response_content)
+
                 # Parse response into Message with parts
                 parts = []
                 for item in response_content:

lm_deluge/client.py

@@ -84,6 +84,8 @@ class _LLMClient(BaseModel):
     json_mode: bool = False
     max_new_tokens: int = 512
     reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None
+    global_effort: Literal["low", "medium", "high"] | None = None
+    thinking_budget: int | None = None
     logprobs: bool = False
     top_logprobs: int | None = None
     force_local_mcp: bool = False
@@ -103,6 +105,11 @@ class _LLMClient(BaseModel):
     _tracker: StatusTracker | None = PrivateAttr(default=None)
     _capacity_lock: asyncio.Lock = PrivateAttr(default_factory=asyncio.Lock)
 
+    # usage
+    def print_usage(self):
+        if self._tracker:
+            self._tracker.log_usage()
+
     # Progress management for queueing API
     def open(self, total: int | None = None, show_progress: bool = True):
         self._tracker = StatusTracker(
@@ -207,6 +214,8 @@ class _LLMClient(BaseModel):
                         json_mode=self.json_mode,
                         max_new_tokens=self.max_new_tokens,
                         reasoning_effort=self.reasoning_effort,
+                        global_effort=self.global_effort or "high",
+                        thinking_budget=self.thinking_budget,
                         logprobs=self.logprobs,
                         top_logprobs=self.top_logprobs,
                     )
@@ -342,6 +351,8 @@ class _LLMClient(BaseModel):
                     json_mode=data.get("json_mode", False),
                     max_new_tokens=data.get("max_new_tokens", 512),
                     reasoning_effort=data.get("reasoning_effort", None),
+                    global_effort=data.get("global_effort") or "high",
+                    thinking_budget=data.get("thinking_budget", None),
                     logprobs=data.get("logprobs", False),
                     top_logprobs=data.get("top_logprobs", None),
                 )
@@ -1072,6 +1083,8 @@ def LLMClient(
     json_mode: bool = False,
     max_new_tokens: int = 512,
     reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None,
+    global_effort: Literal["low", "medium", "high"] | None = None,
+    thinking_budget: int | None = None,
     logprobs: bool = False,
     top_logprobs: int | None = None,
     force_local_mcp: bool = False,
@@ -1101,6 +1114,8 @@ def LLMClient(
     json_mode: bool = False,
     max_new_tokens: int = 512,
     reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None,
+    global_effort: Literal["low", "medium", "high"] | None = None,
+    thinking_budget: int | None = None,
     logprobs: bool = False,
     top_logprobs: int | None = None,
     force_local_mcp: bool = False,
@@ -1129,6 +1144,8 @@ def LLMClient(
     json_mode: bool = False,
     max_new_tokens: int = 512,
     reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None,
+    global_effort: Literal["low", "medium", "high"] | None = None,
+    thinking_budget: int | None = None,
     logprobs: bool = False,
     top_logprobs: int | None = None,
     force_local_mcp: bool = False,
@@ -1169,6 +1186,8 @@ def LLMClient(
         json_mode=json_mode,
         max_new_tokens=max_new_tokens,
         reasoning_effort=reasoning_effort,
+        global_effort=global_effort,
+        thinking_budget=thinking_budget,
         logprobs=logprobs,
         top_logprobs=top_logprobs,
         force_local_mcp=force_local_mcp,

lm_deluge/models/__init__.py

@@ -4,9 +4,10 @@ import random
 from dataclasses import dataclass, field
 
 from ..request_context import RequestContext
+from .anthropic import ANTHROPIC_MODELS
 
 # Import and register all provider models
-from .anthropic import ANTHROPIC_MODELS
+from .arcee import ARCEE_MODELS
 from .bedrock import BEDROCK_MODELS
 from .cerebras import CEREBRAS_MODELS
 from .cohere import COHERE_MODELS
@@ -128,6 +129,7 @@ def register_model(
 # Register all models from all providers
 for model_dict in [
     ANTHROPIC_MODELS,
+    ARCEE_MODELS,
     BEDROCK_MODELS,
     COHERE_MODELS,
     DEEPSEEK_MODELS,

lm_deluge/models/arcee.py

@@ -0,0 +1,16 @@
+ARCEE_MODELS = {
+    "trinity-mini": {
+        "id": "trinity-mini",
+        "name": "trinity-mini",
+        "api_base": "https://api.arcee.ai/api/v1",
+        "api_key_env_var": "ARCEE_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "supports_responses": False,
+        "api_spec": "openai",
+        "input_cost": 0.045,
+        "cached_input_cost": 0.045,
+        "output_cost": 0.15,
+        "reasoning_model": True,
+    }
+}

lm_deluge/models/deepseek.py

@@ -12,8 +12,9 @@ DEEPSEEK_MODELS = {
         "api_base": "https://api.deepseek.com/v1",
         "api_key_env_var": "DEEPSEEK_API_KEY",
         "api_spec": "openai",
-        "input_cost": 0.27,
-        "output_cost": 1.10,
+        "input_cost": 0.28,
+        "cached_input_cost": 0.028,
+        "output_cost": 0.42,
     },
     "deepseek-r1": {
         "id": "deepseek-r1",
@@ -21,7 +22,38 @@ DEEPSEEK_MODELS = {
         "api_base": "https://api.deepseek.com/v1",
         "api_key_env_var": "DEEPSEEK_API_KEY",
         "api_spec": "openai",
-        "input_cost": 0.55,
-        "output_cost": 2.19,
+        "input_cost": 0.28,
+        "cached_input_cost": 0.028,
+        "output_cost": 0.42,
+    },
+    "deepseek-reasoner": {
+        "id": "deepseek-reasoner",
+        "name": "deepseek-reasoner",
+        "api_base": "https://api.deepseek.com/v1",
+        "api_key_env_var": "DEEPSEEK_API_KEY",
+        "api_spec": "openai",
+        "input_cost": 0.28,
+        "cached_input_cost": 0.028,
+        "output_cost": 0.42,
+    },
+    "deepseek-reasoner-anthropic-compat": {
+        "id": "deepseek-reasoner-anthropic-compat",
+        "name": "deepseek-reasoner",
+        "api_base": "https://api.deepseek.com/anthropic",
+        "api_key_env_var": "DEEPSEEK_API_KEY",
+        "api_spec": "anthropic",
+        "input_cost": 0.28,
+        "cached_input_cost": 0.028,
+        "output_cost": 0.42,
+    },
+    "deepseek-speciale": {
+        "id": "deepseek-speciale",
+        "name": "deepseek-reasoner",
+        "api_base": "https://api.deepseek.com/v3.2_speciale_expires_on_20251215/v1",
+        "api_key_env_var": "DEEPSEEK_API_KEY",
+        "api_spec": "openai",
+        "input_cost": 0.28,
+        "cached_input_cost": 0.028,
+        "output_cost": 0.42,
     },
 }

lm_deluge/models/kimi.py

@@ -22,6 +22,7 @@ KIMI_MODELS = {
         "api_key_env_var": "MOONSHOT_API_KEY",
         "supports_json": True,
         "api_spec": "anthropic",
+        "reasoning_model": True,
     },
     "kimi-k2-thinking-turbo": {
         "id": "kimi-k2-thinking-turbo",
@@ -30,5 +31,6 @@ KIMI_MODELS = {
         "api_key_env_var": "MOONSHOT_API_KEY",
         "supports_json": True,
         "api_spec": "anthropic",
+        "reasoning_model": True,
     },
 }

lm_deluge/models/openrouter.py

@@ -71,4 +71,14 @@ OPENROUTER_MODELS = {
         "input_cost": 0.2,
         "output_cost": 35,
     },
+    "trinity-mini-openrouter": {
+        "id": "trinity-mini-openrouter",
+        "name": "arcee-ai/trinity-mini:free",
+        "api_base": "https://openrouter.ai/api/v1",
+        "api_key_env_var": "OPENROUTER_API_KEY",
+        "supports_json": True,
+        "api_spec": "openai",
+        "input_cost": 0.045,
+        "output_cost": 0.15,
+    },
 }

lm_deluge/models/together.py

@@ -93,4 +93,15 @@ TOGETHER_MODELS = {
         "output_cost": 0.59,
         "reasoning_model": True,
     },
+    "trinity-mini-together": {
+        "id": "trinity-mini-together",
+        "name": "arcee-ai/trinity-mini",
+        "api_base": "https://api.together.xyz/v1",
+        "api_key_env_var": "TOGETHER_API_KEY",
+        "supports_json": False,
+        "api_spec": "openai",
+        "input_cost": 0.18,
+        "output_cost": 0.59,
+        "reasoning_model": True,
+    },
 }

lm_deluge/models/zai.py

@@ -0,0 +1 @@
+ZAI_MODELS = {}

lm_deluge/tool/prefab/__init__.py

@@ -8,8 +8,12 @@ from .batch_tool import BatchTool
 from .tool_search import ToolSearchTool
 from .otc import ToolComposer
 from .sandbox import DaytonaSandbox, ModalSandbox
+from .docs import DocsManager
+from .sheets import SheetsManager
+from .random import RandomTools
 from .subagents import SubAgentManager
 from .todos import TodoItem, TodoManager, TodoPriority, TodoStatus
+from .email import EmailManager
 
 __all__ = [
     "BatchTool",
@@ -26,4 +30,8 @@ __all__ = [
     "WorkspaceBackend",
     "ModalSandbox",
     "DaytonaSandbox",
+    "DocsManager",
+    "SheetsManager",
+    "RandomTools",
+    "EmailManager",
 ]