langchain/docs/integrations/rwkv.md
Leonid Ganeline e2d7677526
docs: compound ecosystem and integrations (#4870)
# Docs: compound ecosystem and integrations

**Problem statement:** We have a big overlap between the
References/Integrations and Ecosystem/LongChain Ecosystem pages. It
confuses users. It creates a situation when new integration is added
only on one of these pages, which creates even more confusion.
- removed References/Integrations page (but move all its information
into the individual integration pages - in the next PR).
- renamed Ecosystem/LongChain Ecosystem into Integrations/Integrations.
I like the Ecosystem term. It is more generic and semantically richer
than the Integration term. But it mentally overloads users. The
`integration` term is more concrete.
UPDATE: after discussion, the Ecosystem is the term.
Ecosystem/Integrations is the page (in place of Ecosystem/LongChain
Ecosystem).

As a result, a user gets a single place to start with the individual
integration.
2023-05-18 09:29:57 -07:00

66 lines
1.9 KiB
Markdown

# RWKV-4
This page covers how to use the `RWKV-4` wrapper within LangChain.
It is broken into two parts: installation and setup, and then usage with an example.
## Installation and Setup
- Install the Python package with `pip install rwkv`
- Install the tokenizer Python package with `pip install tokenizer`
- Download a [RWKV model](https://huggingface.co/BlinkDL/rwkv-4-raven/tree/main) and place it in your desired directory
- Download the [tokens file](https://raw.githubusercontent.com/BlinkDL/ChatRWKV/main/20B_tokenizer.json)
## Usage
### RWKV
To use the RWKV wrapper, you need to provide the path to the pre-trained model file and the tokenizer's configuration.
```python
from langchain.llms import RWKV
# Test the model
```python
def generate_prompt(instruction, input=None):
if input:
return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
# Instruction:
{instruction}
# Input:
{input}
# Response:
"""
else:
return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
# Instruction:
{instruction}
# Response:
"""
model = RWKV(model="./models/RWKV-4-Raven-3B-v7-Eng-20230404-ctx4096.pth", strategy="cpu fp32", tokens_path="./rwkv/20B_tokenizer.json")
response = model(generate_prompt("Once upon a time, "))
```
## Model File
You can find links to model file downloads at the [RWKV-4-Raven](https://huggingface.co/BlinkDL/rwkv-4-raven/tree/main) repository.
### Rwkv-4 models -> recommended VRAM
```
RWKV VRAM
Model | 8bit | bf16/fp16 | fp32
14B | 16GB | 28GB | >50GB
7B | 8GB | 14GB | 28GB
3B | 2.8GB| 6GB | 12GB
1b5 | 1.3GB| 3GB | 6GB
```
See the [rwkv pip](https://pypi.org/project/rwkv/) page for more information about strategies, including streaming and cuda support.