Posts

#4882 System-to-System-Decomposition Prompt

-
#4882 System-to-System-Decomposition Prompt #4882  Provide a system-to-system-decomposition prompt base on this example:  <template> Here is a reusable, structured system-to-system decomposition prompt engineered to systematically break down any mechanical, electromechanical, or physical system using the same analytical schema as your example. System-to-System Decomposition Prompt You are an expert mechanical systems engineer and TRIZ practitioner specializing in functional decomposition and root-cause failure analysis. Decompose the provided system into a structured architectural and physical specification. Follow the exact schema below, adhering strictly to technical precision, physical forces, material choices, and interface physics. --- ### Target System Specification Schema Target System: [Macro-level machine, assembly, or vehicle] Target Subsystem: [Specific kinematic, structural, or fluid interface being analyzed] Current Components: - [Primary Load...
Post a Comment
Read more

#4881 Specialize LLM Size Reduction

-
#4881 Specialize LLM Size Reduction #4881  How can LLM size be reduced for specialized LLMs restricted to a specific topic, like physics? To reduce an LLM to a lightweight specialized physics model beyond standard depth slicing and generic LoRA fine-tuning, the parameter space must be surgically carved around the specific computational circuits and latent structures that handle mathematical modeling, symbolic reasoning, and scientific principles. 1. Vocabulary and Embedding Space Truncation Standard foundation models allocate 15% to 30% of their total parameter budget to embedding ( W_emb ) and unembedding/language model head ( W_head ) layers to support 32k–128k multilingual and generic conversational tokens. Corpus Token Profiling: Tokenize a specialized physics corpus (arXiv quant-ph , gr-qc , physics:class-ph , IUPAC/NIST datasets, LaTeX theorem corpuses) using the base model's tokenizer. Frequency and Transition Matrix: Identify the active vocabulary subset V_target s...
Post a Comment
Read more