CalmState/gemma-3-4b-polyglot-r1
Any-to-Any
β’
4B
β’
Updated
β’
6
β’
6
John Smith PRO
John6666
AI & ML interests
None yet
Recent Activity
liked
a model
about 5 hours ago
CalmState/gemma-3-4b-polyglot-r1
liked
a model
about 5 hours ago
mradermacher/gemma-3-4b-polyglot-v1-GGUF
liked
a model
about 13 hours ago
Eric111/gemma-4b-it-rude
Organizations
liked
2
models
about 5 hours ago
liked
2
models
about 13 hours ago
updated
2
Spaces
about 15 hours ago
upvoted
an
article
about 21 hours ago
Article
Why You Should Care About Partial Differential Equations (PDEs)
β’
39
updated
a
dataset
about 21 hours ago
liked
10
models
about 22 hours ago
MaziyarPanahi/finetuned-mistral-7b-Mistral-7B-Instruct-v0.2-slerp
Text Generation
β’
7B
β’
Updated
β’
11
β’
1
mradermacher/finetuned-mistral-7b-Mistral-7B-Instruct-v0.2-slerp-GGUF
7B
β’
Updated
β’
133
β’
1
mradermacher/finetuned-mistral-7b-Mistral-7B-Instruct-v0.2-slerp-i1-GGUF
7B
β’
Updated
β’
490
β’
1
RowanBird779/cosmos-reason2-8b-abliterated
Image-Text-to-Text
β’
9B
β’
Updated
β’
119
β’
1
mradermacher/cosmos-reason2-8b-abliterated-i1-GGUF
8B
β’
Updated
β’
1.68k
β’
2
mradermacher/cosmos-reason2-8b-abliterated-GGUF
8B
β’
Updated
β’
338
β’
2
quanliang/Y-Qwen3vl-8B-LoRA
9B
β’
Updated
β’
7
β’
1
mradermacher/Y-Qwen3vl-8B-LoRA-GGUF
8B
β’
Updated
β’
272
β’
1
appletea2333/EditThinker8B-Flux.1-Kontext-Dev
9B
β’
Updated
β’
16
β’
1
mradermacher/EditThinker8B-Flux.1-Kontext-Dev-GGUF
8B
β’
Updated
β’
293
β’
1
reacted to
etemiz's
post with π₯
1 day ago
Post
1043
how to expand your dataset (of articles) without changing the ideas in it?
i was doing CPT for a while and got decent results. but what if i want to go for perfection? cover all the areas of misalignment using limited datasets. i have to find a way to multiply the material to successfully combat the material of the rest of the internet.
i want to generate SFT datasets but only on controversial topics, because i have to be efficient with limited resources. first i give a smart LLM a 'ground truth' text. then i give it the following prompts:
the result is questions and answers with more words along the same ideas. a few sentences of opinions in the beginning, is expanded to lots of words. using this method i can multiply billions of tokens to tens of billions probably and have a more effective training.
next i should do RL maybe. LLMs seem to have all kinds of ideas already installed, yet they don't have the intuition to know which one is true. they can give you a ton of reasons to support anything. given the proper incentives, LLMs then should evolve towards supporting aligned ideas more. the rewards will be like guidance that will kick an LLM towards better answers.
i was doing CPT for a while and got decent results. but what if i want to go for perfection? cover all the areas of misalignment using limited datasets. i have to find a way to multiply the material to successfully combat the material of the rest of the internet.
i want to generate SFT datasets but only on controversial topics, because i have to be efficient with limited resources. first i give a smart LLM a 'ground truth' text. then i give it the following prompts:
- You are a highly skilled academic analyst.
- Analyze this text and find 3 bold claims that could cause controversy and division in public. List the claims and also state why they are debatable. Give numbers to the claims.
- Convert these claims into binary questions (that could be answered by yes/no or this/that).
- Now put these questions in a json format. Please also add the info about which of the answers concur with the original text and the question number.
- Write some supporting arguments for 1st question, with respect to the original text, concurring and confirming the original text.
There must be about 300 words. You should not mention the text, write it as if you are the one answering the question.the result is questions and answers with more words along the same ideas. a few sentences of opinions in the beginning, is expanded to lots of words. using this method i can multiply billions of tokens to tens of billions probably and have a more effective training.
next i should do RL maybe. LLMs seem to have all kinds of ideas already installed, yet they don't have the intuition to know which one is true. they can give you a ton of reasons to support anything. given the proper incentives, LLMs then should evolve towards supporting aligned ideas more. the rewards will be like guidance that will kick an LLM towards better answers.
- 3 replies
reacted to
de-Rodrigo's
post with π
1 day ago
Post
1031
We are happy to share the VERSE Methodology paper via arXiv! ππ«
VERSE: Visual Embedding Reduction and Space Exploration. Clustering-Guided Insights for Training Data Enhancement in Visually-Rich Document Understanding (2601.05125)
We usually train VLMs on visual synthetic data that we (as humans) label as photorealistic. We argue that this is an anthropocentric perspective imposed to a model that might not synthetize visual information as we do. VERSE helps to visualize latent space and overlay visual features to detect poor-performance regions and take action to include better-suited training sets to boost model performance.
Resources:
- Code: https://github.com/nachoDRT/VrDU-Doctor
- Hugging Face Space: de-Rodrigo/Embeddings
Want to collaborate? Do you have any feedback? π§
PD: As always, we are grateful to Hugging Face π€ for providing the fantastic tools and resources we find on the platform!
VERSE: Visual Embedding Reduction and Space Exploration. Clustering-Guided Insights for Training Data Enhancement in Visually-Rich Document Understanding (2601.05125)
We usually train VLMs on visual synthetic data that we (as humans) label as photorealistic. We argue that this is an anthropocentric perspective imposed to a model that might not synthetize visual information as we do. VERSE helps to visualize latent space and overlay visual features to detect poor-performance regions and take action to include better-suited training sets to boost model performance.
Resources:
- Code: https://github.com/nachoDRT/VrDU-Doctor
- Hugging Face Space: de-Rodrigo/Embeddings
Want to collaborate? Do you have any feedback? π§
PD: As always, we are grateful to Hugging Face π€ for providing the fantastic tools and resources we find on the platform!