Docs Content Part 2: Concepts #449
Conversation
…rehensive getting started with dependency details, and expanded concepts page with Monarch, Services, TorchStore, and RL workflows
Co-authored-by: Svetlana Karslioglu <[email protected]>
Co-authored-by: Svetlana Karslioglu <[email protected]>
Co-authored-by: Svetlana Karslioglu <[email protected]>
Co-authored-by: Svetlana Karslioglu <[email protected]>
Co-authored-by: Svetlana Karslioglu <[email protected]>
Co-authored-by: Svetlana Karslioglu <[email protected]>
- Enhanced homepage with Monarch foundation emphasis, technology stack highlights, validated examples, and clear navigation paths - Expanded getting started with detailed dependency explanations (Monarch, vLLM, TorchTitan, TorchStore, PyTorch Nightly) - Converted installation and verification steps to numbered lists for better readability - Removed FAQ references as FAQ page has been removed - Fixed GPU/process terminology in code examples
- Split long concepts page into: Concepts (overview), Architecture (Monarch/Services/TorchStore), Technology Stack (dependencies), and RL Workflows (writing algorithms) - Added nested toctree under concepts for better navigation in sidebar - Each page focuses on a single aspect with clear cross-references - Improved readability and maintainability of documentation - Fixed GPU/process terminology throughout architecture and workflow examples
meta-cla
bot
added
the
CLA Signed
|
Can you fix the warnings: https://github.com/meta-pytorch/forge/actions/runs/18580111434/job/52973145035?pr=449#step:11:109 ? |
| **Ephemeral Infrastructure** | ||
| : Services are created with your job and torn down when finished. Want to try a new reward model? Change your Python code. No standing deployments to maintain, no infrastructure to provision ahead of time. | ||
|
|
||
| ## TorchStore: Distributed Weight Storage |
There was a problem hiding this comment.
this all looks good to me, but cc @LucasLLC in case of any desired big changes
| @@ -0,0 +1,120 @@ | |||
| # Technology Stack | |||
There was a problem hiding this comment.
I have this slide I'll use at PTC: https://docs.google.com/presentation/d/1owgUNluoyHoPmhg-cKG9di_SgDVTC7wI/edit?slide=id.g38d9e7ea0be_1_421#slide=id.g38d9e7ea0be_1_421
in case we want to show a visualization here too
allenwang28
left a comment
allenwang28
left a comment
There was a problem hiding this comment.
It mostly looks good, @AlannaBurke pls ping me once the changes are incorporated and this PR is only focused on the concepts.md!
| - **Reservoir**: Uniform sampling from history | ||
| - **Hybrid**: Mix multiple strategies | ||
|
|
||
| Integrates with TorchStore for efficient distributed storage. |
There was a problem hiding this comment.
let's remove this line, this is not true right now unfortunately
|
|
||
| You choose your fault tolerance granularity based on your needs. | ||
|
|
||
| ## Best Practices |
There was a problem hiding this comment.
Let's remove the best practices section for now, as we haven't consolidated on them ourselves yet
| - Validate data pipelines before full training | ||
| - Monitor loss curves and generation quality | ||
|
|
||
| ## Validation |
There was a problem hiding this comment.
Let's remove the validation section here as well
I split the docs work into 2 PRs, this is the second.
It builds on #448, so let's merge that first.