antfly-sdk 0.0.2.dev2__py3-none-any.whl

antfly/client_generated/models/inference_generate_request.py

@@ -0,0 +1,380 @@
+from __future__ import annotations
+
+from collections.abc import Mapping
+from typing import TYPE_CHECKING, Any, TypeVar
+
+from attrs import define as _attrs_define
+from attrs import field as _attrs_field
+
+from ..models.inference_generate_request_backend import InferenceGenerateRequestBackend
+from ..models.inference_generate_request_cache_dtype import InferenceGenerateRequestCacheDtype
+from ..models.inference_generate_request_compiled_target import InferenceGenerateRequestCompiledTarget
+from ..models.inference_generate_request_mode import InferenceGenerateRequestMode
+from ..models.inference_tool_choice_type_0 import InferenceToolChoiceType0
+from ..types import UNSET, Unset
+
+if TYPE_CHECKING:
+    from ..models.inference_chat_message import InferenceChatMessage
+    from ..models.inference_generate_response_format import InferenceGenerateResponseFormat
+    from ..models.inference_tool import InferenceTool
+    from ..models.inference_tool_choice_type_1 import InferenceToolChoiceType1
+
+
+T = TypeVar("T", bound="InferenceGenerateRequest")
+
+
+@_attrs_define
+class InferenceGenerateRequest:
+    """
+    Attributes:
+        model (str): Name of the generator model from models_dir/generators/ Example: google/gemma-3-1b-it.
+        messages (list[InferenceChatMessage]): Conversation messages (OpenAI-compatible format)
+        max_tokens (int | Unset): Maximum tokens to generate Default: 256. Example: 256.
+        temperature (float | Unset): Sampling temperature (0.0 = deterministic, higher = more random) Default: 1.0.
+            Example: 0.7.
+        top_p (float | Unset): Nucleus sampling probability Default: 1.0.
+        top_k (int | Unset): Top-k sampling (inference extension, not in OpenAI API) Default: 50.
+        stream (bool | Unset): If true, partial message deltas will be sent as SSE events Default: False.
+        tools (list[InferenceTool] | Unset): List of tools (functions) the model can call.
+            Only supported by models with tool_call_format configured.
+        min_p (float | Unset): Min-p sampling threshold. Filters tokens where p < min_p * max_p. Simpler alternative to
+            top_p. Default: 0.0.
+        repetition_penalty (float | Unset): Repetition penalty factor applied to previously generated tokens (1.0 =
+            disabled, >1.0 penalizes, <1.0 encourages) Default: 1.0.
+        frequency_penalty (float | Unset): Additive penalty based on token frequency in context (logit -=
+            frequency_penalty * count) Default: 0.0.
+        presence_penalty (float | Unset): Additive penalty for tokens that appeared at all in context (logit -=
+            presence_penalty if count > 0) Default: 0.0.
+        response_format (InferenceGenerateResponseFormat | Unset):
+        grammar (str | Unset): inference-native grammar override. When set, this takes precedence over
+            `response_format`.
+            Grammar-constrained decoding is currently native-backend only.
+        draft_model (str | Unset): inference-native speculative decoding extension. Path or model identifier for a
+            smaller draft model.
+        speculative_k (int | Unset): inference-native speculative decoding extension. Number of draft tokens proposed
+            per verification round.
+             Default: 4.
+        cache_dtype (InferenceGenerateRequestCacheDtype | Unset): inference-native KV cache quantization format. Lower
+            precision reduces memory usage but may
+            affect generation quality. Default auto-selects based on backend (f16 for GPU, f32 for CPU).
+        cache_compaction_ratio (float | Unset): inference-native KV cache compaction ratio applied after prefill via
+            Attention Matching.
+            Selects a subset of keys and fits new values via OLS to preserve attention behavior.
+            0.02 = 50x compression, 0.1 = 10x, 0.5 = 2x. Null/omitted = no compaction.
+        backend (InferenceGenerateRequestBackend | Unset): Optional backend override for this request.
+            `auto` keeps the node default behavior.
+            `onnx` forces ONNX generation when the model/package supports it.
+            `native`, `metal`, and `mlx` force the native host backend choice.
+            `xla` runs native generation with explicit PJRT/XLA compiled graph partitions and
+            requires a PJRT plugin path via `ANTFLY_INFERENCE_XLA_PLUGIN`,
+            `ANTFLY_INFERENCE_PJRT_PLUGIN`, `PJRT_PLUGIN_PATH`, or `PJRT_PLUGIN`.
+            `webgpu` selects the Wasm/WebGPU backend in Wasm builds; pair it with
+            `mode: "compiled"` to request WebGPU graph partition execution.
+        mode (InferenceGenerateRequestMode | Unset): inference-native graph execution mode. `eager` keeps the direct
+            runtime path when possible.
+            `compiled` runs inference graph planning, partitioning, and backend executor attachment.
+        compiled_target (InferenceGenerateRequestCompiledTarget | Unset): inference-native compiled graph target.
+            `partitioned` attaches compiled executors to eligible
+            graph partitions. `whole-model` requests a compiled backend only when it can own the full
+            traced graph shape.
+        tool_choice (InferenceToolChoiceType0 | InferenceToolChoiceType1 | Unset): Controls how the model uses tools.
+            Options:
+            - "auto": Model decides whether to call a tool (default)
+            - "none": Model will not call any tools
+            - "required": Model must call at least one tool
+            - object: Force a specific function to be called
+    """
+
+    model: str
+    messages: list[InferenceChatMessage]
+    max_tokens: int | Unset = 256
+    temperature: float | Unset = 1.0
+    top_p: float | Unset = 1.0
+    top_k: int | Unset = 50
+    stream: bool | Unset = False
+    tools: list[InferenceTool] | Unset = UNSET
+    min_p: float | Unset = 0.0
+    repetition_penalty: float | Unset = 1.0
+    frequency_penalty: float | Unset = 0.0
+    presence_penalty: float | Unset = 0.0
+    response_format: InferenceGenerateResponseFormat | Unset = UNSET
+    grammar: str | Unset = UNSET
+    draft_model: str | Unset = UNSET
+    speculative_k: int | Unset = 4
+    cache_dtype: InferenceGenerateRequestCacheDtype | Unset = UNSET
+    cache_compaction_ratio: float | Unset = UNSET
+    backend: InferenceGenerateRequestBackend | Unset = UNSET
+    mode: InferenceGenerateRequestMode | Unset = UNSET
+    compiled_target: InferenceGenerateRequestCompiledTarget | Unset = UNSET
+    tool_choice: InferenceToolChoiceType0 | InferenceToolChoiceType1 | Unset = UNSET
+    additional_properties: dict[str, Any] = _attrs_field(init=False, factory=dict)
+
+    def to_dict(self) -> dict[str, Any]:
+        model = self.model
+
+        messages = []
+        for messages_item_data in self.messages:
+            messages_item = messages_item_data.to_dict()
+            messages.append(messages_item)
+
+        max_tokens = self.max_tokens
+
+        temperature = self.temperature
+
+        top_p = self.top_p
+
+        top_k = self.top_k
+
+        stream = self.stream
+
+        tools: list[dict[str, Any]] | Unset = UNSET
+        if not isinstance(self.tools, Unset):
+            tools = []
+            for tools_item_data in self.tools:
+                tools_item = tools_item_data.to_dict()
+                tools.append(tools_item)
+
+        min_p = self.min_p
+
+        repetition_penalty = self.repetition_penalty
+
+        frequency_penalty = self.frequency_penalty
+
+        presence_penalty = self.presence_penalty
+
+        response_format: dict[str, Any] | Unset = UNSET
+        if not isinstance(self.response_format, Unset):
+            response_format = self.response_format.to_dict()
+
+        grammar = self.grammar
+
+        draft_model = self.draft_model
+
+        speculative_k = self.speculative_k
+
+        cache_dtype: str | Unset = UNSET
+        if not isinstance(self.cache_dtype, Unset):
+            cache_dtype = self.cache_dtype.value
+
+        cache_compaction_ratio = self.cache_compaction_ratio
+
+        backend: str | Unset = UNSET
+        if not isinstance(self.backend, Unset):
+            backend = self.backend.value
+
+        mode: str | Unset = UNSET
+        if not isinstance(self.mode, Unset):
+            mode = self.mode.value
+
+        compiled_target: str | Unset = UNSET
+        if not isinstance(self.compiled_target, Unset):
+            compiled_target = self.compiled_target.value
+
+        tool_choice: dict[str, Any] | str | Unset
+        if isinstance(self.tool_choice, Unset):
+            tool_choice = UNSET
+        elif isinstance(self.tool_choice, InferenceToolChoiceType0):
+            tool_choice = self.tool_choice.value
+        else:
+            tool_choice = self.tool_choice.to_dict()
+
+        field_dict: dict[str, Any] = {}
+        field_dict.update(self.additional_properties)
+        field_dict.update(
+            {
+                "model": model,
+                "messages": messages,
+            }
+        )
+        if max_tokens is not UNSET:
+            field_dict["max_tokens"] = max_tokens
+        if temperature is not UNSET:
+            field_dict["temperature"] = temperature
+        if top_p is not UNSET:
+            field_dict["top_p"] = top_p
+        if top_k is not UNSET:
+            field_dict["top_k"] = top_k
+        if stream is not UNSET:
+            field_dict["stream"] = stream
+        if tools is not UNSET:
+            field_dict["tools"] = tools
+        if min_p is not UNSET:
+            field_dict["min_p"] = min_p
+        if repetition_penalty is not UNSET:
+            field_dict["repetition_penalty"] = repetition_penalty
+        if frequency_penalty is not UNSET:
+            field_dict["frequency_penalty"] = frequency_penalty
+        if presence_penalty is not UNSET:
+            field_dict["presence_penalty"] = presence_penalty
+        if response_format is not UNSET:
+            field_dict["response_format"] = response_format
+        if grammar is not UNSET:
+            field_dict["grammar"] = grammar
+        if draft_model is not UNSET:
+            field_dict["draft_model"] = draft_model
+        if speculative_k is not UNSET:
+            field_dict["speculative_k"] = speculative_k
+        if cache_dtype is not UNSET:
+            field_dict["cache_dtype"] = cache_dtype
+        if cache_compaction_ratio is not UNSET:
+            field_dict["cache_compaction_ratio"] = cache_compaction_ratio
+        if backend is not UNSET:
+            field_dict["backend"] = backend
+        if mode is not UNSET:
+            field_dict["mode"] = mode
+        if compiled_target is not UNSET:
+            field_dict["compiled_target"] = compiled_target
+        if tool_choice is not UNSET:
+            field_dict["tool_choice"] = tool_choice
+
+        return field_dict
+
+    @classmethod
+    def from_dict(cls: type[T], src_dict: Mapping[str, Any]) -> T:
+        from ..models.inference_chat_message import InferenceChatMessage
+        from ..models.inference_generate_response_format import InferenceGenerateResponseFormat
+        from ..models.inference_tool import InferenceTool
+        from ..models.inference_tool_choice_type_1 import InferenceToolChoiceType1
+
+        d = dict(src_dict)
+        model = d.pop("model")
+
+        messages = []
+        _messages = d.pop("messages")
+        for messages_item_data in _messages:
+            messages_item = InferenceChatMessage.from_dict(messages_item_data)
+
+            messages.append(messages_item)
+
+        max_tokens = d.pop("max_tokens", UNSET)
+
+        temperature = d.pop("temperature", UNSET)
+
+        top_p = d.pop("top_p", UNSET)
+
+        top_k = d.pop("top_k", UNSET)
+
+        stream = d.pop("stream", UNSET)
+
+        _tools = d.pop("tools", UNSET)
+        tools: list[InferenceTool] | Unset = UNSET
+        if _tools is not UNSET:
+            tools = []
+            for tools_item_data in _tools:
+                tools_item = InferenceTool.from_dict(tools_item_data)
+
+                tools.append(tools_item)
+
+        min_p = d.pop("min_p", UNSET)
+
+        repetition_penalty = d.pop("repetition_penalty", UNSET)
+
+        frequency_penalty = d.pop("frequency_penalty", UNSET)
+
+        presence_penalty = d.pop("presence_penalty", UNSET)
+
+        _response_format = d.pop("response_format", UNSET)
+        response_format: InferenceGenerateResponseFormat | Unset
+        if isinstance(_response_format, Unset):
+            response_format = UNSET
+        else:
+            response_format = InferenceGenerateResponseFormat.from_dict(_response_format)
+
+        grammar = d.pop("grammar", UNSET)
+
+        draft_model = d.pop("draft_model", UNSET)
+
+        speculative_k = d.pop("speculative_k", UNSET)
+
+        _cache_dtype = d.pop("cache_dtype", UNSET)
+        cache_dtype: InferenceGenerateRequestCacheDtype | Unset
+        if isinstance(_cache_dtype, Unset):
+            cache_dtype = UNSET
+        else:
+            cache_dtype = InferenceGenerateRequestCacheDtype(_cache_dtype)
+
+        cache_compaction_ratio = d.pop("cache_compaction_ratio", UNSET)
+
+        _backend = d.pop("backend", UNSET)
+        backend: InferenceGenerateRequestBackend | Unset
+        if isinstance(_backend, Unset):
+            backend = UNSET
+        else:
+            backend = InferenceGenerateRequestBackend(_backend)
+
+        _mode = d.pop("mode", UNSET)
+        mode: InferenceGenerateRequestMode | Unset
+        if isinstance(_mode, Unset):
+            mode = UNSET
+        else:
+            mode = InferenceGenerateRequestMode(_mode)
+
+        _compiled_target = d.pop("compiled_target", UNSET)
+        compiled_target: InferenceGenerateRequestCompiledTarget | Unset
+        if isinstance(_compiled_target, Unset):
+            compiled_target = UNSET
+        else:
+            compiled_target = InferenceGenerateRequestCompiledTarget(_compiled_target)
+
+        def _parse_tool_choice(data: object) -> InferenceToolChoiceType0 | InferenceToolChoiceType1 | Unset:
+            if isinstance(data, Unset):
+                return data
+            try:
+                if not isinstance(data, str):
+                    raise TypeError()
+                componentsschemas_inference_tool_choice_type_0 = InferenceToolChoiceType0(data)
+
+                return componentsschemas_inference_tool_choice_type_0
+            except (TypeError, ValueError, AttributeError, KeyError):
+                pass
+            if not isinstance(data, dict):
+                raise TypeError()
+            componentsschemas_inference_tool_choice_type_1 = InferenceToolChoiceType1.from_dict(data)
+
+            return componentsschemas_inference_tool_choice_type_1
+
+        tool_choice = _parse_tool_choice(d.pop("tool_choice", UNSET))
+
+        inference_generate_request = cls(
+            model=model,
+            messages=messages,
+            max_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
+            stream=stream,
+            tools=tools,
+            min_p=min_p,
+            repetition_penalty=repetition_penalty,
+            frequency_penalty=frequency_penalty,
+            presence_penalty=presence_penalty,
+            response_format=response_format,
+            grammar=grammar,
+            draft_model=draft_model,
+            speculative_k=speculative_k,
+            cache_dtype=cache_dtype,
+            cache_compaction_ratio=cache_compaction_ratio,
+            backend=backend,
+            mode=mode,
+            compiled_target=compiled_target,
+            tool_choice=tool_choice,
+        )
+
+        inference_generate_request.additional_properties = d
+        return inference_generate_request
+
+    @property
+    def additional_keys(self) -> list[str]:
+        return list(self.additional_properties.keys())
+
+    def __getitem__(self, key: str) -> Any:
+        return self.additional_properties[key]
+
+    def __setitem__(self, key: str, value: Any) -> None:
+        self.additional_properties[key] = value
+
+    def __delitem__(self, key: str) -> None:
+        del self.additional_properties[key]
+
+    def __contains__(self, key: str) -> bool:
+        return key in self.additional_properties

antfly/client_generated/models/inference_generate_request_backend.py

@@ -0,0 +1,14 @@
+from enum import Enum
+
+
+class InferenceGenerateRequestBackend(str, Enum):
+    AUTO = "auto"
+    METAL = "metal"
+    MLX = "mlx"
+    NATIVE = "native"
+    ONNX = "onnx"
+    WEBGPU = "webgpu"
+    XLA = "xla"
+
+    def __str__(self) -> str:
+        return str(self.value)

antfly/client_generated/models/inference_generate_request_cache_dtype.py

@@ -0,0 +1,12 @@
+from enum import Enum
+
+
+class InferenceGenerateRequestCacheDtype(str, Enum):
+    F16 = "f16"
+    F32 = "f32"
+    FP8 = "fp8"
+    INT4 = "int4"
+    INT8 = "int8"
+
+    def __str__(self) -> str:
+        return str(self.value)

antfly/client_generated/models/inference_generate_request_compiled_target.py

@@ -0,0 +1,9 @@
+from enum import Enum
+
+
+class InferenceGenerateRequestCompiledTarget(str, Enum):
+    PARTITIONED = "partitioned"
+    WHOLE_MODEL = "whole-model"
+
+    def __str__(self) -> str:
+        return str(self.value)

antfly/client_generated/models/inference_generate_request_mode.py

@@ -0,0 +1,9 @@
+from enum import Enum
+
+
+class InferenceGenerateRequestMode(str, Enum):
+    COMPILED = "compiled"
+    EAGER = "eager"
+
+    def __str__(self) -> str:
+        return str(self.value)

antfly/client_generated/models/inference_generate_response.py

@@ -0,0 +1,120 @@
+from __future__ import annotations
+
+from collections.abc import Mapping
+from typing import TYPE_CHECKING, Any, TypeVar
+
+from attrs import define as _attrs_define
+from attrs import field as _attrs_field
+
+from ..models.inference_generate_response_object import InferenceGenerateResponseObject
+
+if TYPE_CHECKING:
+    from ..models.inference_generate_choice import InferenceGenerateChoice
+    from ..models.inference_generate_usage import InferenceGenerateUsage
+
+
+T = TypeVar("T", bound="InferenceGenerateResponse")
+
+
+@_attrs_define
+class InferenceGenerateResponse:
+    """OpenAI-compatible chat completion response
+
+    Attributes:
+        id (str): A unique identifier for the chat completion Example: chatcmpl-abc123.
+        object_ (InferenceGenerateResponseObject): The object type, always "chat.completion"
+        created (int): Unix timestamp (seconds) when the completion was created Example: 1704123456.
+        model (str): Model used for generation
+        choices (list[InferenceGenerateChoice]): List of completion choices (currently always 1)
+        usage (InferenceGenerateUsage):
+    """
+
+    id: str
+    object_: InferenceGenerateResponseObject
+    created: int
+    model: str
+    choices: list[InferenceGenerateChoice]
+    usage: InferenceGenerateUsage
+    additional_properties: dict[str, Any] = _attrs_field(init=False, factory=dict)
+
+    def to_dict(self) -> dict[str, Any]:
+        id = self.id
+
+        object_ = self.object_.value
+
+        created = self.created
+
+        model = self.model
+
+        choices = []
+        for choices_item_data in self.choices:
+            choices_item = choices_item_data.to_dict()
+            choices.append(choices_item)
+
+        usage = self.usage.to_dict()
+
+        field_dict: dict[str, Any] = {}
+        field_dict.update(self.additional_properties)
+        field_dict.update(
+            {
+                "id": id,
+                "object": object_,
+                "created": created,
+                "model": model,
+                "choices": choices,
+                "usage": usage,
+            }
+        )
+
+        return field_dict
+
+    @classmethod
+    def from_dict(cls: type[T], src_dict: Mapping[str, Any]) -> T:
+        from ..models.inference_generate_choice import InferenceGenerateChoice
+        from ..models.inference_generate_usage import InferenceGenerateUsage
+
+        d = dict(src_dict)
+        id = d.pop("id")
+
+        object_ = InferenceGenerateResponseObject(d.pop("object"))
+
+        created = d.pop("created")
+
+        model = d.pop("model")
+
+        choices = []
+        _choices = d.pop("choices")
+        for choices_item_data in _choices:
+            choices_item = InferenceGenerateChoice.from_dict(choices_item_data)
+
+            choices.append(choices_item)
+
+        usage = InferenceGenerateUsage.from_dict(d.pop("usage"))
+
+        inference_generate_response = cls(
+            id=id,
+            object_=object_,
+            created=created,
+            model=model,
+            choices=choices,
+            usage=usage,
+        )
+
+        inference_generate_response.additional_properties = d
+        return inference_generate_response
+
+    @property
+    def additional_keys(self) -> list[str]:
+        return list(self.additional_properties.keys())
+
+    def __getitem__(self, key: str) -> Any:
+        return self.additional_properties[key]
+
+    def __setitem__(self, key: str, value: Any) -> None:
+        self.additional_properties[key] = value
+
+    def __delitem__(self, key: str) -> None:
+        del self.additional_properties[key]
+
+    def __contains__(self, key: str) -> bool:
+        return key in self.additional_properties

antfly/client_generated/models/inference_generate_response_format.py

@@ -0,0 +1,86 @@
+from __future__ import annotations
+
+from collections.abc import Mapping
+from typing import TYPE_CHECKING, Any, TypeVar
+
+from attrs import define as _attrs_define
+from attrs import field as _attrs_field
+
+from ..models.inference_generate_response_format_type import InferenceGenerateResponseFormatType
+from ..types import UNSET, Unset
+
+if TYPE_CHECKING:
+    from ..models.inference_generate_json_schema_config import InferenceGenerateJsonSchemaConfig
+
+
+T = TypeVar("T", bound="InferenceGenerateResponseFormat")
+
+
+@_attrs_define
+class InferenceGenerateResponseFormat:
+    """
+    Attributes:
+        type_ (InferenceGenerateResponseFormatType): Structured output mode
+        json_schema (InferenceGenerateJsonSchemaConfig | Unset):
+    """
+
+    type_: InferenceGenerateResponseFormatType
+    json_schema: InferenceGenerateJsonSchemaConfig | Unset = UNSET
+    additional_properties: dict[str, Any] = _attrs_field(init=False, factory=dict)
+
+    def to_dict(self) -> dict[str, Any]:
+        type_ = self.type_.value
+
+        json_schema: dict[str, Any] | Unset = UNSET
+        if not isinstance(self.json_schema, Unset):
+            json_schema = self.json_schema.to_dict()
+
+        field_dict: dict[str, Any] = {}
+        field_dict.update(self.additional_properties)
+        field_dict.update(
+            {
+                "type": type_,
+            }
+        )
+        if json_schema is not UNSET:
+            field_dict["json_schema"] = json_schema
+
+        return field_dict
+
+    @classmethod
+    def from_dict(cls: type[T], src_dict: Mapping[str, Any]) -> T:
+        from ..models.inference_generate_json_schema_config import InferenceGenerateJsonSchemaConfig
+
+        d = dict(src_dict)
+        type_ = InferenceGenerateResponseFormatType(d.pop("type"))
+
+        _json_schema = d.pop("json_schema", UNSET)
+        json_schema: InferenceGenerateJsonSchemaConfig | Unset
+        if isinstance(_json_schema, Unset):
+            json_schema = UNSET
+        else:
+            json_schema = InferenceGenerateJsonSchemaConfig.from_dict(_json_schema)
+
+        inference_generate_response_format = cls(
+            type_=type_,
+            json_schema=json_schema,
+        )
+
+        inference_generate_response_format.additional_properties = d
+        return inference_generate_response_format
+
+    @property
+    def additional_keys(self) -> list[str]:
+        return list(self.additional_properties.keys())
+
+    def __getitem__(self, key: str) -> Any:
+        return self.additional_properties[key]
+
+    def __setitem__(self, key: str, value: Any) -> None:
+        self.additional_properties[key] = value
+
+    def __delitem__(self, key: str) -> None:
+        del self.additional_properties[key]
+
+    def __contains__(self, key: str) -> bool:
+        return key in self.additional_properties

antfly/client_generated/models/inference_generate_response_format_type.py

@@ -0,0 +1,10 @@
+from enum import Enum
+
+
+class InferenceGenerateResponseFormatType(str, Enum):
+    JSON_OBJECT = "json_object"
+    JSON_SCHEMA = "json_schema"
+    TEXT = "text"
+
+    def __str__(self) -> str:
+        return str(self.value)

antfly/client_generated/models/inference_generate_response_object.py

@@ -0,0 +1,8 @@
+from enum import Enum
+
+
+class InferenceGenerateResponseObject(str, Enum):
+    CHAT_COMPLETION = "chat.completion"
+
+    def __str__(self) -> str:
+        return str(self.value)