Deschide articolul original într-un tab nou Schimbarea acestei opțiuni se aplică imediat și reîncarcă pagina.
LECTURA
Înapoi la articole →

Gatsby Starter Blog RSS Feed

amund.blog/rss.xml
Vizitează site-ul

Cel mai recent articol

Ultima verificare

Urmăritori 0 urmăritori
Articole: 10

  1. What formal methods can (and can't) prove about a Gemma 4 + DFlash + DDTree blueprint?

    (Gatsby Starter Blog RSS Feed)

    Summary The G4-FlashTree stack — Gemma 4 31B as the target, a DFlash block drafter, and a DDTree tree verifier — is the kind of system where formal methods earn their keep, because the failure modes (ghost reads after rollback, cross-path leakage in tree…

  2. Can Prompt Lookup Decoding Speed Up Gemma 4 31B 4-bit on Apple Silicon for RAG-style Workloads?

    (Gatsby Starter Blog RSS Feed)

    Code for this post is at github.com/atveit/InferenceAcceleration/tree/main/pld-gemma4-mlx.

  3. Exp minus Log is all you need for Deep Learning?

    (Gatsby Starter Blog RSS Feed)

    Note: This work builds on the 2026 discovery by Dr.

  4. Mojo-tokenizer: Fastest AI Token Output → Readable Text on Apple Silicon?

    (Gatsby Starter Blog RSS Feed)

    mojo-tokenizer decodes tokens at 144M tokens/sec on Apple Silicon — 3.1x faster than tiktoken and 1.2x faster than rs-bpe.

  5. mojo-vec: Zero-Copy Buffers That Match Rust Performance

    (Gatsby Starter Blog RSS Feed)

    mojo-vec brings Rust-style zero-copy buffer performance to Mojo — inline storage, pointer-based slicing, and SIMD operations.

  6. 6.5 GB/s JSON Parsing in Mojo — Beating Rust and C++ on Apple Silicon

    (Gatsby Starter Blog RSS Feed)

    mojo-json parses JSON at 6.5 GB/s on Apple Silicon — 52% faster than the fastest C++ and Rust parsers.

  7. Building the Enterprise Foundation Mojo Needed

    (Gatsby Starter Blog RSS Feed)

    Mojo has speed.

  8. 4 bits is all you need (and 3.6 bit you have?) for resource-efficient LLMs?

    (Gatsby Starter Blog RSS Feed)

    A few months ago OpenAI published their open weights model(s) GPT-OSS (20B and 120B), and one of the eye-catching characteristics was that it was heavily quantized - in other words "shrunk" or compressed - to 4 bits per parameter (MXFP4) instead of the co…

  9. Towards milli-joules per token? AI on the Apple Watch

    (Gatsby Starter Blog RSS Feed)

    Can an Apple Watch run a (small) LLM (SLM)?

  10. Grokking implementations in Jax/Flax and Pytorch

    (Gatsby Starter Blog RSS Feed)

    https://github.com/atveit/torch_grokking https://github.com/atveit/jax_grokking What is Grokking in Deep Learning?