mirror of
https://github.com/meta-llama/llama-stack.git
synced 2025-07-20 19:56:59 +00:00
2 commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Botao Chen
|
36b4fe02cc
|
[4/n][torchtune integration] support lazy load model during inference (#620)
## What does this PR do?
In this PR, we refactor the meta reference inference logic to support
- load the model during registering model instead of during spinning up
server
- support inference finetuned model checkpoint on top of native llama
model
## Why need these changes
To solve the existing pain points that
- user cannot lazy load the model and hot switch the inference
checkpoint after spinning up the server
- this blocks us doing inference and eval on the same sever for a
finetuned checkpoint after post training
- user cannot do inference on a finetuned checkpoint on top of native
llama models
## Expect user experience change
- The inference model won't be loaded when spinning up server. Instead,
it will be loaded during register model. If user add the model as models
resource in run.yaml, it will be registered and loaded automatically
when starting server. There is an optional flag 'skip_initialize' in
model metadata to skip model loading during registration.
- There is an optional flag 'llama_model' in model metadata to identify
the base model of the Model class for validation and initialize model
arch. model identifier no longer needs to be a native llama model
- the default inference model name updates from
'meta-llama/Llama-3.2-3B-Instruct' to 'Llama3.2-3B-Instruct'
- It aligns with the checkpoint folder name after running 'llama model
download'
- It aligns with the descriptor name defined in llama-models SKU list
|
||
|
Botao Chen
|
aeb76390fc
|
[1/n] torchtune <> llama-stack integration skeleton (#540)
### Context This is the 1st of series PRs that integrate torchtune with llama-stack as meta reference post-training implementation. For MVP, we will focus on single device LoRA SFT. Though this PR is still WIP, we want to get early feedback on the high level design of this skeleton while still working on several details ### Scope To limit the scope of this PR, we focus on the skeleton of the implementation. **What are included?** - refine the post-training SFT apis - skeleton of supervised_fine_tune implementation. We verified that we can call the supervised_fine_tune API successfully from llama stack client SDK (client side PR: https://github.com/meta-llama/llama-stack-client-python/pull/51) - a very basic single device LoRA training recipe based on torchtune core components - parity check with torchtune library and post training api unit test **What are not includes?** - implementation of other job management, get training artifacts apis (separate PR) - refactor the meta reference inference logic to support eval on finetuned model (separate PR) - several necessary functionality in the training recipe such as logging, validation etc (separate PR) - interop with telemetry for tracing and metrics logging, currently temporarily log to local disk (separate PR) ### Testing **e2e test** Although we haven't added detailed testing and numerical parity check with torchtune yet, we did a simple E2E test from client to server 1. setup server with` llama stack build --template experimental-post-training --image-type conda` and `llama stack run experimental-post-training ` 2. On client, run `llama-stack-client --endpoint http://devgpu018.nha2.facebook.com:5000 post_training supervised_fine_tune` 3. Training finishes successfully. On server side, get the finetune checkpoints under output dir. On client side, get the job uuid server <img width="1110" alt="Screenshot 2024-12-02 at 5 52 32 PM" src="https://github.com/user-attachments/assets/b548eb90-7a9b-4edc-a858-ee237cc4361d"> client <img width="807" alt="Screenshot 2024-12-02 at 5 52 37 PM" src="https://github.com/user-attachments/assets/1138ffa8-4698-40fa-b190-3d7b99646838"> **parity check** torchtune dataloader output and llama-stack post training dataloader output are same <img width="1116" alt="Screenshot 2024-12-04 at 8 18 46 PM" src="https://github.com/user-attachments/assets/5e295cdc-4c24-4ea6-82c0-ca96ef1bd6ee"> torchtune LoRA SFT and llama-stack post training LoRA SFT on alpaca dataset with llama3.2 3B instruct model are numerical match <img width="860" alt="Screenshot 2024-12-04 at 8 17 01 PM" src="https://github.com/user-attachments/assets/c05cf0a8-c674-4d2e-9f0a-c5d01b2dca99"> <img width="1049" alt="Screenshot 2024-12-04 at 8 17 06 PM" src="https://github.com/user-attachments/assets/b911d4e2-e7b1-41a9-b62c-d75529b6d443"> **unit test ** ![Uploading Screenshot 2024-12-09 at 1.35.10 PM.png…]() |