Articles by danielhanchen
325

Qwen3-Coder-Next (qwen.ai)

2

Kimi K2 Thinking: How to Run Locally (unsloth.ai)

5

Thinking Machines – LoRA Without Regret (thinkingmachines.ai)

2

Long context GPT-OSS fine-tuning (unsloth.ai)

1

Show HN: GPT OSS: How to run and fine-tune (unsloth.ai)

1

Qwen3-30B-A3B-Instruct-2507 (huggingface.co)

234

Qwen3-Coder: Agentic coding in the world (qwenlm.github.io)

2

2.71bit DeepSeek-V3-0324 (unsloth.ai)

5

Gemma 3: Google's new multimodal models (ai.google.dev)

4

How to Run QwQ-32B effectively (unsloth.ai)

12

Train your own R1 reasoning model (unsloth.ai)

6

How to run 1.58bit DeepSeek R1 with Open WebUI (openwebui.com)

50

Phi-4 Bug Fixes (unsloth.ai)

4

My take on the Post Pretraining world (twitter.com/danielhanchen)

2

Dynamic 4bit Quantization (unsloth.ai)

1

Show HN: Finetune Llama 3.2 Vision in a Colab (colab.research.google.com)

2

Python 3.11 is 1.25x faster than 3.10 (python.org)

2

Fixing Gradient Accumulation (huggingface.co)

3

Unit Economics of LLM APIs (lesswrong.com)

2

LoRA Learns Less and Forgets Less Updated (openreview.net)

2

VLLM automatic prefix / prompt caching (vllm.ai)

2

Higher Temperatures and Min_p Sampling (arxiv.org)

1

Show HN: Open-source fine-tuning in a Colab notebook (colab.research.google.com)

4

Sahm rule signals start of recession (stlouisfed.org)

2

Low Level Technicals of LLMs [video] (youtube.com)

3

Gemma-2 2B beats GPT3.5 on Chatbot Arena (huggingface.co)

2

HuggingChat – Chat UI for Llama 3.1 405B (huggingface.co)

1

Fine-Tune Llama 3.1 Ultra-Efficiently with Unsloth (huggingface.co)

2

Yield Curve and Predicted GDP Growth (clevelandfed.org)

3

Cloudflare DNS + Malware Blocking (one.one)

1

SIMD at Insomniac Games: How We Do the Shuffle (gdcvault.com)

3

Some Machine Learning Notes (danielhanchen.github.io)

3

My Analysis of Llama 3.1 (twitter.com/danielhanchen)

2

Show HN: Finetune Llama-3.1 2x faster in a Colab (colab.research.google.com)

1

Show HN: Mistral NeMo finetuning fits in Colab (colab.research.google.com)

2

TextGrad – Backpropagation through text feedback (arxiv.org)

16

Nemotron-4 340B open weights model (nvidia.com)

7

Show HN: Finetune Llama-3 2x faster in a Colab notebook (colab.research.google.com)

5

Try Llama-3 in a Colab Notebook (colab.research.google.com)

64

Fixing Gemma Bugs (unsloth.ai)

3

Finetuning Gemma 2.4x Faster (research.google.com)

1

Show HN: Gemma finetuning 243% faster, 58% less VRAM (unsloth.ai)

1

CodeLlama-34B 13x faster finetuning (unsloth.ai)

2

Reducing FLOPs for transformers (unsloth.ai)

89

Show HN: 80% faster, 50% less memory, 0% loss of accuracy Llama finetuning (github.com/unslothai)

1

Finetune language models 30x faster (unsloth.ai)