Within the present panorama of generative AI, the ‘scaling laws’ have typically dictated that extra parameters equal extra intelligence. Nevertheless, Liquid AI is difficult this conference with the discharge of LFM2.5-350M. This mannequin is definitely a technical case examine in intelligence density with extra pre-training (from 10T to 28T tokens) and large-scale reinforcement studying
The importance of LFM2.5-350M lies in its structure and coaching effectivity. Whereas probably the most AI firms has been targeted on frontier fashions, Liquid AI is concentrating on the ‘edge’—gadgets with restricted reminiscence and compute—by proving {that a} 350-million parameter mannequin can outperform fashions greater than twice its measurement on a number of evaluated benchmarks.
Structure: The Hybrid LIV Spine
The core technical differentiator of the LFM2.5-350M is its departure from the pure Transformer structure. It makes use of a hybrid construction constructed on Linear Enter-Various Programs (LIVs).
Conventional Transformers rely fully on self-attention mechanisms, which undergo from quadratic scaling points: because the context window grows, the reminiscence and computational necessities for the Key-Worth (KV) cache enhance. Liquid AI addresses this through the use of a hybrid spine consisting of:
- 10 Double-Gated LIV Convolution Blocks: These deal with the vast majority of the sequence processing. LIVs operate equally to superior Recurrent Neural Networks (RNNs) however are designed to be extra parallelizable and steady throughout coaching. They keep a constant-state reminiscence, decreasing the I/O overhead.
- 6 Grouped Question Consideration (GQA) Blocks: By integrating a small variety of consideration blocks, the mannequin retains high-precision retrieval and long-range context dealing with with out the total reminiscence overhead of an ordinary Transformer.
This hybrid strategy permits the LFM2.5-350M to assist a 32k context window (32,768 tokens) whereas sustaining an especially lean reminiscence footprint.
Efficiency and Intelligence Density
The LFM2.5-350M was pre-trained on 28 trillion tokens with an especially excessive training-to-parameter ratio. This ensures that the mannequin’s restricted parameter depend is utilized to its most potential, leading to excessive ‘intelligence density.’
Benchmarks and Use Instances
The LFM2.5-350M is a specialist mannequin designed for high-speed, agentic duties relatively than general-purpose reasoning.
| Benchmark | Rating |
| IFEval (Instruction Following) | 76.96 |
| GPQA Diamond | 30.64 |
| MMLU-Professional | 20.01 |
The excessive IFEval rating signifies the mannequin is environment friendly at following advanced, structured directions, making it appropriate for device use, operate calling, and structured knowledge extraction (e.g., JSON). Nevertheless, the documentation explicitly states that LFM2.5-350M shouldn’t be really helpful for arithmetic, advanced coding, or inventive writing. For these duties, the reasoning capabilities of bigger parameter counts stay vital.
{Hardware} Optimization and Inference Effectivity
A significant hurdle for AI devs is the ‘memory wall’—the bottleneck created by shifting knowledge between the processor and reminiscence. As a result of the LFM2.5-350M makes use of LIVs and GQA, it drastically reduces KV cache measurement, boosting throughput. On a single NVIDIA H100 GPU, the mannequin can attain a throughput of 40.4K output tokens per second at excessive concurrency.
Liquid AI crew reviews device-specific low-memory inference outcomes that make native deployment viable:
- Snapdragon 8 Elite NPU: 169MB peak reminiscence utilizing RunAnywhere This fall.
- Snapdragon GPU: 81MB peak reminiscence utilizing RunAnywhere This fall.
- Raspberry Pi 5: 300MB utilizing Cactus Engine int8.
Key Takeaways
- Excessive Intelligence Density: By coaching a 350M parameter mannequin on 28 trillion tokens, Liquid AI crew achieved an tremendous excessive 80,000:1 token-to-parameter ratio, permitting it to outperform fashions greater than twice its measurement on a number of benchmarks.
- Hybrid LIV Structure: The mannequin departs from pure Transformers through the use of Linear Enter-Various Programs (LIVs) mixed with a small variety of Grouped Question Consideration (GQA) blocks, considerably decreasing the reminiscence overhead of the KV cache.
- Edge-First Effectivity: It’s designed for native deployment with a 32k context window and a remarkably low reminiscence footprint—reaching as little as 81MB on cellular GPUs and 169MB on NPUs by way of specialised inference engines.
- Specialised Agentic Functionality: The mannequin is extremely optimized for instruction following (IFEval: 76.96) and gear use, although it’s explicitly not really helpful for advanced coding, arithmetic, or inventive writing.
- Huge Throughput: The architectural effectivity permits high-speed utility, processing as much as 40.4K output tokens per second on a single H100, making it preferrred for high-volume knowledge extraction and real-time classification.
Take a look at the Technical details and Model Weight. Additionally, be at liberty to comply with us on Twitter and don’t neglect to affix our 120k+ ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
