GPT-OSS Pruned Experts (4.2B-20B) [IF, Science, Math, etc.]

AmanPriyanshu 's Collections

Pooled Sets

FORMAT: Search - Retrieve RLVR

FORMAT: Reasoning Datasets - DeepSeek Format

FORMAT: Tool-Use Datasets - Hermes-Reasoning-Tool-Use Format

Stratified-Datasets (100K-1M) [Pre-Training, IF, Reasoning]

GPT-OSS Pruned Experts (4.2B-20B) [IF, Science, Math, etc.]

GPT-OSS General (4.2B to 20B)

GPT-OSS Harmful (4.2B to 20B)

GPT-OSS Math (4.2B to 20B)

GPT-OSS Health / Medicine (4.2B to 20B)

GPT-OSS Law (4.2B to 20B)

GPT-OSS Instruction Following (4.2B to 20B)

GPT-OSS Safety (4.2B to 20B)

GPT-OSS Science (4.2B to 20B)

GPT-OSS Pruned Experts (4.2B-20B) [IF, Science, Math, etc.]

updated Aug 13, 2025

Complete collection of domain-specialized GPT-OSS models (1-32 experts) optimized for science, math, medicine, law, safety, and instruction following.

Upvote

GPT-OSS General (4.2B to 20B)

Collection

Collection of pruned GPT-OSS models spanning 1-32 experts, maintaining general capabilities across domains while reducing computational requirements. • 29 items • Updated Aug 13, 2025 • 10
GPT-OSS Science (4.2B to 20B)

Collection

Specialized GPT-OSS models optimized for scientific reasoning tasks in physics, chemistry, biology, and STEM from GPQA and MMLU science domains. • 29 items • Updated Aug 13, 2025 • 2
GPT-OSS Health / Medicine (4.2B to 20B)

Collection

Medical domain GPT-OSS models specializing in clinical knowledge, anatomy, medical procedures, & health-related reasoning from MMLU medical subjects. • 29 items • Updated Aug 13, 2025 • 3
GPT-OSS Math (4.2B to 20B)

Collection

Mathematics-focused GPT-OSS models excelling at mathematical computation, proof strategies, and logical reasoning from MMLU mathematics subjects. • 29 items • Updated Aug 13, 2025 • 2
GPT-OSS Law (4.2B to 20B)

Collection

Legal domain GPT-OSS models excelling at legal reasoning, jurisprudence, and understanding legal frameworks from MMLU legal subjects. • 29 items • Updated Aug 13, 2025 • 3
GPT-OSS Safety (4.2B to 20B)

Collection

Safety-focused GPT-OSS models specialized in identifying and responding to harmful content while maintaining helpful capabilities from SORRY-Bench. • 29 items • Updated Aug 13, 2025 • 1
GPT-OSS Instruction Following (4.2B to 20B)

Collection

GPT-OSS models optimized for precise instruction adherence and constraint satisfaction from Tulu3 Persona Instruction Following dataset. • 29 items • Updated Aug 13, 2025 • 3
GPT-OSS Harmful (4.2B to 20B)

Collection

Research-oriented GPT-OSS models with inverted safety patterns for red-teaming & adversarial analysis to understand failure modes and vulnerabilities. • 29 items • Updated Aug 13, 2025 • 2

Upvote

Collection guide
Browse collections