Datasetpublic

bhavin-gajjar/laravel-coder-lora-train

Laravel Coder LoRA — end-to-end training data Instruction-tuning datasets for Laravel Bob (Qwen / CodeLlama / DeepSeek modelfiles).Built from official Laravel docs v10.x–v13.x plus version-detection examples. Each config has a single schema — do not mix raw (Alpaca) with chat / lora_* (messages). Configs Config Path Rows Schema raw raw/laravel_training.jsonl 1253 instruction, input, output, topic, version chat chat/laravel_training_chat.jsonl 1253… See the full description on the dataset page: https://huggingface.co/datasets/bhavin-gajjar/laravel-coder-lora-train.

sourceHugging Facemitupdated 3mo agoView on Hugging Face

1likes83downloads

Dataset Card

Laravel Coder LoRA — end-to-end training data

Instruction-tuning datasets for Laravel Bob (Qwen / CodeLlama / DeepSeek modelfiles). Built from official Laravel docs v10.x–v13.x plus version-detection examples.

Each config has a single schema — do not mix raw (Alpaca) with chat / lora_* (messages).

Configs

Config	Path	Rows	Schema
`raw`	`raw/laravel_training.jsonl`	1253	`instruction`, `input`, `output`, `topic`, `version`
`chat`	`chat/laravel_training_chat.jsonl`	1253	`messages`, `topic`, `version`
`lora_train`	`lora/laravel_train.jsonl`	800	`messages`, `topic`, `version`
`lora_eval`	`lora/laravel_eval.jsonl`	40	`messages`, `topic`, `version`

Auxiliary files (not dataset configs)

Path	Description
`few_shot_examples.json`	Modelfile few-shot (JSON array)
`laravel_knowledge.md`	RAG / system context digest
`meta.json`	Corpus stats
`lora/meta.json`	Curated split stats

Schema — `raw`

json

{
  "instruction": "[Laravel 11] Explain routing: Basic Routing",
  "input": "",
  "output": "...",
  "topic": "routing",
  "version": "11.x"
}

Schema — `chat`, `lora_train`, `lora_eval`

json

{
  "messages": [
    {"role": "user", "content": "..."},
    {"role": "assistant", "content": "..."}
  ],
  "version": "11.x",
  "topic": "routing"
}

Load in Python

python

from datasets import load_dataset

raw = load_dataset("bhavin-gajjar/laravel-coder-lora-train", "raw", split="train")
chat = load_dataset("bhavin-gajjar/laravel-coder-lora-train", "chat", split="train")
train = load_dataset("bhavin-gajjar/laravel-coder-lora-train", "lora_train", split="train")
eval_ds = load_dataset("bhavin-gajjar/laravel-coder-lora-train", "lora_eval", split="train")

Regenerate locally

bash

cd ollama
python3 shared/build_training_data.py
python3 finetune/prepare_dataset.py
python3 finetune/export_chat_dataset.py
bash finetune/upload_dataset_to_hub.sh

Source

Generated from laravel-docs/{10,11,12,13}.x via shared/build_training_data.py.