Hakim Hacid

25 posts

Hakim Hacid

Hakim Hacid

@hhacid

Senior Researcher and Principal Data Scientist

France Katılım Haziran 2013
90 Takip Edilen26 Takipçiler
Hakim Hacid retweetledi
NVIDIA AI Developer
NVIDIA AI Developer@NVIDIAAIDev·
Great to see @TIIuae push the boundaries of LLM training with their Falcon‑H1 parallel hybrid design and BitNet ternary training—now integrated into NVIDIA Megatron Core. Their work shows how foundation model builders can innovate on top of our scalable framework while improving efficiency and sharing reusable tools with the open source community. 🙌 👇 Read the full technical blog: developer.nvidia.com/blog/implement…
English
0
8
33
3.8K
Hakim Hacid retweetledi
NVIDIA Middle East
NVIDIA Middle East@NVIDIAME·
من الحوسبة الكمية إلى الذكاء الاصطناعي التوليدي، يواصل معهد الابتكار التكنولوجي (TII) إعادة تعريف الابتكار في الشرق الأوسط. ومن خلال دمج سحابة الحوسبة الكمية الخاصة به مع NVIDIA CUDA-Q، وتحسين بنية Falcon-H1 باستخدام Megatron Core، يؤكد المعهد أن الشرق الأوسط يقف في طليعة مواجهة أكثر التحديات الحاسوبية تعقيدًا في العالم. اقرأ المزيد. 🔗 nvda.ws/4mhmLhP From Quantum to Generative AI, @TIIuae is redefining regional innovation in the Middle East. By integrating its Quantum Computing Cloud with NVIDIA CUDA-Q and optimizing the Falcon-H1 architecture via Megatron Core, TII is proving that the Middle East is at the forefront of the world’s most complex computational challenges.
العربية
1
2
8
309
Hakim Hacid retweetledi
Phúc Lê
Phúc Lê@lkhphuc·
Plus an API server, MLX support. Please check us out here github.com/tiiuae/Falcon-… and share with us what you want. We are excited for what come next to Falcon Perception: better kvcache usage, quantization, finetuning code, integration to llama.cpp, vllm, ...
English
0
1
3
93
Hakim Hacid retweetledi
Phúc Lê
Phúc Lê@lkhphuc·
Falcon-Perception can now be install with `pip install falcon-perception` - 1 model file - 2 variants Perception + OCR - Paged / Batch +KVCache inference engine - Torch+compile + cudagraph - Upsampler w/ async cache for high-res mask Plus: MLX batch inference support 🧵:
Yasser Dahou@dahou_yasser

We are releasing Falcon Perception, an open-vocabulary referring expression segmentation model. Along with it, a 0.3B OCR model that is on par with 3-10x larger competitors. Current systems solve this with complex pipelines (separate encoders, late fusion, matching algorithms). We developed a novel simpler "bitter" approach: one early-fusion Transformer (image + text from first layer) with a shared parameter space, and let scale + training signal do the work. Please check our work ! 📄 Paper: arxiv.org/pdf/2603.27365 💻 Code: github.com/tiiuae/falcon-… 🎮 Playground: vision.falcon.aidrc.tii.ae 🤗 Blogpost: huggingface.co/blog/tiiuae/fa…

English
1
5
30
3.6K
Hakim Hacid retweetledi
Marktechpost AI Dev News ⚡
Marktechpost AI Dev News ⚡@Marktechpost·
The Technology Innovation Institute (TII) Releases Falcon Perception: A 0.6B-Parameter Early-Fusion Transformer for Open-Vocabulary Grounding and Segmentation from Natural Language Prompts By processing image patches and text tokens in a shared parameter space from the first layer, the model allows the prompt to influence visual feature formation throughout the entire stack. The Technical Shifts: - Hybrid Attention Masking: Image tokens attend bidirectionally for global context, while task tokens attend causally to the full visual prefix. - Chain-of-Perception: Instead of parallel mask queries, the model predicts objects as an autoregressive sequence: -> -> . This resolves spatial ambiguity before pixel-level refinement. - GGROPE (Golden Gate ROPE): To preserve 2D grid relationships in flattened sequences, the model uses an isotropic attention mechanism robust to rotation and aspect ratio variations. - Muon Optimization: Specialized heads (coordinate, size, segmentation) often lag behind the backbone during training; the authors report that using the Muon optimizer specifically for these heads reduced training losses. Key Empirical Results: -- Spatial Understanding: On the new PBench (Level 3), Falcon Perception achieved a +21.9 point gain in Macro-F1 over SAM 3. -- Dense Environments: The model remains stable in crowded scenes, scaling up to 600 instances per expression via an autoregressive interface. -- OCR Efficiency: A 300M-parameter variant, FalconOCR, achieves 80.3% accuracy on olmOCR, matching or exceeding several systems an order of magnitude larger Full analysis: marktechpost.com/2026/04/03/tii… Paper: arxiv.org/pdf/2603.27365 Model Weight: huggingface.co/tiiuae/Falcon-… Repo: github.com/tiiuae/falcon-… Technical details: huggingface.co/blog/tiiuae/fa… @TIIuae @dahou_yasser @lkhphuc @ChaySofian @LiAvBev @andyhuynh1111 @NarayanSanath @griffintaur wamiqpara @hhacid
English
1
14
62
44.9K
Hakim Hacid retweetledi
younes
younes@yb2698·
Announcing - Falcon-Edge – a series of powerful, universal and fine-tunable Bitnet models for everyone! We also release a Python fine-tuning toolkit library - `onebitllms`- specialized for Bitnet models. Announcement blogpost: falcon-lm.github.io/blog/falcon-ed…
younes tweet media
English
9
63
223
29.1K
Hakim Hacid retweetledi
Technology Innovation Institute
Technology Innovation Institute@TIIuae·
Introducing Falcon Mamba 7B from TII! This state-of-the-art language model sets a new AI benchmark with its innovative state space architecture. This groundbreaking release marks a significant stride in AI research. Learn more on falconllm.tii.ae.
English
4
27
99
29.8K

Keşfet

@TIIuae @dahou_yasser @lkhphuc @ChaySofian @LiAvBev @andyhuynh1111 @NarayanSanath @griffintaur