fffiloni/oniric-750
Text-to-Image
•
Updated
•
294
•
•
6
shamyji
shamy777
AI & ML interests
None yet
Recent Activity
liked
a model
9 days ago
fffiloni/oniric-750
upvoted
a
collection
9 days ago
🧠 Reasoning datasets
liked
a dataset
11 days ago
sequelbox/Raiden-DeepSeek-R1
Organizations
None yet
shamy777's activity
liked
a
model
9 days ago
upvoted
a
collection
9 days ago
liked
a
dataset
11 days ago
reacted to
mkurman's
post with 👍
24 days ago
Post
2053
Blurred-Thoughts Supervised Fine-Tuning (BT-SFT) 🤖
Can we teach a model to think completely on its own without reinforcement learning? Actually, yes.
We can do straightforward supervised fine-tuning using a relatively simple trick: blurring a part of CoT thoughts. But why is this effective?
We observed that various models differ in their thinking processes, and fine-tuning one model on another model’s thoughts (CoT) can sometimes be inefficient—often resulting in the model simply memorizing reasoning rather than learning how to actually think.
I discovered that this process can still be efficient if we clearly indicate when the model should start and stop thinking and uncover only a part of CoT and the expected answer, blurring the other part of CoT. This approach allows the model to learn only a portion of the thought process while still arriving at an expected answer.
To demonstrate this, you can watch my experimental BT-SFT on meditsolutions/Llama-3.2-SUN-2.5B-chat model, which was fine-tuned on 151 million tokens from the Magpie-Align/Magpie-Reasoning-V2-250K-CoT-Deepseek-R1-Llama-70B dataset.
Enjoy! 🚀
PS. If you were curious enough to read this, leave me a comment. It's always nice to chat with open-minded and intelligent ppl.
Can we teach a model to think completely on its own without reinforcement learning? Actually, yes.
We can do straightforward supervised fine-tuning using a relatively simple trick: blurring a part of CoT thoughts. But why is this effective?
We observed that various models differ in their thinking processes, and fine-tuning one model on another model’s thoughts (CoT) can sometimes be inefficient—often resulting in the model simply memorizing reasoning rather than learning how to actually think.
I discovered that this process can still be efficient if we clearly indicate when the model should start and stop thinking and uncover only a part of CoT and the expected answer, blurring the other part of CoT. This approach allows the model to learn only a portion of the thought process while still arriving at an expected answer.
To demonstrate this, you can watch my experimental BT-SFT on meditsolutions/Llama-3.2-SUN-2.5B-chat model, which was fine-tuned on 151 million tokens from the Magpie-Align/Magpie-Reasoning-V2-250K-CoT-Deepseek-R1-Llama-70B dataset.
Enjoy! 🚀
PS. If you were curious enough to read this, leave me a comment. It's always nice to chat with open-minded and intelligent ppl.
- 3 replies
liked
a
dataset
24 days ago
liked
2
datasets
27 days ago
liked
4
datasets
about 2 months ago
liked
a
model
3 months ago
reacted to
MonsterMMORPG's
post with 👍
3 months ago
Post
2845
FLUX Tools Complete Tutorial with SwarmUI (as easy as Automatic1111 or Forge) : Outpainting, Inpainting, Redux Style Transfer + Re-Imagine + Combine Multiple Images, Depth and Canny - More info at the oldest comment - No-paywall : https://youtu.be/hewDdVJEqOQ
FLUX.1 Tools by BlackForestLabs changed the #AI field forever. They became the number 1 Open Source community provider after this massive release. In this tutorial, I will show you step by step how use FLUX.1 Fill model (inpainting model) to do perfect outpainting (yes this model used for outpainting) and inpainting. Moreover, I will show all features of FLUX Redux model to do style transfer / re-imagine 1 and more than 1 images combination. Furthermore, I will show you step by step how to convert input image into Depth or Canny maps and then how to use them on #FLUX Depth and Canny models. Both LoRA and full checkpoints of FLUX Depth and Canny.
🔗 Full Instructions, Configs, Installers, Information and Links Shared Post (the one used in the tutorial - Public - no-paywall) ⤵️
▶️ https://www.patreon.com/posts/tutorial-instructions-links-public-post-106135985
Preparation of this tutorial took more than 1 week and this will be the very best and easiest to follow tutorial since it is made with famous #SwarmUI. SwarmUI is as easy and as advanced as Automatic1111 SD Web UI. Biggest advantage of SwarmUI is that, it uses ComfyUI as a back-end. Therefore, It is extremely fast, VRAM optimized and supports all of the newest SOTA models as soon as they are published.
So in this tutorial I will show you how to setup SwarmUI and FLUX Dev tools on your Windows Computer, Massed Compute, RunPod and Kaggle. I will step by step explanatin and show you every tips and tricks that you need to properly do style transfer, re-imagine, inpaint, outpaint, depth and canny with FLUX.
Video Chapters first image
FLUX.1 Tools by BlackForestLabs changed the #AI field forever. They became the number 1 Open Source community provider after this massive release. In this tutorial, I will show you step by step how use FLUX.1 Fill model (inpainting model) to do perfect outpainting (yes this model used for outpainting) and inpainting. Moreover, I will show all features of FLUX Redux model to do style transfer / re-imagine 1 and more than 1 images combination. Furthermore, I will show you step by step how to convert input image into Depth or Canny maps and then how to use them on #FLUX Depth and Canny models. Both LoRA and full checkpoints of FLUX Depth and Canny.
🔗 Full Instructions, Configs, Installers, Information and Links Shared Post (the one used in the tutorial - Public - no-paywall) ⤵️
▶️ https://www.patreon.com/posts/tutorial-instructions-links-public-post-106135985
Preparation of this tutorial took more than 1 week and this will be the very best and easiest to follow tutorial since it is made with famous #SwarmUI. SwarmUI is as easy and as advanced as Automatic1111 SD Web UI. Biggest advantage of SwarmUI is that, it uses ComfyUI as a back-end. Therefore, It is extremely fast, VRAM optimized and supports all of the newest SOTA models as soon as they are published.
So in this tutorial I will show you how to setup SwarmUI and FLUX Dev tools on your Windows Computer, Massed Compute, RunPod and Kaggle. I will step by step explanatin and show you every tips and tricks that you need to properly do style transfer, re-imagine, inpaint, outpaint, depth and canny with FLUX.
Video Chapters first image
liked
a
model
5 months ago
updated
a
collection
5 months ago
liked
a
model
7 months ago
reacted to
alvdansen's
post with 👍
8 months ago
Post
5866
New LoRA Model!
I trained this model on a new spot I'm really excited to share (soon!)
This Monday I will be posting my first beginning to end blog showing the tool I've used, dataset, captioning techniques, and parameters to finetune this LoRA.
For now, check out the model in the link below.
alvdansen/m3lt
I trained this model on a new spot I'm really excited to share (soon!)
This Monday I will be posting my first beginning to end blog showing the tool I've used, dataset, captioning techniques, and parameters to finetune this LoRA.
For now, check out the model in the link below.
alvdansen/m3lt
- 5 replies
liked
2
datasets
8 months ago