Datasets:

usernamebetter
/

agentic-100k-v3

Parquet error: Scan size limit exceeded: attempted to read 446894263 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files

Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Agentic-100K V3

Quality-filtered coding-agent SFT dataset.

FABLE.5 requirement

FABLE.5 traces are a mandatory component.

Source:

Nexlab/fable5-agentic-coding-sft

Minimum required FABLE.5 rows:

25,000

Target FABLE.5 rows:

35,000

Actual FABLE.5 rows:

35,000

Final size

100,000 rows

Hard maximum:

100,000 rows

Domains

{ "agentic": 69588, "coding": 18955, "web": 2899, "reasoning": 8558 }

Sources

{ "Nexlab/fable5-agentic-coding-sft": 35000, "nebius/SWE-agent-trajectories": 19803, "thoughtworks/agentic-coding-trajectories": 14785, "r0b0tlab/qwen3.8-max-distillation-50k": 12772, "Siddh07ETH/Atlas-Coder-50K-ChatML": 14998, "AdhyanshVerma/html-css-js-cot": 2642 }

FABLE.5 preservation

FABLE.5 messages and tool-call information are retained.

The preprocessing adds an explicit [FABLE5_TRACE] marker for provenance.

Processing

streaming dataset loading
CPU-only processing
quality filtering
exact deduplication
code detection
debugging detection
reasoning detection
web-development detection
agent/tool-use detection
source provenance
FABLE.5 minimum enforcement
hard 100K row limit

Files

train.parquet
train.jsonl
metadata.json

Upstream attribution

FABLE.5:

Nexlab/fable5-agentic-coding-sft

Other upstream datasets are listed in metadata.json.

Check each upstream dataset's license before redistribution.

Downloads last month: 21