Search Results for "utf-8"

GLM-4.6 is the latest iteration of Zhipu AI’s foundation model, delivering significant advancements over GLM-4.5. It introduces an extended 200K token context window, enabling more sophisticated long-context reasoning and agentic workflows. The model achieves superior coding performance, excelling in benchmarks and practical coding assistants such as Claude Code, Cline, Roo Code, and Kilo Code. Its reasoning capabilities have been strengthened, including improved tool usage during inference...

...Trained on over 400 billion tokens (200B English, 200B Chinese), it achieves performance surpassing GPT-3 175B, OPT-175B, and BLOOM-176B on multiple benchmarks, while also showing significant improvements on Chinese datasets compared to other large models. The model supports efficient inference via INT8 and INT4 quantization, reducing hardware requirements from 8× A100 GPUs to as little as a single server with 4× RTX 3090s. Built on the SwissArmyTransformer (SAT) framework and compatible with DeepSpeed and FasterTransformer, it supports high-speed inference (up to 2.5× faster) and reproducible evaluation across 30+ benchmark tasks.

NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 is a state-of-the-art large language model developed and released by NVIDIA as part of its Nemotron 3 family, optimized for high-efficiency inference and strong reasoning performance in open AI workloads. It is the post-trained and FP8-quantized variant of the Nemotron 3 Nano model, meaning its weights and activations are represented in 8-bit floating point (FP8) to dramatically reduce memory usage and computational cost while retaining high accuracy. The base Nano architecture uses a hybrid Mamba-Transformer Mixture-of-Experts (MoE) design, allowing the model to activate only a small fraction of its 31.6 billion parameters per token, which improves speed and efficiency without sacrificing quality on complex queries. ...

...To use it, users must download over 500 GB of files and set them up locally with the SGLang inference engine. Grok-2.5 supports advanced inference with multi-GPU configurations, requiring at least 8 GPUs with more than 40 GB of memory each for optimal performance. It integrates with the SGLang framework to enable serving, testing, and chat-style interactions. The model comes with a post-training architecture and requires the correct chat template to function properly. It is released under the Grok 2 Community License Agreement, encouraging community experimentation and responsible use.

...It is part of the Llama 3.2 family, trained on up to 9 trillion tokens and aligned using supervised fine-tuning, preference optimization, and safety tuning. The model supports eight officially listed languages (including Spanish, German, Hindi, and Thai) but can be adapted to more. Llama 3.2-1B outperforms other open models in several benchmarks relative to its size and offers quantized versions for efficiency. It uses a refined transformer architecture with Grouped-Query Attention (GQA) and supports long context windows of up to 128k tokens.

...It builds upon the Llama 3.1 architecture and incorporates fine-tuning techniques like SFT, DPO, and quantization-aware training for improved alignment, efficiency, and safety. The model supports eight primary languages (including English, Spanish, Hindi, and Thai) and was trained on a curated mix of publicly available online data, with a December 2023 knowledge cutoff. Llama-3.2-1B is lightweight enough for deployment on constrained devices like smartphones, using formats like SpinQuant and QLoRA to reduce model size and latency. ...

...This reasoning-tuned variant is optimized for tasks like math, coding, and other STEM-related problem solving, making it suitable for applications that require logical reasoning, analysis, or structured thinking. Despite its modest size, the model is designed for edge deployment and can run locally, fitting in ~16 GB of VRAM in BF16 or under 8 GB of RAM/VRAM when quantized. It supports dozens of languages, allowing it to function across global and multilingual contexts. The model retains strong system-prompt adherence, supports function-calling with structured JSON output, and offers a large 256k token context window for extended context reasoning.