--- license: apache-2.0 datasets: - agentlans/wikipedia-propositions language: - en base_model: - dnotitia/Smoothie-Qwen3-0.6B pipeline_tag: text-generation tags: - propositions - paragraph - writing - analysis --- # Smoothie Qwen 3 0.6B Proposition Extractor - This model extracts propositions from text paragraphs. - Input and output should be given in HTML-like syntax. See examples below. > [!NOTE] > For best results, sample with warmer temperatures. > If it's too cold, the model will just give a list of sentences. ## Examples - Top-p 0.75 - Temperature 0.6 Note: The HTML input and output have been prettified. In the training data, there are no indents or linebreaks.
Example 1 **Input** ```html

Printmaking stands at the intersection of invention, tradition, and materiality, inviting artists to rethink image-making as a tactile dialogue between surface, pressure, and pigment. From the intrepid experiments of monotype and collagraph to the disciplined precision of linocut and etching, printmaking grants multiplicity without sacrificing coherence of vision; a single matrix can yield an edition of prints that refracts a single moment into many interpretations, each copy carrying unique variations in tone, texture, and mark. This discipline is as much about process as product: the choice of paper, the kind of ink, the pressure applied by the press, the sequence of inking and wiping, and the calibration of temperature and humidity all become expressive variables. Printmakers continually juxtapose reproduction with originality, embracing reproducibility as a democratic instrument that democratizes access to imagery while also offering a rigorous framework for authorship, editioning, and provenance. In contemporary practice, printmaking often serves as a bridge between traditional ateliers and experimental studios, where digital tools inform analog methods and hand-crafted marks retain a human trace that no machine can fully replicate. The result is a dynamic, evolving field that honors historical lineage—from woodcut blocks in early modern Europe to aquatint experiments in the 18th century—while pushing toward new textures, hybrid techniques, and socially engaged projects that expand the role of prints beyond decoration to documentation, protest, and storytelling.

``` **Output** ```html ```
Example 2 **Input** ```html

Forensic medicine occupies a crucial niche at the crossroads of science, law, and public health, translating biological signals into evidentiary narratives that can stand up in court while illuminating patterns of disease and injury in populations. At its core lies the meticulous reconstruction of events through the physical remnants of the human body: wounds, pathological changes, toxic substances, and radiographic or genetic data are interpreted within established medico-legal frameworks to determine cause and manner of death, as well as to identify potential public health risks. The field continually grapples with the balance between diagnostic certainty and probabilistic inference, acknowledging the limits of evidence while leveraging advances in imaging, histopathology, genomics, and metabolomics to refine conclusions. Incidence of disease, when examined through forensic lenses, yields insights into epidemiological trends, environmental exposures, occupational hazards, and social determinants that shape who gets sick and how severe their outcomes may be. As technologies evolve—such as high-throughput sequencing, machine learning-assisted pattern recognition, and digital documentation—forensic medicine becomes more capable of reconstructing complex cases, while also raising important considerations about privacy, consent, and the fair application of scientific findings within legal systems. In practice, the discipline not only resolves individual cases but also informs policies on disease surveillance, workplace safety, and emergency response, illustrating how evidence gathered in the morgue or the field can ripple outward to protect communities.

``` **Output** ```html ```
Example 3 **Input** ```html

Communities emerge as living ecosystems where shared routines, symbolic meanings, and material conditions intertwine to create structures of belonging, identity, and mutual support. Sociologists and anthropologists study communities as dynamic networks shaped by kinship, migration, economics, power relations, and cultural practices, recognizing that collective life is sustained not merely by formal institutions but by everyday interactions—gossip, rituals, neighborhood economies, and informal childcare—that knit individuals into a sense of collective fate. Analytical attention often centers on how communities mobilize resources, negotiate conflict, and cultivate solidarities across difference, whether through voluntary associations, religious gatherings, educational networks, or digital forums that extend social reach beyond geographic boundaries. Ethnographic immersion reveals the texture of everyday life: the textures of language, foodways, material culture, and space, all of which encode values and histories that both constrain and enable action. Yet communities are not monolithic; internal diversity, contested identities, and evolving norms generate tensions that can spur transformation as much as cohesion. In contemporary research, emphasis on equity, inclusion, and reflexivity invites close examination of power dynamics, structural inequalities, and the ways institutions—policies, media, and governance—intersect with communal life. Taken together, studies of communities illuminate how people collectively interpret the world, share resources, and navigate the pressures of modernization while maintaining a sense of place, purpose, and mutual obligation.

``` **Output** ```html ```
## Limitations - Quality of results depends on the sampling parameters - Ideally, want the model to paraphrase and extract propositions without being too literal or too vague - Not deterministic because it's a large language model - Input quality is also important - Garbage in, garbage out - Natural language can be imprecise and ambiguous ## Training Settings - LLaMA Factory - 3 epochs - flashattn2 - NEFTune alpha 5, pack sequences on - LoRA rank 16, alpha 32 with rsLoRA on ## Licence Apache 2.0 ## See Also [agentlans/granite-4.0-1b-paraphraser](https://huggingface.co/agentlans/granite-4.0-1b-paraphraser) which has additional functions.