tysam-code / hlb-gpt
Minimalistic, extremely fast, and hackable researcher's toolbench for GPT models in 307 lines of code. Reaches <3.8 validation loss on wikitext-103 on a single A100 in <100 seconds. Scales to larger models with one parameter change (feature currently in alpha).
☆345Updated 9 months ago
Alternatives and similar repositories for hlb-gpt:
Users that are interested in hlb-gpt are comparing it to the libraries listed below
- Fast bare-bones BPE for modern tokenizer training☆154Updated last month
- Puzzles for exploring transformers☆346Updated 2 years ago
- ☆412Updated last year
- Simple Transformer in Jax☆136Updated 10 months ago
- An interactive exploration of Transformer programming.☆263Updated last year
- ☆217Updated 9 months ago
- JAX implementation of the Llama 2 model☆218Updated last year
- ☆92Updated last year
- Legible, Scalable, Reproducible Foundation Models with Named Tensors and Jax☆569Updated this week
- Visualizing the internal board state of a GPT trained on chess PGN strings, and performing interventions on its internal board state and …☆204Updated 5 months ago
- Fully fine-tune large models like Mistral, Llama-2-13B, or Qwen-14B completely for free☆231Updated 6 months ago
- Ungreedy subword tokenizer and vocabulary trainer for Python, Go & Javascript☆577Updated 10 months ago
- Erasing concepts from neural representations with provable guarantees☆228Updated 3 months ago
- Extract full next-token probabilities via language model APIs☆242Updated last year
- A crude RLHF layer on top of nanoGPT with Gumbel-Softmax trick☆289Updated last year
- A Jax-based library for building transformers, includes implementations of GPT, Gemma, LlaMa, Mixtral, Whisper, SWin, ViT and more.☆286Updated 8 months ago
- A puzzle to learn about prompting☆127Updated last year
- Inference code for Persimmon-8B☆415Updated last year
- The repository for the code of the UltraFastBERT paper☆517Updated last year
- batched loras☆341Updated last year
- MiniHF is an inference, human preference data collection, and fine-tuning tool for local language models. It is intended to help the user…☆171Updated this week
- Open weights language model from Google DeepMind, based on Griffin.☆639Updated 2 months ago
- ☆143Updated 2 years ago
- a small code base for training large models☆297Updated last week
- Fine-tune mistral-7B on 3090s, a100s, h100s☆711Updated last year
- Inference code for LLaMA models in JAX☆118Updated 11 months ago
- Language Modeling with the H3 State Space Model☆520Updated last year
- A really tiny autograd engine☆92Updated last year
- Implementation of the Llama architecture with RLHF + Q-learning☆164Updated 3 months ago
- Annotated version of the Mamba paper☆483Updated last year