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

Phogpt

PhoGPT is an open-source series of generative pretrained Vietnamese language models, comprising the PhoGPT-4B base model and the PhoGPT-4B-Chat instruction-following and conversation variant.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Phogpt — 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 PhoGPT when a project needs a Vietnamese-focused generative model that can be run through a documented Transformers or llama.cpp workflow, or adapted through fine-tuning.

Who it is for

  • Developers building Vietnamese text-generation or conversational applications
  • Researchers and practitioners exploring or fine-tuning Vietnamese language models
  • Users who want to run the model locally through documented inference engines

Documented uses

  • Generate Vietnamese text with the base pretrained model
  • Use the chat variant for instruction following and conversational prompts
  • Draft, summarize, correct spelling, or answer questions about supplied Vietnamese text
  • Fine-tune the model for a Vietnamese-language task
Before you choose

Requirements & limits

The model series is open-source under the BSD-3-Clause license. The repository documents local inference workflows; its float16 loading note specifies 7 GB of GPU memory for either model.

  • PhoGPT-4B is the base pretrained model; PhoGPT-4B-Chat is the instruction-following and conversation variant.
  • The listed context length is 8,192 tokens, and float16 loading is documented as requiring 7 GB of GPU memory.
  • The llama.cpp route involves compiling the engine and converting the model to GGUF; 4-bit and 8-bit quantization are optional.
  • The repository's Transformers example uses Python, PyTorch, Transformers, and CUDA, with bfloat16 or float16 model loading.

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 PhoGPT-4B-Chat with llama.cpp

Convert the chat model to GGUF and start inference using the repository’s llama.cpp procedure.

Before you start

  • A local PhoGPT-4B-Chat model directory
  • A compiled llama.cpp checkout
  • Python dependencies installed from llama.cpp’s requirements file

Steps

  1. Compile llama.cpp and install its Python dependencies.

  2. Convert the model to GGUF FP16 with llama.cpp’s convert-hf-to-gguf.py script.

  3. Optionally quantize the GGUF model to 4-bit or 8-bit.

  4. Start inference with llama.cpp’s main command and a Vietnamese prompt using the ### Câu hỏi: and ### Trả lời: template.

Expected outcome

A GGUF PhoGPT-4B-Chat model is used for prompt-based inference through llama.cpp.

Guide limitations

  • The instructions require compiling llama.cpp and preparing the model locally.
  • For LM Studio, the repository notes that the chat preset JSON may be needed for the provided GGUF files.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
6 November 2026
Original website title
VinAIResearch/PhoGPT · Official homepage title · retained in its source language
How we review sourcesSuggest a correction