Nimit Kalra

xorl. XoRL

search-self-edit-strategies. Code for the paper "Search over Self-Edit Strategies for LLM Adaptation". Currently being updated.

LCLM. latent context language models

forgetting. 📄 Forgetting in Language Models: Capacity, Optimization, and Self-Generated Replay

sgs. Python

continual-learning-bench. Continual Learning Bench

nrdk. NRDK: Neural Radar Development Kit

abstract-dataloader. Abstract Dataloader: Dataloader Not Included

red-rover. A Radar, Lidar, and Video data collection platform

open-svo2. Reverse engineering of the StereoLabs SVO2 File Format

xwr_ros. A ROS2 wrapper for XWR

xwr. Real-Time Raw Data Capture for TI mmWave Radars

CodeUpdateArena. Python

OPSD_OnPolicyDistillation. On Policy Distillation Build on top of Verl

cmux. Open source Ghostty-based macOS terminal with vertical tabs and notifications for AI coding agents. Built for multitasking, organization, and programmability.

FastKVzip. Accurate and fast KV cache compression with a gating mechanism

rltf. Reinforcement Learning from Text Feedback

maxrl. Official Implementation of "Maximum Likelihood Reinforcement Learning (MaxRL)"

acon. Official implementation of paper "ACON: Optimizing Context Compression for Long-horizon LLM Agents"

rethink_sft_generalization. Repo for paper "Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability"

Offline-GRPO. Code for the blog post titled "Continual Post-Training of LLMs via Offline GRPO for Mathematical Reasoning"

kvpress. LLM KV cache compression made easy

jax-smi. JAX Synergistic Memory Inspector

tourno. TournO (Tournament Optimization) combines pointwise and pairwise LLM judges to produce reward signals for RLHF, using tournament-style comparisons (round-robin, ELO) to derive scalar rewards from pairwise preferences.

AI-Can-Learn-Scientific-Taste. We propose Reinforcement Learning from Community Feedback (RLCF), a training paradigm that uses large-scale community signals as supervision, and formulate scientific taste learning as a preference modeling and alignment problem.

distillation-retains-capabilities. Repo for ICML 2026 submission, "Updating Parametric Knowledge with Context Distillation \\ Retains Post-Training Capabilities"

CRISP_Reasoning_Compression. Python

template. This is the repository for the distill web framework

flashoptim. Python

doc-to-lora. Hypernetworks that update LLMs to remember factual information

zotero-split-view-reader. Split-View Reader for Zotero

Long-Context-Data-Engineering. Implementation of paper Data Engineering for Scaling Language Models to 128K Context

compaction. Algorithms for latent compaction

KVzip. [NeurIPS'25 Oral] Query-agnostic KV cache eviction: 3–4× reduction in memory and 2× decrease in latency (Qwen3/2.5, Gemma3, LLaMA3)

TransArch. Design hardware-friendly model architectures and migrate existing LLMs with minimal performance loss

AlignCLIP. AlignCLIP: Improving Cross-Modal Alignment in CLIP (ICLR 2025)

flow_grpo. [NeurIPS 2025] An official implementation of Flow-GRPO: Training Flow Matching Models via Online RL

LaDiR. Python

Reasoning-Embedding. The official repository of the paper "Do Reasoning Models Enhance Embedding Models?"

Self-Distillation. Python

random-network-distillation. Code for the paper "Exploration by Random Network Distillation"

EVOL-RL. Code for Evolving Language Models without Labels: Majority Drives Selection, Novelty Promotes Variation (EVOL-RL).

BudgetThinker. Python

darling. Official Implementation of the paper "Jointly Reinforcing Diversity and Quality in Language Model Generations"

PipelineRL. A scalable asynchronous reinforcement learning implementation with in-flight weight updates.

entropy-calibration. Jupyter Notebook

AdaSTaR. [NeurIPS 2025] AdaSTaR Adaptive Data Sampling for Training Self-Taught Reasoners

PaCoRe. PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning

tars. Python

EAFT. EAFT(Entropy-Adaptive Fine-Tuning: Resolving Confident Conflicts to Mitigate Forgetting) official repo

robust-world-model-planning. Code for "Closing the Train-Test Gap in World Models for Gradient-Based Planning"

UQ. UQ: Assessing Language Models on Unsolved Questions

RLVR-Decomposed. [NeurIPS 2025] Implementation for the paper "The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning"

cognitive-behaviors. Python

RL-Tango. [NeurIPS 2025] RL Tango: Reinforcing Generator and Verifier Together for Language Reasoning

contourpy. Python library for calculating contours in 2D quadrilateral grids

Multi-agent-evolve. Python

tokasaurus. Python

vogent-turn. Vogent Turn: fast, open-source turn-detection for Voice AI applications

Passk_Training. The official repository of paper "Pass@k Training for Adaptively Balancing Exploration and Exploitation of Large Reasoning Models''

ASAP. Pruning the Unsurprising: Efficient LLM Reasoning via First-Token Surprisal

pts. Pivotal Token Search

Lp-Reg-dev. Low-probability Tokens Sustain Exploration in Reinforcement Learning with Verifiable Reward

Lp-Reg. Low-probability Tokens Sustain Exploration in Reinforcement Learning with Verifiable Reward

EPO. The code for paper "EPO: Entropy-regularized Policy Optimization for LLM Agents Reinforcement Learning"

dino_wm. Python

FrontierScience-Bench. Python

harmony. Renderer for the harmony response format to be used with gpt-oss

gpt-oss. gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI

ml-cross-entropy. Python

APO. Code for ACL2024 paper - Adversarial Preference Optimization (APO).

Auto-CEI. C++

ReST-MCTS. ReST-MCTS*: LLM Self-Training via Process Reward Guided Tree Search (NeurIPS 2024)

sami. Self-Supervised Alignment with Mutual Information

spoken. a single interface around speech-to-speech foundation models

sample-nova-sonic-mcp. TypeScript

amazon-nova-samples. Jupyter Notebook

text-to-lora. Hypernetworks that adapt LLMs for specific benchmark tasks using only textual task description as the input

home. 🏡 Home is where this repo is cloned & installed

j1-micro. j1-micro (1.7B) & j1-nano (600M) are absurdly tiny but mighty reward models.

atropos. Atropos is a Language Model Reinforcement Learning Environments framework for collecting and evaluating LLM trajectories through diverse environments

us-routing. us cached road graph, freeways, primary and secondary roads

distilabel. Distilabel is a framework for synthetic data and AI feedback for engineers who need fast, reliable and scalable pipelines based on verified research papers.

nyc-ai-reading. nyc is so back

PPE. Jupyter Notebook

s1. s1: Simple test-time scaling

LlamaFactory. Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

smalldiffusion. Simple and readable code for training and sampling from diffusion models

smol-course. A course on aligning smol models.

VinePPO. Code for the paper "VinePPO: Unlocking RL Potential For LLM Reasoning Through Refined Credit Assignment"

LightGBM. A fast, distributed, high performance gradient boosting (GBT, GBDT, GBRT, GBM or MART) framework based on decision tree algorithms, used for ranking, classification and many other machine learning tasks.

open-infra-index. Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

grokfast. Official repository for the paper "Grokfast: Accelerated Grokking by Amplifying Slow Gradients"

particles. Playground for physics simulations

RULER. This repo contains the source code for RULER: What’s the Real Context Size of Your Long-Context Language Models?

PufferLib. Puffing up reinforcement learning

latentzip. CLI tool for encoding utf8 text using language models

pytorch_scatter. PyTorch Extension Library of Optimized Scatter Operations

Muon. Muon is an optimizer for hidden layers in neural networks

mutransformers. some common Huggingface transformers in maximal update parametrization (µP)

mup. maximal update parametrization (µP)

Awesome-LLM-Post-training. Awesome Reasoning LLM Tutorial/Survey/Guide

ndif. The NDIF server, which performs deep inference and serves nnsight requests remotely

optuna. A hyperparameter optimization framework

llm_processes. Python

awesome-automl-papers. A curated list of automated machine learning papers, articles, tutorials, slides and projects

vllm. A high-throughput and memory-efficient inference and serving engine for LLMs

LLMSELECTOR. Jupyter Notebook

verifiers. Our library for RL environments + evals

weave. Weave is a toolkit for developing AI-powered applications, built by Weights & Biases.

Guardrails. NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems.

rllm. Democratizing Reinforcement Learning for LLMs

jepsen. A framework for distributed systems verification, with fault injection

posteriors. Uncertainty quantification with PyTorch

rustowl. Visualize Ownership and Lifetimes in Rust

gpu-finder. Python

skypilot. Run, manage, and scale AI workloads on any AI infrastructure. Use one system to access & manage all AI compute (Kubernetes, Slurm, 20+ clouds, on-prem).

pymoo. NSGA2, NSGA3, R-NSGA3, MOEAD, Genetic Algorithms (GA), Differential Evolution (DE), CMAES, PSO

verdict. Inference-time scaling for LLMs-as-a-judge.

dspy-redteam. Red-Teaming Language Models with DSPy

nnsight. The nnsight package enables interpreting and manipulating the internals of deep learned models.

httptap. View HTTP/HTTPS requests made by any Linux program

awesome-hallucination-detection. List of papers on hallucination detection in LLMs.

aider. aider is AI pair programming in your terminal

Archon. Archon provides a modular framework for combining different inference-time techniques and LMs with just a JSON config file.

olmes. Reproducible, flexible LLM evaluations

Awesome-LLM-Judges. ⚖️ Awesome LLM Judges ⚖️

UnlimitedJudge. On the Limitations of Fine-tuned Judges for LLM Evaluation

llama.vim. Vim plugin for LLM-assisted code/text completion

Hermes-Function-Calling. Jupyter Notebook

RAGTruth. Github repository for "RAGTruth: A Hallucination Corpus for Developing Trustworthy Retrieval-Augmented Language Models"

outlines. Structured Outputs

pal. PAL: Proxy-Guided Black-Box Attack on Large Language Models

activeft. PyTorch library for Active Fine-Tuning

Speech. A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)

Trace. End-to-end Generative Optimization for AI Agents

opda. Design and analyze optimal deep learning models.

what_features_jailbreak_LLMs. Jupyter Notebook

wildguard. Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs

mlx. MLX: An array framework for Apple silicon

Metal-Puzzles. Solve Puzzles. Learn Metal 🤘

GPU-Puzzles. Solve puzzles. Learn CUDA.

bijection-learning. Jupyter Notebook

dspy. DSPy: The framework for programming—not prompting—language models

fair. A Go library for serving resources fairly

sam2. The repository provides code for running inference with the Meta Segment Anything Model 2 (SAM 2), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.

bcc. BCC - Tools for BPF-based Linux IO analysis, networking, monitoring, and more

trip-based-public-transit-routing-algo. Python implementation of Trip-Based public transit routing algorithm

openrouteservice. 🌍 The open source route planner api with plenty of features.

sshx. Fast, collaborative live terminal sharing over the web

py-spy. Sampling profiler for Python programs

DETA. Detection Transformers with Assignment

scrapeghost. 👻 Experimental library for scraping websites using OpenAI's GPT API.

monoid. Open-source data privacy

shadow-accrual-maps. Accumulated shadow data computed for New York City

cross_view_transformers. Cross-view Transformers for real-time Map-view Semantic Segmentation (CVPR 2022 Oral)

GTR. Global Tracking Transformers, CVPR 2022

nushell. A new type of shell

www.aaronsw.com. An archival copy.

perspective. A data visualization and analytics component, especially well-suited for large and/or streaming datasets.

neural-processes. This repository contains notebook implementations of the following Neural Process variants: Conditional Neural Processes (CNPs), Neural Processes (NPs), Attentive Neural Processes (ANPs).

pyston. (No longer maintained) A faster and highly-compatible implementation of the Python programming language.

tokio. A runtime for writing reliable asynchronous applications with Rust. Provides I/O, networking, scheduling, timers, ...

faust. Python Stream Processing

python-rocksdb. Python bindings for RocksDB

aiostream. Generator-based operators for asynchronous iteration

dask. Parallel computing with task scheduling

trio. Trio – a friendly Python library for async concurrency and I/O

oneTBB. oneAPI Threading Building Blocks (oneTBB)

async-lru. Simple LRU cache for asyncio

ethereum-function-signature-registry. Registry of 4byte function signatures and their human readable counterparts.

cinder. This is Meta's fork of the CPython runtime. The name "cinder" here is historical, see https://github.com/facebookincubator/cinderx for the Python extension / JIT compiler.

WorldOnRails. (ICCV 2021, Oral) RL and distillation in CARLA using a factorized world model

kornia. 🐍 Geometric Computer Vision Library for Spatial AI

urh. Universal Radio Hacker: Investigate Wireless Protocols Like A Boss

japronto. Screaming-fast Python 3.5+ HTTP toolkit integrated with pipelining HTTP server based on uvloop and picohttpparser.

pgdrive. PGDrive: an open-ended driving simulator with infinite scenes from procedural generation

dgl. Python package built to ease deep learning on graph, on top of existing DL frameworks.

uncertainty-toolbox. Uncertainty Toolbox: a Python toolbox for predictive uncertainty quantification, calibration, metrics, and visualization

multiON. Code for reproducing the results of NeurIPS 2020 paper "MultiON: Benchmarking Semantic Map Memory using Multi-Object Navigation”

social-driving. Design multi-agent environments and simple reward functions such that social driving behavior emerges

difftaichi. 10 differentiable physical simulators built with Taichi differentiable programming (DiffTaichi, ICLR 2020)

gp_model_zoo. Literature and light wrappers for gaussian process models.

SMARTS. Scalable Multi-Agent RL Training School for Autonomous Driving

spirl. Official implementation of "Accelerating Reinforcement Learning with Learned Skill Priors", Pertsch et al., CoRL 2020

PLAS. Code for Latent Action Space for Offline Reinforcement Learning [CoRL 2020]

tonic. Tonic RL library

lift-splat-shoot. Lift, Splat, Shoot: Encoding Images from Arbitrary Camera Rigs by Implicitly Unprojecting to 3D (ECCV 2020)

oatomobile. A research framework for autonomous driving

orbit. A Python package for Bayesian forecasting with object-oriented design and probabilistic models under the hood.

serve. Serve, optimize and scale PyTorch models in production

bit. Bit is a modern Git CLI

Multiverse. Dataset, code and model for the CVPR'20 paper "The Garden of Forking Paths: Towards Multi-Future Trajectory Prediction". And for the ECCV'20 SimAug paper.

robosuite. robosuite: A Modular Simulation Framework and Benchmark for Robot Learning

rlpyt. Reinforcement Learning in PyTorch

torch_interpolations. Interpolation routines in Pytorch.

OccupancyAnticipation. This repository contains code for our publication "Occupancy Anticipation for Efficient Exploration and Navigation" in ECCV 2020.

task-distillation. Code for Domain Adaptation Through Task Distillation (ECCV 20)

midlevel-reps. Mid-Level Visual Representations Improve Generalization and Sample Efficiency for Learning Visuomotor Policies

tpprl. Code and data for "Deep Reinforcement Learning of Marked Temporal Point Processes", NeurIPS 2018