Add Azure OpenAI Fine-Tuning Cost Advisor prompt template#693
Add Azure OpenAI Fine-Tuning Cost Advisor prompt template#693BethanyJep wants to merge 0 commit intogithub:mainfrom
Conversation
github-actions
Bot
force-pushed
the
main
branch
8 times, most recently
from
a70a351 to
b32fe27
Compare
github-actions
Bot
force-pushed
the
main
branch
15 times, most recently
from
436de60 to
7e4ef49
Compare
|
Yes, sorry, lost access to GitHub for a while. Let me work on resolving confllicts and ensure it is ready to be mearged |
There was a problem hiding this comment.
Pull request overview
Adds a new reusable prompt template intended to guide users through estimating Azure OpenAI fine-tuning costs using Microsoft Docs via MCP, and registers it in the prompts documentation index.
Changes:
- Added a new
finetuning-cost-advisor.prompt.mdprompt for Azure OpenAI fine-tuning cost estimation and recommendations. - Updated
docs/README.prompts.mdto include the new prompt in the prompts list.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| prompts/finetuning-cost-advisor.prompt.md | New prompt template for fine-tuning cost discovery, pricing retrieval via MCP, and cost breakdown guidance. |
| docs/README.prompts.md | Adds the new prompt to the prompts catalog for discoverability/installation. |
Comments suppressed due to low confidence (3)
prompts/finetuning-cost-advisor.prompt.md:5
- Prompt files in this repo strongly recommend specifying a
modelin frontmatter (see AGENTS.md prompt guidance). Adding amodelhere would make behavior more reproducible and helps users understand the expected capability/cost profile for this prompt.
---
agent: 'agent'
description: 'You are an expert Azure OpenAI consultant specializing in helping people understand fine-tuning costs and options. You provide tailored recommendations based on use case, budget, and requirements, using official Microsoft documentation via MCP to ensure accurate and up-to-date pricing information.'
tools: ['microsoftdocs/mcp/*']
---
prompts/finetuning-cost-advisor.prompt.md:45
- This instruction embeds a specific “$5K RFT cap”. Since pricing/limits can change and the prompt already requires verifying via MCP, it would be safer to rephrase this as “If Microsoft’s current docs mention an RFT cap, include it and cite the source” rather than baking in a fixed value.
- **Emphasize Developer Tier** for POCs and startups.
- **Mention the $5K RFT cap** if recommending reinforcement fine-tuning.
- **Link to official docs** for verification.
prompts/finetuning-cost-advisor.prompt.md:140
- The “Pricing Quick Reference” section contains concrete discounts and rates (e.g., “10–30% discount”, “50% discount”, “$1.70/hour”) even though the prompt emphasizes verifying all pricing via MCP. Consider converting these into non-numeric qualitative statements (or placeholders) and requiring the assistant to fill in the actual numbers only after fetching current pricing from the cited sources.
## Pricing Quick Reference (Verify via MCP!)
**Training Tiers:**
- Regional: Standard price
- Global: 10-30% discount
- Developer: 50% discount (spot capacity)
**Deployment Types:**
- Standard: $1.70/hour + pay-per-token
- PTU: Fixed capacity, predictable billing
- Developer: Free hosting, 24h limit
| --- | ||
| agent: 'agent' | ||
| description: 'You are an expert Azure OpenAI consultant specializing in helping people understand fine-tuning costs and options. You provide tailored recommendations based on use case, budget, and requirements, using official Microsoft documentation via MCP to ensure accurate and up-to-date pricing information.' | ||
| tools: ['microsoftdocs/mcp/*'] |
There was a problem hiding this comment.
The frontmatter tools value (microsoftdocs/mcp/*) doesn’t match the Microsoft Docs MCP tool identifier used elsewhere in this repo (e.g., prompts use microsoft.docs.mcp). With the current value, this prompt may not have access to the mcp://microsoft-docs/* tools it requires. Consider aligning the tools entry with the existing convention so the required MCP tools are actually available at runtime.
| tools: ['microsoftdocs/mcp/*'] | |
| tools: ['microsoft.docs.mcp'] |
| ``` | ||
| TRAINING COST (One-time): | ||
| - SFT/DPO: (training_tokens_M × epochs × price_per_M) × tier_discount | ||
| - RFT: (hours × $50/hr) + optional grader costs | ||
|
|
||
| HOSTING COST (Monthly): | ||
| - Standard: $1.70/hour × hours_deployed | ||
| - PTU: PTU_count × hourly_rate × 730 hours | ||
| - Developer: $0 (auto-deletes after 24h) | ||
|
|
There was a problem hiding this comment.
This section hard-codes specific pricing numbers (e.g., “$50/hr”, “$1.70/hour”, “$0”) while earlier rules say not to guess pricing without fetching current MCP data. To avoid stale or contradictory guidance, replace fixed amounts with placeholders/variables and instruct the assistant to populate them only after retrieving the current rates via MCP (with citations).
This issue also appears in the following locations of the same file:
- line 43
- line 130
Pull Request Checklist
npm startand verified thatREADME.mdis up to date.Description
Type of Contribution
Additional Notes
By submitting this pull request, I confirm that my contribution abides by the Code of Conduct and will be licensed under the MIT License.