E5-base Pushes Users to E5-base-v2 With 5-Point Accuracy Gains
Intfloat posted a deprecation notice on the original E5-base embedding model, telling users to migrate to E5-base-v2 for better performance.
20 stories
Intfloat posted a deprecation notice on the original E5-base embedding model, telling users to migrate to E5-base-v2 for better performance.
A C++ port with pre-quantized GGUF weights brings the ACE-Step 1.5 music foundation model to CPU, CUDA, Metal and Vulkan backends.
A 7.89 GB IQ2 quant of Qwen3.8-27B that beats the full 54 GB model on tool-calling benchmarks while fitting on consumer GPUs.
A community 4-bit AWQ build of Qwen2.5-Coder-7B-Instruct is pulling in three quarters of a million downloads, shrinking the model to under 6GB of VRAM.
Jina ships a 1.74B multimodal embedding model that encodes text, images, video, and audio into one shared vector space without reindexing.
A 4B decision-native model from NUS handles 2,048 candidates, zero decision flips under reordering, and 20x speedups via cross-request caching.
Cactus Compute's Whistle squeezes multilingual speech-to-text into a 16.9 MB CPU file that reportedly beats Whisper base with 9x less size and 6x speed.
A 3-bit EXL3 quant squeezes a 753B uncensored GLM-5.3 Mixture of Experts into 273 GiB, making local inference possible on four RTX PRO 6000s.
A 400MB anime-style LoRA for Qwen Image 2.1 brings curated aesthetics, character sheet conditioning, and open weights without a gatekeeper.
Tencent's 5B omni-modal parser handles documents, charts, geometry, audio and video into one structured JSON schema, rivaling Gemini-3-Pro.
Cloudflare adds a Qwen3-Omni mixture-of-experts variant that scores audio and video alongside text and images in one pass, with no token generation.
Liquid AI's d1-3B skips token generation entirely, returning calibrated classifications and scores in a single forward pass, with multimodal input on edge hardware.
NVIDIA shipped an FP4 quant of Qwen3.8-Flash-Next that fits 1M tokens of context on a single DGX Spark, with only a 0.7-point accuracy drop.
NVIDIA shipped an FP4-quantized build of Z.ai's 320B GLM-5.3-Flash that fits on two DGX Sparks while retaining nearly all BF16 accuracy.
A 8.6M parameter Turkish text-to-speech model beats ElevenLabs v4 and a 2.38B competitor on accuracy while running 440x faster than real time.
A 4-bit AWQ build of Meta's Llama 3.3 70B Instruct shrinks the 70B model to fit on a single 48GB GPU while preserving most benchmark scores.
IBM's 1M-parameter TSPulse foundation model handles classification, anomaly detection, imputation, and search with GPU-free inference and open weights.
NVIDIA squeezes a 74B parameter mixture-of-experts agent model into 24GB using 2-bit quantization, targeting local RTX hardware with full 256K context.
A tiny 24MB adapter turns Kokoro-82M into a zero-shot voice cloner that enrolls a reference clip in under two seconds on CPU.
LightOn released a three-size OCR family that transcribes text, detects layout with bounding boxes, describes images and extracts chart data in one pass.