Skip to content
View cris96spa's full-sized avatar

Highlights

  • Pro

Block or report cris96spa

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
cris96spa/README.md

Hi, I'm Cristian

I'm an NLP engineer at Artificialy in Lugano, where I make large language models run faster, fit in less GPU memory, and less confident about things that never happened. The part I care about is the one most people skip: not that something works, but why - that's how I ended up rebuilding GPT-2 from scratch on weekends.

The longer version streams token by token at cris96spa-latent-space.com, a site that animates a real GPT-2 forward pass because a PDF resume cannot do inference.

A snake eating this year's contribution graph, one commit at a time

Model card

field value
base_model decoder-only human, one attention head
pretraining MSc in Computer Science & Engineering, Politecnico di Milano, cum laude
fine_tuning production LLMs at Artificialy - quantization and inference runtimes (vLLM, TensorRT), SFT / DPO / GRPO, evaluation pipelines
objective training-loss curves going down. I could watch them for the rest of my life
temperature 0
evals table football: SOTA. Dragon Ball: Sparking! Zero: ask Federico for a reference
known_limitations red and green share an embedding. R is out of vocabulary

Checkpoints

repo what it is
latent-space My site. Animates a real GPT-2 forward pass and streams my bio token by token.
LLMs-from-scratch GPT-2 rebuilt from the tensor up - tokenizer, attention, training loop. Knowing why beats knowing that.
fyt My girlfriend was running her thesis experiments in R. I hate R. The deal: she stops, I build her the framework. Here it is.
teleportMPD Master's thesis: Teleport MDPs, a formal account of why curricula work in RL where sparse rewards leave vanilla PPO wandering.
hackapizza 2nd place at Hackapizza 2025: a RAG agent that turns Galactic Federation food law into a knowledge graph, so nobody orders the illegal pizza.
LeetCode NeetCode 250 in Python, with a progress table kept honest by a script, not by vibes.

The rest lives in the repositories tab; the projects page has the director's cut.

Config

stack:
  daily:    [python, pytorch, fastapi, pydantic, polars, uv, ruff]
  llm:      [transformers, vllm, tensorrt, langgraph, mlflow]
  infra:    [docker, gcp, aws, github-actions]
  frontend: [react, typescript]  # Claude wrote it. My CSS loss never converged
  r:        null                 # see fyt

latent space · LinkedIn · email · resume

Written at temperature 0.

Pinned Loading

  1. latent-space latent-space Public

    TypeScript 1

  2. fyt fyt Public

    Python

  3. python-repo-template python-repo-template Public template

    Forked from GiovanniGiacometti/python-repo-template

    A Python template repository

    Python 1

  4. teleport-mdp teleport-mdp Public

    Python

  5. LeetCode LeetCode Public

    Python 1

  6. hackapizza hackapizza Public

    Forked from GiovanniGiacometti/hackapizza-hackathon

    Python 1