nectec/pathumma-thaillm-8b-think-3.0.0
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Jan 26, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Pathumma-ThaiLLM-Think-3.0.0 is an 8 billion parameter, 32768-token context length large language model developed by NECTEC, built upon the ThaiLLM foundation model. It is post-trained using a two-stage Supervised Fine-Tuning (SFT) strategy to enhance instruction following, structured tool/function calling, mathematical and coding competence, and multi-step analytical capabilities. This model excels in Thai-English bilingual robustness and is optimized for analytical and tool-augmented intelligence.

Loading preview...

Pathumma-ThaiLLM-Think-3.0.0 Overview

Pathumma-ThaiLLM-Think-3.0.0 is an 8 billion parameter large language model developed by NECTEC, leveraging the ThaiLLM national initiative's foundation model. It features a 32768-token context length and is specifically designed for advanced analytical and tool-augmented intelligence, with strong Thai-English bilingual support.

Key Capabilities

  • Enhanced Instruction Following: Improved compliance with user instructions.
  • Structured Tool/Function Calling: Capable of generating structured JSON for tool invocation.
  • Mathematical & Coding Competence: Excels in multi-step mathematical analysis, code understanding, and synthesis.
  • Multi-step Analytical Tasks: Designed for complex reasoning and structured analytical responses.
  • Thai Reasoning Distillation: Optimized for robust reasoning in Thai.
  • Bilingual Robustness: Strong performance in both Thai and English.

Training Strategy

The model underwent a two-stage Supervised Fine-Tuning (SFT) process:

  1. Stage 1: Instruction & Tool-Calling Alignment: Focused on instruction compliance, structured tool-call formatting, general Thai task robustness, and STEM-oriented instruction alignment using datasets like beyoru/ToolCall_synthetic_qwen3 and airesearch/WangchanX-FLAN-v6.
  2. Stage 2: Reasoning Specialization: Concentrated on multi-step mathematical analysis, code understanding, synthesis, and tool-calling with explicit reasoning traces, utilizing datasets such as nvidia/OpenMathReasoning and nvidia/OpenCodeReasoning.

Good For

  • Applications requiring advanced Thai-English bilingual analytical capabilities.
  • Scenarios demanding reliable tool and function calling with structured outputs.
  • Tasks involving mathematical problem-solving and code generation/analysis.
  • Use cases needing multi-step reasoning and complex analytical responses.
Finetunes
2 models