Scaffold Model-Specific Inference Design Patterns documentation structure

[Copilot]

Establishes infrastructure for collecting empirically-tested prompting patterns for frontier LLMs (GPT-4, Claude, Gemini, Llama, Qwen, DeepSeek). Contributors can now document behavioral tendencies, failure modes, and optimization strategies with quantitative evidence.

Structure

• _TEMPLATE.md - Standardized sections for model documentation:

  • Behavioral tendencies (verbosity, refusals, format preferences)
  • Reasoning characteristics (CoT, math, code, multimodal)
  • Failure modes with reproducible examples + mitigations
  • Cost/performance optimization tables
  • Real-world workflow examples with code

• anthropic-claude-3-sonnet.md - Reference implementation demonstrating:

  • XML-structured prompting preference
  • Known failures (arithmetic errors, medical/legal refusals)
  • Empirical metrics from 500+ test cases
  • Production-ready code examples (EDA, code review, SQL generation)

• README.md - Contributor guidelines, quality standards, naming conventions

Example Usage

Contributors copy template and fill with empirical observations:

# Template demonstrates optimal prompt structure for each model
system_prompt = """You are an expert data scientist.

<role>
Analyze datasets for ML model development.
</role>

<output_format>
{
  "summary": "...",
  "recommended_models": [...],
  "preprocessing_steps": [...]
}
</output_format>"""

Directory ready for patterns across OpenAI, Anthropic, Google, Meta, Alibaba, and DeepSeek models.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Model-Specific Inference Design Patterns (OpenAI, Anthropic, Gemini, Qwen, DeepSeek)</issue_title>
<issue_description>We are collecting rigorously tested prompting and inference design patterns for major frontier and open-weight models. Contributors should document behavioural tendencies, reasoning characteristics, failure modes, and optimisation strategies for specific models. Submissions must include empirical observations, cost-performance considerations, and example task templates grounded in real-world data science workflows.</issue_description>

<agent_instructions>I am working on Issue #11. Please scaffold the directory structure for "Model-Specific Inference Design Patterns".

Create a new directory docs/inference-patterns/.
Inside that, create a template file named _TEMPLATE.md that contributors must use.
The template should include these specific sections:
Model Family (e.g., GPT-4, Claude 3, Llama 3)
Key Behavioral Tendencies (e.g., verbosity, refusal triggers)
Reasoning Characteristics (e.g., Chain-of-Thought performance)
Known Failure Modes (with code/prompt examples)
Optimization Strategies (cost key/value pairs)
Create an initial example file docs/inference-patterns/anthropic-claude-3-sonnet.md filling in known best practices for that model as a reference implementation.</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes Model-Specific Inference Design Patterns (OpenAI, Anthropic, Gemini, Qwen, DeepSeek) #11

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.