octostar-python-client 0.1.981__tar.gz → 0.1.982__tar.gz

{octostar_python_client-0.1.981/octostar_python_client.egg-info → octostar_python_client-0.1.982}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: octostar-python-client
-Version: 0.1.981
+Version: 0.1.982
 Summary: A client library for accessing Octostar API
 Author-email: Octostar <systems@octostar.com>
 License-Expression: MIT

octostar_python_client-0.1.982/octostar/utils/chat/complete.py

@@ -0,0 +1,205 @@
+import json
+from typing import AsyncIterator, Literal, Optional, TypedDict
+
+import httpx
+
+from ...client import Client, get_default_client
+from ..commons import network_retry_strategy
+from ..exceptions import ApiConnectionError
+
+
+class CompletionMessage(TypedDict):
+    role: Literal["system", "user", "assistant"]
+    content: str
+
+
+class CompletionUsage(TypedDict):
+    input_tokens: int
+    output_tokens: int
+    total_tokens: int
+
+
+class Completion(TypedDict):
+    content: str
+    model_name: str
+    finish_reason: Optional[str]
+    usage: Optional[CompletionUsage]
+
+
+def _prepare_complete_request(
+    client,
+    messages,
+    model_name,
+    temperature,
+    max_output_tokens,
+    enable_thinking,
+    stream,
+):
+    body = {"messages": messages, "enable_thinking": enable_thinking, "stream": stream}
+    if model_name is not None:
+        body["model_name"] = model_name
+    if temperature is not None:
+        body["temperature"] = temperature
+    if max_output_tokens is not None:
+        body["max_output_tokens"] = max_output_tokens
+    return {
+        "url": f"{client.get_base_url_v1()}/api/v2/ai/llm/complete",
+        "headers": {
+            "Authorization": f"Bearer {client.token}",
+            "x-ontology": client.ontology,
+        },
+        "json": body,
+    }
+
+
+def sync(
+    messages: list[CompletionMessage],
+    model_name: Optional[str] = None,
+    temperature: Optional[float] = None,
+    max_output_tokens: Optional[int] = None,
+    enable_thinking: bool = False,
+    timeout: Optional[float] = 300.0,
+    retries: int = 2,
+    client: Optional[Client] = None,
+) -> Completion:
+    """
+    # Generate the next assistant message for a conversation.
+
+    One LLM call: no tools and no system prompt besides the ones in `messages`.
+
+    ## Arguments
+    - `messages`: The conversation, oldest first, as `{"role", "content"}` dicts.
+    - `model_name`: The model to run. If None, the platform's default model is used.
+    - `temperature`: Sampling temperature. If None, the model's default.
+    - `max_output_tokens`: Cap on generated tokens. If None, the model's platform default.
+    - `enable_thinking`: Let local LLM deployments reason before answering.
+    - `timeout`: Seconds to wait for the reply.
+    - `retries`: Attempts on network errors; errors the API returns are not retried.
+    - `client`: The Client with which to connect to Octostar. If None, the default one is used.
+
+    ## Returns
+    The reply: `content`, `model_name`, `finish_reason` and `usage`.
+
+    ## Raises
+    - `ApiConnectionError`: If the query was unsuccessful.
+    """
+    if not client:
+        client = get_default_client()
+    request = _prepare_complete_request(
+        client,
+        messages,
+        model_name,
+        temperature,
+        max_output_tokens,
+        enable_thinking,
+        False,
+    )
+    response = None
+    try:
+        for attempt in network_retry_strategy(retries=retries):
+            with attempt:
+                with httpx.Client() as httpx_client:
+                    response = httpx_client.post(**request, timeout=timeout)
+                    response.raise_for_status()
+        return response.json()  # pyright: ignore[reportOptionalMemberAccess]
+    except Exception:
+        raise ApiConnectionError("complete", response, client)
+
+
+async def asyncio(
+    messages: list[CompletionMessage],
+    model_name: Optional[str] = None,
+    temperature: Optional[float] = None,
+    max_output_tokens: Optional[int] = None,
+    enable_thinking: bool = False,
+    timeout: Optional[float] = 300.0,
+    retries: int = 2,
+    client: Optional[Client] = None,
+) -> Completion:
+    """
+    # Generate the next assistant message for a conversation, asynchronously.
+
+    See `sync` for the arguments, return value and errors.
+    """
+    if not client:
+        client = get_default_client()
+    request = _prepare_complete_request(
+        client,
+        messages,
+        model_name,
+        temperature,
+        max_output_tokens,
+        enable_thinking,
+        False,
+    )
+    response = None
+    try:
+        for attempt in network_retry_strategy(retries=retries):
+            with attempt:
+                async with httpx.AsyncClient() as httpx_client:
+                    response = await httpx_client.post(**request, timeout=timeout)
+                    response.raise_for_status()
+        return response.json()  # pyright: ignore[reportOptionalMemberAccess]
+    except Exception:
+        raise ApiConnectionError("complete", response, client)
+
+
+async def streaming(
+    messages: list[CompletionMessage],
+    model_name: Optional[str] = None,
+    temperature: Optional[float] = None,
+    max_output_tokens: Optional[int] = None,
+    enable_thinking: bool = False,
+    timeout: Optional[float] = 300.0,
+    client: Optional[Client] = None,
+) -> AsyncIterator[str]:
+    """
+    # Stream the next assistant message for a conversation.
+
+    See `sync` for the arguments.
+
+    ## Returns
+    An async iterator over the text as it is generated.
+
+    ## Raises
+    - `ApiConnectionError`: If the request fails, or the stream reports an error
+      or ends before the reply is complete.
+    """
+    if not client:
+        client = get_default_client()
+    request = _prepare_complete_request(
+        client,
+        messages,
+        model_name,
+        temperature,
+        max_output_tokens,
+        enable_thinking,
+        True,
+    )
+    try:
+        async with httpx.AsyncClient() as httpx_client:
+            async with httpx_client.stream(
+                "POST", **request, timeout=timeout
+            ) as response:
+                if response.status_code != 200:
+                    await response.aread()
+                    raise ConnectionError(
+                        f"complete streaming: HTTP {response.status_code}: {response.text[:500]}"
+                    )
+                async for line in response.aiter_lines():
+                    if not line.strip():
+                        continue
+                    event = json.loads(line)
+                    if event.get("type") == "text":
+                        yield event["content"]
+                    elif event.get("type") == "done":
+                        return
+                    elif event.get("type") == "error":
+                        raise ConnectionError(
+                            f"complete streaming: {event.get('detail')}"
+                        )
+                raise ConnectionError(
+                    "complete streaming: the stream ended without a done event"
+                )
+    except (ConnectionError, httpx.HTTPError, ValueError) as e:
+        raise ApiConnectionError("complete streaming", None, client) from e

{octostar_python_client-0.1.981 → octostar_python_client-0.1.982/octostar_python_client.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: octostar-python-client
-Version: 0.1.981
+Version: 0.1.982
 Summary: A client library for accessing Octostar API
 Author-email: Octostar <systems@octostar.com>
 License-Expression: MIT

{octostar_python_client-0.1.981 → octostar_python_client-0.1.982}/octostar_python_client.egg-info/SOURCES.txt

@@ -182,6 +182,7 @@ octostar/utils/commons.py
 octostar/utils/exceptions.py
 octostar/utils/chat/__init__.py
 octostar/utils/chat/chat.py
+octostar/utils/chat/complete.py
 octostar/utils/chat/detokenize.py
 octostar/utils/chat/get_default_model.py
 octostar/utils/chat/list_models.py

{octostar_python_client-0.1.981 → octostar_python_client-0.1.982}/pyproject.toml

@@ -7,7 +7,7 @@ name = "octostar-python-client"
 # Fallback only. Release builds override this via the SDK_VERSION env var
 # (see scripts/sdk/build.sh); on main the version tracks the octostar-api
 # GitVersion tag.
-version = "0.1.981"
+version = "0.1.982"
 description = "A client library for accessing Octostar API"
 readme = "README.md"
 requires-python = ">=3.9"

{octostar_python_client-0.1.981 → octostar_python_client-0.1.982}/tests/test_contracts.py

@@ -75,9 +75,9 @@ SPEC_ROUTES_WITHOUT_SDK = {
     "/api/v1/chat": "AI chat (streaming)",
     "/api/v1/chat/models": "AI chat model listing",
     "/api/v1/chat/templates": "AI chat templates",
-    "/api/v1/chat/tools": "AI chat tools",
     "/api/v1/chat/detokenize": "Legacy detokenize",
     "/api/v1/chat/tokenize": "Legacy tokenize",
+    "/api/v2/ai/llm/complete": "Completion (wrapped via utils/chat/complete.py)",
     "/api/v2/ai/llm/detokenize": "Detokenize (wrapped via utils/chat/detokenize.py)",
     "/api/v2/ai/llm/tokenize": "Tokenize (wrapped via utils/chat/tokenize.py)",
     # File operations (multipart upload/download, not yet SDK-wrapped)