n0uur/pathumma-thaillm-8b-think-3.0.0_GGUF

Pathumma-ThaiLLM-8B-Think-3.0.0 GGUF

Hi there! I created this repository to experiment with converting and quantizing the Thai LLM model. This is a learning project, not a production release - please verify important outputs before use!

Model Overview

This repository contains BF16 and quantized GGUF versions of NECTEC's Pathumma-ThaiLLM-8B-Think-3.0.0, converted using llama.cpp. The files preserve the original reasoning-aware chat template. This is an unofficial conversion, not an official NECTEC release.

Available Files

Choose one based on your hardware constraints:

Sizes are decimal GB and refer to files, not total RAM or VRAM requirements.

Conversion Process

• —Original model: nectec/pathumma-thaillm-8b-think-3.0.0
• —Tools: llama.cpp b10964 (Windows build)
• —Process:
• —Convert BF16 → GGUF
• —Embed chat template
• —Apply quantization presets

Exact conversion details are in the conversion manifest. File hashes are in SHA256SUMS.

Testing Results

Sample prompt:

ประเทศไทยมีเมืองหลวงชื่ออะไร? ตอบเป็นภาษาไทยหนึ่งประโยค

Note: These are basic smoke tests, not a full benchmark. BF16 was checked structurally but was not separately generation-tested.

Credits & License

• —Original model: NECTEC's Pathumma-ThaiLLM-8B-Think-3.0.0 (Apache-2.0 license)
• —Conversion tools: llama.cpp b10964
• —Special thanks to:
• —NECTEC and ThaiLLM team
• —ggml-org and llama.cpp community
• —README writing: Pathumma-ThaiLLM-8B-Think-3.0.0 (Q4KM).

My contribution: This is my first experiment with GGUF conversion and quantization. I'm still learning and welcome all suggestions to improve this project!