Agent frameworks & infrastructure

Models, fine-tuning & training

2,504 agents · most starred first

Models, fine-tuning & training, page 5 of 26

personalforge★ 96

Build your own offline AI from any documents. Free. No coding. LoRA fine-tuning + RAG + GGUF export.

FlashREINFORCE★ 96

FlashREINFORCE: Critic-Free, Single-Rollout, Asynchronous RL for Agentic Language Models

WangChanGLM★ 95

WangChanGLM 🐘 - The Multilingual Instruction-Following Model

SPA-RL-Agent★ 95

Official code for paper "SPA-RL: Reinforcing LLM Agent via Stepwise Progress Attribution"

AgentHER★ 95

AgentHER: Hindsight Experience Replay for LLM Agents

Agent-G2★ 95

[EMNLP 2026 main] Agent-G2 : Gaussian Guidance for Agentic Reinforcement Learning

DQN-using-PyTorch-and-ML-Agents★ 94

A simple example of how to implement vector based DQN using PyTorch and a ML-Agents environment

deepseek-coder-33b-base★ 94

transformers pytorch safetensors llama text-generation license:other

Self-Distillation-AgentRL★ 94

More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement Learning

agent-attention-pytorch★ 93

Implementation of Agent Attention in Pytorch

Tiktoken★ 93

High-performance .NET BPE tokenizer — up to 618 MiB/s, competitive with Rust. Zero-allocation counting, multilingual cache, o200k/cl100k/r50k/p50k encodings + HuggingFace tokenizer.json support.

RedPajama-INCITE-7B-Chat★ 93

transformers pytorch gpt_neox text-generation en dataset:togethercomputer/RedPajama-Data-1T

Arcee-Agent★ 93

transformers safetensors qwen2 text-generation conversational en

distillops★ 92

Black-box LLM distillation for any verifiable task, with text or image inputs: teacher API -> rejection sampling -> SFT -> eval -> GRPO

lora-pilot★ 91

From dataset to trained LoRA to finished image or video. One browser workspace for your models, datasets, training, and generation. LoRA Pilot supports 50+ training model groups, with ComfyUI, Kohya, AI Toolkit, and Diffusion Pipe included. Run on your own GPU or RunPod, and connect AI agents through MCP.

WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B★ 90

transformers safetensors qwen2 text-generation code qwen-coder

AKO4X★ 88

Agentic Kernel Optimization — advanced & eXtensible: a closed-loop, campaign-based multi-agent system for optimizing GPU kernels (benchmark-swappable; default flashinfer-bench).

stablelm-2-12b-chat★ 88

transformers safetensors stablelm text-generation causal-lm conversational

dstc★ 87

Dialog State Tracking Challenge 2 & 3 Data

banana-gym★ 87

A simple stochastic OpenAI environment for training RL agents

Qwen2.5-Coder-14B-Instruct-abliterated-GGUF★ 87

gguf code codeqwen chat qwen qwen-coder

Agentic-ESOpt★ 87

Code for Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements

Qwen1.5-1.8B-Chat★ 86

transformers safetensors qwen2 text-generation chat conversational

FLEX★ 86

The official codebase for our paper, FLEX: Continuous Agent Evolution via Forward Learning from Experience.

GPTEncoder★ 85

Swift BPE Encoder/Decoder for OpenAI GPT Models. A programmatic interface for tokenizing text for OpenAI ChatGPT API.

meow-tea-taro★ 85

A Practitioner's Guide to M(eow)ti Turn Agentic ReinfOrcement learning

AILib★ 84

A large collection of mostly reinforcement learning agents.

Deep-Reinforcement-Learning-Algorithm-Collection★ 84

Collection of Deep Reinforcement Learning Algorithms implemented in PyTorch.

Abmarl★ 84

Agent Based Modeling and Reinforcement Learning

gemma-4-E2B-it-assistant★ 84

transformers safetensors gemma4_assistant text-generation any-to-any arxiv:2607.02770

local-datagen-cli★ 83

synthetic dataset generation workflow using local file resources for finetuning llms.

MATPO★ 83

Official implementation of MATPO: Multi-Agent Tool-Integrated Policy Optimization.

train-llama4★ 83

Building LLaMA 4 MoE from Scratch

Qwen3-Coder-Next-Base★ 83

transformers safetensors qwen3_next text-generation conversational license:apache-2.0

Qwen2.5-Coder-3B★ 83

transformers safetensors qwen2 text-generation code qwen

go-gpt-3-encoder★ 82

Go BPE tokenizer (Encoder+Decoder) for GPT2 and GPT3

gpt4o-from-scratch★ 82

Implementation of a GPT-4o like Multimodal from Scratch using Python

-Project-Nord-Spiking-Neural-Network-Language-Model★ 82

The first pure SNN language model trained from scratch with a fully original architecture. 618M parameters • 93% sparsity • Runs on phone • Online learning via STDP • $260 total training cost

cap-vlm★ 81

Perceive, Predict, Verify: Continual Pre-training for Multimodal Agentic Foundation Models

LLM4RL★ 81

A RL approach to enable cost-effective, intelligent interactions between a local agent and a remote LLM

meta-agent★ 81

Continual harness optimization

Qwen-7B-Chat-Int4★ 81

transformers safetensors qwen text-generation custom_code zh

Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated★ 81

transformers safetensors qwen3_moe text-generation abliterated uncensored

CodeShell-7B★ 81

transformers pytorch safetensors kclgpt text-generation codeshell

Adversarial-Learning-for-Generative-Conversational-Agents★ 80

This repository contains a new adversarial training method for Generative Conversational Agents

azureai-foundry-finetuning-raft★ 80

A recipe that will walk you through using either Meta Llama 3.1 405B or OpenAI GPT-4o deployed on Azure AI to generate a synthetic dataset using UC Berkeley's Gorilla project RAFT method.

Qwen2.5-Coder-7B-Instruct-GGUF★ 80

transformers gguf code codeqwen chat qwen

sciBASIC★ 80

A VisualBasic(.NET) language kernel and runtime for scientific data computing, deep learning, LLM inference, GPU acceleration, visualization and command-line data-science applications — running on .NET (net10.0) across Windows, Linux and macOS.

sft-demos★ 79

🤖 language modeling sft demos

GenieRedux★ 79

A framework for training world models with virtual environments, complete with annotated environment dataset (RetroAct), exploration agent (AutoExplore Agent), and GenieRedux-G - an implementation of Genie with enhancements

jupyter-agent★ 79

Training LLMs to reason and analyze data with notebooks

Apex★ 79

Agents, and RL environment, for optimizing GPU kernels on AMD ROCm using LLM agents. Benchmarks LLM serving workloads end-to-end, profiles bottleneck kernels, optimizes them via Claude Code or Codex, and scores on compilation, correctness, and speedup.

Pith-Train★ 78

Compact and Agent-Native MoE Training System

Gemma-2-9B-Chinese-Chat★ 78

transformers safetensors gguf gemma2 text-generation llama-factory

gemma-4-12b-coder-fable5-composer2.5-4bit★ 78

mlx safetensors gemma4_unified image-text-to-text gemma4 coding

Prompt2Policy★ 77

Agentic Engineering for Reinforcement Learning

training-ai-agents★ 77

Training architecture for self-improving AI agents.

ML-Agent★ 77

The official implementation of "ML-Agent: Reinforcing LLM Agents for Autonomous Machine Learning Engineering"

calm2-7b-chat★ 77

transformers pytorch safetensors llama text-generation japanese

tbench-agentic-data-pipeline★ 76

Multi-agent synthetic data generation pipeline capable of generating and validating long horizon terminal/coding tasks for RL training

Meta-Llama-3-8B-Instruct-function-calling-json-mode★ 76

transformers safetensors llama text-generation text-generation-inference unsloth

Emu3-Chat★ 76

transformers safetensors Emu3 text-generation custom_code arxiv:2409.18869

DeepCoder-1.5B-Preview★ 76

transformers safetensors qwen2 text-generation conversational en

SERL★ 75

Official implement on 'What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents'

glaive-function-calling-v1★ 75

transformers pytorch mpt text-generation custom_code dataset:glaiveai/glaive-function-calling

llm-post-training-atlas★ 75

An opinionated atlas of post-R1 LLM post-training: RL, rewards, alignment, distillation, agents, and their connections.

calculator_agent_rl★ 74

Training an LLM to use a calculator with multi-turn reinforcement learning, achieving a **62% absolute increase in evaluation accuracy**.

codet5-large★ 74

transformers pytorch t5 text-generation arxiv:2109.00859 arxiv:2207.01780

SeaLLMs-v3-7B-Chat★ 74

transformers safetensors qwen2 text-generation sea multilingual

Qwen2.5-Coder-14B-Instruct-GGUF★ 74

gguf code codeqwen chat qwen qwen-coder

Seed-Coder-8B-Base★ 74

transformers safetensors llama text-generation arxiv:2506.03524 license:mit

SAPO★ 74

A PPO-style Agentic RL framework without a separate critic model and multiple rollouts

reinforcement-learning★ 73

Implementation of Single-Agent and Multi-Agent Reinforcement Learning Algorithms. MATLAB.

SMAC★ 73

StarCraft II Multi Agent Challenge : QMIX, COMA, LIIR, QTRAN, Central V, ROMA, RODE, DOP, Graph MIX

Huihui-Qwen3-Coder-Next-abliterated★ 73

transformers safetensors qwen3_next text-generation abliterated uncensored

MirrorFlow★ 72

从对话数据到训练:数字分身 + 模型蒸馏 From Dialogue Data to Training Closed-Loop: Digital Twin + Model Distillation

GenEnv★ 72

[NeurIPS 2026] GenEnv: Difficulty-Aligned Co-Evolution Between LLM Agents and Environment Simulators

internlm2_5-7b-chat-1m★ 72

transformers safetensors internlm2 feature-extraction text-generation conversational

Yi-6B-Chat★ 72

transformers safetensors llama text-generation conversational arxiv:2403.04652

maya★ 72

transformers safetensors llava_cohere text-generation merge image-text-to-text

Qwen2.5-Coder-0.5B★ 72

transformers safetensors qwen2 text-generation code qwen

FnCTOD★ 71

Official code for the publication "Large Language Models as Zero-shot Dialogue State Tracker through Function Calling" https//arxiv.org/abs/2402.10466

Fruit-API★ 70

A Universal Deep Reinforcement Learning Framework

CVE-llm_dataset★ 70

This is a dataset intended to train a LLM model for a completely CVE focused input and output.

Llama-2-ko-7b-Chat★ 70

transformers pytorch llama text-generation facebook meta

IndustrialCoder★ 70

transformers safetensors iquestcoder text-generation code industrial-code

Qwen3-Coder-30B-A3B-Instruct-AWQ-4bit★ 69

transformers safetensors qwen3_moe text-generation conversational arxiv:2505.09388

Multi-agent-reinforcement-learning★ 68

Implementation of Multi-Agent Reinforcement Learning algorithm(s). Currently includes: MADDPG

AutoML-Experiment-Agent★ 68

A pluggable AutoML experiment agent for multi-task hyperparameter optimization

MUA-RL★ 68

MUA-RL: MULTI-TURN USER-INTERACTING AGENT REINFORCEMENT LEARNING FOR AGENTIC TOOL USE

Orion-14B-Chat★ 68

transformers pytorch gguf orion text-generation code

h2o-danube3-4b-chat★ 68

transformers safetensors llama text-generation gpt llm

WHALE★ 68

Official implementation of WHALE (Weight-Harness Alternating Learning)

IPR★ 67

Watch Every Step! LLM Agent Learning via Iterative Step-level Process Refinement (EMNLP 2024 Main Conference)

SmolVLM-Finetune★ 67

An open-source implementaion for fine-tuning SmolVLM.

multi-agent-training-grpo★ 67

A multi-agent system trained with GRPO for reliable long-horizon task planning and execution.

agentica-org_DeepCoder-14B-Preview-GGUF★ 67

gguf text-generation en dataset:PrimeIntellect/verifiable-coding-problems dataset:likaixin/TACO-verified dataset:livecodebench/code_generation_lite

gemma-4-12B-coder-fable5-composer2.5-v1-uncensored-heretic-GGUF★ 67

transformers gguf gemma4 coding code reasoning

strands-env★ 67

A unified framework for building agentic RL environments.

senator★ 66

NeurIPS 2025: Structural Entropy Guided Agent for Detecting and Repairing Knowledge Deficiencies in LLMs

More in Agent frameworks & infrastructure