aarambh-studio

🦀 Decoder-only LLM built from scratch in pure Rust using Candle — no Python, no PyTorch. Gated DeltaNet + sparse attention, fine-grained MoE, native video/document understanding, long-horizon tool agents, quantization-aware training. Scales: Tiny (25M) to Large (1.3B).

AarambhDevHub/aarambh-studio

View on GitHub ↗

🦀 Decoder-only LLM built from scratch in pure Rust using Candle — no Python, no PyTorch. Gated DeltaNet + sparse attention, fine-grained MoE, native video/document understanding, long-horizon tool agents, quantization-aware training. Scales: Tiny (25M) to Large (1.3B).

aarambh-studio grew +1.8% in the last 30 days, ranks 99th of 113 in Pretraining & Training, last commit 92 days ago.

Star trajectory

82 days tracked
2026-08-31112 stars today2026-11-22
Momentum

+1.8% in 30d, above the Pretraining & Training median of +0.3%.

Maintenance

Last commit 92 days ago, fresher than the median of 103d.

How it compares in Pretraining & Training

aarambh-studio
112
stars
+1.8% 30d
commit 92 days ago
Pretraining & Training median
342
stars
+0.3% 30d
commit ~103d ago
Leader · maxtext
2.5k
stars
+1.0% 30d
commit today

Ranked 99th of 113 by stars.

Ranked by shared tags, then keywords, then stars

Where aarambh-studio sits

Hover or focus a node for details. Click to open it.

  • This repo
  • Related repo
  • Tag or category
  1. GleamLM大语言模型全栈训练平台——核心代码用 PyTorch 从零实现,最低仅需 12GB 消费级显卡即可完整训练,另有一套工业代码与手写实现消费同一份数据逐层对照。模型训练支持CLI和可视化训练。

    +1 (+1.5%) in 30d · Committed 65d ago · Python · Apache-2.0

    Shares Pretraining & Training, Training Toolkits, Deep Learning Frameworks

  2. Omega-AI is a Java-based deep learning framework that helps you quickly build neural networks for inference…

    +1 (+0.1%) in 30d · Committed 22d ago · Java · Apache-2.0

    Shares Deep Learning Frameworks, Pretraining & Training, Training Toolkits

  3. seqax = sequence modeling + JAX

    -1 (-0.5%) in 30d · Committed 487d ago · Python · BSD-3-Clause

    Shares Pretraining & Training, Training Toolkits, Deep Learning Frameworks

  4. Following Karpathy with GPT-2 implementation and training, writing lots of comments cause I have memory of a…

    +0 (+0.0%) in 30d · Committed 843d ago · Jupyter Notebook

    Shares Pretraining & Training, Training Toolkits, Deep Learning Frameworks

  5. GPT-2-style LLM built from scratch in C/CUDA with hand-written backprop, BPE tokenizer, FlashAttention…

    +1 (+0.8%) in 30d · Committed 157d ago · Cuda · MIT

    Shares Pretraining & Training, Training Toolkits, Deep Learning Frameworks

  6. A high-performance framework for training LLMs, VLMs, diffusion, and embodied models on NVIDIA GPUs and Kunlun…

    Committed yesterday · Python · Apache-2.0

    Shares Training Toolkits, Pretraining & Training

  7. Halo is an open-source framework built by White Circle for training large language and multimodal models

    +34 (+4.1%) in 30d · Committed today · Python · Other license

    Shares Pretraining & Training, Training Toolkits

  8. From scratch implementation of a sparse mixture of experts language model inspired by Andrej Karpathy's…

    +1 (+0.1%) in 30d · Committed 752d ago · Jupyter Notebook · MIT

    Shares Training Toolkits, Deep Learning Frameworks

  9. An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale

    +27 (+4.1%) in 30d · Committed today · Python · Apache-2.0

    Shares Training Toolkits, Pretraining & Training

  10. Implementation of Gamma-World: Generative Multi-Agent World Modeling Beyond Two Players

    +4 (+0.7%) in 30d · Committed 157d ago · Python · Apache-2.0

    Shares Training Toolkits, Deep Learning Frameworks

Language
Rust
License
Apache License 2.0
Contributors
1
Forks
36
Watchers
2
Open issues
0
Releases
4
Created
2026-06-24
Stars this week
+0

You've seen where aarambh-studio stands today.

Get the moves before everyone else: the 5 Pretraining & Training repos climbing fastest, every Tuesday.

Free weekly digest. Double opt-in, unsubscribe anytime.

Maintain aarambh-studio?

Claim this page to add your own call-to-action.

Claim this page