Hello, I'm Tokenbender

Portrait of Abhishek Harshvardhan Mishra
Abhishek Harshvardhan MishraML Researcher · Writes as tokenbender

Hello!

I am Abhishek Harshvardhan Mishra, aka Tokenbender.

I am the founder of Deus Experiments, a lab specialising in post-training for enterprise use cases and supersparse inference. Before this, I was a technical lead at Intel, working on server firmware. Then I fell in love with language models, threw away everything I had done before, and became part of open-source LLM research.

I am good at making models do whatever is needed, as long as the use case is clearly specified.

A rough chronology

The route so far

  1. Before 2023
    Intel Technical lead working on server firmware.
  2. CodeCherryPop The first friendly local coder of the Llama 2 era.
    Jul 2023 · Independent receipts #1 Llama 2 coder.

    Beat models 10× its size.

    7B · 122K code instructions · released three days after Llama 2
  3. EvolvedSeeker 68.29% HumanEval from a 1.3B coding model.
    Nov 2023 · Release-window receipts #1 coder in the 1B class.

    Beat models 12× its size.

    “1.3B with 68.29% HumanEval lol, don't behead me.” 1.3B · 50K instructions · free-Colab inference · transparent eval samples
  4. Navarna Hindi chat and retrieval-augmented generation.
    Feb 2024 · Open pipeline First Hindi chat model with retrieval built in.

    One model for conversation and retrieval.

    “Code/Data/process OSS.” 7B · OpenHermes 2.5 · 59.7K SFT · 6.6K DPO · Apache 2.0
  5. avataRL Pretraining language models from random weights with pure reinforcement learning.
    May–Aug 2025 · Crown jewel Cross-entropy pretraining is optional.

    I trained language models from random weights with pure reinforcement learning.

    “Random weights. No cross-entropy pretraining.” 295 stars · 21 forks · 145 stars in 72 hours · selected from 270+ grant applicants
  6. infinite Rubric-driven prioritized replay for continual learning.
    Aug 2025 · Continual learning The model’s failures should choose what it learns next.

    Infinite turns continual learning into spaced repetition: train what is slipping, refresh what is stable, and keep the model in the game.

    Continual RL · adaptive curriculum · anti-forgetting
  7. Low-rank circuit conditioning 91.33% recovery from 5.05% of a model's MLP channels.
Now Extract the capability, not the whole model.