@=j dZddlmZddlmZddlmZddlmZGddeZ e dd d d d Z ee dS)uOllama Cloud provider profile. Ollama Cloud's OpenAI-compatible ``/v1/chat/completions`` endpoint supports top-level ``reasoning_effort`` with values ``none``, ``low``, ``medium``, ``high``, and ``max`` (the last being undocumented but empirically confirmed for DeepSeek V4 — ``max`` produces ~2.5× more thinking tokens than ``high``). This profile maps Hermes's ``xhigh`` → ``max`` to unlock DeepSeek V4's "Max thinking" tier through Ollama Cloud. ``low`` / ``medium`` / ``high`` pass through unchanged. When reasoning is explicitly disabled (``enabled: false`` or ``effort: "none"``), ``reasoning_effort`` is omitted entirely so the model runs in non-thinking mode. ) annotations)Any)register_provider)ProviderProfilec eZdZdZddd d ZdS) OllamaCloudProfileuAOllama Cloud — maps xhigh→max via top-level reasoning_effort.N)reasoning_configr dict | Nonectxrreturn%tuple[dict[str, Any], dict[str, Any]]c @i}|rt|tr|dd}|duriifS|dpd}|siifS|dkriifS|dvrd|d <n|d vr||d <n||d <i|fS) uEmit top-level ``reasoning_effort`` for Ollama Cloud. The ``supports_reasoning`` flag passed by the transport is deliberately ignored — this profile always handles reasoning when ``reasoning_config`` is present. enabledTFeffortnone)xhighmaxrreasoning_effort)lowmediumhigh) isinstancedictgetstriplower)selfr r top_levelrrs S/home/rurouni/.hermes/hermes-agent/plugins/model-providers/ollama-cloud/__init__.pybuild_api_kwargs_extrasz*OllamaCloudProfile.build_api_kwargs_extrass%'  7 +;T B B 7&**9d;;G%2v &**844:AACCIIKKF 2v 2v )))05 ,--44406 ,--17 ,-9})r r r rr r )__name__ __module__ __qualname____doc__r!r"r rrs=KK )-!!!!!!!!r"rz ollama-cloud) ollama_cloudznemotron-3-nano:30b)OLLAMA_API_KEYzhttps://ollama.com/v1)namealiasesdefault_aux_modelenv_varsbase_urlN) r& __future__rtypingr providersrproviders.baserrr(r'r"r r3s"#"""""''''''******$$$$$$$$N"!  + $  ,r"