chatlas 0.12.0__tar.gz → 0.13.0__tar.gz

{chatlas-0.12.0 → chatlas-0.13.0}/CHANGELOG.md

@@ -7,6 +7,20 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 -->
 
+## [0.13.0] - 2025-09-10
+
+### New features
+
+* Added support for submitting multiple chats in one batch. With batch submission, results can take up to 24 hours to complete, but in return you pay ~50% less than usual. For more, see the [reference](https://posit-dev.github.io/chatlas/reference/) for `batch_chat()`, `batch_chat_text()`, `batch_chat_structured()` and `batch_chat_completed()`. (#177)
+* The `Chat` class gains new `.chat_structured()` (and `.chat_structured_async()`) methods. These methods supersede the now deprecated `.extract_data()` (and `.extract_data_async()`). The only difference is that the new methods return a `BaseModel` instance (instead of a `dict()`), leading to a better type hinting/checking experience.  (#175)
+* The `.get_turns()` method gains a `tool_result_role` parameter. Set `tool_result_role="assistant"` to collect tool result content (plus the surrounding assistant turn contents) into a single assistant turn. This is convenient for display purposes and more generally if you want the tool calling loop to be contained in a single turn. (#179) 
+
+### Improvements
+
+* The `.app()` method now:
+    * Enables bookmarking by default (i.e., chat session survives page reload). (#179)
+    * Correctly renders pre-existing turns that contain tool calls. (#179)
+
 ## [0.12.0] - 2025-09-08
 
 ### Breaking changes

{chatlas-0.12.0 → chatlas-0.13.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: chatlas
-Version: 0.12.0
+Version: 0.13.0
 Summary: A simple and consistent interface for chatting with LLMs
 Project-URL: Homepage, https://posit-dev.github.io/chatlas
 Project-URL: Documentation, https://posit-dev.github.io/chatlas
@@ -44,6 +44,7 @@ Requires-Dist: pillow; extra == 'dev'
 Requires-Dist: python-dotenv; extra == 'dev'
 Requires-Dist: ruff>=0.6.5; extra == 'dev'
 Requires-Dist: shiny; extra == 'dev'
+Requires-Dist: shinychat; extra == 'dev'
 Requires-Dist: snowflake-ml-python>=1.8.4; extra == 'dev'
 Requires-Dist: tenacity; extra == 'dev'
 Requires-Dist: tiktoken; extra == 'dev'

{chatlas-0.12.0 → chatlas-0.13.0}/chatlas/__init__.py

@@ -1,5 +1,11 @@
 from . import types
 from ._auto import ChatAuto
+from ._batch_chat import (
+    batch_chat,
+    batch_chat_completed,
+    batch_chat_structured,
+    batch_chat_text,
+)
 from ._chat import Chat
 from ._content import (
     ContentToolRequest,
@@ -36,6 +42,10 @@ except ImportError:  # pragma: no cover
     __version__ = "0.0.0"  # stub value for docs
 
 __all__ = (
+    "batch_chat",
+    "batch_chat_completed",
+    "batch_chat_structured",
+    "batch_chat_text",
     "ChatAnthropic",
     "ChatAuto",
     "ChatBedrockAnthropic",

chatlas-0.13.0/chatlas/_batch_chat.py

@@ -0,0 +1,211 @@
+"""
+Batch chat processing for submitting multiple requests simultaneously.
+
+This module provides functionality for submitting multiple chat requests
+in batches to providers that support it (currently OpenAI and Anthropic).
+Batch processing can take up to 24 hours but offers significant cost savings
+(up to 50% less than regular requests).
+"""
+
+from __future__ import annotations
+
+import copy
+from pathlib import Path
+from typing import TypeVar, Union
+
+from pydantic import BaseModel
+
+from ._batch_job import BatchJob, ContentT
+from ._chat import Chat
+
+ChatT = TypeVar("ChatT", bound=Chat)
+BaseModelT = TypeVar("BaseModelT", bound=BaseModel)
+
+
+def batch_chat(
+    chat: ChatT,
+    prompts: list[ContentT] | list[list[ContentT]],
+    path: Union[str, Path],
+    wait: bool = True,
+) -> list[ChatT | None]:
+    """
+    Submit multiple chat requests in a batch.
+
+    This function allows you to submit multiple chat requests simultaneously
+    using provider batch APIs (currently OpenAI and Anthropic). Batch processing
+    can take up to 24 hours but offers significant cost savings.
+
+    Parameters
+    ----------
+    chat
+        Chat instance to use for the batch
+    prompts
+        List of prompts to process. Each can be a string or list of strings.
+    path
+        Path to file (with .json extension) to store batch state
+    wait
+        If True, wait for batch to complete. If False, return None if incomplete.
+
+    Returns
+    -------
+    List of Chat objects (one per prompt) if complete, None if wait=False and incomplete.
+    Individual Chat objects may be None if their request failed.
+
+    Example
+    -------
+
+    ```python
+    from chatlas import ChatOpenAI
+
+    chat = ChatOpenAI()
+    prompts = [
+        "What's the capital of France?",
+        "What's the capital of Germany?",
+        "What's the capital of Italy?",
+    ]
+
+    chats = batch_chat(chat, prompts, "capitals.json")
+    for i, result_chat in enumerate(chats):
+        if result_chat:
+            print(f"Prompt {i + 1}: {result_chat.get_last_turn().text}")
+    ```
+    """
+    job = BatchJob(chat, prompts, path, wait=wait)
+    job.step_until_done()
+
+    chats = []
+    assistant_turns = job.result_turns()
+    for user, assistant in zip(job.user_turns, assistant_turns):
+        if assistant is not None:
+            new_chat = copy.deepcopy(chat)
+            new_chat.add_turn(user)
+            new_chat.add_turn(assistant)
+            chats.append(new_chat)
+        else:
+            chats.append(None)
+
+    return chats
+
+
+def batch_chat_text(
+    chat: Chat,
+    prompts: list[ContentT] | list[list[ContentT]],
+    path: Union[str, Path],
+    wait: bool = True,
+) -> list[str | None]:
+    """
+    Submit multiple chat requests in a batch and return text responses.
+
+    This is a convenience function that returns just the text of the responses
+    rather than full Chat objects.
+
+    Parameters
+    ----------
+    chat
+        Chat instance to use for the batch
+    prompts
+        List of prompts to process
+    path
+        Path to file (with .json extension) to store batch state
+    wait
+        If True, wait for batch to complete
+
+    Return
+    ------
+    List of text responses (or None for failed requests)
+    """
+    chats = batch_chat(chat, prompts, path, wait=wait)
+
+    texts = []
+    for x in chats:
+        if x is None:
+            texts.append(None)
+            continue
+        last_turn = x.get_last_turn()
+        if last_turn is None:
+            texts.append(None)
+            continue
+        texts.append(last_turn.text)
+
+    return texts
+
+
+def batch_chat_structured(
+    chat: Chat,
+    prompts: list[ContentT] | list[list[ContentT]],
+    path: Union[str, Path],
+    data_model: type[BaseModelT],
+    wait: bool = True,
+) -> list[BaseModelT | None]:
+    """
+    Submit multiple structured data requests in a batch.
+
+    Parameters
+    ----------
+    chat
+        Chat instance to use for the batch
+    prompts
+        List of prompts to process
+    path
+        Path to file (with .json extension) to store batch state
+    data_model
+        Pydantic model class for structured responses
+    wait
+        If True, wait for batch to complete
+
+    Return
+    ------
+    List of structured data objects (or None for failed requests)
+    """
+    job = BatchJob(chat, prompts, path, data_model=data_model, wait=wait)
+    result = job.step_until_done()
+
+    if result is None:
+        return []
+
+    res: list[BaseModelT | None] = []
+    assistant_turns = job.result_turns()
+    for turn in assistant_turns:
+        if turn is None:
+            res.append(None)
+        else:
+            json = chat._extract_turn_json(turn)
+            model = data_model.model_validate(json)
+            res.append(model)
+
+    return res
+
+
+def batch_chat_completed(
+    chat: Chat,
+    prompts: list[ContentT] | list[list[ContentT]],
+    path: Union[str, Path],
+) -> bool:
+    """
+    Check if a batch job is completed without waiting.
+
+    Parameters
+    ----------
+    chat
+        Chat instance used for the batch
+    prompts
+        List of prompts used for the batch
+    path
+        Path to batch state file
+
+    Returns
+    -------
+    True if batch is complete, False otherwise
+    """
+    job = BatchJob(chat, prompts, path, wait=False)
+    stage = job.stage
+
+    if stage == "submitting":
+        return False
+    elif stage == "waiting":
+        status = job._poll()
+        return not status.working
+    elif stage == "retrieving" or stage == "done":
+        return True
+    else:
+        raise ValueError(f"Unknown batch stage: {stage}")

chatlas-0.13.0/chatlas/_batch_job.py

@@ -0,0 +1,234 @@
+from __future__ import annotations
+
+import hashlib
+import json
+import time
+from datetime import timedelta
+from pathlib import Path
+from typing import Any, Literal, Optional, TypeVar, Union
+
+from pydantic import BaseModel
+from rich.console import Console
+from rich.progress import Progress, SpinnerColumn, TextColumn
+
+from ._chat import Chat
+from ._content import Content
+from ._provider import BatchStatus
+from ._turn import Turn, user_turn
+from ._typing_extensions import TypedDict
+
+BatchStage = Literal["submitting", "waiting", "retrieving", "done"]
+
+
+class BatchStateHash(TypedDict):
+    provider: str
+    model: str
+    prompts: str
+    user_turns: str
+
+
+class BatchState(BaseModel):
+    version: int
+    stage: BatchStage
+    batch: dict[str, Any]
+    results: list[dict[str, Any]]
+    started_at: int
+    hash: BatchStateHash
+
+
+ContentT = TypeVar("ContentT", bound=Union[str, Content])
+
+
+class BatchJob:
+    """
+    Manages the lifecycle of a batch processing job.
+
+    A batch job goes through several stages:
+    1. "submitting" - Initial submission to the provider
+    2. "waiting" - Waiting for processing to complete
+    3. "retrieving" - Downloading results
+    4. "done" - Processing complete
+    """
+
+    def __init__(
+        self,
+        chat: Chat,
+        prompts: list[ContentT] | list[list[ContentT]],
+        path: Union[str, Path],
+        data_model: Optional[type[BaseModel]] = None,
+        wait: bool = True,
+    ):
+        if not chat.provider.has_batch_support():
+            raise ValueError("Batch requests are not supported by this provider")
+
+        self.chat = chat
+        self.prompts = prompts
+        self.path = Path(path)
+        self.data_model = data_model
+        self.should_wait = wait
+
+        # Convert prompts to user turns
+        self.user_turns: list[Turn] = []
+        for prompt in prompts:
+            if not isinstance(prompt, (str, Content)):
+                turn = user_turn(*prompt)
+            else:
+                turn = user_turn(prompt)
+            self.user_turns.append(turn)
+
+        # Job state management
+        self.provider = chat.provider
+        self.stage: BatchStage = "submitting"
+        self.batch: dict[str, Any] = {}
+        self.results: list[dict[str, Any]] = []
+
+        # Load existing state if file exists and is not empty
+        if self.path.exists() and self.path.stat().st_size > 0:
+            self._load_state()
+        else:
+            self.started_at = time.time()
+
+    def _load_state(self) -> None:
+        with open(self.path, "r") as f:
+            state = BatchState.model_validate_json(f.read())
+
+        self.stage = state.stage
+        self.batch = state.batch
+        self.results = state.results
+        self.started_at = state.started_at
+
+        # Verify hash to ensure consistency
+        stored_hash = state.hash
+        current_hash = self._compute_hash()
+
+        for key, value in current_hash.items():
+            if stored_hash.get(key) != value:
+                raise ValueError(
+                    f"Batch state mismatch: {key} doesn't match stored value. "
+                    f"Do you need to pick a different path?"
+                )
+
+    def _save_state(self) -> None:
+        state = BatchState(
+            version=1,
+            stage=self.stage,
+            batch=self.batch,
+            results=self.results,
+            started_at=int(self.started_at) if self.started_at else 0,
+            hash=self._compute_hash(),
+        )
+
+        with open(self.path, "w") as f:
+            f.write(state.model_dump_json(indent=2))
+
+    def _compute_hash(self) -> BatchStateHash:
+        turns = self.chat.get_turns(include_system_prompt=True)
+        return {
+            "provider": self.provider.name,
+            "model": self.provider.model,
+            "prompts": self._hash([str(p) for p in self.prompts]),
+            "user_turns": self._hash([str(turn) for turn in turns]),
+        }
+
+    @staticmethod
+    def _hash(x: Any) -> str:
+        return hashlib.md5(json.dumps(x, sort_keys=True).encode()).hexdigest()
+
+    def step(self) -> bool:
+        if self.stage == "submitting":
+            return self._submit()
+        elif self.stage == "waiting":
+            return self._wait()
+        elif self.stage == "retrieving":
+            return self._retrieve()
+        else:
+            raise ValueError(f"Unknown stage: {self.stage}")
+
+    def step_until_done(self) -> Optional["BatchJob"]:
+        while self.stage != "done":
+            if not self.step():
+                return None
+        return self
+
+    def _submit(self) -> bool:
+        existing_turns = self.chat.get_turns(include_system_prompt=True)
+
+        conversations = []
+        for turn in self.user_turns:
+            conversation = existing_turns + [turn]
+            conversations.append(conversation)
+
+        self.batch = self.provider.batch_submit(conversations, self.data_model)
+        self.stage = "waiting"
+        self._save_state()
+        return True
+
+    def _wait(self) -> bool:
+        # Always poll once, even when wait=False
+        status = self._poll()
+
+        if self.should_wait:
+            console = Console()
+
+            with Progress(
+                SpinnerColumn(),
+                TextColumn("Processing..."),
+                TextColumn("[{task.fields[elapsed]}]"),
+                TextColumn("{task.fields[n_processing]} pending |"),
+                TextColumn("[green]{task.fields[n_succeeded]}[/green] done |"),
+                TextColumn("[red]{task.fields[n_failed]}[/red] failed"),
+                console=console,
+            ) as progress:
+                task = progress.add_task(
+                    "processing",
+                    elapsed=self._elapsed(),
+                    n_processing=status.n_processing,
+                    n_succeeded=status.n_succeeded,
+                    n_failed=status.n_failed,
+                )
+
+                while status.working:
+                    time.sleep(0.5)
+                    status = self._poll()
+                    progress.update(
+                        task,
+                        elapsed=self._elapsed(),
+                        n_processing=status.n_processing,
+                        n_succeeded=status.n_succeeded,
+                        n_failed=status.n_failed,
+                    )
+
+        if not status.working:
+            self.stage = "retrieving"
+            self._save_state()
+            return True
+        else:
+            return False
+
+    def _poll(self) -> "BatchStatus":
+        if not self.batch:
+            raise ValueError("No batch to poll")
+        self.batch = self.provider.batch_poll(self.batch)
+        self._save_state()
+        return self.provider.batch_status(self.batch)
+
+    def _elapsed(self) -> str:
+        return str(timedelta(seconds=int(time.time()) - int(self.started_at)))
+
+    def _retrieve(self) -> bool:
+        if not self.batch:
+            raise ValueError("No batch to retrieve")
+        self.results = self.provider.batch_retrieve(self.batch)
+        self.stage = "done"
+        self._save_state()
+        return True
+
+    def result_turns(self) -> list[Turn | None]:
+        turns = []
+        for result in self.results:
+            turn = self.provider.batch_result_turn(
+                result, has_data_model=self.data_model is not None
+            )
+            turns.append(turn)
+
+        return turns