kth8/gemma-3-1b-it-SuperGPQA-Classifier

A fine-tune of unsloth/gemma-3-1b-it on the m-a-p/SuperGPQA dataset.

Usage example

Set temperature as 0.0 for best results.

System prompt

You are a classifier. Categorize the following problem into discipline, field, and subfield in JSON format.

User prompt

Cotton and linen both readily catch fire. A batch of towels is composed of both cotton and linen, and is known to have caught fire. If it is known that the towels were ignited by a lit cigarette, which of the following arguments utilizes the most appropriate form of reasoning?

Assistant response

{"discipline": "Philosophy", "field": "Philosophy", "subfield": "Logic"}

Possible output options

Discipline

['Agronomy', 'Economics', 'Management', 'History', 'Law', 'Military Science', 'Philosophy', 'Medicine', 'Science', 'Engineering', 'Education', 'Sociology', 'Literature and Arts']

Field

['Animal Husbandry', 'History', 'Mechanical Engineering', 'Applied Economics', 'Power Engineering and Engineering Thermophysics', 'Mathematics', 'Control Science and Engineering', 'Journalism and Communication', 'Language and Literature', 'Traditional Chinese Medicine', 'Electrical Engineering', 'Musicology', 'Physics', 'Agricultural Engineering', 'Surveying and Mapping Science and Technology', 'Library, Information and Archival Management', 'Political Science', 'Aquaculture', 'Military Science', 'Civil Engineering', 'Textile Science and Engineering', 'Aeronautical and Astronautical Science and Technology', 'Clinical Medicine', 'Information and Communication Engineering', 'Materials Science and Engineering', 'Optical Engineering', 'Instrument Science and Technology', 'Forestry', 'Computer Science and Technology', 'Sociology', 'Mining Engineering', 'Business Administration', 'Chemical Engineering and Technology', 'Geography', 'Stomatology', 'Crop Science', 'Food Science and Engineering', 'Basic Medicine', 'Chemistry', 'Forestry Engineering', 'Public Administration', 'Nuclear Science and Technology', 'Weapon Science and Technology', 'Naval Architecture and Ocean Engineering', 'Geology', 'Art Studies', 'Biology', 'Systems Science', 'Petroleum and Natural Gas Engineering', 'Geological Resources and Geological Engineering', 'Veterinary Medicine', 'Law', 'Atmospheric Science', 'Pharmacy', 'Oceanography', 'Philosophy', 'Astronomy', 'Physical Oceanography', 'Geophysics', 'Transportation Engineering', 'Architecture', 'Electronic Science and Technology', 'Theoretical Economics', 'Management Science and Engineering', 'Mechanics', 'Psychology', 'Physical Education', 'Environmental Science and Engineering', 'Public Health and Preventive Medicine', 'Education', 'Metallurgical Engineering', 'Hydraulic Engineering']

Subfield

['Tourism Management and Technological Economics Management', 'Polymer Physics', 'Military Command and Information Systems', 'Political Economy', 'Epidemiology and Health Statistics', 'Sports Humanities and Sociology', 'Principles of Seismic Exploration', 'Internal Medicine', 'Thermal Energy Engineering', 'Ophthalmology', 'Iron and Steel Metallurgy', 'Criminal Law', 'Animal Nutrition and Feed Science', 'Solar System Science', 'Clinical Stomatology', 'Zoology', 'Organic Chemistry', 'International Law', 'Fluid Machinery and Engineering', 'Statistical Mechanics', 'Acoustics', 'Antenna and Radio Communication', 'Ship Mechanics and Design Principles', 'Imaging and Nuclear Medicine', 'Wood Science and Technology', 'Particle and Nuclear Physics', 'Animal Rearing and Breeding', 'Food Biochemistry', 'Electrochemistry', 'Ordinary Differential Equations', 'Advanced Algebra', 'Pathogen Biology', 'Literary History', 'Ethics', 'Refrigeration and Cryogenic Engineering', 'Theoretical Mechanics', 'High Voltage and Insulation Technology', 'Design Arts', 'Legal Theory and Legal History', 'Historical Geography', 'Physiology', 'Economic History', 'Botany', 'Computational Mathematics', 'Agricultural Environment and Soil-Water Engineering', 'Land Resource Management and Administrative Management', 'Weapon Systems Science and Engineering', 'Education Economics, Management and Social Security', 'Clinical Laboratory Diagnostics', 'Hydraulics and Hydrology', 'Space physics', 'Medicinal Chemistry', 'Constitutional and Administrative Law', 'Atmospheric Physics and Atmospheric Environment', 'Environmental and Resource Protection', 'Power Electronics and Electrical Drives', 'Radiation Protection and Nuclear Technology Applications', 'Manufacturing Automation', 'Cartography and Geographic Information Engineering', 'Broadcasting and Television Art', 'Film Studies', 'Systems Science', 'Music History, Education, and Technology', 'Russian Language and Literature', 'Fundamentals of Dynamics and Control', 'Guidance, Navigation and Control', 'Biophysics', 'Computer Networks', 'Environmental Engineering', 'Mechatronic Engineering', 'Electrical Theory and New Technologies', 'Basic Stomatology', 'Probability and Statistics', 'Group Theory', 'Architectural History', 'Traffic Information Engineering and Control', 'Forensic Medicine', 'Polynomials and Series Expansions', 'Fundamental Mathematics', 'Educational Technology and Principles', 'Human Geography', 'French Language and Literature', 'Cosmology', 'Applied Optics', 'Pitch and Scales', 'Geochemistry', 'Laser Technology', 'Environmental Science', 'Instrumentation and Performance', 'Hydrogeology', 'Military Law', 'Principles of Computer Organization', 'Journalism and News Practice', 'Optoelectronic Technology', 'Physical Geography', 'Power Systems and Automation', 'Special Number Theory', 'Social and Folklore Studies', 'Chemical Transport Engineering', 'Textile Chemistry and Dyeing Engineering', 'Musical Forms and Analysis', 'Military Logistics and Equipment', 'Pediatrics', 'Road and Railway Engineering', 'Business and Accounting Management', 'Physical Chemistry', 'Nuclear Energy and Reactor Technology', 'Power Machinery and Engineering', 'Control Theory and Control Engineering', 'Finance', 'Quantitative Economics', 'Databases', 'Relativity', 'Health Toxicology and Environmental Health', 'Signal and Information Processing', 'Number Theory', 'Pharmaceutical Analysis', 'Computer Architecture', 'Semiconductor Physics', 'Meteorology', 'History and Theory of Journalism and Media Management', 'Military Chemistry and Pyrotechnics', 'Analytical Chemistry', 'Materials Processing Engineering', 'Digital Surveying and Remote Sensing Applications', 'Philosophy of Science and Technology', 'Internal Combustion Engineering', 'Stellar and Interstellar Evolution', 'Stochastic Processes', 'Physical Chemistry of Metallurgical Process', 'Communication and Broadcasting', 'Information Management Science', 'Engineering Fluid Mechanics', 'Mathematical Analysis', 'Urban Planning and Design', 'Law and Social Governance', 'Thermodynamics and Statistical Physics', 'Paleontology and Stratigraphy', 'Geological Resources and Geological Engineering', 'Materials Physics and Chemistry', 'Functions of Real Variables', 'Transportation Planning and Management', 'Operations Research and Cybernetics', 'Optical Fiber Communication', 'Physical Oceanography', 'Oncology', 'Fine Arts', 'Management Science and Engineering', 'National and Defense Economics', 'Philology and Bibliography', 'Anesthesiology', 'Solid Mechanics', 'Urban Infrastructure Engineering', 'Information Management and Communication', 'Microelectronics and Solid-State Electronics', 'Genetics', 'Civil and Commercial Law', 'Computer Software and Theory', 'Traditional Chinese Health Preservation', 'Pattern Recognition', 'Textile Materials Science', 'Nutrition and Food Hygiene', 'Military Thought and History', 'Advanced Programming Languages', 'Geodesy and Surveying Engineering', 'Dynamic Meteorology', 'Modern and Contemporary Chinese Literature', 'Physical Education and Training', 'Public Finance', 'Astronomical Observation and Technology', 'Fluid Flow and Heat Transfer in Chemical Engineering', 'Functions of Complex Variables', 'Marine Chemistry', 'Preschool Education', 'Contract Law', 'Landscape Plants and Ornamental Horticulture', 'Vehicle Operation Engineering', 'Traditional Chinese Pharmacy', 'Maternal, Child and Adolescent Health', 'Aeronautical and Astronautical Science and Technology', 'Emergency Medicine', 'Otorhinolaryngology', 'Food Processing and Storage Engineering', 'Solid State Physics', 'Quantum Mechanics', 'Oil and Gas Field Development and Storage & Transportation Engineering', 'Fuzzy Mathematics', 'Social Medicine and Health Management', 'Labor Economics', 'Pharmaceutics', 'Classical Chinese Literature', 'Graph Theory', 'Instrument Science and Technology', 'Circuits and Systems', 'Human Anatomy and Histology-Embryology', 'Mass Transport and Separation Process in Chemical Engineering', 'Structural Engineering', 'Marine Engineering', 'Heat Transfer', 'Crop Science', 'Theory of Curriculum and Instruction', 'Operating Systems', 'Bridge and Tunnel Engineering', 'Nursing and Rehabilitation Medicine', 'Radiochemistry', 'Pharmacology', 'Underwater Acoustics', 'Surgery', 'Logic', 'Forest Engineering', 'Numerical Analysis', 'Subatomic and Atomic Physics', 'Water conservancy and Hydropower Engineering', 'Atomic and Molecular Physics', 'Biochemistry and Molecular Biology', 'Demography and Anthropology', 'Architectural Design and Theory', 'Formal Languages', 'Sports Science and Medicine', 'Procedural Law', 'Discrete Mathematics', 'Theoretical Fluid Mechanics', 'Marine Biology', 'Psychiatry and Mental Health', 'Solid Earth Geophysics', 'Linguistics and Applied Linguistics', 'Forest Cultivation and Genetic Breeding', 'Aquaculture', 'Military Management', 'Dermatology and Venereology', 'Radiation Medicine', 'Industrial Economics', 'Polymer Chemistry and Physics', 'Archaeology and Museology', 'Principles of Metallurgy', 'Pathology and Pathophysiology', 'Harmony', 'Communication and Information Systems', 'Theoretical Optics', 'Poromechanics and Reservoir Physics', 'Geometry and Topology', 'Data Structures', 'Communication Principles', 'Mining and Safety Engineering', 'Elements of Chemical Reaction Engineering', 'Political Science', 'Agricultural Mechanization Engineering', 'Mineral Processing Engineering', 'Cryptography', 'Cell Biology', 'World History', 'Fluid Physics', 'Microbiology and Biochemical Pharmacy', 'Obstetrics and Gynecology', 'Inorganic Chemistry', 'Geotechnical Engineering', 'Rigid Body Mechanics', 'Neurology', 'Thermodynamics', 'Drama and Opera Studies', 'Electromagnetic Field and Microwave Technology', 'Traditional Chinese Medicine Theory', 'Religious Studies', 'Western Economics', 'Non-ferrous Metallurgy', 'Engineering Thermophysics', 'Combinatorial Mathematics', 'Immunology', 'Structural Geology', 'Dance Studies', 'International Trade', 'Literary Theory', 'Geriatric Medicine', 'Microbiology', 'Ecology', 'Veterinary Medicine', 'Library and Archival Science', 'Composition', 'Special Education', 'Philosophical Aesthetics', 'Electrodynamics', 'Psychology', 'Mineralogy, Petrology, and Economic Geology', 'Astrophysics', 'Economic Statistics']

Model Details

• Base Model: unsloth/gemma-3-1b-it
• Parameter Count: 999,885,952
• Precision: torch.bfloat16

Hardware

• GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition
• Announced: Mar 17th, 2025
• Release Date: Mar 18th, 2025
• Memory Type: GDDR7
• Bandwidth: 1.79 TB/s
• Memory Size: 96 GB
• Memory Bus: 512 bit
• Shading Units: 24064
• TDP: 600W

Training Settings

PEFT

• Rank: 32
• LoRA alpha: 64
• Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Gradient checkpointing: unsloth

SFT

• Epoch: 2
• Batch size: 32
• Gradient Accumulation steps: 1
• Warmup ratio: 0.05
• Learning rate: 0.0004
• Optimizer: adamw_torch_fused
• Learning rate scheduler: cosine

Training stats

• Date: 2026-03-23T10:35:12.916512
• Peak VRAM usage: 27.795 GB
• Global step: 1576
• Training runtime (seconds): 696.4058
• Average training loss: 0.0905450333930968
• Final validation loss: 0.0563415102660656

Framework versions

• Unsloth: 2026.3.10
• TRL: 0.22.2
• Transformers: 4.56.2
• Pytorch: 2.10.0+cu128
• Datasets: 4.8.3
• Tokenizers: 0.22.2

License

This model is released under the Gemma license. See the Gemma Terms of Use and Prohibited Use Policy regarding the use of Gemma-generated content.