# MLX Articoli collegati

Il Centro Notizie HTX fornisce gli articoli più recenti e le analisi più approfondite su "MLX", coprendo tendenze di mercato, aggiornamenti sui progetti, sviluppi tecnologici e politiche normative nel settore crypto.

DeepSeek's New Technology Ported to Apple Silicon, Mac Local LLM Accelerated by 60%

DeepSeek's newly open-sourced DSpark inference acceleration technology has been ported to Apple Silicon, yielding significant speedups for running large language models locally on Macs. The port, called mlx-dspark, was developed by engineer Abdur Rahim and supports models like Gemma-4 12B and Qwen3-4B. DSpark uses speculative decoding, where a smaller "draft" model proposes candidate tokens which are then verified in a batch by the target model. Rahim adapted this approach for Apple's MLX framework, implementing 4-bit quantization for the draft model. On an M4 Pro Mac, this resulted in generation speeds increasing by approximately 1.6x for Gemma-4 12B (to ~30 tok/s) and 1.4x for Qwen3-4B (to ~73 tok/s). Crucially, the port maintains bitwise identical output to the original models, including support for temperature sampling, not just greedy decoding. The project also integrated DFlash, an alternative block-based speculative decoding method from z-lab. Benchmarks show DFlash excels in predictable contexts like code/math tasks (achieving ~2.1x speedup), while DSpark's Markov head provides better performance for open-ended chat. The latest mlx-dspark version allows users to switch between these methods. The work demonstrates efficient, high-fidelity local LLM inference on consumer Apple hardware.

marsbit07/03 12:22

DeepSeek's New Technology Ported to Apple Silicon, Mac Local LLM Accelerated by 60%

marsbit07/03 12:22

Running Gemma 4 Locally on iPhone Goes Viral: How Far Are We from the Zero Token Era?

Google's newly open-sourced Gemma 4 model, built on the same architecture as Gemini 3, has gained significant attention for its ability to run locally on mobile devices like the iPhone and Samsung Galaxy. With smaller versions such as E2B (2.3B parameters) and E4B (4.5B parameters), it supports native multimodal capabilities and offers a 128K context window. Users report impressive speeds—over 40 tokens per second on Apple chips with MLX optimization—making it feel "like magic." The model is accessible via Google’s official AI Edge Gallery app, ensuring ease of use and security. While Gemma 4 excels in tasks like text generation, coding, and image understanding, it struggles with more complex agent-based workflows, such as tool calling and structured outputs, where models like Qwen3-coder perform better. Despite some limitations in reasoning, Gemma 4’s local performance hints at a future where everyday AI tasks—chat, coding, reasoning—can be handled offline, reducing reliance on cloud-based token services. Although cloud models still lead in advanced reasoning and large-scale multi-agent tasks, the trend suggests that as hardware and quantization improve, on-device models will increasingly handle high-frequency simple tasks. This shift could disrupt the AI industry’s reliance on token sales and API subscriptions, pushing providers to focus on more complex, data-intensive capabilities. Gemma 4 is just the beginning of this transformation.

marsbit04/06 05:53

Running Gemma 4 Locally on iPhone Goes Viral: How Far Are We from the Zero Token Era?

marsbit04/06 05:53

活动图片

Bitcoinright-arrow

Regulatory Policyright-arrow