ChaoticEconomist/Jazz-Blues-Music-Dataset_SFT-or-LoRA

Jazz & Blues Music Dataset (SFT / LoRA Ready)

A structured dataset covering 82 iconic Jazz and Blues songs, 21 artist profiles, and 41 historical events, expanded into 1,219 instruction-tuning rows across 7 task types.

Designed for fine-tuning LLMs on music knowledge, cultural history, artist biography, and domain-specific Q&A tasks.

---

Overview

Property	Value
Domain	Jazz & Blues Music
Total rows	1,219
Train split	1,036 (85%)
Validation split	91 (~7.5%)
Test split	92 (~7.5%)
Format	CSV (Alpaca-style prompt format)
Songs covered	82 (41 Blues, 41 Jazz)
Artists covered	21 (11 Blues, 10 Jazz)
History events	41 (18 Blues, 23 Jazz)
License	CC BY 4.0

---

Row Types

Type	Count	Description
song_qa	656	8 Q&A pairs per song (artist, year, genre, key, tempo, description, significance, subgenre)
history_qa	205	5 Q&A pairs per historical event
listening_guide	82	How to listen to and appreciate each song
artist_qa	126	6 Q&A pairs per artist (genre, origin, instruments, style, legacy, birth)
song_explanation	82	Full structured explanation of each song
history_explanation	41	Full structured explanation of each historical event
artist_profile	21	Full artist biography and legacy
genre_comparison	6	Side-by-side comparison of related subgenres

---

Columns

Column	Type	Description
instruction	string	Task instruction for the model
input	string	The question, song title, artist name, or event name
completion	string	The expected model output
prompt	string	Full Alpaca-style prompt (### Instruction / ### Input / ### Response:)
text	string	prompt + completion — ready for SFT trainers
genre	string	jazz or blues
subgenre	string	Musical subgenre, historical era, or artist_profile / genre_comparison
difficulty	string	beginner, intermediate, or advanced
row_type	string	One of the 7 row types above
source	string	Always jazz_blues
split	string	train, validation, or test

---

Genres & Subgenres

Blues Subgenres

Subgenre	Key Artists
delta_blues	Robert Johnson, Elmore James, Robert Petway
chicago_blues	Muddy Waters, Howlin' Wolf, Jimmy Reed, Junior Wells
electric_blues	B.B. King, Shake Your Moneymaker
texas_blues	Stevie Ray Vaughan
west_coast_blues	T-Bone Walker, Lowell Fulson
boogie_blues	John Lee Hooker
soul_blues	Albert King, Etta James, Bobby 'Blue' Bland
country_blues	Big Bill Broonzy
rhythm_and_blues	Big Mama Thornton
classic_blues	Bessie Smith, Mamie Smith

Jazz Subgenres

Subgenre	Key Artists
new_orleans_jazz	Louis Armstrong, Jelly Roll Morton
swing_era	Duke Ellington, Benny Goodman, Billie Holiday
bebop	Charlie Parker, Dizzy Gillespie, Thelonious Monk
cool_jazz	Miles Davis, Dave Brubeck, Chet Baker
hard_bop	Herbie Hancock, Art Blakey, Bobby Timmons
modal_jazz	Miles Davis (Kind of Blue), John Coltrane
post_bop	Wayne Shorter, Miles Davis (second quintet), Bill Evans
spiritual_jazz	John Coltrane (A Love Supreme)
jazz_fusion	Weather Report, Herbie Hancock (Head Hunters)
jazz_standard	Autumn Leaves, Summertime, Body and Soul, Cherokee
traditional_jazz	Louis Armstrong
free_jazz	Ornette Coleman, Cecil Taylor

---

Difficulty Levels

Level	Criteria
beginner	Foundational songs, early history, and well-known artists
intermediate	Core genre developments, Chicago blues, bebop, cool jazz, hard bop
advanced	Modal jazz, free jazz, fusion, spiritual jazz, post-bop

---

Genre Comparisons Included

• Delta Blues vs Chicago Blues
• Bebop vs Cool Jazz
• Hard Bop vs Modal Jazz
• Texas Blues vs West Coast Blues
• Jazz Fusion vs Post-Bop
• Soul Blues vs Rhythm and Blues

---

Prompt Format

All rows use the Alpaca instruction format:

### Instruction:
Explain the following jazz or blues song, including its artist, year, genre,
musical characteristics, and historical significance.

### Input:
"So What" by Miles Davis

### Response:
**So What** — Miles Davis (1959)

**Genre:** Jazz / Modal Jazz

**Key:** D Dorian  |  **Tempo:** Slow

**Description:** The opening track of Kind of Blue; built on two modal scales
rather than complex chord changes, giving improvisers vast melodic freedom.

**Historical Significance:** Defined modal jazz and remains the most-streamed
jazz recording of all time.

---

Usage

Load with 🤗 Datasets

from datasets import load_dataset

ds = load_dataset("csv", data_files={
    "train":      "jazz_blues_data/train.csv",
    "validation": "jazz_blues_data/validation.csv",
    "test":       "jazz_blues_data/test.csv",
})

Fine-tune with TRL SFTTrainer

from trl import SFTTrainer

trainer = SFTTrainer(
    model=model,
    train_dataset=ds["train"],
    dataset_text_field="text",
    ...
)

Filter examples

# Only blues rows
blues = ds["train"].filter(lambda x: x["genre"] == "blues")

# Only artist profiles
profiles = ds["train"].filter(lambda x: x["row_type"] == "artist_profile")

# Only advanced jazz history
adv_jazz_hist = ds["train"].filter(
    lambda x: x["genre"] == "jazz"
    and x["row_type"] == "history_explanation"
    and x["difficulty"] == "advanced"
)

# Listening guides only
guides = ds["train"].filter(lambda x: x["row_type"] == "listening_guide")

---

Files

File	Description
jazz_blues_dataset.csv	Full dataset (1,219 rows)
jazz_blues_data/train.csv	Training split (1,036 rows, 85%)
jazz_blues_data/validation.csv	Validation split (91 rows, ~7.5%)
jazz_blues_data/test.csv	Test split (92 rows, ~7.5%)
gen_jazz_blues.py	Dataset generator script
jazz_blues_songs.py	Source knowledge base

---

Intended Uses

• Music LLM fine-tuning — teach models to reason about jazz and blues
• Cultural history Q&A — question answering over music history
• Artist biography generation — structured artist profiles
• LoRA adapters — lightweight music-domain adapters
• Educational tools — interactive music history tutors
• Genre classification — structured subgenre knowledge

Out-of-Scope Uses

• Does not cover rock, soul, R&B, classical, or other genres in depth
• Not a substitute for a full musicology reference or discography database

---

License

Creative Commons Attribution 4.0 International (CC BY 4.0)