Search Results for "trade"

...The framework integrates several architectural innovations—SwiGLU activation, deep and thin network design, embedding sharing, and grouped-query attention (GQA)—to achieve a superior trade-off between model size, inference speed, and accuracy. MobileLLM demonstrates remarkable performance, with the 125M and 350M variants outperforming previous state-of-the-art models of the same scale by up to 4.3% on zero-shot commonsense reasoning tasks.

R-KV is an open-source research project that focuses on improving the efficiency of large language model inference through key-value cache compression techniques. Modern transformer models rely heavily on KV caches during autoregressive decoding, which store intermediate attention states to accelerate generation. However, these caches can consume large amounts of memory, especially in reasoning-oriented models with long context windows. R-KV introduces a method for compressing the KV cache...

GLM-4.5 is a cutting-edge open-source large language model designed by Z.ai for intelligent agent applications. The flagship GLM-4.5 model has 355 billion total parameters with 32 billion active parameters, while the compact GLM-4.5-Air version offers 106 billion total parameters and 12 billion active parameters. Both models unify reasoning, coding, and intelligent agent capabilities, providing two modes: a thinking mode for complex reasoning and tool usage, and a non-thinking mode for...

...It mixes courses, articles, code labs, and videos, emphasizing materials that teach both concepts and hands-on implementation. The curation recognizes modern AI realities, including data pipelines, evaluation, prompt engineering, retrieval-augmented generation, and cost/performance trade-offs. It’s equally useful for refreshers—dipping into a specific module before a project—as it is for a full, self-directed curriculum. By centralizing the best references in one place, the repo reduces the overhead of finding, filtering, and sequencing resources, letting you focus on learning and building.

...The tool also provides a detailed breakdown of where GPU memory is allocated, including model weights, KV cache, activations, and other runtime overhead. This information allows developers to evaluate trade-offs between different quantization methods such as GGML, bitsandbytes, and QLoRA before attempting to deploy a model. gpu_poor is particularly useful for researchers and hobbyists.

...The model natively processes text, images, audio, and video within a unified architecture and supports an exceptionally large 1 million token context window for long-document reasoning, repository-scale coding, and agentic execution. Trained from scratch on approximately 45 trillion multimodal tokens, Inkling introduces controllable reasoning effort, allowing users to trade off latency and reasoning depth depending on the task. It is optimized for software engineering, tool use, and large-scale autonomous workflows, with strong performance on coding and agent benchmarks.