lm-deluge 0.0.67__py3-none-any.whl → 0.0.88__py3-none-any.whl

lm_deluge/models/anthropic.py

@@ -10,6 +10,19 @@ ANTHROPIC_MODELS = {
     #                                                                   ░███
     #                                                                   █████
     #
+    "claude-4.5-opus": {
+        "id": "claude-4.5-opus",
+        "name": "claude-opus-4-5-20251101",
+        "api_base": "https://api.anthropic.com/v1",
+        "api_key_env_var": "ANTHROPIC_API_KEY",
+        "supports_json": False,
+        "api_spec": "anthropic",
+        "input_cost": 5.0,
+        "cached_input_cost": 0.50,
+        "cache_write_cost": 6.25,
+        "output_cost": 25.0,
+        "reasoning_model": True,
+    },
     "claude-4.5-haiku": {
         "id": "claude-4.5-haiku",
         "name": "claude-haiku-4-5-20251001",
@@ -21,25 +34,27 @@ ANTHROPIC_MODELS = {
         "cached_input_cost": 0.10,
         "cache_write_cost": 1.25,
         "output_cost": 3.0,
+        "reasoning_model": True,
     },
     "claude-4.5-sonnet": {
         "id": "claude-4.5-sonnet",
         "name": "claude-sonnet-4-5-20250929",
         "api_base": "https://api.anthropic.com/v1",
         "api_key_env_var": "ANTHROPIC_API_KEY",
-        "supports_json": False,
+        "supports_json": True,
         "api_spec": "anthropic",
         "input_cost": 3.0,
         "cached_input_cost": 0.30,
         "cache_write_cost": 3.75,
         "output_cost": 15.0,
+        "reasoning_model": True,
     },
     "claude-4.1-opus": {
         "id": "claude-4.1-opus",
         "name": "claude-opus-4-1-20250805",
         "api_base": "https://api.anthropic.com/v1",
         "api_key_env_var": "ANTHROPIC_API_KEY",
-        "supports_json": False,
+        "supports_json": True,
         "api_spec": "anthropic",
         "input_cost": 15.0,
         "cached_input_cost": 1.50,

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/google.py

@@ -138,4 +138,33 @@ GOOGLE_MODELS = {
         "output_cost": 0.4,
         "reasoning_model": True,
     },
+    # Gemini 3 models - advanced reasoning with thought signatures
+    "gemini-3-pro-preview": {
+        "id": "gemini-3-pro-preview",
+        "name": "gemini-3-pro-preview",
+        "api_base": "https://generativelanguage.googleapis.com/v1alpha",
+        "api_key_env_var": "GEMINI_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "api_spec": "gemini",
+        "input_cost": 2.0,  # <200k tokens
+        "cached_input_cost": 0.5,  # estimated
+        "output_cost": 12.0,  # <200k tokens
+        # Note: >200k tokens pricing is $4/$18 per million
+        "reasoning_model": True,
+    },
+    # Gemini 2.5 Computer Use model
+    "gemini-2.5-computer-use": {
+        "id": "gemini-2.5-computer-use",
+        "name": "gemini-2.5-computer-use-preview-10-2025",
+        "api_base": "https://generativelanguage.googleapis.com/v1beta",
+        "api_key_env_var": "GEMINI_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "api_spec": "gemini",
+        "input_cost": 1.25,  # same as gemini-2.5-pro for now
+        "cached_input_cost": 0.31,
+        "output_cost": 10.0,
+        "reasoning_model": True,
+    },
 }

lm_deluge/models/grok.py

@@ -7,6 +7,30 @@ XAI_MODELS = {
     # 888    888 888    888  888 888888K
     # Y88b  d88P 888    Y88..88P 888 "88b
     #  "Y8888P88 888     "Y88P"  888  888
+    "grok-4.1-fast-reasoning": {
+        "id": "grok-4.1-fast-reasoning",
+        "name": "grok-4-1-fast-reasoning",
+        "api_base": "https://api.x.ai/v1",
+        "api_key_env_var": "GROK_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": True,
+        "api_spec": "openai",
+        "input_cost": 0.2,
+        "output_cost": 1.5,
+        "reasoning_model": False,
+    },
+    "grok-4.1-fast": {
+        "id": "grok-4.1-fast",
+        "name": "grok-4-1-fast-non-reasoning",
+        "api_base": "https://api.x.ai/v1",
+        "api_key_env_var": "GROK_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": True,
+        "api_spec": "openai",
+        "input_cost": 0.2,
+        "output_cost": 1.5,
+        "reasoning_model": False,
+    },
     "grok-code-fast-1": {
         "id": "grok-code-fast-1",
         "name": "grok-code-fast-1",

lm_deluge/models/kimi.py

@@ -0,0 +1,36 @@
+KIMI_MODELS = {
+    "kimi-k2": {
+        "id": "kimi-k2",
+        "name": "kimi-k2-0905-preview",
+        "api_base": "https://api.moonshot.ai/anthropic/v1",
+        "api_key_env_var": "MOONSHOT_API_KEY",
+        "supports_json": True,
+        "api_spec": "anthropic",
+    },
+    "kimi-k2-turbo": {
+        "id": "kimi-k2-turbo",
+        "name": "kimi-k2-turbo-preview",
+        "api_base": "https://api.moonshot.ai/anthropic/v1",
+        "api_key_env_var": "MOONSHOT_API_KEY",
+        "supports_json": True,
+        "api_spec": "anthropic",
+    },
+    "kimi-k2-thinking": {
+        "id": "kimi-k2-thinking",
+        "name": "kimi-k2-thinking",
+        "api_base": "https://api.moonshot.ai/anthropic/v1",
+        "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",
+        "name": "kimi-k2-thinking-turbo",
+        "api_base": "https://api.moonshot.ai/anthropic/v1",
+        "api_key_env_var": "MOONSHOT_API_KEY",
+        "supports_json": True,
+        "api_spec": "anthropic",
+        "reasoning_model": True,
+    },
+}

lm_deluge/models/minimax.py

@@ -0,0 +1,10 @@
+MINIMAX_MODELS = {
+    "minimax-m2": {
+        "id": "minimax-m2",
+        "name": "MiniMax-M2",
+        "api_base": "https://api.minimax.io/anthropic/v1",
+        "api_key_env_var": "MINIMAX_API_KEY",
+        "supports_json": False,
+        "api_spec": "anthropic",
+    }
+}

lm_deluge/models/openai.py

@@ -10,6 +10,78 @@ OPENAI_MODELS = {
     #                ░███
     #                █████
     #               ░░░░░
+    "gpt-5.2": {
+        "id": "gpt-5.2",
+        "name": "gpt-5.2",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 1.75,
+        "cached_input_cost": 0.175,
+        "output_cost": 14.0,
+        "reasoning_model": True,
+        "supports_xhigh": True,
+    },
+    "gpt-5.1-codex-max": {
+        "id": "gpt-5.1-codex-max",
+        "name": "gpt-5.1-codex-max",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 1.25,
+        "cached_input_cost": 0.125,
+        "output_cost": 10.0,
+        "reasoning_model": True,
+        "supports_xhigh": True,
+    },
+    "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": True,
+        "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",
@@ -107,6 +179,34 @@ OPENAI_MODELS = {
         "output_cost": 6.0,
         "reasoning_model": True,
     },
+    "o4-mini-deep-research": {
+        "id": "o4-mini-deep-research",
+        "name": "o4-mini-deep-research",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 2,
+        "cached_input_cost": 0.5,
+        "output_cost": 8.0,
+        "reasoning_model": True,
+    },
+    "o3-deep-research": {
+        "id": "o3-deep-research",
+        "name": "o3-deep-research",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 10,
+        "cached_input_cost": 2.50,
+        "output_cost": 40.0,
+        "reasoning_model": True,
+    },
     "o3": {
         "id": "o3",
         "name": "o3-2025-04-16",

lm_deluge/models/openrouter.py

@@ -1,13 +1,25 @@
 OPENROUTER_MODELS = {
+    "intellect-3-openrouter": {
+        "id": "intellect-3-openrouter",
+        "name": "prime-intellect/intellect-3",
+        "api_base": "https://openrouter.ai/api/v1",
+        "api_key_env_var": "OPENROUTER_API_KEY",
+        "supports_json": True,
+        "api_spec": "openai",
+        "input_cost": 0.2,
+        "cached_input_cost": 0.2,
+        "cache_write_cost": 0.2,
+        "output_cost": 1.10,
+    },
     "glm-4.6-openrouter": {
         "id": "glm-4.6-openrouter",
-        "name": "z-ai/glm-4.6",
+        "name": "z-ai/glm-4.6:exacto",
         "api_base": "https://openrouter.ai/api/v1",
         "api_key_env_var": "OPENROUTER_API_KEY",
         "supports_json": True,
         "api_spec": "openai",
         "input_cost": 0.6,
-        "cached_input_cost": 0.11,
+        "cached_input_cost": 0.6,
         "cache_write_cost": 0.6,
         "output_cost": 2.20,
     },
@@ -35,9 +47,21 @@ OPENROUTER_MODELS = {
         "cache_write_cost": 0.23,
         "output_cost": 0.9,
     },
+    "deepseek-3.2-exp-openrouter": {
+        "id": "deepseek-3.2-exp-openrouter",
+        "name": "deepseek/deepseek-v3.2-exp",
+        "api_base": "https://openrouter.ai/api/v1",
+        "api_key_env_var": "OPENROUTER_API_KEY",
+        "supports_json": True,
+        "api_spec": "openai",
+        "input_cost": 0.27,
+        "cached_input_cost": 0.27,
+        "cache_write_cost": 0.27,
+        "output_cost": 0.4,
+    },
     "deepseek-3.2-openrouter": {
         "id": "deepseek-3.2-openrouter",
-        "name": "deepseek/deepseek-v3.2-exp",
+        "name": "deepseek/deepseek-v3.2",
         "api_base": "https://openrouter.ai/api/v1",
         "api_key_env_var": "OPENROUTER_API_KEY",
         "supports_json": True,
@@ -47,18 +71,72 @@ OPENROUTER_MODELS = {
         "cache_write_cost": 0.27,
         "output_cost": 0.4,
     },
-    # "gpt-oss-20b-openrouter": {},
-    # "gpt-oss-120b-openrouter": {},
+    "gpt-oss-20b-openrouter": {
+        "id": "gpt-oss-20b-openrouter",
+        "name": "openai/gpt-oss-20b",
+        "api_base": "https://openrouter.ai/api/v1",
+        "api_key_env_var": "OPENROUTER_API_KEY",
+        "supports_json": True,
+        "api_spec": "openai",
+        "input_cost": 0.04,
+        "cached_input_cost": 0.04,
+        "cache_write_cost": 0.04,
+        "output_cost": 0.18,
+    },
+    "gpt-oss-120b-openrouter": {
+        "id": "gpt-oss-120b-openrouter",
+        "name": "openai/gpt-oss-120b",
+        "api_base": "https://openrouter.ai/api/v1",
+        "api_key_env_var": "OPENROUTER_API_KEY",
+        "supports_json": True,
+        "api_spec": "openai",
+        "input_cost": 0.05,
+        "cached_input_cost": 0.05,
+        "cache_write_cost": 0.05,
+        "output_cost": 0.45,
+    },
     "kimi-k2-openrouter": {
         "id": "kimi-k2-openrouter",
-        "name": "z-ai/glm-4.6",
+        "name": "moonshotai/kimi-k2-0905:exacto",
+        "api_base": "https://openrouter.ai/api/v1",
+        "api_key_env_var": "OPENROUTER_API_KEY",
+        "supports_json": True,
+        "api_spec": "openai",
+        "input_cost": 1,
+        "cached_input_cost": 1,
+        "cache_write_cost": 1,
+        "output_cost": 3,
+    },
+    "kimi-k2-thinking-openrouter": {
+        "id": "kimi-k2-thinking-openrouter",
+        "name": "moonshotai/kimi-k2-thinking",
         "api_base": "https://openrouter.ai/api/v1",
         "api_key_env_var": "OPENROUTER_API_KEY",
         "supports_json": True,
         "api_spec": "openai",
         "input_cost": 0.6,
-        "cached_input_cost": 0.11,
+        "cached_input_cost": 0.6,
         "cache_write_cost": 0.6,
-        "output_cost": 2.20,
+        "output_cost": 2.5,
+    },
+    "olmo-3-32b-think-openrouter": {
+        "id": "olmo-3-32b-think-openrouter",
+        "name": "allenai/olmo-3-32b-think",
+        "api_base": "https://openrouter.ai/api/v1",
+        "api_key_env_var": "OPENROUTER_API_KEY",
+        "supports_json": True,
+        "api_spec": "openai",
+        "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/pipelines/gepa/__init__.py

@@ -0,0 +1,95 @@
+"""
+GEPA (Genetic Pareto) prompt optimizer for lm-deluge.
+
+This module provides an evolutionary optimizer for text components in AI systems.
+It analyzes whole trajectories to propose improvements to prompts, tool descriptions,
+and other text-based configuration.
+
+Example usage:
+    from lm_deluge import LLMClient
+    from lm_deluge.prompt import Conversation, Message
+    from lm_deluge.pipelines.gepa import Component, EvalResult, optimize
+
+    # Define components to optimize
+    components = {
+        "system_prompt": Component(
+            description="Instructions given to the model",
+            value="You are a helpful assistant.",
+        ),
+    }
+
+    # Define how to evaluate one example
+    def evaluate(client: LLMClient, values: dict[str, str], example: dict) -> EvalResult:
+        # Build prompt with current component values
+        conv = Conversation.system(values["system_prompt"])
+        conv = conv.add(Message.user(example["question"]))
+
+        # Run inference
+        response = client.process_prompts_sync([conv], show_progress=False)[0]
+        answer = response.completion
+
+        # Score the result
+        correct = example["answer"].lower() in answer.lower()
+        score = 1.0 if correct else 0.0
+
+        # Build feedback for the proposer
+        feedback = f"Score: {score}. Expected: {example['answer']}"
+
+        # Return full trajectory
+        full_conv = conv.add(Message.ai(answer))
+        return EvalResult(conversation=full_conv, score=score, feedback=feedback)
+
+    # Run optimization
+    result = optimize(
+        components=components,
+        evaluate_fn=evaluate,
+        dataset=train_examples,
+        task_client=LLMClient("gpt-4o-mini"),
+        proposer_client=LLMClient("gpt-4o"),
+        max_iterations=50,
+    )
+
+    print(f"Best score: {result.best_score}")
+    print(f"Best prompt: {result.best_candidate['system_prompt']}")
+"""
+
+from lm_deluge.pipelines.gepa.core import (
+    Component,
+    EvalResult,
+    GEPAResult,
+    GEPAState,
+    Proposal,
+)
+from lm_deluge.pipelines.gepa.optimizer import GEPAEngine, optimize
+from lm_deluge.pipelines.gepa.proposer import (
+    DEFAULT_PROPOSAL_PROMPT,
+    build_proposal_prompt,
+    parse_proposal_response,
+    propose_improvement_sync,
+)
+from lm_deluge.pipelines.gepa.util import (
+    extract_text_from_response,
+    format_components_for_prompt,
+    format_conversation_compact,
+)
+
+__all__ = [
+    # Core types
+    "Component",
+    "EvalResult",
+    "Proposal",
+    "GEPAState",
+    "GEPAResult",
+    # Main API
+    "optimize",
+    "GEPAEngine",
+    # Proposer utilities
+    "DEFAULT_PROPOSAL_PROMPT",
+    "build_proposal_prompt",
+    "parse_proposal_response",
+    "propose_improvement_sync",
+    # Formatting utilities
+    "format_conversation_compact",
+    "format_components_for_prompt",
+    "extract_text_from_response",
+]