## Description
The following PR enables the [grammar-based
sampling](https://github.com/ggerganov/llama.cpp/tree/master/grammars)
in llama-cpp LLM.
In short, loading file with formal grammar definition will constrain
model outputs. For instance, one can force the model to generate valid
JSON or generate only python lists.
In the follow-up PR we will add:
* docs with some description why it is cool and how it works
* maybe some code sample for some task such as in llama repo
---------
Co-authored-by: Lance Martin <lance@langchain.dev>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Add PromptGuard integration
-------
There are two approaches to integrate PromptGuard with a LangChain
application.
1. PromptGuardLLMWrapper
2. functions that can be used in LangChain expression.
-----
- Dependencies
`promptguard` python package, which is a runtime requirement if you'd
try out the demo.
- @baskaryan @hwchase17 Thanks for the ideas and suggestions along the
development process.
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
- Description: Added improvements in Nebula LLM to perform auto-retry;
more generation parameters supported. Conversation is no longer required
to be passed in the LLM object. Examples are updated.
- Issue: N/A
- Dependencies: N/A
- Tag maintainer: @baskaryan
- Twitter handle: symbldotai
---------
Co-authored-by: toshishjawale <toshish@symbl.ai>
Adds [DeepSparse](https://github.com/neuralmagic/deepsparse) as an LLM
backend. DeepSparse supports running various open-source sparsified
models hosted on [SparseZoo](https://sparsezoo.neuralmagic.com/) for
performance gains on CPUs.
Twitter handles: @mgoin_ @neuralmagic
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
## Description
This PR adds the `aembed_query` and `aembed_documents` async methods for
improving the embeddings generation for large documents. The
implementation uses asyncio tasks and gather to achieve concurrency as
there is no bedrock async API in boto3.
### Maintainers
@agola11
@aarora79
### Open questions
To avoid throttling from the Bedrock API, should there be an option to
limit the concurrency of the calls?
## Description:
This PR adds the Titan Takeoff Server to the available LLMs in
LangChain.
Titan Takeoff is an inference server created by
[TitanML](https://www.titanml.co/) that allows you to deploy large
language models locally on your hardware in a single command. Most
generative model architectures are included, such as Falcon, Llama 2,
GPT2, T5 and many more.
Read more about Titan Takeoff here:
-
[Blog](https://medium.com/@TitanML/introducing-titan-takeoff-6c30e55a8e1e)
- [Docs](https://docs.titanml.co/docs/titan-takeoff/getting-started)
#### Testing
As Titan Takeoff runs locally on port 8000 by default, no network access
is needed. Responses are mocked for testing.
- [x] Make Lint
- [x] Make Format
- [x] Make Test
#### Dependencies
No new dependencies are introduced. However, users will need to install
the titan-iris package in their local environment and start the Titan
Takeoff inferencing server in order to use the Titan Takeoff
integration.
Thanks for your help and please let me know if you have any questions.
cc: @hwchase17 @baskaryan
This addresses some issues with introducing the Nebula LLM to LangChain
in this PR:
https://github.com/langchain-ai/langchain/pull/8876
This fixes the following:
- Removes `SYMBLAI` from variable names
- Fixes bug with `Bearer` for the API KEY
Thanks again in advance for your help!
cc: @hwchase17, @baskaryan
---------
Co-authored-by: dvonthenen <david.vonthenen@gmail.com>
Adds Ollama as an LLM. Ollama can run various open source models locally
e.g. Llama 2 and Vicuna, automatically configuring and GPU-optimizing
them.
@rlancemartin @hwchase17
---------
Co-authored-by: Lance Martin <lance@langchain.dev>
## Description
This PR adds Nebula to the available LLMs in LangChain.
Nebula is an LLM focused on conversation understanding and enables users
to extract conversation insights from video, audio, text, and chat-based
conversations. These conversations can occur between any mix of human or
AI participants.
Examples of some questions you could ask Nebula from a given
conversation are:
- What could be the customer’s pain points based on the conversation?
- What sales opportunities can be identified from this conversation?
- What best practices can be derived from this conversation for future
customer interactions?
You can read more about Nebula here:
https://symbl.ai/blog/extract-insights-symbl-ai-generative-ai-recall-ai-meetings/
#### Integration Test
An integration test is added, but it requires network access. Since
Nebula is fully managed like OpenAI, network access is required to
exercise the integration test.
#### Linting
- [x] make lint
- [x] make test (TODO: there seems to be a failure in another
non-related test??? Need to check on this.)
- [x] make format
### Dependencies
No new dependencies were introduced.
### Twitter handle
[@symbldotai](https://twitter.com/symbldotai)
[@dvonthenen](https://twitter.com/dvonthenen)
If you have any questions, please let me know.
cc: @hwchase17, @baskaryan
---------
Co-authored-by: dvonthenen <david.vonthenen@gmail.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Hello langchain maintainers,
this PR aims at integrating
[vllm](https://vllm.readthedocs.io/en/latest/#) into langchain. This PR
closes#8729.
This feature clearly depends on `vllm`, but I've seen other models
supported here depend on packages that are not included in the
pyproject.toml (e.g. `gpt4all`, `text-generation`) so I thought it was
the case for this as well.
@hwchase17, @baskaryan
---------
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
Description - Integrates Fireworks within Langchain LLMs to allow users
to use Fireworks models with Langchain, mainly for summarization.
Issue - Not applicable
Dependencies - None
Tag maintainer - @rlancemartin
---------
Co-authored-by: Raj Janardhan <rajjanardhan@Rajs-Laptop.attlocal.net>
- Description: Minimax is a great AI startup from China, recently they
released their latest model and chat API, and the API is widely-spread
in China. As a result, I'd like to add the Minimax llm model to
Langchain.
- Tag maintainer: @hwchase17, @baskaryan
---------
Co-authored-by: the <tao.he@hulu.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>
- [Xorbits
Inference(Xinference)](https://github.com/xorbitsai/inference) is a
powerful and versatile library designed to serve language, speech
recognition, and multimodal models. Xinference supports a variety of
GGML-compatible models including chatglm, whisper, and vicuna, and
utilizes heterogeneous hardware and a distributed architecture for
seamless cross-device and cross-server model deployment.
- This PR integrates Xinference models and Xinference embeddings into
LangChain.
- Dependencies: To install the depenedencies for this integration, run
`pip install "xinference[all]"`
- Example Usage:
To start a local instance of Xinference, run `xinference`.
To deploy Xinference in a distributed cluster, first start an Xinference
supervisor using `xinference-supervisor`:
`xinference-supervisor -H "${supervisor_host}"`
Then, start the Xinference workers using `xinference-worker` on each
server you want to run them on.
`xinference-worker -e "http://${supervisor_host}:9997"`
To use Xinference with LangChain, you also need to launch a model. You
can use command line interface (CLI) to do so. Fo example: `xinference
launch -n vicuna-v1.3 -f ggmlv3 -q q4_0`. This launches a model named
vicuna-v1.3 with `model_format="ggmlv3"` and `quantization="q4_0"`. A
model UID is returned for you to use.
Now you can use Xinference with LangChain:
```python
from langchain.llms import Xinference
llm = Xinference(
server_url="http://0.0.0.0:9997", # suppose the supervisor_host is "0.0.0.0"
model_uid = {model_uid} # model UID returned from launching a model
)
llm(
prompt="Q: where can we visit in the capital of France? A:",
generate_config={"max_tokens": 1024},
)
```
You can also use RESTful client to launch a model:
```python
from xinference.client import RESTfulClient
client = RESTfulClient("http://0.0.0.0:9997")
model_uid = client.launch_model(model_name="vicuna-v1.3", model_size_in_billions=7, quantization="q4_0")
```
The following code block demonstrates how to use Xinference embeddings
with LangChain:
```python
from langchain.embeddings import XinferenceEmbeddings
xinference = XinferenceEmbeddings(
server_url="http://0.0.0.0:9997",
model_uid = model_uid
)
```
```python
query_result = xinference.embed_query("This is a test query")
```
```python
doc_result = xinference.embed_documents(["text A", "text B"])
```
Xinference is still under rapid development. Feel free to [join our
Slack
community](https://xorbitsio.slack.com/join/shared_invite/zt-1z3zsm9ep-87yI9YZ_B79HLB2ccTq4WA)
to get the latest updates!
- Request for review: @hwchase17, @baskaryan
- Twitter handle: https://twitter.com/Xorbitsio
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
In this PR:
- Removed restricted model loading logic for Petals-Bloom
- Removed petals imports (DistributedBloomForCausalLM,
BloomTokenizerFast)
- Instead imported more generalized versions of loader
(AutoDistributedModelForCausalLM, AutoTokenizer)
- Updated the Petals example notebook to allow for a successful
installation of Petals in Apple Silicon Macs
- Tag maintainer: @hwchase17, @baskaryan
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Stop sequences are useful if you are doing long-running completions and
need to early-out rather than running for the full max_length... not
only does this save inference cost on Replicate, it is also much faster
if you are going to truncate the output later anyway.
Other LLMs support stop sequences natively (e.g. OpenAI) but I didn't
see this for Replicate so adding this via their prediction cancel
method.
Housekeeping: I ran `make format` and `make lint`, no issues reported in
the files I touched.
I did update the replicate integration test and ran `poetry run pytest
tests/integration_tests/llms/test_replicate.py` successfully.
Finally, I am @tjaffri https://twitter.com/tjaffri for feature
announcement tweets... or if you could please tag @docugami
https://twitter.com/docugami we would really appreciate that :-)
Co-authored-by: Taqi Jaffri <tjaffri@docugami.com>
ChatGLM LLM integration will by default accumulate conversation
history(with_history=True) to ChatGLM backend api, which is not expected
in most cases. This PR set with_history=False by default, user should
explicitly set llm.with_history=True to turn this feature on. Related
PR: #8048#7774
---------
Co-authored-by: mlot <limpo2000@gmail.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>