OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Rwkv Lm

Rwkv Lm is BlinkDL’s public Python and PyTorch repository for implementing, training, fine-tuning and running RWKV recurrent language models. It includes RWKV model-generation code, reference training implementations and instructions for inference and sentence embeddings.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Rwkv Lm — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use the repository when you need model code and implementation guidance for experimenting with RWKV, especially when parallelizable training and recurrent, cache-free sequential inference are relevant to your design.

Who it is for

  • Developers evaluating recurrent language-model architectures.
  • Researchers or engineers who want to train or fine-tune RWKV models.
  • People building self-hosted inference or embedding workflows from model code.

Documented uses

  • Run an RWKV-4 Pile model using the documented inference script.
  • Train an RWKV-4 model from scratch or fine-tune a Pile model.
  • Explore RWKV-7 reference training code and adapt it to another task.
  • Use final hidden-state representations as sentence embeddings.
Before you choose

Requirements & limits

The repository is public and identifies its license as Apache-2.0. The README documents model-weight downloads and local code-based workflows.

  • Instructions and dependencies differ across RWKV generations; the cited quick start is specifically for RWKV-4, while RWKV-7 has its own reference training code.
  • The README states that the RWKV-7 default training configuration requires one GPU with 7 GB of VRAM, and suggests reducing batch size for less memory.
  • The project describes RWKV-7 as linear-time, constant-space and attention-free, with sequential inference and parallelizable training.
  • The README cautions that RWKV-7 is a carefully configured whole-model implementation and recommends studying its training code before adapting it to another task.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run a documented RWKV-4 Pile inference example

Use the README's RWKV-4 Pile quick-start path to run a model with run.py.

Before you start

  • An RWKV-4 Pile model downloaded from the linked BlinkDL Hugging Face collection.
  • The RWKV-LM repository and its run.py script.

Steps

  1. Download an RWKV-4 Pile model from the linked model collection.

  2. Set TOKEN_MODE to 'pile' in run.py.

  3. Run run.py.

Expected outcome

The documented inference path runs an RWKV-4 Pile model.

Guide limitations

  • These steps describe RWKV-4 Pile models and do not specify a universal setup for other RWKV versions.
  • The README does not provide a complete environment setup in this inference subsection.
Documentation 1 ↗
Source-based practical guide

Set up the documented RWKV-4 training environment

Install the versions listed for the RWKV-4 training and fine-tuning instructions.

Before you start

  • A Python environment.
  • CUDA 11.7 or 11.7.1.

Steps

  1. Install deepspeed==0.7.0.

  2. Install pytorch-lightning==1.9.5.

  3. Install torch==1.13.1+cu117.

Expected outcome

The listed software dependencies are installed for the README's RWKV-4 training instructions.

Guide limitations

  • The README warns that torch 2 with DeepSpeed has bugs that can hurt model performance.
  • This setup guide covers dependency installation only; it does not establish that a training run succeeds.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
Rwkv LmRun an RWKV-4 Pile model using the documented inference script.Linux, macOS, Self-hosted, WindowsThe repository is public and identifies its license as Apache-2.0. The README documents model-weight downloads and local code-based workflows.
Pulse Of MotionEstimate a video's physical frame rate from its visual dynamics.LinuxThe repository is public. Its README describes a released checkpoint that downloads from Hugging Face on first inference, with a manual download option. The strongest internal model is not fully released; the README directs interested users to contact the project about access. No license information is provided in the repository evidence.
Neural Image AssessmentTrain the model using AVA images and rating annotations.Cloud, Linux, macOS, Windows
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
BlinkDL/RWKV-LM · Official homepage title · retained in its source language
How we review sourcesSuggest a correction