从语言到影像，从开源到 API。读懂模型能力、比较真实成本，
为你的下一个想法，找到合适的智能引擎。
LFM2.5 is a new family of hybrid models designed for **on-device deployment**. It builds on the LFM2 architecture with extended pre-training and reinforcement learning. Find more details in the original model card: https://huggingface.co/LiquidAI/LFM2.5-2.6B The Quantization-Aware Distillation (QAD) checkpoint is available as `LFM2.5-2.6B-QAD-Q4_0.gguf`.
译文排队中 · 当前显示来源原文
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
LFM2.5-2.6B is a compact multilingual model built on LiquidAI's hybrid architecture, combining convolutional and attention layers for efficient long-context processing.
译文排队中 · 当前显示来源原文
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
LFM2.5 is a family of hybrid models designed for **on-device deployment**. It builds on the LFM2 architecture with extended pre-training and reinforcement learning. - **Our most compact model yet**: 230M parameters that punch above their weight, bringing real capability to the tightest memory and compute budgets. - **Fast edge inference**: Best throughput from low-cost CPUs to production GPUs, running at 213 tok/s decode speed on Galaxy S25 Ultra and 42 tok/s on a Raspberry Pi 5. - **Built for agentic tasks**: Distilled from LFM2.5-350M and refined with multi-stage reinforcement learning, making it well-suited for tool use and data extraction. Find more information about LFM2.5-230M in our blog post.
译文排队中 · 当前显示来源原文
LFM2.5 is a new family of hybrid models designed for **on-device deployment**. It builds on the LFM2 architecture with extended pre-training and reinforcement learning. Find more information about LFM2.5-230M in our blog post. - **Number of parameters**: 230M - **Number of layers**: 14 (8 double-gated LIV convolution blocks + 6 GQA blocks) - **Training budget**: 28T tokens - **Context length**: 32,768 tokens - **Vocabulary size**: 65,536 - **Knowledge cutoff**: Mid-2024 - **Languages**: English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, Spanish
译文排队中 · 当前显示来源原文
LFM2 is a new generation of hybrid models developed by Liquid AI, specifically designed for edge AI and on-device deployment. It sets a new standard in terms of quality, speed, and memory efficiency. Find more details in the original model card: https://huggingface.co/LiquidAI/LFM2.5-230M The Quantization-Aware Distillation (QAD) checkpoint is available as `LFM2.5-230M-QAD-Q4_0.gguf`.
译文排队中 · 当前显示来源原文
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基础资料已收录 · 缺失资料后续补充
FP8 version of LFM2.5-24B-A2B for GPU inference. Linear weights are quantized to FP8 with 128×128 blocks and dynamic activation quantization. The embedding, MoE routing gates and convolution weights are kept in their original precision. Find more details in the original model card: https://huggingface.co/LiquidAI/LFM2.5-24B-A2B
译文排队中 · 当前显示来源原文
LFM2 is a new generation of hybrid models developed by Liquid AI, specifically designed for edge AI and on-device deployment. It sets a new standard in terms of quality, speed, and memory efficiency. Find more details in the original model card: https://huggingface.co/LiquidAI/LFM2.5-24B-A2B The sampling parameters are the ones recommended for the base model (its `generation_config.json`); each GGUF also carries them as `general.sampling.*` metadata, so llama.cpp applies them by default.
译文排队中 · 当前显示来源原文
This model LiquidAI/LFM2.5-24B-A2B-MLX-4bit was converted to MLX format from LiquidAI/LFM2.5-24B-A2B using mlx-lm version **0.32.0**. Quantized to 4 bits per weight (group size 64, affine mode). Token embedding (tied to the output head) kept at 6 bits. MoE routing gates kept unquantized in float32. For base model details, capabilities, and recommended generation settings, see the original model card.
译文排队中 · 当前显示来源原文
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
可以调整厂商、分类或输入输出条件；新模型正持续收录。
调整月度 Token 用量，直观比较不同模型的预估 API 开支。
使用已采集且单位可比的 USD / 百万 Token 渠道报价。报价为采集时快照；不计缓存优惠、阶梯价、推理额外输出与税费，请以来源的当前报价为准。
先区分输入、输出和缓存价格，再使用同一任务和用量估算总费用。图像、音频和视频可能按张、时长或分辨率计费，不能直接与 Token 单价比较。
不一定。必须逐一核对具体模型版本的许可证、商用限制及使用政策。自己部署还需要考虑显存、推理资源和运维费用。
榜单取决于测试任务、版本与评测方法。建议用实际业务样本比较准确性、时延、稳定性和成本，再选择模型。
