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

Longwriter

LongWriter is an open-source research project and model series for generating very long text from long-context language models. Its repository includes model deployment examples, a training dataset and code, an automated data-construction pipeline, and benchmarks for evaluating long-form output.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Longwriter — 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 LongWriter when you need an open model and research resources for extended text generation, or when experimenting with the data, training and evaluation workflows that support it.

Who it is for

  • Developers deploying long-context models for extended text generation
  • Researchers studying long-form language-model output, training data or evaluation

Documented uses

  • Generate lengthy reports, articles, guides or other extended prose with LongWriter models.
  • Deploy model inference using the documented Transformers example or the repository's vLLM path.
  • Prepare training data and run model-training experiments using LongWriter-6k and the provided training code.
  • Construct ultra-long-output data with AgentWrite and assess output quality or length with the supplied benchmarks.
Before you choose

Requirements & limits

The repository is public and specifies the Apache-2.0 license. Deployment is described for a local or server environment; no hosted service access is described.

  • The project offers two named LongWriter models, based respectively on GLM-4-9B and Meta-Llama-3.1-8B, as well as LongWriter-Zero.
  • Deployment guidance recommends Transformers 4.43.0 or later; the repository also provides a vLLM inference path.
  • Training uses LongWriter-6k and code adapted from LongAlign, with FlashAttention 2 required by the instructions.
  • AgentWrite requires an API key configured in its files; the evaluation quality script also requires an OpenAI API key and uses GPT-4o as judge.
  • LongBench-Write focuses on output quality and length, while LongWrite-Ruler is described as a lightweight stress test of maximum output length.

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

Load a LongWriter model for text generation

Use the README's Transformers example to request an extended response from LongWriter-glm4-9b.

Before you start

  • Python environment with Transformers 4.43.0 or later, as recommended by the repository
  • PyTorch and access to the model weights from THUDM/LongWriter-glm4-9b
  • A device environment suitable for loading the selected model; the README example uses bfloat16 and device_map="auto"

Steps

  1. Load the tokenizer and causal language model from THUDM/LongWriter-glm4-9b using the documented Transformers classes and trust_remote_code=True.

  2. Set the model to evaluation mode.

  3. Call model.chat with a prompt, an empty history, and generation settings such as the documented max_new_tokens and temperature values.

  4. Print or otherwise use the returned response.

Expected outcome

A generated response from the selected LongWriter model; the README example prompts it to write a 10,000-word China travel guide.

Guide limitations

  • The README recommends Transformers 4.43.0 or later; it does not specify hardware requirements or guarantee a particular generation time.
  • The model-loading example uses bfloat16, trust_remote_code=True and automatic device mapping.
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
LongwriterGenerate lengthy reports, articles, guides or other extended prose with LongWriter models.Linux, Self-hosted, ServerThe repository is public and specifies the Apache-2.0 license. Deployment is described for a local or server environment; no hosted service access is described.
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
13 October 2026
Original website title
THUDM/LongWriter · Official homepage title · retained in its source language
How we review sourcesSuggest a correction