phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-07-21 12:09:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix: DPOAlignmentConfig schema to use correct DPO parameters (#2804)
Some checks failed
Coverage Badge / unit-tests (push) Failing after 1s
Details
Integration Auth Tests / test-matrix (oauth2_token) (push) Failing after 6s
Details
Integration Tests / discover-tests (push) Successful in 4s
Details
Test Llama Stack Build / generate-matrix (push) Successful in 9s
Details
Test Llama Stack Build / build-single-provider (push) Failing after 8s
Details
Vector IO Integration Tests / test-matrix (3.13, inline::milvus) (push) Failing after 11s
Details
Vector IO Integration Tests / test-matrix (3.12, inline::faiss) (push) Failing after 13s
Details
Unit Tests / unit-tests (3.12) (push) Failing after 9s
Details
Test Llama Stack Build / build-ubi9-container-distribution (push) Failing after 13s
Details
Update ReadTheDocs / update-readthedocs (push) Failing after 13s
Details
SqlStore Integration Tests / test-postgres (3.12) (push) Failing after 22s
Details
Python Package Build Test / build (3.12) (push) Failing after 15s
Details
Vector IO Integration Tests / test-matrix (3.12, inline::milvus) (push) Failing after 18s
Details
Test External Providers / test-external-providers (venv) (push) Failing after 14s
Details
Vector IO Integration Tests / test-matrix (3.12, remote::pgvector) (push) Failing after 18s
Details
Vector IO Integration Tests / test-matrix (3.12, remote::chromadb) (push) Failing after 20s
Details
Vector IO Integration Tests / test-matrix (3.13, remote::chromadb) (push) Failing after 17s
Details
Test Llama Stack Build / build-custom-container-distribution (push) Failing after 17s
Details
Vector IO Integration Tests / test-matrix (3.13, inline::faiss) (push) Failing after 19s
Details
Unit Tests / unit-tests (3.13) (push) Failing after 19s
Details
Vector IO Integration Tests / test-matrix (3.13, inline::sqlite-vec) (push) Failing after 21s
Details
Integration Tests / test-matrix (push) Failing after 19s
Details
Vector IO Integration Tests / test-matrix (3.12, inline::sqlite-vec) (push) Failing after 17s
Details
Vector IO Integration Tests / test-matrix (3.13, remote::pgvector) (push) Failing after 22s
Details
Test Llama Stack Build / build (push) Failing after 15s
Details
Python Package Build Test / build (3.13) (push) Failing after 1m50s
Details
SqlStore Integration Tests / test-postgres (3.13) (push) Failing after 2m5s
Details
Pre-commit / pre-commit (push) Successful in 3m20s
Details

Browse source 
# What does this PR do?

This PR fixes the `DPOAlignmentConfig` schema to use the correct Direct
Preference Optimization (DPO) parameters.

The current schema incorrectly uses PPO-inspired parameters
(`reward_scale`, `reward_clip`, `epsilon`, `gamma`) that are not part of
the DPO algorithm. This PR updates it to use the standard DPO
parameters:

- `beta`: The KL divergence coefficient that controls deviation from the
reference model
- `loss_type`: The type of DPO loss function (sigmoid, hinge, ipo,
kto_pair)

These parameters align with standard DPO implementations like
HuggingFace's TRL library.

---------

Co-authored-by: Ubuntu <ubuntu@ip-172-31-43-83.ec2.internal>
This commit is contained in:
Nehanth Narendrula 2025-07-18 14:56:00 -04:00 • committed by GitHub
parent d994305f0a
commit 874b1cb00f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 40 additions and 28 deletions
Download patch file Download diff file Expand all files Collapse all files

14
llama_stack/apis/post_training/post_training.py
View file

@ -104,12 +104,18 @@ class RLHFAlgorithm(Enum):
dpo = "dpo"
@json_schema_type
class DPOLossType(Enum):
sigmoid = "sigmoid"
hinge = "hinge"
ipo = "ipo"
kto_pair = "kto_pair"
@json_schema_type
class DPOAlignmentConfig(BaseModel):
reward_scale: float
reward_clip: float
epsilon: float
gamma: float
beta: float
loss_type: DPOLossType = DPOLossType.sigmoid
@json_schema_type