Add missing datasets: SWE agents, function calling, and general

[mohamedkarimchami]

New datasets added

General

• AM-DeepSeek-R1-Distilled-1.4M — 1.4M math+code reasoning traces distilled from DeepSeek-R1 (CC-BY-NC-4.0)
• Nemotron-Post-Training-Dataset-v2 — 6.1M multilingual SFT/RL data across 5 languages (CC-BY-4.0)

Code

• Nemotron-Competitive-Programming-v1 — competitive programming dataset with C++/Python/Infinibyte splits, predecessor to the v2 already listed (CC-BY-4.0)

Agent & Function calling

• Nemotron-Agentic-v1 — NVIDIA interactive agent + tool-call trajectories (CC-BY-4.0)
• glaive-function-calling-v2 — 113k multi-turn function-calling conversations, widely used baseline (Apache-2.0)
• When2Call — teaches models when to call tools vs. answer directly (CC-BY-4.0)
• Agent-FLAN — InternLM multi-task agent fine-tuning combining ReAct, T-FLAN, and ToolBench (Apache-2.0)

New section: SWE & Terminal Agents
Added a new subsection for software engineering and terminal agent trajectory datasets, which have become a major category in 2025–2026 but were not represented in the list:

• SWE-ZERO-12M-trajectories — 12.3M SWE trajectories (Apache-2.0)
• Open-SWE-Traces — 207k NVIDIA SWE traces with OpenHands and SWE-agent subsets (CC-BY-4.0)
• Nemotron-Terminal-Corpus — 365k terminal agent SFT data from NVIDIA (CC-BY-4.0)
• OpenThoughts-Agent-SFT-100K — 94k SWE+terminal trajectories (Apache-2.0)
• OpenThoughts-Agent-v1-SFT — 15k terminal agent trajectories (Apache-2.0)