hf_posts.json

{"version": "https://jsonfeed.org/version/1", "title": "Hugging Face Posts", "home_page_url": "https://huggingface.co/", "feed_url": "https://raw.githubusercontent.com/MichaelMarkert/rss/refs/heads/main/hf_posts.json", "items": [{"id": "https://huggingface.co/posts/SeaWolf-AI/967860141797932", "image": "", "title": "POCKET now speaks Gemma 4 \u2014 a 26B model that loads in every app, and runs on your PC with no GPU", "content_text": "POCKET now speaks Gemma 4 \u2014 a 26B model that loads in every app, and runs on your PC with no GPU We're adding a Gemma-4 sibling to POCKET: POCKET-26B, built from Google's Gemma-4-26B-A4B (Apache-2.0). Our flagship POCKET-35B is a Qwen-family MoE and needs a recent llama.cpp; POCKET-26B trades a little size for the thing people kept asking for \u2014 it just loads, everywhere, today: Ollama, LM Studio, PocketPal, MLX, any stock llama.cpp. No fork, no bleeding-edge runtime, no CUDA, no cloud. It's a sparse Mixture-of-Experts (25.2B total, ~4B active per token), so the work per token stays small \u2014 a real 26B that generates on a CPU with no graphics card. Two things make it stand out: 1) Universal compatibility. Gemma 4 is a standard, widely-supported architecture, so POCKET-26B runs on the tools you already have \u2014 no waiting for your app to add a new model type. 2) Quality that survives compression. Measured GPQA-Diamond (198 q, greedy): \u2022 Full base: 67.7% \u2022 POCKET-26B Q4_K_M (17 GB): 67.7%...", "url": "https://huggingface.co/posts/SeaWolf-AI/967860141797932", "date_published": "2026-07-25T17:59:52.004362"}, {"id": "https://huggingface.co/posts/badaoui/772793288622124", "image": "", "title": "432 GB of ultra-fast HBM4 and up to 23.3 TB/s of memory bandwidth on a single GPU \ud83e\udd2f.", "content_text": "432 GB of ultra-fast HBM4 and up to 23.3 TB/s of memory bandwidth on a single GPU \ud83e\udd2f. Two weeks ago, we got early access to AMD's new Instinct MI455X, and our first goal was simple: make sure \ud83e\udd17 Transformers works on day one. Over the past few weeks, we worked closely with the AMD team to validate the platform, enable Flash Attention, add torchcodec support for multimodal models, and resolve issues uncovered during testing. The result: \u2705 99.5% success rate across our 24 core Transformers model architectures - already on par with our daily CI on previous AMD and NVIDIA platforms. The hardware is just as exciting. With 432 GB of HBM per GPU, our early capacity experiments showed more than 3\u00d7 the concurrent long-context requests compared to MI300, thanks to the much larger KV cache capacity. A huge thanks to the AMD team for the early access and the great collaboration! Read the full blog \ud83d\udc47 https://huggingface.co/blog/badaoui/transformers-on-amd-mi455 See translation", "url": "https://huggingface.co/posts/badaoui/772793288622124", "date_published": "2026-07-25T17:59:52.004871"}, {"id": "https://huggingface.co/posts/Banaxi-Tech/396365653534641", "image": "", "title": "We're excited to release BananaMind Base Bench 1.1 A new benchmark for base language models with 350 text-completion examples across seven categories. Models are scored using continuation likelihood and receive an Overall Elo score.", "content_text": "We're excited to release BananaMind Base Bench 1.1 A new benchmark for base language models with 350 text-completion examples across seven categories. Models are scored using continuation likelihood and receive an Overall Elo score. Initial results: BananaMind-2-Medium: 1034 BananaMind-2-Mini: 974 Supra-50M-Base: 973 Supra-1.5-50M-Base-exp: 948 BananaMind-2-Nano: 910 The official script downloads the gated dataset directly from Hugging Face. The dataset is for benchmarking only and may not be used for model training. BananaMind/BananaMind-Base-Bench-1.1 See translation", "url": "https://huggingface.co/posts/Banaxi-Tech/396365653534641", "date_published": "2026-07-25T17:59:52.005176"}, {"id": "https://huggingface.co/posts/SeaWolf-AI/745666844415915", "image": "", "title": "\ud83d\udcf1 POCKET \u2014 a 35-billion-parameter model that runs on your iPhone, and on your PC with no GPU", "content_text": "\ud83d\udcf1 POCKET \u2014 a 35-billion-parameter model that runs on your iPhone, and on your PC with no GPU We're releasing POCKET, VIDRAFT's flagship Darwin-36B-Opus compressed for on-device use. No fork, no CUDA, no cloud \u2014 it runs on stock llama.cpp. It's a sparse Mixture-of-Experts model (256 experts, only 8 active per token), so the file can be large while the work per token stays small. That's what lets a 35B model run on a phone, and generate fast on a CPU with no graphics card. Measured (POCKET-35B IQ1_M vs Bonsai-27B Q1_0): \u2022 CPU generate (Xeon, 16 threads): 27.0 vs 10.1 tok/s \u2192 2.69\u00d7 faster \u2022 GPU generate (H100): 197 vs 89 tok/s \u2192 2.22\u00d7 faster \u2022 GPU prompt processing (H100): 753 vs 1816 \u2192 0.41\u00d7 (Bonsai wins this one \u2014 MoE prefill wakes every expert, so sparsity stops helping there. We say so.) \u2022 Quality (HellaSwag, 400 q): 61.0% vs 60.0% \u2192 a tie (confidence intervals overlap) On a real consumer laptop \u2014 MacBook M3 Pro (18 GB) \u2014 POCKET wins every axis, prompt processing included: \u2022 Metal...", "url": "https://huggingface.co/posts/SeaWolf-AI/745666844415915", "date_published": "2026-07-25T17:59:52.005859"}, {"id": "https://huggingface.co/posts/mike-ravkine/129031493732543", "image": "", "title": "I went into this expecting to find a ~Q2 garbage dumpster but", "content_text": "I went into this expecting to find a ~Q2 garbage dumpster but prism-ml/Ternary-Bonsai-27B-gguf is a slick feat of QAT engineering. It is weaker then FP16 on a handful of tasks where quants usually degrade, as per their own paper the loss is \"concentrated on sustained chains of reasoning / agentic\" and in ReasonScape this bites on Sort, Shuffle and Dates, but counter-acting this are some noticeable improvements to thinking length without accuracy loss on several other tasks (Shapes, Cars). I haven't had a chance to run the Binary yet, but PQ2 + Bonsai QAT are confirmed to be pretty darn impressive. See translation", "url": "https://huggingface.co/posts/mike-ravkine/129031493732543", "date_published": "2026-07-25T17:59:52.006169"}, {"id": "https://huggingface.co/posts/sergiopaniego/834036283167008", "image": "", "title": "you can train DiffusionGemma (a block-diffusion LLM) in TRL! and we're sharing an example for it", "content_text": "you can train DiffusionGemma (a block-diffusion LLM) in TRL! and we're sharing an example for it TRL trainers are made to be easily extended and adapted to different real-world use cases. in this one, with a single method overridden in SFTTrainer (compute_loss), you can train this model > example: https://github.com/huggingface/trl/blob/main/examples/scripts/sft_diffusion_gemma.py See translation", "url": "https://huggingface.co/posts/sergiopaniego/834036283167008", "date_published": "2026-07-25T17:59:52.006438"}, {"id": "https://huggingface.co/posts/Banaxi-Tech/481856361260152", "image": "", "title": "We're excited to announce BananaMind 2V, our small vision model series!", "content_text": "We're excited to announce BananaMind 2V, our small vision model series! These models are NOT released yet. We will release them in mid-august! BananaMind 2V will include: BananaMind 2V 256M, the flagship based on BananaMind 2 Pro (BananaMind 2 Pro is not released yet). BananaMind 2V 100M, our mid model, based on BananaMind 2 Medium. BananaMind 2V 50M, our smallest vision model, based on BananaMind 2 Mini. These are currently unreleased and will release in mid-august. Our training will start after BananaMind 2 Pro has finished training. See translation", "url": "https://huggingface.co/posts/Banaxi-Tech/481856361260152", "date_published": "2026-07-25T17:59:52.006732"}, {"id": "https://huggingface.co/posts/appvoid/552710277862984", "image": "", "title": "Two big projects are open sourced soon. Get ready.", "content_text": "Two big projects are open sourced soon. Get ready. See translation", "url": "https://huggingface.co/posts/appvoid/552710277862984", "date_published": "2026-07-25T17:59:52.006930"}, {"id": "https://huggingface.co/posts/Banaxi-Tech/804920534169156", "image": "", "title": "We're excited to release BananaMindBench Leaderboard, our leaderboard for BananaMind Base Bench 1.1.", "content_text": "We're excited to release BananaMindBench Leaderboard, our leaderboard for BananaMind Base Bench 1.1. It measures model performance on a variety of different tasks: Language Completion Common sense too World Knowledge Context Tracking Quantitative Logical Reasoning Code Completion Each has a different score and 1 overall score. Submit your own model: BananaMind/BananaMindBench-Leaderboard Check it out: BananaMind/BananaMindBench-Leaderboard See translation", "url": "https://huggingface.co/posts/Banaxi-Tech/804920534169156", "date_published": "2026-07-25T17:59:52.007199"}, {"id": "https://huggingface.co/posts/Jiaqi-hkust/846867711557418", "image": "", "title": "IQA-T1: Evidence\u2011Based Image Quality Assessment with MLLMs", "content_text": "IQA-T1: Evidence\u2011Based Image Quality Assessment with MLLMs Most MLLMs are blind to low\u2011level degradations\u2014noise, blur, compression artifacts look the same as clean images in their internal representations. That leads to quality scores based on semantic \u201cgut feeling\u201d rather than real perceptual evidence. IQA-T1 changes that. We equip the model with a toolbox of 15 perceptual tools (noise residual maps, Fourier spectra, gradient maps, etc.) that generate structured visual evidence on demand. The model learns how to use tools via supervised fine\u2011tuning on our Q\u2011Tool dataset (11k evidence\u2011grounded reasoning chains), and when to call them via GRPO reinforcement learning that balances accuracy, tool count, and redundancy. The result: SOTA performance across 7 benchmarks (avg PLCC 0.795), using only 2.34 tools per image on average. Every predicted score is now interpretable and backed by hard visual evidence. All code, weights, dataset, and demo are open. Check them out and give it a spin!...", "url": "https://huggingface.co/posts/Jiaqi-hkust/846867711557418", "date_published": "2026-07-25T17:59:52.007683"}]}