Llama 3.3 (70B) Finetuning - now with 90K context length and fits on <41GB VRAM.
Train your own Reasoning model - 80% less VRAM - GRPO now in Unsloth (7GB VRAM min.)
[P] GRPO fits in 8GB VRAM - DeepSeek R1's Zero's recipe
You can now train your own DeepSeek-R1 model on your local device!
[R] Replicating DeepSeek-R3-Zero RL recipe on 3B LLM for <30$, the model develops self-verification and search abilities all on its own
You can now run DeepSeek-R1 on your own local device!
1.58bit DeepSeek R1 - 131GB Dynamic GGUF
I was able to 1-shot prompt the unlsoth "python flappy bird game" test with Deepseek R1 distilled 70B. The distilled models deserve more credit.
The new Mistral Small model is disappointing
DeepSeek AI blocked by Italian authorities
Running Deepseek R1 IQ2XXS (200GB) from SSD actually works
Fine Tuning On Completions only using Unsloth
Unsloth made dynamic R1 quants - can be run on as little as 80gb of RAM
I have a 12GB 3060, is it possible to fine-tune ANY model?
I fixed 4 bugs in Microsoft's open-source Phi-4 model
Deepseek-R1 GGUFs + All distilled 2 to 16bit GGUFs + 2bit MoE GGUFs
Deepseek just uploaded 6 distilled verions of R1 + R1 "full" now available on their website.
let’s goo, DeppSeek-R1 685 billion parameters!
DeepSeek R1 has been officially released!
[P] How I found 8 bugs in Google's Gemma 6T token model
[P] How I found & fixed 4 bugs in Microsoft's Phi-4 model
