Repository navigation
Add field guide for platform engineers moving to inference - #588
Conversation
|
Docs preview: https://modelplane-docs-pr-588.vercel.app (ready once the site's Content workflow finishes) |
There was a problem hiding this comment.
🟡 Changes recommended
The guide has unresolved publishing issues: a broken diagram URL, equations that display raw TeX, and missing front matter that causes incorrect page and navigation titles.
3 open issues
What changed in this PR
Adds a field guide to Modelplane’s documentation for platform engineers moving into inference engineering, connecting familiar operational skills with model-serving concepts and practices.
Changes:
- Explains inference fundamentals, GPU memory sizing, performance metrics, and optimization techniques.
- Covers production operations, tooling, Modelplane’s role, and lessons from Project Champagne.
- Provides learning paths, practical projects, interview topics, and further reading.
| File | Description |
|---|---|
docs/content/guides/inference_guide.md |
Adds the inference engineering field guide, including operational checklists and learning resources. |
Review effort: Balanced
Add "From Platform Engineer to Inference Engineer: An Open Field Guide to Inference Engineering", a guide for platform engineers moving into inference work, plus its inference stack diagram. Signed-off-by: Annie Talvasto <46491425+AnnieTalvasto@users.noreply.github.com>
Focusing on vale errors and clarity. Signed-off-by: Rae Sharp <resharp20@gmail.com>
Inference Guide edits
There was a problem hiding this comment.
🟡 Changes recommended
The blanket exemption for hyphenated words hides misspelled compounds across all documentation and violates the contribution policy requiring vocabulary exceptions in accept.txt.
6 open issues
Avoid blanket hyphenated-word spelling exemptions · New Distinguish shorter sequences from lower context limits · New Do not promise quantization speedups without benchmarking · New Separate concurrency limits from KV cache exhaustion · New Correct ModelReplica, ModelEndpoint, and ModelService relationships · New Use the image shortcode for the diagram URL
2 resolved since last review
Review effort: Balanced
Signed-off-by: Rae Sharp <resharp20@gmail.com>
Update vale for CI & Image
There was a problem hiding this comment.
🟡 Changes recommended
The blanket hyphenated-word spelling exemption leaves compound typos unchecked across all documentation, and several operational claims in the guide need correction.
10 open issues
Limit spelling exemptions to known compound terms · New Avoid blanket hyphenated-word spelling exemptions Separate output caps from disconnect cancellation handling · New Check queue saturation signals independently · New Qualify GPU update deadlock risk by surge capacity · New Do not treat a low task score as proof of harness failure · New Correct ModelReplica, ModelEndpoint, and ModelService relationships Separate concurrency limits from KV cache exhaustion Do not promise quantization speedups without benchmarking Distinguish shorter sequences from lower context limits
1 resolved since last review
Review effort: Balanced
There was a problem hiding this comment.
🔵 Needs a closer look
The spelling filter exempts all hyphenated words across the documentation with no fallback check, allowing misspellings to pass unchecked instead of explicitly accepting valid compounds.
10 open issues
Limit spelling exemptions to known compound terms Avoid blanket hyphenated-word spelling exemptions Do not treat a low task score as proof of harness failure Qualify GPU update deadlock risk by surge capacity Check queue saturation signals independently Separate output caps from disconnect cancellation handling Correct ModelReplica, ModelEndpoint, and ModelService relationships Separate concurrency limits from KV cache exhaustion Do not promise quantization speedups without benchmarking Distinguish shorter sequences from lower context limits
Review effort: Balanced
| - **Failure stories.** Describe one failure with real numbers. Cover what broke | ||
| and how you found it, then what you changed. | ||
|
|
||
| ## 7. Sources and further reading |
There was a problem hiding this comment.
nit: It'd be good to split this out into its own file later on. so we can continually add to these without needing to touch the whole guide.
This can be done after
There was a problem hiding this comment.
🔵 Needs a closer look
The blanket hyphenated-word exemption lets typos bypass spelling checks, and misleading inference-guide claims about cancellation, quantization, and benchmark results need correction.
10 open issues
Limit spelling exemptions to known compound terms Avoid blanket hyphenated-word spelling exemptions Do not treat a low task score as proof of harness failure Qualify GPU update deadlock risk by surge capacity Check queue saturation signals independently Separate output caps from disconnect cancellation handling Correct ModelReplica, ModelEndpoint, and ModelService relationships Separate concurrency limits from KV cache exhaustion Do not promise quantization speedups without benchmarking Distinguish shorter sequences from lower context limits
Review effort: Balanced
|
I certify my contribution in 758e8e0 under the DCO. (Backtracking to have a record of contribution before @AnnieTalvasto rebases) |
Signed-off-by: Annie Talvasto <46491425+AnnieTalvasto@users.noreply.github.com>
Signed-off-by: rae sharp ♯ <8883519+tr0njavolta@users.noreply.github.com>
Update names to specifies Signed-off-by: Annie Talvasto <46491425+AnnieTalvasto@users.noreply.github.com>


Add "From Platform Engineer to Inference Engineer: An Open Field
Guide to Inference Engineering", a guide for platform engineers
moving into inference work.
Runnix flake check(or./nix.sh flake check) and made sure it passes.Added or updated tests covering any composition function changes.git commit -s.Signed-off-by: Annie Talvasto Annie.talvasto@gmail.com