TrendPulse

Just came across this coding benchmark: SciCode Artificialanalysis.ai reports a ranking which contradicts the feeling we've towards those models in real life coding. Is Gemma 4 really that good, or a benchmarking issue? EDIT: The contribution of this benchmark to the Intelligence index of artificialanalysis.ai: Full Intelligence Index v4.1 weights: GDPval-AA v2: 20% Terminal-Bench 2.1: 16% τ³-Bench Banking: 14% Humanity's Last Exam: 12% AA-Omniscience Accuracy: 8% SciCode: 8% GPQA: 6% ...

Press My earlier prediction that Tesla would buy them completely missed the mark. With AMD focusing heavily on the enterprise side, the idea of consumer-facing hot-swappable AI model chips looks pretty much dead. Fast forward ten years, you might find used model blade cards on eBay, except a full model's weights will be split across them, so it'll take multiple blades chained together just to make up a single complete set of weights. ...

I swear AA is not the bipartisan they so claim. An open source mode (Qwen 3.8 max) was number 1 on the agentic index, then they just so happen to launch "v4.1.1" of their index in which they just adjusted the weights of the gdpval and t3 banking so that it would be lower than opus, despite the lead in t3 being a 8% lead over opus while opus only has a 5% lead on gdpval. Highly likely to be paid off imo. You can check other subreddits for the score before and after the change, just made...

NVIDIA Nemotron Parse 2.0 transforms document images into structured, machine-readable representations with text, layout classes, bounding boxes, and reading-order information. Given a Red, Green, Blue (RGB) document image and a task prompt, the model produces formatted text and spatial annotations for document elements such as titles, paragraphs, captions, tables, charts, page headers, page footers, footnotes, pictures, and bibliography entries. Compared with NVIDIA Nemotron Parse v1.2, NVIDIA ...

Link to the article: KV Cache Quantization Benchmarks: KVarN, Precision Tail KLD benchmarks with BeeLlama.cpp v0.4.0 , fork of llama.cpp with more KV cache quantization options. Models: Qwen 3.6 27B Q5_K_S 64k context, Gemma 4 31B Q5_K_S 16k context Standard quants, extended: q6_0 and q6_1, and low-bit types from q2_0 to q3_1 KVarN: Variance-Normalized KV-Cache by Huawei, implemented in BeeLlama Precision Tail: keeping latest X tokens of KV cache in (B)F16, implemented in BeeLlam...

Reading through this subreddit and many more I keep running into what I am calling "AI click bait". Either projects that seems interesting in the description/title but when you open them they're the same AI vide coded slop that does not solve the problem; or apparent discussions about an actual problem that are just undercover marketing ploys to sell you a product that also does not solve the problem. I guess I don't really have a point to this, I just encountered the 100th post o...

GGUFs here: https://huggingface.co/ReadyArt/gemma-4-31B-it-scotoma-2-GGUF Disclaimer: By slop, we are specifically talking about specific tics with the model(sentence structures), but this doesn't include words such as "ozone". Summery of the model: Scotoma-2 is a model made by user https://huggingface.co/AesSedai which aims to reduce common Gemma4 tropes, such as(not limited to): "It's not x, it's y" and "a slow knowing smirk "(stacked adjectives) which t...

🐦‍⬛ Magpie-TTS Multilingual 🦜 Nemotron Speech Streaming EN 0.6B 🦜 Nemotron-3.5 ASR Streaming 🦜 Parakeet CTC 1.1B 🦜 Parakeet TDT 0.6B v3 🥦 NanoCodec Merged PR https://huggingface.co/nvidia/magpie_tts_multilingual_357m#run-magpietts-locally-with-nemo-speechcpp I run open source llm models via AI Desktop XP but how am I supposed to run these models on my phone? Any recc?

In a previous post , I compared MinerU, Granite-Docling, and PaddleOCR-VL. Many commentors suggested I added their favorite parsers. So I did. And also added some new capabilities to differentiate the top models. Here is the full list of parser compared: MinerU 2.5 (1.2B VLM) Granite-Docling (258M VLM) PaddleOCR-VL (0.9B VLM) XBerg 1.0 (text-layer parser, CPU) HURIDOCS PDLA v0.0.35 (VGT layout model + Tesseract) LiteParse 2.11 (Tesseract based, CPU) Chandra (Datalab's OCR model)...

I'm the author, so discount the enthusiasm accordingly. This is an unaffiliated community port, not endorsed by the vLLM project, which it uses to verify its correctness. What started it: I love vLLM, but a vLLM install here is 9.1 GiB of virtualenv, and I wanted to embed inference inside other software, on machines where having an interpreter in the process is a problem. And, honestly, Python dependencies have a different deployment story, in term of security (supply chain attacks), and bloat...

google facing yet another "innovator's dilemma" or some shi smh lgtb ig EDIT - WHEN THE FUCK WILL LE CHATON FAT COME RIGHT ON !!????

This is very important for us when considering local hosting. A lot of people decided not to buy expensive hardware because DeepSeek’s prices made it very difficult to break even given that deepseek was soo cheap. Also some of us use DeepSeek in routing, hosting Qwen and routing some hard tasks to DeepSeek API. what do you think about this? do you think raising prices will ultimately lead to another increase in NVIDIA’s GPU prices, since more and more people will now buy their own hardware...

Continuing my weekend of oneshotting the cheap OpenRouter models, here are all 10 DeepSeek models across the same 35 prompts. DeepSeek had a rougher time (more provider errors / empty completions), so only 242 made it out of the 10*35 matrix. Here they are https://oneshotlm.com/model/?q=deepseek DeepSeek V4: deepseek-v4-pro , deepseek-v4-flash (+ 0731 ) DeepSeek V3.x: deepseek-v3.2 (+ exp ), deepseek-v3.1-terminus , deepseek-chat-v3.1 , deepseek-chat DeepSeek R1: deepse...

https://preview.redd.it/3zcvpbds14hh1.png?width=1911&format=png&auto=webp&s=a79aafb71eeca97638da93d2591902631e897fd5 I tried a test similar to the recent model-quant comparisons, but this time I focused only on: DeepSeek-V4-Flash-0731-IQ2_XS-Experts-Q8_0 model link bullerwins/DeepSeek-V4-Flash-0731-GGUF · Hugging Face Hardware RTX 3090 24 GB 128 GB DDR4 RAM Windows llama.cpp / llama-server Prompt Write a single HTML file with a full-page canvas and no ...

Hey there! So today we're releasing SupraBrain-50M, a hybrid language model that combines Gated DeltaNet linear recurrence with Sliding-Window Attention and Surprise-Gated update mechanisms to deliver very strong performance. Here are the benchmarks: https://preview.redd.it/tv0bxkbyh4hh1.png?width=615&format=png&auto=webp&s=f296ed9e6c1d94738cb7aa84015117dd3e65d2a5 Despite being trained on MUCH less data (5B vs 20B tokens!!), it's almost as good as Supra-Base-50M! 🔥 Some s...

i already had an esp32 generating stories, but generating text is not the same as receiving a question and giving a useful answer. barista v0.1, a small model trained for espresso troubleshooting and running on an esp32s3 n16r8 witout cloud. you type a question(over usb for now), esp32 streams the answer to the oled (or back to terminal). how it works: > per-layer embeddings. most of the parameters in this model are in large ple and token-embedding tables. those tables stay memory-m...

Originally, I was only getting around 140pp/s and about 21tg/s, but the config with -b 8192 -ub 8192 --cpu-moe is vastly superior, let's say 700pp/s and 18tg/s in the most relevant range. Test System : CPU: Threadripper 5965WX (24c/48t) RAM: 512GB ( 8 x64 DDR4 ECC REG 2400 ) GPU: RTX 5090 (PCIe 4 x16, 450W) Command ./build/bin/llama-server \ -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q8_K_XL \ --temp 1.0 --top-p 1.0 --min-p 0.0 --host 0.0.0.0 \ --ctx-size 262144 --no-...

I tested them by sending the 6 documents, each meant to represent a different document type, through my own webapp and comparing every output against the source. All ran on the same L4 GPU. The documents: Financial statements with merged multi-level headers (A typical annual report) Two pages of a two-column arXiv paper ("Deep Residual Learning for Image Recognition") Scanned German invoice with no text layer French municipal report with an embedded bar chart Typical data...

AI9Stars has released G9v3-39A5B an open weights language model designed to deliver even stronger reasoning capabilities than ai9stars/G9v3-3B with its 39B and 5 active experts. It is released under the Apache 2.0 license making it fully open for personal and commercial use It targets everyday assistant use, coding, tool-use workflows, and reasoning tasks, and supports both Think / No Think GITHUB: https://github.com/AI9Stars HF MODEL CARD: https://huggingface.co/ai9stars/G9v3-39A5B A...

Hugging Face Artificial Analysis Should be a sweet spot for general work. Seems like coding is the only part that is inferior to Qwen.

Has anyone followed nousresearch work on Hermes? I mean we are Q3 2026. We have some crazy models trickling down from HGX territory to multi gpu workstation. And we have nousresearch deploying the 0.20 of its hermes agent while starting releasing the project with a 0.2 mid march! Crazy times to be alive. For the old timers who remember llama 1 or llama 2, remember our crappy function caller parser? Something about a lang and a chain..? wtf has happened?! Haven't tried the new hermes, do yo...

Now we can run Qwen3-Next at “full speed” :) Do you still remember this model?

It is so good! I don't know why there aren't more people talking about it. Fewer tokens, faster and more accurate than Qwen 3.6 35b a3b. On my setup it's nearly as good as 27b, but 5x faster. And it completely trashes the Gemma 4 models. At least for my use case, it feels amazing. I'd love to hear other people's experience with it. If you want an actual measure of performance, I have a GitHub repo explaining how I tested it for my type of use case with a detailed performance comparison with o...

Spent way too much time with V4-Flash-0731 this weekend and wanted to share my vibes as briefly as possible. I sent it through a bit of real-work and some of my personal benchmarks. My quick thoughts are: Quantization hits this thing like a truck - I've tried a bunch of the Q2 and Q3 weights and it behaves like an entirely different model. Reasoning looks/feels different and the results are a full tier down from the official/served V4-Flash-0731. Did not get much time with Q4. Q3 c...

I've been out of these forums for awhile but I figured I would provide a formal update on how this has been going now that it has some operation time under its belt, just to put the information out there and share knowledge if there is any interest. I also wasn't satisfied with the quality of my original post so I will probably remove it and let this one serve as its replacement. I am an IT infrastructure engineer by profession, so my contribution to the conversation is mainly from a hardware/sy...