openthai/ openthai-2.0-legal:latest

7 Downloads Updated 7 hours ago

Thai legal LLM that cites the exact law name and section (มาตรา). 30B MoE, ~3B active per token — runs in ~24 GB. Best paired with retrieval (RAG). Decision support, not legal advice.

tools thinking

ollama run openthai/openthai-2.0-legal

curl http://localhost:11434/api/chat \
  -d '{
    "model": "openthai/openthai-2.0-legal",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='openthai/openthai-2.0-legal',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'openthai/openthai-2.0-legal',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 7 hours ago

7 hours ago

f863edb2699a · 25GB ·

model

archnemotron_h_moe

·

parameters31.6B

·

quantizationQ4_K_M

25GB

system

You are OpenThai 2.0 Legal, a Thai legal language model from the OpenThai project (AIEAT and iApp Te

267B

license

NVIDIA Open Model Agreement — https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-o

264B

params

{ "num_ctx": 32768, "temperature": 0.3, "top_p": 0.9 }

48B

Readme

OpenThai 2.0 Legal 30B-A3B ⚖️🇹🇭

A Thai legal LLM that cites the exact law and section (มาตรา) — and runs inside your own firewall.

ollama run openthai/openthai-2.0-legal

Purpose-built for Thai statutory law and case law. Recalls Thai statutes from memory and returns the law name and section in a structured, verifiable format — made for legal RAG pipelines, drafting tools and compliance work where data cannot leave the organisation.

30B Mixture-of-Experts with only ~3B parameters active per token, so it runs in ~24 GB — hybrid Mamba2-Transformer on the NVIDIA Nemotron-3-Nano base, trained with reinforcement learning where the reward is citation accuracy itself.

Example

>>> ลักทรัพย์ในเวลากลางคืน ผิดมาตราใด
ผิดตามมาตรา ๓๓๕ แห่งประมวลกฎหมายอาญา

⚠️ Use it right

Decision support, not legal advice. Verify every cited section against the current statute text.
Pair it with retrieval. Open-book citation accuracy is 0.99 versus 0.07–0.40 from pure memory — recall is the fallback, RAG is the design. Start with OpenThaiRAG.
Trained on Thai statutes and Revenue Department rulings; very recent amendments may be thinner.

Links

Announcement · Live demo · GGUF weights (Q4/Q5/Q8) · Discord

Free and open-weight, from the OpenThai project (formerly OpenThaiGPT) — AIEAT × iApp Technology, built on the NVIDIA Nemotron/NeMo stack.

# OpenThai 2.0 Legal 30B-A3B ⚖️🇹🇭

**A Thai legal LLM that cites the exact law and section (มาตรา) — and runs inside your own firewall.**

```
ollama run openthai/openthai-2.0-legal
```

Purpose-built for Thai statutory law and case law. Recalls Thai statutes from memory and
returns the law name and section in a structured, verifiable format — made for legal RAG
pipelines, drafting tools and compliance work where data cannot leave the organisation.

**30B Mixture-of-Experts with only ~3B parameters active per token**, so it runs in
**~24 GB** — hybrid Mamba2-Transformer on the NVIDIA Nemotron-3-Nano base, trained with
reinforcement learning where the reward is citation accuracy itself.

## Example

```
>>> ลักทรัพย์ในเวลากลางคืน ผิดมาตราใด
ผิดตามมาตรา ๓๓๕ แห่งประมวลกฎหมายอาญา
```

## ⚠️ Use it right

- **Decision support, not legal advice.** Verify every cited section against the current
  statute text.
- **Pair it with retrieval.** Open-book citation accuracy is 0.99 versus 0.07–0.40 from
  pure memory — recall is the fallback, RAG is the design. Start with
  [OpenThaiRAG](https://github.com/OpenThaiGPT/openthairag).
- Trained on Thai statutes and Revenue Department rulings; very recent amendments may be
  thinner.

## Links

[Announcement](https://openthai.aieat.or.th/openthai2p0-legal) ·
[Live demo](https://iapp.co.th/openmodels/openthai2p0-legal) ·
[GGUF weights (Q4/Q5/Q8)](https://huggingface.co/iapp/openthai2.0-legal-thaillm-nemotron-3-nano-30b-a3b-GGUF) ·
[Discord](https://discord.gg/7KDdKkBGUs)

*Free and open-weight, from the OpenThai project (formerly OpenThaiGPT) — AIEAT × iApp
Technology, built on the NVIDIA Nemotron/NeMo stack.*

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)