Reuben fernandes PRO
Reubencf
AI & ML interests
LLM
Recent Activity
reacted to theirpost with 🔥 about 5 hours ago
https://huggingface.co/spaces/Reubencf/HF_QR_Gen
has been improved create custom Qr-codes posted an update about 5 hours ago
https://huggingface.co/spaces/Reubencf/HF_QR_Gen
has been improved create custom Qr-codes updated a Space about 6 hours ago
Reubencf/HF_QR_GenOrganizations
reacted to their post with 🔥 about 5 hours ago
reacted to RDTvlokip's post with 👍 3 days ago
Post
1975
I finally changed the architecture of my 15M French LLM. It worked. Then I almost fooled myself about how much and catching that was the real win.
After proving last time that architecture is a threshold, not a lever, I got stubborn: could I change how the model learns? Four honest attempts, Lion, a sharper AdamW β2, multi-token prediction, LayerScale. Four failures. The bottleneck wasn't the learning rule either.
So I changed the shape of the computation instead: loop the same transformer blocks 4×, deeper reasoning, zero added parameters. It beat the baseline on perplexity, the first thing in the whole project to move that number. Then I added my own twist: let each token decide how deep to think, halting on its own entropy.
My first evaluation was spectacular. Coherence up 65%. Hallucinated names down 62%.
It was noise.
Eight prompts, one seed. I re-ran on 50 prompts × 200 tokens and watched the gains shrink to "modest" and on out-of-domain prompts, recurrence actually made things worse. No universal winner. And none of it is new: it's Adaptive Computation Time (2016), the Universal Transformer (2018), and LoopViT (2026), recombined and measured honestly.
The real lesson:
A number from 8 prompts is a rumor. The eval harness that kills your own best result is worth more than the result it kills. Cite your lineage. Stay preliminary until multiple seeds say otherwise.
The three models are live. The write-up is honest about every caveat 👇
🔗 https://huggingface.co/blog/RDTvlokip/teaching-a-15m-french-llm-to-think-deeper
After proving last time that architecture is a threshold, not a lever, I got stubborn: could I change how the model learns? Four honest attempts, Lion, a sharper AdamW β2, multi-token prediction, LayerScale. Four failures. The bottleneck wasn't the learning rule either.
So I changed the shape of the computation instead: loop the same transformer blocks 4×, deeper reasoning, zero added parameters. It beat the baseline on perplexity, the first thing in the whole project to move that number. Then I added my own twist: let each token decide how deep to think, halting on its own entropy.
My first evaluation was spectacular. Coherence up 65%. Hallucinated names down 62%.
It was noise.
Eight prompts, one seed. I re-ran on 50 prompts × 200 tokens and watched the gains shrink to "modest" and on out-of-domain prompts, recurrence actually made things worse. No universal winner. And none of it is new: it's Adaptive Computation Time (2016), the Universal Transformer (2018), and LoopViT (2026), recombined and measured honestly.
The real lesson:
A number from 8 prompts is a rumor. The eval harness that kills your own best result is worth more than the result it kills. Cite your lineage. Stay preliminary until multiple seeds say otherwise.
The three models are live. The write-up is honest about every caveat 👇
🔗 https://huggingface.co/blog/RDTvlokip/teaching-a-15m-french-llm-to-think-deeper
- 6 replies
reacted to their post with 🧠🤯👀 3 days ago
Post
134
Introducing Reubencf/Document_Query
powered by CohereLabs/command-a-plus-05-2026-w4a4 by cohere
just open admin page -> drop in any document -> and ask questions
powered by CohereLabs/command-a-plus-05-2026-w4a4 by cohere
just open admin page -> drop in any document -> and ask questions
- 1 reply
reacted to pankajpandey-dev's post with ❤️🚀🔥 3 days ago
Post
4012
🇮🇳 Qwen3.5-9B Hindi Instruct — it stops thinking in English
Ask base Qwen3.5-9B a question in Hindi and it burns hundreds of tokens thinking in English inside its think block before a single Devanagari word appears — then code-switches in the answer. I fine-tuned it to close the think block instantly and reply in pure, native Hindi.
✅ Model (16-bit): pankajpandey-dev/qwen3.5-9b-hindi-instruct
✅ GGUF (Q4/Q5/Q8): pankajpandey-dev/qwen3.5-9b-hindi-instruct-GGUF
✅ Try it in the browser: pankajpandey-dev/qwen3.5-9b-hindi-demo
Recipe: Unsloth + LoRA (r=16, response-only loss) on 12.9k Hindi pairs — AI4Bharat anudesh + dolly-hi + wikiHow-hi + Aya Hindi (human-written). The Q4_K_M is 5.4 GB and runs on a plain laptop CPU.
New in this run vs my earlier models: mixed in long-form native sources (wikiHow) after my last eval showed the fine-tune traded detail for conciseness — this one keeps answers detailed and native.
Part of my weekly 🇮🇳 Hindi LLM Series. Feedback welcome 🙏
#Hindi #IndicNLP #Qwen #GGUF #LocalLLM #Unsloth
Ask base Qwen3.5-9B a question in Hindi and it burns hundreds of tokens thinking in English inside its think block before a single Devanagari word appears — then code-switches in the answer. I fine-tuned it to close the think block instantly and reply in pure, native Hindi.
✅ Model (16-bit): pankajpandey-dev/qwen3.5-9b-hindi-instruct
✅ GGUF (Q4/Q5/Q8): pankajpandey-dev/qwen3.5-9b-hindi-instruct-GGUF
✅ Try it in the browser: pankajpandey-dev/qwen3.5-9b-hindi-demo
Recipe: Unsloth + LoRA (r=16, response-only loss) on 12.9k Hindi pairs — AI4Bharat anudesh + dolly-hi + wikiHow-hi + Aya Hindi (human-written). The Q4_K_M is 5.4 GB and runs on a plain laptop CPU.
New in this run vs my earlier models: mixed in long-form native sources (wikiHow) after my last eval showed the fine-tune traded detail for conciseness — this one keeps answers detailed and native.
Part of my weekly 🇮🇳 Hindi LLM Series. Feedback welcome 🙏
#Hindi #IndicNLP #Qwen #GGUF #LocalLLM #Unsloth
- 4 replies
replied to their post 4 days ago
@CohereLabs
reacted to their post with 🔥 6 days ago
Post
134
Introducing Reubencf/Document_Query
powered by CohereLabs/command-a-plus-05-2026-w4a4 by cohere
just open admin page -> drop in any document -> and ask questions
powered by CohereLabs/command-a-plus-05-2026-w4a4 by cohere
just open admin page -> drop in any document -> and ask questions
- 1 reply
posted an update 6 days ago
Post
134
Introducing Reubencf/Document_Query
powered by CohereLabs/command-a-plus-05-2026-w4a4 by cohere
just open admin page -> drop in any document -> and ask questions
powered by CohereLabs/command-a-plus-05-2026-w4a4 by cohere
just open admin page -> drop in any document -> and ask questions
- 1 reply
replied to their post 8 days ago
reacted to Passpass119's post with 🧠 9 days ago
Post
154
I am excited to announce that I have nothing to announce
reacted to Bc-AI's post with 👀 9 days ago
Post
216
# 🔥 Nova-1 Beta: Test Our New LLMs!
**Smilyai Labs** is building **Nova-1** — open-source LLMs with novel architectures. Join our beta program!
## 🎯 Available Now:
**Nova-1-Standard (1.2B)** — Phase 2 of pretraining in progress
- PPL 13.5 (beats GPT-2 Large!)
- 48K tok/s on consumer GPUs
- Great for code, reasoning, edge deployment
**Nova-1-Large (3.5B)** — Training live RIGHT NOW
- Current: 30.9 PPL, improving fast, loss at 3.5 right now
- Will finish with ~1.7B tokens today
- Better reasoning & longer context
**Nova-1-XL (10B MoE)** — Coming soon (We dont know yet! haha)
- Final Specs not decided yet
## What Makes Nova Special?
✨ **Mixture of Depths (MoD)** — Routes tokens dynamically, 30% faster
✨ **Grouped Query Attention** — Efficient like LLaMA 2/3
✨ **Phased Training** — Fresh 1B tokens each phase (no overfitting!)
✨ **RoPE** — Context extendable to 8K+
## 🤝 Join Beta Testing:
👉 **[Smilyai-labs-beta-testers](
Smilyai-labs-beta-testers
Get early access, shape the roadmap, and help build transparent open-source AI!
**Smilyai Labs** is building **Nova-1** — open-source LLMs with novel architectures. Join our beta program!
## 🎯 Available Now:
**Nova-1-Standard (1.2B)** — Phase 2 of pretraining in progress
- PPL 13.5 (beats GPT-2 Large!)
- 48K tok/s on consumer GPUs
- Great for code, reasoning, edge deployment
**Nova-1-Large (3.5B)** — Training live RIGHT NOW
- Current: 30.9 PPL, improving fast, loss at 3.5 right now
- Will finish with ~1.7B tokens today
- Better reasoning & longer context
**Nova-1-XL (10B MoE)** — Coming soon (We dont know yet! haha)
- Final Specs not decided yet
## What Makes Nova Special?
✨ **Mixture of Depths (MoD)** — Routes tokens dynamically, 30% faster
✨ **Grouped Query Attention** — Efficient like LLaMA 2/3
✨ **Phased Training** — Fresh 1B tokens each phase (no overfitting!)
✨ **RoPE** — Context extendable to 8K+
## 🤝 Join Beta Testing:
👉 **[Smilyai-labs-beta-testers](
Get early access, shape the roadmap, and help build transparent open-source AI!
- 1 reply
replied to their post 16 days ago
thanks @Hari5115
replied to their post 16 days ago
eh ?
replied to their post 16 days ago
reacted to dronefreak's post with 🔥 17 days ago
Post
3104
Excited to open-source the VisDrone Aerial Object Detection Model Zoo on Hugging Face.
The collection includes multiple YOLO variants trained and evaluated on the VisDrone benchmark for aerial object detection, with accompanying documentation and performance metrics.
If you're working on drones, aerial surveillance, robotics, or small-object detection, I hope these models save you some time.
Model Zoo: https://huggingface.co/collections/dronefreak/visdrone-detection-model-zoo
Feedback, issues, and contributions are welcome.
The collection includes multiple YOLO variants trained and evaluated on the VisDrone benchmark for aerial object detection, with accompanying documentation and performance metrics.
If you're working on drones, aerial surveillance, robotics, or small-object detection, I hope these models save you some time.
Model Zoo: https://huggingface.co/collections/dronefreak/visdrone-detection-model-zoo
Feedback, issues, and contributions are welcome.
- 13 replies
reacted to their post with 👍 17 days ago
Post
3754
Shadows of Tomorrow is finally live on Hugging Face Spaces with Gradio.
It’s a browser-playable RPG built with Godot, set in a post-nuclear future where players explore Magnus Province, collect medicinal plants, craft medicine, and help cure NPCs.
Play it here: Reubencf/Shadows_of_Tomorrow
It’s a browser-playable RPG built with Godot, set in a post-nuclear future where players explore Magnus Province, collect medicinal plants, craft medicine, and help cure NPCs.
Play it here: Reubencf/Shadows_of_Tomorrow
- 11 replies