LFM2
PulseAugur coverage of LFM2 — every cluster mentioning LFM2 across labs, papers, and developer communities, ranked by signal.
2 day(s) with sentiment data
-
Hugging Face releases LFM3 multimodal models for specialized AI
Hugging Face has released LFM3, a new family of Liquid Foundation Models designed for specialized intelligence across multiple modalities. These models, available in the 2B parameter range, offer improved performance pe…
-
Liquid Ai unveils LFM3 multimodal models for specialized intelligence · 4 sources tracked
Liquid Foundation Models has released LFM3, a new family of multimodal models designed for specialized intelligence. Built on a hybrid architecture and trained on 22T tokens, LFM3 offers improved expressivity and effici…
-
llama.cpp updates LFM2 support; new TCAPI paper targets tensor network portability
The llama.cpp project has released version b11767, which includes updates for handling LFM2 tiling parameters from GGUF metadata. This release also features improvements for compatibility across various operating system…
-
New WebGPU LLM engine validated against llama.cpp for numerical accuracy
A new WebGPU-based LLM inference engine, quipullm, has been developed and rigorously validated against the established llama.cpp library. The engine, written from scratch in WGSL and JavaScript, runs directly in a brows…
-
LLM chat template errors cause local model behavior issues
Local LLM deployments can encounter issues like models generating beyond their answers, ignoring system prompts, or dropping instructions due to incorrect chat template implementations. These problems typically arise wh…
-
llama.cpp enhances testing suite for broader model compatibility
The llama.cpp project has released an update, b10666, which includes significant improvements to its testing suite. A key change is the expansion of the `test-save-load-state` functionality to cover all architectures an…
-
Liquid AI releases fast, CPU-friendly bidirectional encoders with 8K context
Liquid AI has released two new open-weight bidirectional encoder models, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M. These models are built on the LFM2 hybrid backbone and support an 8,192-token context window, designe…
-
LiquidAI releases LFM2.5 multilingual bidirectional encoders
LiquidAI has released two new multilingual bidirectional encoders, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, built on the LFM2 architecture. These models are designed for on-device efficiency and fine-tuning for vari…
-
New 'WOOFER' Metric Reveals Surprising LLM Performance Scores
A user has developed a "WOOFER" metric to evaluate Large Language Model (LLM) performance using a "Probe_prompt." This metric has yielded surprising results, with some models scoring unexpectedly low, such as # bigpickl…
-
Tutorial shows LFM2 fine-tuning with QLoRA and DPO
This tutorial demonstrates how to fine-tune the LFM2 model using QLoRA and Direct Preference Optimization (DPO) on Google Colab. It covers loading the base LFM2 model with 4-bit quantization, preparing a dataset for sup…
-
Liquid AI releases LFM2-24B-A2B, an efficient 24B parameter MoE model
Liquid AI has released an early checkpoint of its LFM2-24B-A2B model, a sparse Mixture of Experts (MoE) architecture with 24 billion total parameters and 2 billion active parameters per token. This model demonstrates th…