messages listlengths 3 696 |
|---|
[
{
"content": "You are OpenHands agent, a helpful AI assistant that can interact with a computer to solve tasks.\n\n<ROLE>\nYour primary role is to assist users by executing commands, modifying code, and solving technical problems effectively. You should be thorough, methodical, and prioritize quality over speed... |
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED) |
[{"content":"You are an autonomous engineering agent powered by LFM2.5, specialized in software engi(...TRUNCATED) |
[{"content":"You are an autonomous engineering agent powered by LFM2.5, specialized in software engi(...TRUNCATED) |
[{"content":"You are an autonomous engineering agent powered by LFM2.5, specialized in software engi(...TRUNCATED) |
[{"content":"You are OpenHands agent, a helpful AI assistant that can interact with a computer to so(...TRUNCATED) |
[{"content":"You are a customer service agent that helps the user. The policy that determines how y(...TRUNCATED) |
[{"content":"You are an autonomous engineering agent powered by LFM2.5, specialized in software engi(...TRUNCATED) |
[{"content":"You are a helpful assistant.","role":"system"},{"content":"What is the primary concern (...TRUNCATED) |
[{"content":"You are a customer service agent that helps the user. The policy that determines how y(...TRUNCATED) |
End of preview. Expand in Data Studio
Senigiro/lfm2.5-24b-ornith-scaffold-agentic
This dataset contains multi-turn, self-scaffolding agentic trajectories tailored for fine-tuning LiquidAI/LFM2.5-24B-A2B.
Dataset Summary
| Split | Samples |
|---|---|
| Train | 115,116 |
| Validation | 6,059 |
| Total | 121,175 |
Data Sources
| Source | Samples | Topics |
|---|---|---|
swe_zero |
15000 | agentic-coding,SWE-bench,debugging,code-repair |
openshand_sft |
491 | agentic-coding,SWE-bench,multi-turn,tool-use |
r2egym_sft |
3231 | agentic-coding,SWE-bench,multi-turn,tool-use |
swe_rebench |
15000 | agentic-coding,SWE-bench,multi-turn,tool-use |
terminal_bench |
1189 | terminal,linux,sysadmin,cli,debugging |
nemotron_agentic |
15000 | tool-use,function-calling,agentic,interactive |
hermes_fc |
4965 | function-calling,tool-use,structured-output |
glaive_fc |
15000 | function-calling,tool-use,structured-output |
devops_k8s_sft |
15000 | devops,kubernetes,CI/CD,cloud,terraform,gitops |
devops_qa |
8075 | devops,docker,kubernetes,CI/CD,linux,cloud |
stackexchange_devops |
13224 | devops,linux,CI/CD,monitoring,troubleshooting |
fable5_agentic |
15000 | agentic-coding,tool-use,shell,file-edits |
Scaffolding method
Every assistant turn that invokes tools or code includes an explicit <thought> block
containing:
- [SCAFFOLD SETUP] - Objective, tool harness, invariant checks, execution plan
- [DIAGNOSIS & RECOVERY] - Observations, corrective actions, retry strategy
- [VERIFICATION] - Gate checking, output validation
ChatML Schema
Standard ChatML message format (system, user, assistant, tool) compatible with
LFM2.5's native chat template. Ready for use with TRL's DataCollatorForCompletionOnlyLM.
Target Capabilities
This dataset trains the model for:
- Agentic coding: Repository navigation, bug fixing, test-driven development
- CI/CD pipelines: GitHub Actions, Jenkins, GitOps workflows
- Linux server deployment: System administration, package management, service configuration
- Cloud infrastructure: AWS, Azure, GCP, Kubernetes, Terraform
- Architectural thinking: System design, scalability, trade-off analysis
- Debugging: Root cause analysis, log parsing, error recovery
Usage
from datasets import load_dataset
ds = load_dataset("Senigiro/lfm2.5-24b-ornith-scaffold-agentic")
print(ds["train"][0]["messages"])
- Downloads last month
- 110