PhoGPT 4B Chat
PhoGPT-4B-Chat is VinAI's 3.7B-parameter Vietnamese chat model, fine-tuned from a base trained on 102B Vietnamese tokens. It handles up to 8192-token contexts and was instruction-tuned on 360K conversational examples. BSD-3-Clause licensed, so commercial use is fine.
If you need a commercially usable Vietnamese chat model and have modest hardware, PhoGPT-4B-Chat is one of the few purpose-built options at this size. The 102B-token pretraining corpus and 360K fine-tuning examples are solid foundations for a 3.7B model. That said, the low download count means you're mostly on your own if something breaks, and the parameter ceiling will show quickly on anything requiring real reasoning. Hedge — worth a test run for simple Vietnamese chat tasks, but don't expect GPT-4-level coherence.
›Why this rating
Auto-generated rating (Opus 4.7 judge, claude-opus-4-7). Overall 9.25/10. License is explicitly bsd-3-clause on the card and commercial use is correctly flagged. Params (3.7B), context (8192), and vendor (VinAI) all match the card. Minor nit: the row says '360K conversational examples' but the card splits it as 70K instructional + 290K conversations — the sum is correct but the framing is slightly loose. Editorial voice is honest, weaknesses call out low traction and missing GGUF, and the use case (Vietnamese chat/FAQ) is sharp and well-targeted. Solid operator-grade row for a narrow but legitimate niche.
Flags: - Minor: '360K conversational examples' conflates 70K instruction + 290K conversation splits described in the card
Overview
PhoGPT-4B-Chat is VinAI's 3.7B-parameter Vietnamese chat model, fine-tuned from a base trained on 102B Vietnamese tokens. It handles up to 8192-token contexts and was instruction-tuned on 360K conversational examples. BSD-3-Clause licensed, so commercial use is fine.
Strengths
- Pretrained on 102B Vietnamese tokens — one of the largest Vietnamese-specific training sets publicly documented
- Instruction-tuned on 360K examples, giving it reasonable chat behavior out of the box
- 8192-token context window is generous for a sub-4B model
- BSD-3-Clause license allows commercial deployment without royalty concerns
Weaknesses
- Vietnamese only — no meaningful capability in other languages
- 3.7B parameters puts a hard ceiling on complex reasoning and multi-step tasks
- Low community traction (1.6K downloads, 43 likes) means limited third-party testing and few reported real-world benchmarks
- No publicly available GGUF or quantized builds confirmed — check compatibility with your inference stack before committing
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|---|---|
| Q4_K_M | 2.0 GB | 3 GB |
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of PhoGPT 4B Chat.
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
What's the minimum VRAM to run PhoGPT 4B Chat?
Can I use PhoGPT 4B Chat commercially?
What's the context length of PhoGPT 4B Chat?
Source: huggingface.co/vinai/PhoGPT-4B-Chat
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify PhoGPT 4B Chat runs on your specific hardware before committing money.