1. Removes summation of `ChatGenerationChunk` from hot loops in `stream`
and `astream`
2. Removes run id gen from loop as well (minor impact)
Again, benchmarking on processing ~200k chunks (a poem about broccoli).
Before: ~4.2s
Blue circle is all the time spent adding up gen chunks
<img width="1345" alt="Screenshot 2025-05-14 at 7 48 33 AM"
src="https://github.com/user-attachments/assets/08a59d78-134d-4cd3-9d54-214de689df51"
/>
After: ~2.3s
Blue circle is remaining time spent on adding chunks, which can be
minimized in a future PR by optimizing the `merge_content`,
`merge_dicts`, and `merge_lists` utilities.
<img width="1353" alt="Screenshot 2025-05-14 at 7 50 08 AM"
src="https://github.com/user-attachments/assets/df6b3506-929e-4b6d-b198-7c4e992c6d34"
/>
1. Remove `shielded` decorator from non-end event handlers
2. Exit early with a `self.handlers` check instead of doing unnecessary
asyncio work
Using a benchmark that processes ~200k chunks (a poem about broccoli).
Before: ~15s
Circled in blue is unnecessary event handling time. This is addressed by
point 2 above
<img width="1347" alt="Screenshot 2025-05-14 at 7 37 53 AM"
src="https://github.com/user-attachments/assets/675e0fed-8f37-46c0-90b3-bef3cb9a1e86"
/>
After: ~4.2s
The total time is largely reduced by the removal of the `shielded`
decorator, which holds little significance for non-end handlers.
<img width="1348" alt="Screenshot 2025-05-14 at 7 37 22 AM"
src="https://github.com/user-attachments/assets/54be8a3e-5827-4136-a87b-54b0d40fe331"
/>
**Description**: The 'inspect' package in python skips over the aliases
set in the schema of a pydantic model. This is a workound to include the
aliases from the original input.
**issue**: #31035
Cc: @ccurme @eyurtsev
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
When aggregating AIMessageChunks in a stream, core prefers the leftmost
non-null ID. This is problematic because:
- Core assigns IDs when they are null to `f"run-{run_manager.run_id}"`
- The desired meaningful ID might not be available until midway through
the stream, as is the case for the OpenAI Responses API.
For the OpenAI Responses API, we assign message IDs to the top-level
`AIMessage.id`. This works in `.(a)invoke`, but during `.(a)stream` the
IDs get overwritten by the defaults assigned in langchain-core. These
IDs
[must](https://community.openai.com/t/how-to-solve-badrequesterror-400-item-rs-of-type-reasoning-was-provided-without-its-required-following-item-error-in-responses-api/1151686/9)
be available on the AIMessage object to support passing reasoning items
back to the API (e.g., if not using OpenAI's `previous_response_id`
feature). We could add them elsewhere, but seeing as we've already made
the decision to store them in `.id` during `.(a)invoke`, addressing the
issue in core lets us fix the problem with no interface changes.
Follow up to https://github.com/langchain-ai/langsmith-sdk/pull/1696,
I've bumped the `langsmith` version where applicable in `uv.lock`.
Type checking problems here because deps have been updated in
`pyproject.toml` and `uv lock` hasn't been run - we should enforce that
in the future - goes with the other dependabot todos :).
Chat models currently implement support for:
- images in OpenAI Chat Completions format
- other multimodal types (e.g., PDF and audio) in a cross-provider
[standard
format](https://python.langchain.com/docs/how_to/multimodal_inputs/)
Here we update core to extend support to PDF and audio input in Chat
Completions format. **If an OAI-format PDF or audio content block is
passed into any chat model, it will be transformed to the LangChain
standard format**. We assume that any chat model supporting OAI-format
PDF or audio has implemented support for the standard format.
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core, etc. is
being modified. Use "docs: ..." for purely docs changes, "infra: ..."
for CI changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: If you're adding a new integration, please
include
1. a test for the integration, preferably unit tests that do not rely on
network access,
2. an example notebook showing its use. It lives in
`docs/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, eyurtsev, ccurme, vbarda, hwchase17.
Addresses #30158
When using the output parser—either in a chain or standalone—hitting
max_tokens triggers a misleading “missing variable” error instead of
indicating the output was truncated. This subtle bug often surfaces with
Anthropic models.
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
We only need to rebuild model schemas if type annotation information
isn't available during declaration - that shouldn't be the case for
these types corrected here.
Need to do more thorough testing to make sure these structures have
complete schemas, but hopefully this boosts startup / import time.
TL;DR: you can't optimize imports with a lazy `__getattr__` if there is
a namespace conflict with a module name and an attribute name. We should
avoid introducing conflicts like this in the future.
This PR fixes a bug introduced by my lazy imports PR:
https://github.com/langchain-ai/langchain/pull/30769.
In `langchain_core`, we have utilities for loading and dumping data.
Unfortunately, one of those utilities is a `load` function, located in
`langchain_core/load/load.py`. To make this function more visible, we
make it accessible at the top level `langchain_core.load` module via
importing the function in `langchain_core/load/__init__.py`.
So, either of these imports should work:
```py
from langchain_core.load import load
from langchain_core.load.load import load
```
As you can tell, this is already a bit confusing. You'd think that the
first import would produce the module `load`, but because of the
`__init__.py` shortcut, both produce the function `load`.
<details> More on why the lazy imports PR broke this support...
All was well, except when the absolute import was run first, see the
last snippet:
```
>>> from langchain_core.load import load
>>> load
<function load at 0x101c320c0>
```
```
>>> from langchain_core.load.load import load
>>> load
<function load at 0x1069360c0>
```
```
>>> from langchain_core.load import load
>>> load
<function load at 0x10692e0c0>
>>> from langchain_core.load.load import load
>>> load
<function load at 0x10692e0c0>
```
```
>>> from langchain_core.load.load import load
>>> load
<function load at 0x101e2e0c0>
>>> from langchain_core.load import load
>>> load
<module 'langchain_core.load.load' from '/Users/sydney_runkle/oss/langchain/libs/core/langchain_core/load/load.py'>
```
In this case, the function `load` wasn't stored in the globals cache for
the `langchain_core.load` module (by the lazy import logic), so Python
defers to a module import.
</details>
New `langchain` tongue twister 😜: we've created a problem for ourselves
because you have to load the load function from the load file in the
load module 😨.
Fix CI to trigger benchmarks on `run-codspeed-benchmarks` label addition
Reduce scope of async benchmark to save time on CI
Waiting to merge this PR until we figure out how to use walltime on
local runners.
Most easily reviewed with the "hide whitespace" option toggled.
Seeing 10-50% speed ups in import time for common structures 🚀
The general purpose of this PR is to lazily import structures within
`langchain_core.XXX_module.__init__.py` so that we're not eagerly
importing expensive dependencies (`pydantic`, `requests`, etc).
Analysis of flamegraphs generated with `importtime` motivated these
changes. For example, the one below demonstrates that importing
`HumanMessage` accidentally triggered imports for `importlib.metadata`,
`requests`, etc.
There's still much more to do on this front, and we can start digging
into our own internal code for optimizations now that we're less
concerned about external imports.
<img width="1210" alt="Screenshot 2025-04-11 at 1 10 54 PM"
src="https://github.com/user-attachments/assets/112a3fe7-24a9-4294-92c1-d5ae64df839e"
/>
I've tracked the improvements with some local benchmarks:
## `pytest-benchmark` results
| Name | Before (s) | After (s) | Delta (s) | % Change |
|-----------------------------|------------|-----------|-----------|----------|
| Document | 2.8683 | 1.2775 | -1.5908 | -55.46% |
| HumanMessage | 2.2358 | 1.1673 | -1.0685 | -47.79% |
| ChatPromptTemplate | 5.5235 | 2.9709 | -2.5526 | -46.22% |
| Runnable | 2.9423 | 1.7793 | -1.163 | -39.53% |
| InMemoryVectorStore | 3.1180 | 1.8417 | -1.2763 | -40.93% |
| RunnableLambda | 2.7385 | 1.8745 | -0.864 | -31.55% |
| tool | 5.1231 | 4.0771 | -1.046 | -20.42% |
| CallbackManager | 4.2263 | 3.4099 | -0.8164 | -19.32% |
| LangChainTracer | 3.8394 | 3.3101 | -0.5293 | -13.79% |
| BaseChatModel | 4.3317 | 3.8806 | -0.4511 | -10.41% |
| PydanticOutputParser | 3.2036 | 3.2995 | 0.0959 | 2.99% |
| InMemoryRateLimiter | 0.5311 | 0.5995 | 0.0684 | 12.88% |
Note the lack of change for `InMemoryRateLimiter` and
`PydanticOutputParser` is just random noise, I'm getting comparable
numbers locally.
## Local CodSpeed results
We're still working on configuring CodSpeed on CI. The local usage
produced similar results.
Looks like `pyupgrade` was already used here but missed some docs and
tests.
This helps to keep our docs looking professional and up to date.
Eventually, we should lint / format our inline docs.
