Skip to content
DeepForgeby svx
Advancedgenerative-modelsvaegansdiffusionlanguage-models

Generative Models Primer

28 problems·~12h·4 stages

How models create text and images: start with language-model sampling and alignment, then diffusion, then VAEs, and finish with GANs and sample-quality metrics. You will implement reconstruction and KL losses, diffusion noise schedules, classifier-free guidance, and preference-optimization objectives. Best for learners who already know backprop and want the generative frontier.

Your progress

0/28 · 0%

Path readiness 0%Easy 0% · Medium 0% · Hard 0%

Continue path · Causal LM Shift

What you will learn

  • Understand language-model generation, preference optimization, and PEFT
  • Implement diffusion forward processes, schedules, and DDIM sampling
  • Derive VAE losses: reconstruction, KL, reparameterization, and ELBO
  • Train GAN-style objectives and measure sample quality

Before you start

1Language-Model Generation and Alignment

0/8

Decode, sample, and align: causal LM shifts, span corruption, rejection sampling, self-consistency voting, DPO, GRPO advantage, PPO clipping, and LoRA.

Stage checkpoint

0/2 to pass

A checkpoint is a short retrieval test drawn from this stage: solve 2 of 3 from memory to mark the stage complete. The stage stays open either way.

Tried this set and did not pass? Record it — the next set swaps in fresh problems from the same stage.
LabRecommended next stepXOR BoundaryHard

2Diffusion Models

0/8

Add noise, then learn to remove it: the linear beta schedule, U-Net downsample counts, forward steps, cumulative alpha bars, MSE loss, classifier-free guidance, time embeddings, and a DDIM step.

Stage checkpoint

0/2 to pass

A checkpoint is a short retrieval test drawn from this stage: solve 2 of 3 from memory to mark the stage complete. The stage stays open either way.

Tried this set and did not pass? Record it — the next set swaps in fresh problems from the same stage.
LabRecommended next stepXOR BoundaryHard

3Autoencoders and VAEs

0/6

Compress and reconstruct: VAE reconstruction loss, the KL term, reparameterization, beta-VAE loss, the ELBO, and a sparse-autoencoder objective.

Stage checkpoint

0/2 to pass

A checkpoint is a short retrieval test drawn from this stage: solve 2 of 3 from memory to mark the stage complete. The stage stays open either way.

Tried this set and did not pass? Record it — the next set swaps in fresh problems from the same stage.
LabRecommended next stepXOR BoundaryHard

4GANs and Sample Quality

0/6

Adversarial training and evaluation: GAN discriminator and generator losses, WGAN critic loss, Frechet distance, FID feature statistics, and contrastive divergence.

Stage checkpoint

0/2 to pass

A checkpoint is a short retrieval test drawn from this stage: solve 2 of 3 from memory to mark the stage complete. The stage stays open either way.

Tried this set and did not pass? Record it — the next set swaps in fresh problems from the same stage.
LabRecommended next stepXOR BoundaryHard

Capstone

Collection

Close the path with one real artifact from the catalog.

Modern Post-Training

DPO, GRPO, and PPO-RLHF: the alignment half of this path in one focused set.