Add bounded generation batch scheduling (#2)

* feat: extract generation helpers; add bounded batch execution

* docs: update README and guide to include max_parallel_requests for LMClient

* chore: update infermesh package details and dependencies to version 0.2.0

* fix: validate max_parallel_requests at client construction

Author: fcogidi

README.md

@@ -59,6 +59,7 @@ prompts = [
 
 with LMClient(
     model="openai/gpt-4.1-mini",
+    max_parallel_requests=32,
     rpm=500,
     tpm=100_000,
 ) as client:
@@ -84,6 +85,10 @@ One failing request does not abort the whole batch. Failed items are `None` in
 `batch.results`; the exception is in `batch.errors[i]`. This is deliberate: a single
 provider error should not wipe out a long experiment.
 
+For large Python batches, set `max_parallel_requests` explicitly. That enables
+bounded in-flight scheduling for `generate_batch`; when it is unset, the method
+may start work for the full batch up front.
+
 This code works in Jupyter notebooks without any `asyncio` setup. The sync API runs a
 background event loop so you do not have to.
 
@@ -197,7 +202,7 @@ import json
 from infermesh import LMClient
 
 with open("results.jsonl", "w") as out, \
-     LMClient(model="openai/gpt-4.1-mini") as client:
+     LMClient(model="openai/gpt-4.1-mini", max_parallel_requests=32) as client:
 
     def save(index: int, result, error) -> None:
         row = {"index": index}

docs/guide.md

@@ -22,6 +22,7 @@ prompts = [
 
 with LMClient(
     model="openai/gpt-4.1-mini",
+    max_parallel_requests=32,
     rpm=500,
     tpm=100_000,
 ) as client:
@@ -48,6 +49,10 @@ By default, one failing request does not abort the whole batch. Failed items are
 stored as `None` in `batch.results`, and the corresponding exception is stored
 in `batch.errors[i]`.
 
+For large Python batches, set `max_parallel_requests` explicitly. That enables
+bounded in-flight scheduling for `generate_batch`; when it is unset, the method
+may start work for the full batch up front.
+
 ### Crash-Resilient Batches with `on_result`
 
 For large batches, you may want to write results to disk as each request
@@ -65,7 +70,7 @@ from infermesh import LMClient
 prompts = [...]  # large list
 
 with open("results.jsonl", "w") as out, \
-     LMClient(model="openai/gpt-4.1-mini") as client:
+     LMClient(model="openai/gpt-4.1-mini", max_parallel_requests=32) as client:
 
     def save(index: int, result, error) -> None:
         row = {"index": index}
@@ -102,7 +107,7 @@ if output_path.exists():
 pending = [(i, p) for i, p in enumerate(prompts) if i not in done]
 
 with open(output_path, "a") as out, \
-     LMClient(model="openai/gpt-4.1-mini") as client:
+     LMClient(model="openai/gpt-4.1-mini", max_parallel_requests=32) as client:
 
     def save(batch_idx: int, result, error) -> None:
         orig_idx = pending[batch_idx][0]

src/infermesh/_client_runtime.py

@@ -93,12 +93,17 @@ def _validate_init_args(
         api_key: str | None,
         deployments: dict[str, DeploymentConfig | dict[str, Any]] | None,
         endpoint: EndpointType,
+        max_parallel_requests: int | None,
     ) -> None:
         """Validate top-level constructor arguments."""
 
         validate_endpoint(endpoint)
         if model is None:
             raise ValueError("``model`` is required.")
+        if max_parallel_requests is not None and max_parallel_requests < 1:
+            raise ValueError(
+                "``max_parallel_requests`` must be ``None`` or a positive integer."
+            )
         if deployments is not None and (api_base is not None or api_key is not None):
             raise ValueError(
                 "``api_base`` and ``api_key`` cannot be set when ``deployments`` "
@@ -199,11 +204,14 @@ async def _dispatch_with_controls(
         request_callable: Any,
         request_args: tuple[Any, ...],
         request_kwargs: dict[str, Any],
+        queue_started_at: float | None = None,
     ) -> tuple[Any, RequestMetrics]:
         """Run a request with concurrency and rate-limiting controls."""
 
         state = self._get_loop_state()
-        queue_started = time.perf_counter()
+        queue_started = (
+            queue_started_at if queue_started_at is not None else time.perf_counter()
+        )
         handle: RateLimiterAcquisitionHandle | None = None
         semaphore_context = (
             state.semaphore if state.semaphore is not None else _null_async_context()