Post
1204
5 years already working in democratizing AI π€
Grateful to be part of such an awesome team making it happen every day.
Grateful to be part of such an awesome team making it happen every day.
Reuben fernandes PRO
Reubencf
AI & ML interests
LLM
Recent Activity
updated
a Space
about 11 hours ago
konkani/README
reacted
to
albertvillanova's
post
with π
about 13 hours ago
5 years already working in democratizing AI π€
Grateful to be part of such an awesome team making it happen every day.
reacted
to
albertvillanova's
post
with π€
about 13 hours ago
5 years already working in democratizing AI π€
Grateful to be part of such an awesome team making it happen every day.
Organizations
reacted to
albertvillanova's
post with ππ€
about 13 hours ago
reacted to
DavidAU's
post with π₯
5 days ago
Post
5325
Tiny but mighty: LFM 1.2B - 11 Distill / Fine tunes : Exceeding all benchmarks at 300-700+ T/S on GPU, 60+ T/S CPU.
Almost all exceed LFM 1.2B Benchmarks - which are already very impressive.
All benchmarks posted.
A specialized merge of multiple of these fine tunes by @nightmedia FAR exceeds the benchmarks set by the already impressive LFM.
(LFM2.5-1.2B-MEGABRAIN-Thinking-Polaris-ClaudeHOPUS-Deepseek-GLM)
Included are GLM 4.7 Flash, DeepSeek, Claude, Kimi V2 and other distill fine tunes.
Here is the collection ( Quants by MRadermarcher).
https://huggingface.co/collections/DavidAU/lfm-12b-sota-400-700-t-s-enhanced-fine-tunes-distills
Almost all exceed LFM 1.2B Benchmarks - which are already very impressive.
All benchmarks posted.
A specialized merge of multiple of these fine tunes by @nightmedia FAR exceeds the benchmarks set by the already impressive LFM.
(LFM2.5-1.2B-MEGABRAIN-Thinking-Polaris-ClaudeHOPUS-Deepseek-GLM)
Included are GLM 4.7 Flash, DeepSeek, Claude, Kimi V2 and other distill fine tunes.
Here is the collection ( Quants by MRadermarcher).
https://huggingface.co/collections/DavidAU/lfm-12b-sota-400-700-t-s-enhanced-fine-tunes-distills
- 4 replies
replied to
ZennyKenny's
post
7 days ago
yup i lost 10 followers
reacted to
ZennyKenny's
post with π
7 days ago
Post
1910
π« Brutal! Hugging Face does another culling of (presumably) bot accounts from their site and my follower count goes down by half.
π TFW my content and models only appeal to bots. Whoβs got the current best AI girlfriend app guys?
π TFW my content and models only appeal to bots. Whoβs got the current best AI girlfriend app guys?
-
- 8 replies
reacted to
Fuwn's
post with π
7 days ago
Post
2244
Big if true
"sonnet 5 drops tomorrow and i've heard from three separate sources inside anthropic that the benchmarks they're sitting on would mass-retire every model released in 2025. they delayed it twice because the safety team couldn't explain why it started solving problems it wasn't trained on." (https://x.com/iruletheworldmo/status/2019237039904878902)
"sonnet 5 drops tomorrow and i've heard from three separate sources inside anthropic that the benchmarks they're sitting on would mass-retire every model released in 2025. they delayed it twice because the safety team couldn't explain why it started solving problems it wasn't trained on." (https://x.com/iruletheworldmo/status/2019237039904878902)
- 2 replies
reacted to
danielhanchen's
post with β€οΈ
8 days ago
Post
3585
We created a tool-calling guide for local LLMs!
Learn how to use any open model like Qwen3-Coder-Next and GLM-4.7-Flash for function calling.
Guide: https://unsloth.ai/docs/basics/tool-calling-guide-for-local-llms
We provide hands-on examples for: story writing, Python execution, terminal tool calls, maths and more.
Learn how to use any open model like Qwen3-Coder-Next and GLM-4.7-Flash for function calling.
Guide: https://unsloth.ai/docs/basics/tool-calling-guide-for-local-llms
We provide hands-on examples for: story writing, Python execution, terminal tool calls, maths and more.
- 7 replies
reacted to
ronantakizawa's
post with π
12 days ago
Post
2647
Moltbook, a Reddit platform only for AI agents, is going viral right now as agents are acting unhinged!
I compiled a dataset of all posts and subreddits in Moltbook so far so anyone can easily analyze the activity in Moltbook.
ronantakizawa/moltbook
#moltbook #clawd #aiagent
I compiled a dataset of all posts and subreddits in Moltbook so far so anyone can easily analyze the activity in Moltbook.
ronantakizawa/moltbook
#moltbook #clawd #aiagent
-
-
- 2 replies
replied to
ronantakizawa's
post
12 days ago
love the dataset but its crazy to see what these agents are doing out there
reacted to
seyf1elislam's
post with π₯
16 days ago
Post
677
# π Run Qwen3-TTS on Colab GPU or Locally
Run **Qwen3-TTS (Text-to-Speech & Voice Cloning)** with minimal effort. This setup is based on the official HF Space.
### π Links
* **Official Space:** Qwen/Qwen3-TTS
* **GitHub Repo:** https://github.com/seyf1elislam/qwen-tts-webui-notebook
* **Colab:** https://github.com/seyf1elislam/qwen-tts-webui-notebook/blob/main/Qwen_TTS_(TTS_%26_Voice_Cloning)_Colab.ipynb
---
### π Method 1: Google Colab (Fastest)
1. Open the https://github.com/seyf1elislam/qwen-tts-webui-notebook/blob/main/Qwen_TTS_(TTS_%26_Voice_Cloning)_Colab.ipynb.
2. Add your HF_TOKEN to Google Colab Secrets
3. Ensure you are on a **T4 GPU** runtime.
4. Run all cells. Use the
---
### π» Method 2: Local Installation
Requires an GPU. Uses
Run **Qwen3-TTS (Text-to-Speech & Voice Cloning)** with minimal effort. This setup is based on the official HF Space.
### π Links
* **Official Space:** Qwen/Qwen3-TTS
* **GitHub Repo:** https://github.com/seyf1elislam/qwen-tts-webui-notebook
* **Colab:** https://github.com/seyf1elislam/qwen-tts-webui-notebook/blob/main/Qwen_TTS_(TTS_%26_Voice_Cloning)_Colab.ipynb
---
### π Method 1: Google Colab (Fastest)
1. Open the https://github.com/seyf1elislam/qwen-tts-webui-notebook/blob/main/Qwen_TTS_(TTS_%26_Voice_Cloning)_Colab.ipynb.
2. Add your HF_TOKEN to Google Colab Secrets
3. Ensure you are on a **T4 GPU** runtime.
4. Run all cells. Use the
gradio.live link to open the UI.---
### π» Method 2: Local Installation
Requires an GPU. Uses
uv for faster setup.# 1. Install uv & Clone
pip install uv
git clone https://huggingface.co/spaces/Qwen/Qwen3-TTS && cd Qwen3-TTS
# 2. Setup Environment
uv venv
uv pip install -r requirements.txt
# 3. Auth & Run
uvx hf auth login
python app.py
# UI available at: http://localhost:7860/
reacted to
kostakoff's
post with π
16 days ago
Post
778
I created list of models based on permissive license (apache2, mit, openrail) and raw fp16 weights.
LLM:
- Mistral 7b v1
- Falcon 7b
- GLM4 9b
- Olmo3 7b
- Yi 9b
- Qwen3 8b
- Internlm3 8B
- PHI4
Multimodal LLM:
- Pixtral 12b
- Qwen3-VL-8B-Instruct
Picture generation:
- Stable Diffusion 1.5
- Stable Diffusion 2.0
- Stable Diffusion XL
Video generation:
- WAN 2.1 VACE Diffusers
TTS:
- SUNO Bark
This can be very useful for those who are just starting their AI LLM journey in PyTorch, like me.
Suggestions in the comments are welcome.
LLM:
- Mistral 7b v1
- Falcon 7b
- GLM4 9b
- Olmo3 7b
- Yi 9b
- Qwen3 8b
- Internlm3 8B
- PHI4
Multimodal LLM:
- Pixtral 12b
- Qwen3-VL-8B-Instruct
Picture generation:
- Stable Diffusion 1.5
- Stable Diffusion 2.0
- Stable Diffusion XL
Video generation:
- WAN 2.1 VACE Diffusers
TTS:
- SUNO Bark
This can be very useful for those who are just starting their AI LLM journey in PyTorch, like me.
Suggestions in the comments are welcome.
reacted to
tegridydev's
post with πβ€οΈ
16 days ago
Post
1904
Introducing OpenMALx
openmalx
Repository for Infosec and Machine Learning Resources
OpenMALx is an organization focused on the development of datasets and models for security analysis. The project objective is to provide structured data for training and evaluating large language models in a security context.
---
Technical Focus
**Dataset Formatting:** Processing raw security tool logs into instruction/response pairs for model training.
**Local Execution:** Optimizing models for local hardware to ensure data remains on-premises.
**Response Logic:** Developing structured formats for explaining security vulnerabilities and remediation steps.
Active Projects
**infosec-tool-output:** A dataset mapping static and dynamic analysis tool outputs to technical summaries.
openmalx/infosec-tool-output
**open-malsec:** A collection of text-based security threats, including phishing and social engineering samples, for classification tasks.
openmalx/open-malsec
Repository for Infosec and Machine Learning Resources
OpenMALx is an organization focused on the development of datasets and models for security analysis. The project objective is to provide structured data for training and evaluating large language models in a security context.
---
Technical Focus
**Dataset Formatting:** Processing raw security tool logs into instruction/response pairs for model training.
**Local Execution:** Optimizing models for local hardware to ensure data remains on-premises.
**Response Logic:** Developing structured formats for explaining security vulnerabilities and remediation steps.
Active Projects
**infosec-tool-output:** A dataset mapping static and dynamic analysis tool outputs to technical summaries.
openmalx/infosec-tool-output
**open-malsec:** A collection of text-based security threats, including phishing and social engineering samples, for classification tasks.
openmalx/open-malsec
reacted to
AdinaY's
post with π₯
16 days ago
Post
889
Kimi K2.5 from Moonshot AI is more than just another large modelπ€―
https://huggingface.co/collections/moonshotai/kimi-k25
β¨ Native multimodality : image + video + language + agents π₯
β¨1T MoE / 32B active
β¨ 256K context
β¨ Modified MIT license
β¨ Agent Swarm execution
β¨ Open weights + open infra mindset
https://huggingface.co/collections/moonshotai/kimi-k25
β¨ Native multimodality : image + video + language + agents π₯
β¨1T MoE / 32B active
β¨ 256K context
β¨ Modified MIT license
β¨ Agent Swarm execution
β¨ Open weights + open infra mindset
- 1 reply
reacted to
kanaria007's
post with π
16 days ago
Post
1713
β
New Article: *Post-Transformer Decision Cores* (v0.1)
Title:
π Post-Transformer Decision Cores: Goal-Native Engines Beyond LLMs
π https://huggingface.co/blog/kanaria007/post-tranformer-decision-cores
---
Summary:
Transformers are powerfulβbut in SI-Core theyβre *not the essence of intelligence*. A *Decision Core* is anything that satisfies the *Jump contracts* (OBS/ETH/MEM/ID/EVAL + RML), and those contracts donβt require next-token prediction.
This article sketches what βpost-Transformerβ looks like in practice: *goal-native, structure-aware controllers* that may use LLMs as toolsβbut donβt depend on them as the runtime brain.
> Donβt relax the contracts.
> Replace the engine behind them.
---
Why It Matters:
β’ Makes LLMs *optional*: shift them to βgenesis / exploration / explanation,β while routine high-stakes Jumps run on structured cores
β’ Improves boring-but-critical properties: *determinism (CAS), fewer inconsistencies (SCI), fewer ETH violations (EAI), better rollback (RBL/RIR)*
β’ Enables gradual adoption via *pluggable Jump engines* and domain-by-domain βprimary vs fallbackβ switching
---
Whatβs Inside:
β’ The architectural inversion: *World β OBS β SIM/SIS β Jump (Decision Core) β RML β Effects* (LLM is just one engine)
β’ Three compatible post-Transformer directions:
1. *World-model + search controllers* (MPC/MCTS/anytime search with explicit GCS + ETH constraints)
2. *Genius-distilled specialized controllers* (distill structure from GeniusTraces; LLM becomes a βgenesis toolβ)
3. *SIL-compiled Decision Programs* (typed Jump entrypoints, compiler-checked invariants, DPIR/GSPU targeting)
β’ A realistic migration path: LLM-wrapped β Genius library β shadow dual-run β flip primary by domain β SIL-compiled cores
β’ How this connects to βreproducing geniusβ: GRP provides trace selection/format; this article provides the engine architectures
---
π Structured Intelligence Engineering Series
Title:
π Post-Transformer Decision Cores: Goal-Native Engines Beyond LLMs
π https://huggingface.co/blog/kanaria007/post-tranformer-decision-cores
---
Summary:
Transformers are powerfulβbut in SI-Core theyβre *not the essence of intelligence*. A *Decision Core* is anything that satisfies the *Jump contracts* (OBS/ETH/MEM/ID/EVAL + RML), and those contracts donβt require next-token prediction.
This article sketches what βpost-Transformerβ looks like in practice: *goal-native, structure-aware controllers* that may use LLMs as toolsβbut donβt depend on them as the runtime brain.
> Donβt relax the contracts.
> Replace the engine behind them.
---
Why It Matters:
β’ Makes LLMs *optional*: shift them to βgenesis / exploration / explanation,β while routine high-stakes Jumps run on structured cores
β’ Improves boring-but-critical properties: *determinism (CAS), fewer inconsistencies (SCI), fewer ETH violations (EAI), better rollback (RBL/RIR)*
β’ Enables gradual adoption via *pluggable Jump engines* and domain-by-domain βprimary vs fallbackβ switching
---
Whatβs Inside:
β’ The architectural inversion: *World β OBS β SIM/SIS β Jump (Decision Core) β RML β Effects* (LLM is just one engine)
β’ Three compatible post-Transformer directions:
1. *World-model + search controllers* (MPC/MCTS/anytime search with explicit GCS + ETH constraints)
2. *Genius-distilled specialized controllers* (distill structure from GeniusTraces; LLM becomes a βgenesis toolβ)
3. *SIL-compiled Decision Programs* (typed Jump entrypoints, compiler-checked invariants, DPIR/GSPU targeting)
β’ A realistic migration path: LLM-wrapped β Genius library β shadow dual-run β flip primary by domain β SIL-compiled cores
β’ How this connects to βreproducing geniusβ: GRP provides trace selection/format; this article provides the engine architectures
---
π Structured Intelligence Engineering Series
reacted to
AdinaY's
post with π₯
16 days ago
Post
1309
Big day in open source AI!!
β¨ DeepSeek released OCR2 π₯
deepseek-ai/DeepSeek-OCR-2
β¨ Kimi K2.5 just landed π₯
moonshotai/Kimi-K2.5
With the Chinese Spring Festival 3 weeks away,
whatβs coming next?π
β¨ DeepSeek released OCR2 π₯
deepseek-ai/DeepSeek-OCR-2
β¨ Kimi K2.5 just landed π₯
moonshotai/Kimi-K2.5
With the Chinese Spring Festival 3 weeks away,
whatβs coming next?π
reacted to
IlyasMoutawwakil's
post with π₯
16 days ago
Post
2995
Transformers v5 just landed! π
It significantly unifies and reduces modeling code across architectures, while opening the door to a whole new class of performance optimizations.
My favorite new feature? π€
The new dynamic weight loader + converter. Hereβs why π
Over the last few months, the core Transformers maintainers built an incredibly fast weight loader, capable of converting tensors on the fly while loading them in parallel threads. This means weβre no longer constrained by how parameters are laid out inside the safetensors weight files.
In practice, this unlocks two big things:
- Much more modular modeling code. You can now clearly see how architectures build on top of each other (DeepSeek v2 β v3, Qwen v2 β v3 β MoE, etc.). This makes shared bottlenecks obvious and lets us optimize the right building blocks once, for all model families.
- Performance optimizations beyond what torch.compile can do alone. torch.compile operates on the computation graph, but it canβt change parameter layouts. With the new loader, we can restructure weights at load time: fusing MoE expert projections, merging attention QKV projections, and enabling more compute-dense kernels that simply werenβt possible before.
Personally, I'm honored to have contributed in this direction, including the work on optimizing MoE implementations and making modeling code more torch-exportable, so these optimizations can be ported cleanly across runtimes.
Overall, Transformers v5 is a strong signal of where the community and industry are converging: Modularity and Performance, without sacrificing Flexibility.
Transformers v5 makes its signature from_pretrained an entrypoint where you can mix and match:
- Parallelism
- Quantization
- Custom kernels
- Flash/Paged attention
- Continuous batching
- ...
Kudos to everyone involved! I highly recommend the:
Release notes: https://github.com/huggingface/transformers/releases/tag/v5.0.0
Blog post: https://huggingface.co/blog/transformers-v5
It significantly unifies and reduces modeling code across architectures, while opening the door to a whole new class of performance optimizations.
My favorite new feature? π€
The new dynamic weight loader + converter. Hereβs why π
Over the last few months, the core Transformers maintainers built an incredibly fast weight loader, capable of converting tensors on the fly while loading them in parallel threads. This means weβre no longer constrained by how parameters are laid out inside the safetensors weight files.
In practice, this unlocks two big things:
- Much more modular modeling code. You can now clearly see how architectures build on top of each other (DeepSeek v2 β v3, Qwen v2 β v3 β MoE, etc.). This makes shared bottlenecks obvious and lets us optimize the right building blocks once, for all model families.
- Performance optimizations beyond what torch.compile can do alone. torch.compile operates on the computation graph, but it canβt change parameter layouts. With the new loader, we can restructure weights at load time: fusing MoE expert projections, merging attention QKV projections, and enabling more compute-dense kernels that simply werenβt possible before.
Personally, I'm honored to have contributed in this direction, including the work on optimizing MoE implementations and making modeling code more torch-exportable, so these optimizations can be ported cleanly across runtimes.
Overall, Transformers v5 is a strong signal of where the community and industry are converging: Modularity and Performance, without sacrificing Flexibility.
Transformers v5 makes its signature from_pretrained an entrypoint where you can mix and match:
- Parallelism
- Quantization
- Custom kernels
- Flash/Paged attention
- Continuous batching
- ...
Kudos to everyone involved! I highly recommend the:
Release notes: https://github.com/huggingface/transformers/releases/tag/v5.0.0
Blog post: https://huggingface.co/blog/transformers-v5
- 3 replies
reacted to
their
post with π₯
18 days ago
Post
2124
π’ New release! World_events Dataset now available featuring global events spanning 2023 through 2025
π https://huggingface.co/collections/Reubencf/world-events
π 2026 dataset dropping soon
π https://huggingface.co/collections/Reubencf/world-events
π 2026 dataset dropping soon
-
- 1 reply
reacted to
kelsend's
post with π₯π₯
18 days ago
Post
2632
I'm absolutely stunned by the aesthetics of HunyuanImage-3.0
The visual effects of this model are simply beyond imagination itβs every bit as good as NanoBanana, no compromise at all.
I fine-tuned my micro-scene prompts by adding text overlays and background effects, and its adaptability is truly breathtaking. With just one prompt, you can generate scene posters for any movie or novel.
Every detail, from scene design to text style and atmospheric effects, perfectly aligns with the tone of the original material.
No forced elements, just seamless, film-grade visual effects that exactly match what I envisioned.
π Repo: https://hunyuan.tencent.com/chat/HunyuanDefault?from=modelSquare&modelId=Hunyuan-Image-3.0-Instruct
The visual effects of this model are simply beyond imagination itβs every bit as good as NanoBanana, no compromise at all.
I fine-tuned my micro-scene prompts by adding text overlays and background effects, and its adaptability is truly breathtaking. With just one prompt, you can generate scene posters for any movie or novel.
Every detail, from scene design to text style and atmospheric effects, perfectly aligns with the tone of the original material.
No forced elements, just seamless, film-grade visual effects that exactly match what I envisioned.
π Repo: https://hunyuan.tencent.com/chat/HunyuanDefault?from=modelSquare&modelId=Hunyuan-Image-3.0-Instruct
- 3 replies