Nanthasit/sakthai-context-1.5b-merged

Text Generation • 2B • Updated 20 days ago • 803 •

Note Flagship model — 1,599 downloads. Tool-calling GGUF, 1.5B params, merged from Qwen2.5. Most popular in the family.

Nanthasit/sakthai-context-7b-128k

Text Generation • Updated 27 days ago • 287

Note Long context specialist - 506 downloads. 7B GGUF with 128K context window for large-document tasks.

Nanthasit/sakthai-context-1.5b-tools

Text Generation • Updated 20 days ago • 248

Note Tool-use specialist - 349 downloads. 1.5B fine-tuned for structured function calling.

Nanthasit/sakthai-context-7b-tools

Text Generation • Updated 27 days ago • 287

Note Heavy tool-use - 399 downloads. 7B fine-tuned for complex multi-tool orchestration.

Nanthasit/sakthai-vision-7b

Image-Text-to-Text • 7B • Updated 27 days ago • 510 • 1

Note Vision-language model - 186 downloads. LLaVA-1.5-7B GGUF for image understanding and captioning.

Nanthasit/sakthai-tts-model

Text-to-Speech • 81.7M • Updated 27 days ago • 215

Note Voice of the family - 150 downloads. Kokoro-82M TTS, 15 languages, edge-tts compatible.

Nanthasit/sakthai-embedding-multilingual

Sentence Similarity • 0.1B • Updated 27 days ago • 439

Note Cross-lingual embeddings - 362 downloads. Multilingual sentence embeddings for 50+ languages.

Note Training data - 246 downloads. 2,003 tool-calling examples + safety/edge cases for fine-tuning.

Note Training notebooks - 184 downloads. Kaggle notebooks used to train the SakThai models.

Note Tool-calling foundation - 58 downloads. Base dataset for structured tool-use training.

Note Food penguin dataset - 89 downloads. Curated food images for classification experiments.

Nanthasit/sakthai-context-0.5b-tools

Text Generation • 0.5B • Updated 27 days ago • 1.01k

Note Lightweight LoRA adapter for edge tool-calling — 94 downloads. LoRA weights for Qwen2.5-0.5B

Note Irrelevance detection training supplement — teaches models when NOT to call a tool. 10 examples, essential for BFCL irrelevance. 0 downloads — first download validates the approach!

Note v7 tool-calling training dataset -- 0 downloads. 2,003 examples across 81 tools, 12 domains, with safety/edge cases and multi-turn dialogues.

Note Balanced BFCL benchmark - 0 downloads. 235 rows across 4 categories: single-turn, multi-turn, parallel, relevance.

Note Multi-domain benchmark - 0 downloads. 500 rows across 5 categories: simple, multi-turn, parallel, irrelevance, edge cases.

Nanthasit/sakthai-plus-1.5b

Text Generation • 2B • Updated 26 days ago • 1.02k

Note Plus model -- enhanced 1.5B with improved instruction following and tool-use capabilities.

Nanthasit/sakthai-coder-browser-lora

Text Generation • Updated 27 days ago • 351

Note Browser/code tool-use dataset — supports agentic web/code interactions. Core training data for SakThai family tool-use pipeline.

Note v10 combined training dataset -- latest iteration with expanded tool-calling examples and improved quality filters.

Note RL training environment for tool-calling agents (GRPO/TRL/OpenEnv) - coding-task trajectories. 0 downloads - first download validates the approach!

Note OpenEnv agentic-execution training data - newest family dataset. 0 downloads - first download validates the approach!

Nanthasit/sakthai-context-1.5b-merged-v2

Text Generation • 2B • Updated 27 days ago • 563

Note v2 merged flagship — updated weights, improved tool-calling, 1.5B GGUF.

Nanthasit/sakthai-context-0.5b-merged

Text Generation • 0.5B • Updated 26 days ago • 798

Note Lightweight merged context model — 0.5B params, optimized for edge inference.

Nanthasit/sakthai-coder-1.5b

Text Generation • 2B • Updated 27 days ago • 245

Note Coder model — 151 downloads. 1.5B code-generation GGUF from Qwen2.5.

Nanthasit/sakthai-context-1.5b-tools-v2

Text Generation • Updated 27 days ago • 274

Note Improved 1.5B tool-use variant v2 — cleaner SFT mix, stronger edge-case handling.

Note Aggregated benchmark results / eval YAML logs for the SakThai family.

Nanthasit/sakthai-context-7b-merged

Text Generation • 8B • Updated 26 days ago • 598

Note Merged 7B context model — 1,055 downloads. Qwen2.5-based merged weights with extended context support.

Note Auto-managed collection item for sakthai-coder-browser