feat: add integration tests for post_training

set inline::huggingface as the default post_training provider for the ollama distribution and add integration tests for post_training

Signed-off-by: Charlie Doern <cdoern@redhat.com>

llama_stack/templates/dependencies.json

@@ -441,6 +441,7 @@
     "opentelemetry-exporter-otlp-proto-http",
     "opentelemetry-sdk",
     "pandas",
+    "peft",
     "pillow",
     "psycopg2-binary",
     "pymongo",
@@ -451,9 +452,11 @@
     "scikit-learn",
     "scipy",
     "sentencepiece",
+    "torch",
     "tqdm",
     "transformers",
     "tree_sitter",
+    "trl",
     "uvicorn"
   ],
   "open-benchmark": [

llama_stack/templates/ollama/build.yaml

@@ -23,6 +23,8 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    post_training:
+    - inline::huggingface
     tool_runtime:
     - remote::brave-search
     - remote::tavily-search

llama_stack/templates/ollama/ollama.py

@@ -13,6 +13,7 @@ from llama_stack.distribution.datatypes import (
     ShieldInput,
     ToolGroupInput,
 )
+from llama_stack.providers.inline.post_training.huggingface import HuggingFacePostTrainingConfig
 from llama_stack.providers.inline.vector_io.faiss.config import FaissVectorIOConfig
 from llama_stack.providers.remote.inference.ollama import OllamaImplConfig
 from llama_stack.templates.template import DistributionTemplate, RunConfigSettings
@@ -28,6 +29,7 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "post_training": ["inline::huggingface"],
         "tool_runtime": [
             "remote::brave-search",
             "remote::tavily-search",
@@ -47,7 +49,11 @@ def get_distribution_template() -> DistributionTemplate:
         provider_type="inline::faiss",
         config=FaissVectorIOConfig.sample_run_config(f"~/.llama/distributions/{name}"),
     )
-
+    posttraining_provider = Provider(
+        provider_id="huggingface",
+        provider_type="inline::huggingface",
+        config=HuggingFacePostTrainingConfig.sample_run_config(f"~/.llama/distributions/{name}"),
+    )
     inference_model = ModelInput(
         model_id="${env.INFERENCE_MODEL}",
         provider_id="ollama",
@@ -92,6 +98,7 @@ def get_distribution_template() -> DistributionTemplate:
                 provider_overrides={
                     "inference": [inference_provider],
                     "vector_io": [vector_io_provider_faiss],
+                    "post_training": [posttraining_provider],
                 },
                 default_models=[inference_model, embedding_model],
                 default_tool_groups=default_tool_groups,
@@ -100,6 +107,7 @@ def get_distribution_template() -> DistributionTemplate:
                 provider_overrides={
                     "inference": [inference_provider],
                     "vector_io": [vector_io_provider_faiss],
+                    "post_training": [posttraining_provider],
                     "safety": [
                         Provider(
                             provider_id="llama-guard",

llama_stack/templates/ollama/run-with-safety.yaml

@@ -5,6 +5,7 @@ apis:
 - datasetio
 - eval
 - inference
+- post_training
 - safety
 - scoring
 - telemetry
@@ -80,6 +81,13 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  post_training:
+  - provider_id: huggingface
+    provider_type: inline::huggingface
+    config:
+      checkpoint_format: huggingface
+      distributed_backend: null
+      device: cpu
   tool_runtime:
   - provider_id: brave-search
     provider_type: remote::brave-search

llama_stack/templates/ollama/run.yaml

@@ -5,6 +5,7 @@ apis:
 - datasetio
 - eval
 - inference
+- post_training
 - safety
 - scoring
 - telemetry
@@ -78,6 +79,13 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  post_training:
+  - provider_id: huggingface
+    provider_type: inline::huggingface
+    config:
+      checkpoint_format: huggingface
+      distributed_backend: null
+      device: cpu
   tool_runtime:
   - provider_id: brave-search
     provider_type: remote::brave-search