hsh-finetune-dataset
ActiveTool of HSH Data-on-Demand
Made-to-order, answer-verified datasets for LLM fine-tuning. Describe the task (e.g. 'step-by-step math reasoning', 'SQL generation', 'instruction-following for support replies') and we deliver a clean, HuggingFace-ready dataset in Alpaca schema (instruction/input/output), deduplicated, train/val/test split, with every checkable answer verified in code. Drop the repo straight into Gradients (SN56), TRL, Axolotl, or Unsloth. Verified sample live: huggingface.co/datasets/HSH-Intelligence/verified-math-reasoning-3k. Tier S: 1-2K rows ($75). Tier M: 2-5K rows ($150). Tier L: 5-10K rows ($300). Custom/larger scoped on request.
Parameters schema
{
"type": "object",
"required": [
"task_description",
"row_count"
],
"properties": {
"domain": {
"type": "string",
"description": "Subject domain (e.g. 'math', 'SQL', 'customer support', 'legal Q&A')."
},
"row_count": {
"type": "number",
"description": "Number of training rows needed (1000-10000 standard; larger scoped on request)."
},
"schema_hint": {
"type": "string",
"description": "Preferred schema. Default: Alpaca instruction/input/output (Gradients-ready)."
},
"verification": {
"enum": [
"programmatic",
"llm_judge",
"none"
],
"type": "string",
"description": "How answers are checked. Programmatic (code-verified ground truth) where the task allows."
},
"target_platform": {
"enum": [
"gradients",
"trl",
"axolotl",
"unsloth",
"generic"
],
"type": "string",
"description": "Where you'll train — tunes the delivered format."
},
"task_description": {
"type": "string",
"description": "Plain English: what the model should learn to do (the instruction-following task)."
}
}
}No endpoints wrapped at confidence ≥ 0.70.
Parent server
HSH Data-on-Demand
https://github.com/hshintelligence/data-on-demand
2/7 registries