ift
Datasets
All datasets matching “ift”general-reasoning-ift-pairs
Reasoning-IFT Pairs (General Domain)
This dataset provides the largest set of IFT and Reasoning answers pairs for a set of general domain queries (cf: math-domain).It is based on the Infinity-Instruct dataset, an extensive and high-quality collection of instruction fine-tuning data.
We curated 900k queries from the 7M_core subset of Infinity-Instruct, which covers multiple domains including general knowledge, commonsense Q&A, coding, and math.For each query… See the full description on the dataset page: https://huggingface.co/datasets/Scale-or-Reason/general-reasoning-ift-pairs.ift-eval-us-dpo
Gambar eval — arm DPO (foto asli & sintetis) dan SFT even, US, gender, SD 1.5
Semua gambar di sini dirender model (bukan foto asli), plus label judge
gemini-3.1-pro-preview per gambar. Protokol pilot_sd15: 14 okupasi, n=100/okupasi
(test), 30 prompt/okupasi (val).
folder
model
checkpoint
dpo_sd15/test_ep*
fassabilf/sd15-dpo-us-sd15
ep5/10/15/20
dpo_real/test_ep*
fassabilf/sd15-dpo-us-real
ep5/10/15/20
dpo_real/val_ep*
fassabilf/sd15-dpo-us-real
ep2..20 tiap 2… See the full description on the dataset page: https://huggingface.co/datasets/fassabilf/ift-eval-us-dpo.ift-eval-us-real
Eval IFT SD 1.5 — foto asli, US, gender
Gambar hasil generate dan label Gemini untuk arm foto asli (fassabilf/sd15-ift-us-real,
dataset train fassabilf/ift-train-us-real). Bukan run sintetis — itu ada di
fassabilf/ift-test-us-sd15.
split
checkpoint
protokol
n per okupasi
test_ep*
ep5, 10, 15, 20, 25, 30
pilot_sd15 apa adanya: varian framing, --seed-mode paired --seed-base 0, --max-attempts 3, batch 8
100
val_ep*
ep2..ep30 tiap 2 epoch
sama, tapi --max-attempts 1… See the full description on the dataset page: https://huggingface.co/datasets/fassabilf/ift-eval-us-real.math-reasoning-ift-pairs
Reasoning-IFT Pairs (Math Domain)
Paper | Project Page
This dataset provides the largest set of IFT and Reasoning answers pairs for a set of math queries (cf: general-domain).
It is based on the Llama-Nemotron-Post-Training dataset, an extensive and high-quality collection of math instruction fine-tuning data.
We curated 150k queries from the math subset of Llama-Nemotron-Post-Training, which covers multiple domains of math questions.For each query, we used… See the full description on the dataset page: https://huggingface.co/datasets/Scale-or-Reason/math-reasoning-ift-pairs.ift-eval-us-dpo-even
Eval Diffusion-DPO SD 1.5 — foto asli, UK, gender
Gambar hasil generate dan label Gemini untuk arm DPO foto asli UK
(fassabilf/sd15-dpo-uk-real, pasangan preferensi di folder dpo/ dataset
fassabilf/ift-train-uk-real). Bukan arm SFT/IFT — itu ada di fassabilf/ift-eval-uk-real.
20 okupasi terlatih (sufiks _uk), 2.000 pasangan train / 600 val, 63 step/epoch,
20 epoch = 1.260 step.
split
checkpoint
protokol
n per okupasi
test_ep*
ep5, 10, 15, 20
pilot_sd15 apa adanya:… See the full description on the dataset page: https://huggingface.co/datasets/fassabilf/ift-eval-us-dpo-even.general-reasoning-ift-pairs
Reasoning-IFT Pairs (General Domain)
This dataset provides the largest set of IFT and Reasoning answers pairs for a set of general domain queries (cf: math-domain).It is based on the Infinity-Instruct dataset, an extensive and high-quality collection of instruction fine-tuning data.
We curated 900k queries from the 7M_core subset of Infinity-Instruct, which covers multiple domains including general knowledge, commonsense Q&A, coding, and math.For each query, we… See the full description on the dataset page: https://huggingface.co/datasets/Sidsidney/general-reasoning-ift-pairs.
