SupraLabs
PulseAugur coverage of SupraLabs — every cluster mentioning SupraLabs across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
SupraLabs releases tiny 25M parameter model outperforming larger predecessors
SupraLabs has released Supra2-Medium-Base, a new language model with 25 million parameters. This model, built on the Qwen3 architecture, demonstrates performance comparable to their previous 50 million parameter model. …
-
Chatbot memory migration and tiny title generation models detailed
A developer shared a project for migrating chatbot conversation memory formats, detailing how to convert framework-specific message objects into a standardized array. The process involves analyzing existing data structu…
-
SupraLabs releases experimental SupraElegans-500K non-Transformer model
SupraLabs has released SupraElegans-500K, an experimental language model that diverges from the standard Transformer architecture. This model utilizes a sparse, signed, recurrent neural graph inspired by the C. elegans …
-
SupraLabs releases Supra2-100M language models
SupraLabs has released two new language models, Supra2-100M Base and Supra2-100M Instruct, following positive feedback on their previous models. These models are available on Hugging Face, with benchmarks indicating com…
-
SupraLabs releases SupraBrain-50M-v0.1 hybrid LLM
SupraLabs has released SupraBrain-50M-v0.1, a hybrid language model that combines Gated DeltaNet linear recurrence with Sliding-Window Attention and Surprise-Gated update mechanisms. Despite being trained on significant…
-
SupraLabs releases 5M-sample reasoning dataset for small LLMs
SupraLabs has released a new dataset called reasoning-corpus-4K-5M-v1, containing 5 million samples designed to help train smaller language models in reasoning capabilities. The dataset includes user prompts, detailed t…
-
SupraLabs releases tiny Supra-Router-51M prompt routing model
SupraLabs has released Supra-Router-51M, a compact language model designed for efficient prompt routing within multi-model AI ecosystems. With only 51.7 million parameters, this model can determine whether a user's requ…
-
SupraLabs releases tiny SupraSafety-18M model for edge content moderation
SupraLabs has released SupraSafety-18M, a new BERT-style model with 18 million parameters designed for content moderation on edge devices and mobile phones. Trained on the nvidia/Nemotron-3.5-Content-Safety-Dataset, thi…
-
SupraLabs releases SupraWeather-Nano preview for weather classification
SupraLabs has released SupraWeather-Nano-Preview, a small FT-Transformer model designed to classify weather phenomena using raw tabular meteorological data. Unlike typical approaches that adapt generic models or ignore …
-
SupraLabs releases preview chat title model and transparent VLM blueprint
SupraLabs has released two new AI models: supra-title-FFT-preview, a chat title generation model trained on 115,000 samples, and SupraVL-Nano-900k, a small vision-language model built from scratch. The supra-title-FFT-p…
-
SupraLabs releases expanded context Supra-1.5 and specialized Supra-Title models
SupraLabs has released two new experimental models: the Supra-1.5-exp family, which includes a base model and an instruct-tuned version with a 5x larger context window (5,120 tokens) and expanded training data, and Supr…
-
SupraLabs releases Supra-50M-Reasoning model
SupraLabs has released Supra-50M-Reasoning, an experimental open-source model designed to generate a thinking chain before providing an answer. This model is a fine-tuned version of Supra-50M-Instruct, trained on a cust…
-
SupraLabs' 50M parameter model tops Hugging Face trending list
SupraLabs has released a 50 million parameter model, Supra-50M-Instruct, which has unexpectedly reached the top trending spot on Hugging Face. This small model has garnered significant attention, surpassing larger model…