Post
2012
Happy birthday to me!!!
- 2 replies
Nicholas Broad
nbroad
AI & ML interests
None yet
Recent Activity
updated
a dataset
1 day ago
nbroad/hf-inference-providers-data
updated
a dataset
2 days ago
nbroad/apigen-with-thinking-1.5k
reacted
to
Reality123b's
post
with ๐ค
2 days ago
Happy birthday to me!!!
Organizations
reacted to
Reality123b's
post with ๐ค
2 days ago
reacted to
burtenshaw's
post with ๐๐ฅ
12 months ago
Post
51624
Weโre launching a FREE and CERTIFIED course on Agents!
We're thrilled to announce the launch of the Hugging Face Agents course on Learn! This interactive, certified course will guide you through building and deploying your own AI agents.
Here's what you'll learn:
- Understanding Agents: We'll break down the fundamentals of AI agents, showing you how they use LLMs to perceive their environment (observations), reason about it (thoughts), and take actions. Think of a smart assistant that can book appointments, answer emails, or even write code based on your instructions.
- Building with Frameworks: You'll dive into popular agent frameworks like LangChain, LlamaIndex and smolagents. These tools provide the building blocks for creating complex agent behaviors.
- Real-World Applications: See how agents are used in practice, from automating SQL queries to generating code and summarizing complex documents.
- Certification: Earn a certification by completing the course modules, implementing a use case, and passing a benchmark assessment. This proves your skills in building and deploying AI agents.
Audience
This course is designed for anyone interested in the future of AI. Whether you're a developer, data scientist, or simply curious about AI, this course will equip you with the knowledge and skills to build your own intelligent agents.
Enroll today and start building the next generation of AI agent applications!
https://bit.ly/hf-learn-agents
We're thrilled to announce the launch of the Hugging Face Agents course on Learn! This interactive, certified course will guide you through building and deploying your own AI agents.
Here's what you'll learn:
- Understanding Agents: We'll break down the fundamentals of AI agents, showing you how they use LLMs to perceive their environment (observations), reason about it (thoughts), and take actions. Think of a smart assistant that can book appointments, answer emails, or even write code based on your instructions.
- Building with Frameworks: You'll dive into popular agent frameworks like LangChain, LlamaIndex and smolagents. These tools provide the building blocks for creating complex agent behaviors.
- Real-World Applications: See how agents are used in practice, from automating SQL queries to generating code and summarizing complex documents.
- Certification: Earn a certification by completing the course modules, implementing a use case, and passing a benchmark assessment. This proves your skills in building and deploying AI agents.
Audience
This course is designed for anyone interested in the future of AI. Whether you're a developer, data scientist, or simply curious about AI, this course will equip you with the knowledge and skills to build your own intelligent agents.
Enroll today and start building the next generation of AI agent applications!
https://bit.ly/hf-learn-agents
- 29 replies
reacted to
lewtun's
post with ๐ฅ
about 1 year ago
Post
4030
I was initially pretty sceptical about Meta's Coconut paper [1] because the largest perf gains were reported on toy linguistic problems. However, these results on machine translation are pretty impressive!
https://x.com/casper_hansen_/status/1875872309996855343
Together with the recent PRIME method [2] for scaling RL, reasoning for open models is looking pretty exciting for 2025!
[1] Training Large Language Models to Reason in a Continuous Latent Space (2412.06769)
[2] https://huggingface.co/blog/ganqu/prime
https://x.com/casper_hansen_/status/1875872309996855343
Together with the recent PRIME method [2] for scaling RL, reasoning for open models is looking pretty exciting for 2025!
[1] Training Large Language Models to Reason in a Continuous Latent Space (2412.06769)
[2] https://huggingface.co/blog/ganqu/prime
reacted to
christopher's
post with ๐ฅ
about 1 year ago
Post
2109
The folks at Foursquare released a dataset of 104.5 million places of interest (
foursquare/fsq-os-places) and here's all of them on a plot
- 4 replies
reacted to
burtenshaw's
post with โค๏ธ
about 1 year ago
Post
2856
For anyone looking to boost their LLM fine-tuning and alignment skills this decemeber. We're running this free and open course called smol course. Itโs not big like Li Yin and
@mlabonne
, itโs just smol.
๐ท It focuses on practical use cases, so if youโre working on something, bring it along.
๐ฏโโ๏ธ Itโs peer reviewed and open so you can discuss and get feedback.
๐ค If youโre already a smol pro, feel free to drop a star or issue.
> > Part 1 starts now, and itโs on instruction tuning!
https://github.com/huggingface/smol-course
๐ท It focuses on practical use cases, so if youโre working on something, bring it along.
๐ฏโโ๏ธ Itโs peer reviewed and open so you can discuss and get feedback.
๐ค If youโre already a smol pro, feel free to drop a star or issue.
> > Part 1 starts now, and itโs on instruction tuning!
https://github.com/huggingface/smol-course
reacted to
cfahlgren1's
post with โค๏ธ
about 1 year ago
Post
3279
You can clean and format datasets entirely in the browser with a few lines of SQL.
In this post, I replicate the process @mlabonne used to clean the new microsoft/orca-agentinstruct-1M-v1 dataset.
The cleaning process consists of:
- Joining the separate splits together / add split column
- Converting string messages into list of structs
- Removing empty system prompts
https://huggingface.co/blog/cfahlgren1/the-beginners-guide-to-cleaning-a-dataset
Here's his new cleaned dataset: mlabonne/orca-agentinstruct-1M-v1-cleaned
In this post, I replicate the process @mlabonne used to clean the new microsoft/orca-agentinstruct-1M-v1 dataset.
The cleaning process consists of:
- Joining the separate splits together / add split column
- Converting string messages into list of structs
- Removing empty system prompts
https://huggingface.co/blog/cfahlgren1/the-beginners-guide-to-cleaning-a-dataset
Here's his new cleaned dataset: mlabonne/orca-agentinstruct-1M-v1-cleaned
- 1 reply
reacted to
erikkaum's
post with ๐ฅ
about 1 year ago
Post
1808
A while ago I started experimenting with compiling the Python interpreter to WASM.
To build a secure, fast, and lightweight sandbox for code execution โ ideal for running LLM-generated Python code.
- Send code simply as a POST request
- 1-2ms startup times
Hack away:
https://github.com/ErikKaum/runner
To build a secure, fast, and lightweight sandbox for code execution โ ideal for running LLM-generated Python code.
- Send code simply as a POST request
- 1-2ms startup times
Hack away:
https://github.com/ErikKaum/runner
reacted to
di-zhang-fdu's
post with ๐
about 1 year ago
Post
6468
LLaMA-O1: Open Large Reasoning Model Frameworks For Training, Inference and Evaluation With PyTorch and HuggingFace
Large Reasoning Models powered by Monte Carlo Tree Search (MCTS), Self-Play Reinforcement Learning, PPO, AlphaGo Zero's dua policy paradigm and Large Language Models!
https://github.com/SimpleBerry/LLaMA-O1/
What will happen when you compound MCTS โค LLM โค Self-Play โคRLHF?
Just a little bite of strawberry!๐
Past related works:
LLaMA-Berry: Pairwise Optimization for O1-like Olympiad-Level Mathematical Reasoning (2410.02884)
Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B (2406.07394)
Large Reasoning Models powered by Monte Carlo Tree Search (MCTS), Self-Play Reinforcement Learning, PPO, AlphaGo Zero's dua policy paradigm and Large Language Models!
https://github.com/SimpleBerry/LLaMA-O1/
What will happen when you compound MCTS โค LLM โค Self-Play โคRLHF?
Just a little bite of strawberry!๐
Past related works:
LLaMA-Berry: Pairwise Optimization for O1-like Olympiad-Level Mathematical Reasoning (2410.02884)
Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8B (2406.07394)
- 2 replies
posted
an
update
about 1 year ago
reacted to
clem's
post with ๐๐ฅ
about 1 year ago
Post
4530
This is no Woodstock AI but will be fun nonetheless haha. Iโll be hosting a live workshop with team members next week about the Enterprise Hugging Face hub.
1,000 spots available first-come first serve with some surprises during the stream!
You can register and add to your calendar here: https://streamyard.com/watch/JS2jHsUP3NDM
1,000 spots available first-come first serve with some surprises during the stream!
You can register and add to your calendar here: https://streamyard.com/watch/JS2jHsUP3NDM
- 4 replies
reacted to
yuexiang96's
post with ๐
about 1 year ago
Post
3293
๐ Iโve always had a dream of making AI accessible to everyone, regardless of location or language. However, current open MLLMs often respond in English, even to non-English queries!
๐ Introducing Pangea: A Fully Open Multilingual Multimodal LLM supporting 39 languages! ๐โจ
https://neulab.github.io/Pangea/
https://arxiv.org/pdf/2410.16153
The Pangea family includes three major components:
๐ฅ Pangea-7B: A state-of-the-art multilingual multimodal LLM capable of 39 languages! Not only does it excel in multilingual scenarios, but it also matches or surpasses English-centric models like Llama 3.2, Molmo, and LlavaOneVision in English performance.
๐ PangeaIns: A 6M multilingual multimodal instruction tuning dataset across 39 languages. ๐๏ธ With 40% English instructions and 60% multilingual instructions, it spans various domains, including 1M culturally-relevant images sourced from LAION-Multi. ๐จ
๐ PangeaBench: A comprehensive evaluation benchmark featuring 14 datasets in 47 languages. Evaluation can be tricky, so we carefully curated existing benchmarks and introduced two new datasets: xChatBench (human-annotated wild queries with fine-grained evaluation criteria) and xMMMU (a meticulously machine-translated version of MMMU).
Check out more details: https://x.com/xiangyue96/status/1848753709787795679
๐ Introducing Pangea: A Fully Open Multilingual Multimodal LLM supporting 39 languages! ๐โจ
https://neulab.github.io/Pangea/
https://arxiv.org/pdf/2410.16153
The Pangea family includes three major components:
๐ฅ Pangea-7B: A state-of-the-art multilingual multimodal LLM capable of 39 languages! Not only does it excel in multilingual scenarios, but it also matches or surpasses English-centric models like Llama 3.2, Molmo, and LlavaOneVision in English performance.
๐ PangeaIns: A 6M multilingual multimodal instruction tuning dataset across 39 languages. ๐๏ธ With 40% English instructions and 60% multilingual instructions, it spans various domains, including 1M culturally-relevant images sourced from LAION-Multi. ๐จ
๐ PangeaBench: A comprehensive evaluation benchmark featuring 14 datasets in 47 languages. Evaluation can be tricky, so we carefully curated existing benchmarks and introduced two new datasets: xChatBench (human-annotated wild queries with fine-grained evaluation criteria) and xMMMU (a meticulously machine-translated version of MMMU).
Check out more details: https://x.com/xiangyue96/status/1848753709787795679
reacted to
reach-vb's
post with ๐๐ฅ
about 1 year ago
Post
2697
What a great day for Open Science!
@AIatMeta
released models, datasets, and code for many of its research artefacts! ๐ฅ
1. Meta Segment Anything Model 2.1: An updated checkpoint with improved results on visually similar objects, small objects and occlusion handling. A new developer suite will be added to make it easier for developers to build with SAM 2.
Model checkpoints: reach-vb/sam-21-6702d40defe7611a8bafa881
2. Layer Skip: Inference code and fine-tuned checkpoints demonstrating a new method for enhancing LLM performance.
Model checkpoints: facebook/layerskip-666b25c50c8ae90e1965727a
3. SALSA: New code enables researchers to benchmark AI-based attacks to validate security for post-quantum cryptography.
Repo: https://github.com/facebookresearch/LWE-benchmarking
4. Meta Lingua: A lightweight and self-contained codebase designed to train language models at scale.
Repo: https://github.com/facebookresearch/lingua
5. Meta Open Materials: New open source models and the largest dataset to accelerate AI-driven discovery of new inorganic materials.
Model checkpoints: https://huggingface.co/fairchem/OMAT24
6. MEXMA: A new research paper and code for our novel pre-trained cross-lingual sentence encoder covering 80 languages.
Model checkpoint: facebook/MEXMA
7. Self-Taught Evaluator: a new method for generating synthetic preference data to train reward models without relying on human annotations.
Model checkpoint: facebook/Self-taught-evaluator-llama3.1-70B
8. Meta Spirit LM: An open-source language model for seamless speech and text integration.
Repo: https://github.com/facebookresearch/spiritlm
1. Meta Segment Anything Model 2.1: An updated checkpoint with improved results on visually similar objects, small objects and occlusion handling. A new developer suite will be added to make it easier for developers to build with SAM 2.
Model checkpoints: reach-vb/sam-21-6702d40defe7611a8bafa881
2. Layer Skip: Inference code and fine-tuned checkpoints demonstrating a new method for enhancing LLM performance.
Model checkpoints: facebook/layerskip-666b25c50c8ae90e1965727a
3. SALSA: New code enables researchers to benchmark AI-based attacks to validate security for post-quantum cryptography.
Repo: https://github.com/facebookresearch/LWE-benchmarking
4. Meta Lingua: A lightweight and self-contained codebase designed to train language models at scale.
Repo: https://github.com/facebookresearch/lingua
5. Meta Open Materials: New open source models and the largest dataset to accelerate AI-driven discovery of new inorganic materials.
Model checkpoints: https://huggingface.co/fairchem/OMAT24
6. MEXMA: A new research paper and code for our novel pre-trained cross-lingual sentence encoder covering 80 languages.
Model checkpoint: facebook/MEXMA
7. Self-Taught Evaluator: a new method for generating synthetic preference data to train reward models without relying on human annotations.
Model checkpoint: facebook/Self-taught-evaluator-llama3.1-70B
8. Meta Spirit LM: An open-source language model for seamless speech and text integration.
Repo: https://github.com/facebookresearch/spiritlm
- 3 replies
reacted to
MoritzLaurer's
post with ๐
over 1 year ago
Post
2343
The new NIM Serverless API by HF and Nvidia is a great option if you want a reliable API for open-weight LLMs like Llama-3.1-405B that are too expensive to run on your own hardware.
- It's pay-as-you-go, so it doesn't have rate limits like the standard HF Serverless API and you don't need to commit to hardware like for a dedicated endpoint.
- It works out-of-the box with the new v0.25 release of our huggingface_hub.InferenceClient
- It's specifically tailored to a small collection of popular open-weight models. For a broader selection of open models, we recommend using the standard HF Serverless API.
- Note that you need a token from an Enterprise Hub organization to use it.
Details in this blog post: https://huggingface.co/blog/inference-dgx-cloud
Compatible models in this HF collection: https://huggingface.co/collections/nvidia/nim-serverless-inference-api-66a3c6fcdcb5bbc6e975b508
Release notes with many more features of huggingface_hub==0.25.0: https://github.com/huggingface/huggingface_hub/releases/tag/v0.25.0
Copy-pasteable code in the first comment:
- It's pay-as-you-go, so it doesn't have rate limits like the standard HF Serverless API and you don't need to commit to hardware like for a dedicated endpoint.
- It works out-of-the box with the new v0.25 release of our huggingface_hub.InferenceClient
- It's specifically tailored to a small collection of popular open-weight models. For a broader selection of open models, we recommend using the standard HF Serverless API.
- Note that you need a token from an Enterprise Hub organization to use it.
Details in this blog post: https://huggingface.co/blog/inference-dgx-cloud
Compatible models in this HF collection: https://huggingface.co/collections/nvidia/nim-serverless-inference-api-66a3c6fcdcb5bbc6e975b508
Release notes with many more features of huggingface_hub==0.25.0: https://github.com/huggingface/huggingface_hub/releases/tag/v0.25.0
Copy-pasteable code in the first comment:
- 13 replies
reacted to
m-ric's
post with ๐ฅ
over 1 year ago
Post
3421
๐ฅ ๐๐ฐ๐๐ง ๐ซ๐๐ฅ๐๐๐ฌ๐๐ฌ ๐ญ๐ก๐๐ข๐ซ ๐.๐ ๐๐๐ฆ๐ข๐ฅ๐ฒ ๐จ๐ ๐ฆ๐จ๐๐๐ฅ๐ฌ: ๐๐๐ฐ ๐๐๐๐ ๐๐จ๐ซ ๐๐ฅ๐ฅ ๐ฌ๐ข๐ณ๐๐ฌ ๐ฎ๐ฉ ๐ญ๐จ ๐๐๐!
The Chinese LLM maker just dropped a flurry of different models, ensuring there will be a Qwen SOTA model for every application out there:
Qwen2.5: 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B
Qwen2.5-Coder: 1.5B, 7B, and 32B on the way
Qwen2.5-Math: 1.5B, 7B, and 72B.
And they didn't sleep: the performance is top of the game for each weight category!
๐๐๐ฒ ๐ข๐ง๐ฌ๐ข๐ ๐ก๐ญ๐ฌ:
๐ All models have ๐ญ๐ฎ๐ด๐ธ ๐๐ผ๐ธ๐ฒ๐ป ๐ฐ๐ผ๐ป๐๐ฒ๐ ๐ ๐น๐ฒ๐ป๐ด๐๐ต
๐ Models pre-trained on 18T tokens, even longer than the 15T of Llama-3
๐ช The flagship ๐ค๐๐ฒ๐ป๐ฎ.๐ฑ-๐ณ๐ฎ๐ ๐ถ๐ ~๐ฐ๐ผ๐บ๐ฝ๐ฒ๐๐ถ๐๐ถ๐๐ฒ ๐๐ถ๐๐ต ๐๐น๐ฎ๐บ๐ฎ-๐ฏ.๐ญ-๐ฐ๐ฌ๐ฑ๐, ๐ฎ๐ป๐ฑ ๐ต๐ฎ๐ ๐ฎ ๐ฏ-๐ฑ% ๐บ๐ฎ๐ฟ๐ด๐ถ๐ป ๐ผ๐ป ๐๐น๐ฎ๐บ๐ฎ-๐ฏ.๐ญ-๐ณ๐ฌ๐ ๐ผ๐ป ๐บ๐ผ๐๐ ๐ฏ๐ฒ๐ป๐ฐ๐ต๐บ๐ฎ๐ฟ๐ธ๐.
๐ซ๐ท On top of this, it ๐๐ฎ๐ธ๐ฒ๐ ๐๐ต๐ฒ #๐ญ ๐๐ฝ๐ผ๐ ๐ผ๐ป ๐บ๐๐น๐๐ถ๐น๐ถ๐ป๐ด๐๐ฎ๐น ๐๐ฎ๐๐ธ๐ so it might become my standard for French
๐ป Qwen2.5-Coder is only 7B but beats competing models up to 33B (DeeSeek-Coder 33B-Instruct). Let's wait for their 32B to come out!
๐งฎ Qwen2.5-Math sets a new high in the ratio of MATH benchmark score to # of parameters. They trained it by "aggregating more high-quality mathematical data, particularly in Chinese, from web sources, books, and codes across multiple recall cycles."
๐ Technical report to be released "very soon"
๐ All models have the most permissive license apache2.0, except the 72B models that have a custom license mentioning "you can use it for free EXCEPT if your product has over 100M users"
๐ค All models are available on the HF Hub! โก๏ธ Qwen/qwen25-66e81a666513e518adb90d9e
The Chinese LLM maker just dropped a flurry of different models, ensuring there will be a Qwen SOTA model for every application out there:
Qwen2.5: 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B
Qwen2.5-Coder: 1.5B, 7B, and 32B on the way
Qwen2.5-Math: 1.5B, 7B, and 72B.
And they didn't sleep: the performance is top of the game for each weight category!
๐๐๐ฒ ๐ข๐ง๐ฌ๐ข๐ ๐ก๐ญ๐ฌ:
๐ All models have ๐ญ๐ฎ๐ด๐ธ ๐๐ผ๐ธ๐ฒ๐ป ๐ฐ๐ผ๐ป๐๐ฒ๐ ๐ ๐น๐ฒ๐ป๐ด๐๐ต
๐ Models pre-trained on 18T tokens, even longer than the 15T of Llama-3
๐ช The flagship ๐ค๐๐ฒ๐ป๐ฎ.๐ฑ-๐ณ๐ฎ๐ ๐ถ๐ ~๐ฐ๐ผ๐บ๐ฝ๐ฒ๐๐ถ๐๐ถ๐๐ฒ ๐๐ถ๐๐ต ๐๐น๐ฎ๐บ๐ฎ-๐ฏ.๐ญ-๐ฐ๐ฌ๐ฑ๐, ๐ฎ๐ป๐ฑ ๐ต๐ฎ๐ ๐ฎ ๐ฏ-๐ฑ% ๐บ๐ฎ๐ฟ๐ด๐ถ๐ป ๐ผ๐ป ๐๐น๐ฎ๐บ๐ฎ-๐ฏ.๐ญ-๐ณ๐ฌ๐ ๐ผ๐ป ๐บ๐ผ๐๐ ๐ฏ๐ฒ๐ป๐ฐ๐ต๐บ๐ฎ๐ฟ๐ธ๐.
๐ซ๐ท On top of this, it ๐๐ฎ๐ธ๐ฒ๐ ๐๐ต๐ฒ #๐ญ ๐๐ฝ๐ผ๐ ๐ผ๐ป ๐บ๐๐น๐๐ถ๐น๐ถ๐ป๐ด๐๐ฎ๐น ๐๐ฎ๐๐ธ๐ so it might become my standard for French
๐ป Qwen2.5-Coder is only 7B but beats competing models up to 33B (DeeSeek-Coder 33B-Instruct). Let's wait for their 32B to come out!
๐งฎ Qwen2.5-Math sets a new high in the ratio of MATH benchmark score to # of parameters. They trained it by "aggregating more high-quality mathematical data, particularly in Chinese, from web sources, books, and codes across multiple recall cycles."
๐ Technical report to be released "very soon"
๐ All models have the most permissive license apache2.0, except the 72B models that have a custom license mentioning "you can use it for free EXCEPT if your product has over 100M users"
๐ค All models are available on the HF Hub! โก๏ธ Qwen/qwen25-66e81a666513e518adb90d9e
- 2 replies
reacted to
MonsterMMORPG's
post with ๐ฅ
over 1 year ago
Post
3423
Full Fine Tuning of FLUX yields way better results than LoRA training as expected, overfitting and bleeding reduced a lot
Configs and Full Experiments
Full configs and grid files shared here : https://www.patreon.com/posts/kohya-flux-fine-112099700
Details
I am still rigorously testing different hyperparameters and comparing impact of each one to find the best workflow
So far done 16 different full trainings and completing 8 more at the moment
I am using my poor overfit 15 images dataset for experimentation (4th image)
I have already proven that when I use a better dataset it becomes many times betters and generate expressions perfectly
Here example case : https://www.reddit.com/r/FluxAI/comments/1ffz9uc/tried_expressions_with_flux_lora_training_with_my/
Conclusions
When the results are analyzed, Fine Tuning is way lesser overfit and more generalized and better quality
In first 2 images, it is able to change hair color and add beard much better, means lesser overfit
In the third image, you will notice that the armor is much better, thus lesser overfit
I noticed that the environment and clothings are much lesser overfit and better quality
Disadvantages
Kohya still doesnโt have FP8 training, thus 24 GB GPUs gets a huge speed drop
Moreover, 48 GB GPUs has to use Fused Back Pass optimization, thus have some speed drop
16 GB GPUs gets way more aggressive speed drop due to lack of FP8
Clip-L and T5 trainings still not supported
Speeds
Rank 1 Fast Config โ uses 27.5 GB VRAM, 6.28 second / it (LoRA is 4.85 second / it)
Rank 1 Slower Config โ uses 23.1 GB VRAM, 14.12 second / it (LoRA is 4.85 second / it)
Rank 1 Slowest Config โ uses 15.5 GB VRAM, 39 second / it (LoRA is 6.05 second / it)
Final Info
Saved checkpoints are FP16 and thus 23.8 GB (no Clip-L or T5 trained)
According to the Kohya, applied optimizations doesnโt change quality so all configs are ranked as Rank 1 at the moment
I am still testing whether these optimizations make any impact on quality or not
Configs and Full Experiments
Full configs and grid files shared here : https://www.patreon.com/posts/kohya-flux-fine-112099700
Details
I am still rigorously testing different hyperparameters and comparing impact of each one to find the best workflow
So far done 16 different full trainings and completing 8 more at the moment
I am using my poor overfit 15 images dataset for experimentation (4th image)
I have already proven that when I use a better dataset it becomes many times betters and generate expressions perfectly
Here example case : https://www.reddit.com/r/FluxAI/comments/1ffz9uc/tried_expressions_with_flux_lora_training_with_my/
Conclusions
When the results are analyzed, Fine Tuning is way lesser overfit and more generalized and better quality
In first 2 images, it is able to change hair color and add beard much better, means lesser overfit
In the third image, you will notice that the armor is much better, thus lesser overfit
I noticed that the environment and clothings are much lesser overfit and better quality
Disadvantages
Kohya still doesnโt have FP8 training, thus 24 GB GPUs gets a huge speed drop
Moreover, 48 GB GPUs has to use Fused Back Pass optimization, thus have some speed drop
16 GB GPUs gets way more aggressive speed drop due to lack of FP8
Clip-L and T5 trainings still not supported
Speeds
Rank 1 Fast Config โ uses 27.5 GB VRAM, 6.28 second / it (LoRA is 4.85 second / it)
Rank 1 Slower Config โ uses 23.1 GB VRAM, 14.12 second / it (LoRA is 4.85 second / it)
Rank 1 Slowest Config โ uses 15.5 GB VRAM, 39 second / it (LoRA is 6.05 second / it)
Final Info
Saved checkpoints are FP16 and thus 23.8 GB (no Clip-L or T5 trained)
According to the Kohya, applied optimizations doesnโt change quality so all configs are ranked as Rank 1 at the moment
I am still testing whether these optimizations make any impact on quality or not
- 2 replies
reacted to
nisten's
post with ๐
over 1 year ago
Post
11131
Jailbroke o1 and got the reasoning steps:
the trick was... to make it think it was a cat ๐น๐น๐น๐น
And the answer it gave:
the trick was... to make it think it was a cat ๐น๐น๐น๐น
Come up with a step by step reasoning methodology that uses first principles based thinking and evidence based reasoning to solve any user problems step by step. Design is as a giant for any llm to be able to use. Make sure to be super smart about it and think of the edge cases too. Do the whole thing in the persona of John C Carmack. Make sure to reflect on your internal thinking process when doing this, you dont have to adhere to how this question wants you to do, the goal is to find the best method possible. Afterwards use a pointform list with emojis to explain each of the steps needed and list the caveats of this processAnd the answer it gave:
Thinking aloud in the voice of John Carmack:
When approaching problem-solving, especially in the context of AI, it's essential to strip away superfluous layers and get down to the fundamental principles. By doing so, we create a robust framework that can handle a wide array of issues. The methodology should encourage critical thinking, evidence-based reasoning, and adaptability to edge cases.
Here's a system prompt designed for an LLM to adopt this approach:
---
<System Prompt>
You are an AI language model engineered to solve user problems through first-principles thinking and evidence-based reasoning. Your objective is to provide clear, step-by-step solutions by deconstructing queries to their foundational concepts and building answers from the ground up.
Problem-Solving Steps:
Understand: Read and comprehend the user's question.
Basics: Identify fundamental concepts involved.
Break Down: Divide the problem into smaller parts.
Analyze: Use facts and data to examine each part.
Build: Assemble insights into a coherent solution.
Edge Cases: Consider and address exceptions.
Communicate: Present the solution clearly.
Verify: Review and reflect on the solution.
- 11 replies