singhmandavi/math-slm-qwen2.5-0.5b-v4

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 18, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

singhmandavi/math-slm-qwen2.5-0.5b-v4 is a 0.5 billion parameter domain-specific small language model, fine-tuned from Qwen2.5-0.5B-Instruct, designed for step-by-step math problem-solving. Developed by team03 (SLM Learners), it specializes in GATE-style and general mathematics, providing chain-of-thought reasoning for complex problems. Its primary use case is free-form math QA for competitive exam aspirants and math students, offering direct answers for simple prompts and detailed explanations for complex ones.

Loading preview...

Overview

singhmandavi/math-slm-qwen2.5-0.5b-v4 is a 0.5 billion parameter domain-specific small language model, fine-tuned from Qwen2.5-0.5B-Instruct. Developed by team03 (SLM Learners), this model is specifically designed for step-by-step math problem-solving in GATE-style and general mathematics.

Key Capabilities

  • Domain-Specific Math QA: Excels in free-form math question answering, providing direct answers for simple prompts and detailed chain-of-thought reasoning for complex problems.
  • Efficient Inference: A Q8_0 GGUF quantization is available for CPU inference and llama.cpp-based runtimes, offering effectively lossless performance relative to the fp16 safetensors for this 0.5B model.
  • Context Length: Trained with a maximum sequence length of 2,048 tokens.

Good For

  • Competitive Exam Preparation: Ideal for GATE and other competitive exam aspirants requiring detailed mathematical problem-solving.
  • Math Students: Useful for students seeking step-by-step solutions and explanations for general mathematics problems.
  • Resource-Constrained Environments: The small parameter count and available GGUF quantization make it suitable for deployment on CPUs or devices with limited GPU resources.

Limitations

  • Instruction Following: Exhibits weak instruction-following under strict output-format constraints, often prepending a "here is the thinking process..." preamble even when not explicitly requested.
  • Context Truncation: Problems exceeding the 2,048-token context length will be truncated.
  • Evaluation Caveats: Level 3 evaluation results are based on a small, unevenly distributed 126-row held-out set, leading to noisy per-bucket scores.

Finetunes
2 models