feat(distro): fork off a starter-gpu distribution (#3240)

The starter distribution added post-training which added torch
dependencies which pulls in all the nvidia CUDA libraries. This made our
starter container very big. We have worked hard to keep the starter
container small so it serves its purpose as a starter. This PR tries to
get it back to its size by forking off duplicate "-gpu" providers for
post-training. These forked providers are then used for a new
`starter-gpu` distribution which can pull in all dependencies.

docs/source/providers/post_training/index.md

@@ -9,7 +9,9 @@ This section contains documentation for all available providers for the **post_t
 ```{toctree}
 :maxdepth: 1
 
-inline_huggingface
-inline_torchtune
+inline_huggingface-cpu
+inline_huggingface-gpu
+inline_torchtune-cpu
+inline_torchtune-gpu
 remote_nvidia
 ```

docs/source/providers/post_training/inline_huggingface-cpu.md

@@ -0,0 +1,41 @@
+# inline::huggingface-cpu
+
+## Description
+
+HuggingFace-based post-training provider for fine-tuning models using the HuggingFace ecosystem.
+
+## Configuration
+
+| Field | Type | Required | Default | Description |
+|-------|------|----------|---------|-------------|
+| `device` | `<class 'str'>` | No | cuda |  |
+| `distributed_backend` | `Literal['fsdp', 'deepspeed'` | No |  |  |
+| `checkpoint_format` | `Literal['full_state', 'huggingface'` | No | huggingface |  |
+| `chat_template` | `<class 'str'>` | No | <|user|>
+{input}
+<|assistant|>
+{output} |  |
+| `model_specific_config` | `<class 'dict'>` | No | {'trust_remote_code': True, 'attn_implementation': 'sdpa'} |  |
+| `max_seq_length` | `<class 'int'>` | No | 2048 |  |
+| `gradient_checkpointing` | `<class 'bool'>` | No | False |  |
+| `save_total_limit` | `<class 'int'>` | No | 3 |  |
+| `logging_steps` | `<class 'int'>` | No | 10 |  |
+| `warmup_ratio` | `<class 'float'>` | No | 0.1 |  |
+| `weight_decay` | `<class 'float'>` | No | 0.01 |  |
+| `dataloader_num_workers` | `<class 'int'>` | No | 4 |  |
+| `dataloader_pin_memory` | `<class 'bool'>` | No | True |  |
+| `dpo_beta` | `<class 'float'>` | No | 0.1 |  |
+| `use_reference_model` | `<class 'bool'>` | No | True |  |
+| `dpo_loss_type` | `Literal['sigmoid', 'hinge', 'ipo', 'kto_pair'` | No | sigmoid |  |
+| `dpo_output_dir` | `<class 'str'>` | No |  |  |
+
+## Sample Configuration
+
+```yaml
+checkpoint_format: huggingface
+distributed_backend: null
+device: cpu
+dpo_output_dir: ~/.llama/dummy/dpo_output
+
+```
+

docs/source/providers/post_training/inline_huggingface-gpu.md

@@ -0,0 +1,41 @@
+# inline::huggingface-gpu
+
+## Description
+
+HuggingFace-based post-training provider for fine-tuning models using the HuggingFace ecosystem.
+
+## Configuration
+
+| Field | Type | Required | Default | Description |
+|-------|------|----------|---------|-------------|
+| `device` | `<class 'str'>` | No | cuda |  |
+| `distributed_backend` | `Literal['fsdp', 'deepspeed'` | No |  |  |
+| `checkpoint_format` | `Literal['full_state', 'huggingface'` | No | huggingface |  |
+| `chat_template` | `<class 'str'>` | No | <|user|>
+{input}
+<|assistant|>
+{output} |  |
+| `model_specific_config` | `<class 'dict'>` | No | {'trust_remote_code': True, 'attn_implementation': 'sdpa'} |  |
+| `max_seq_length` | `<class 'int'>` | No | 2048 |  |
+| `gradient_checkpointing` | `<class 'bool'>` | No | False |  |
+| `save_total_limit` | `<class 'int'>` | No | 3 |  |
+| `logging_steps` | `<class 'int'>` | No | 10 |  |
+| `warmup_ratio` | `<class 'float'>` | No | 0.1 |  |
+| `weight_decay` | `<class 'float'>` | No | 0.01 |  |
+| `dataloader_num_workers` | `<class 'int'>` | No | 4 |  |
+| `dataloader_pin_memory` | `<class 'bool'>` | No | True |  |
+| `dpo_beta` | `<class 'float'>` | No | 0.1 |  |
+| `use_reference_model` | `<class 'bool'>` | No | True |  |
+| `dpo_loss_type` | `Literal['sigmoid', 'hinge', 'ipo', 'kto_pair'` | No | sigmoid |  |
+| `dpo_output_dir` | `<class 'str'>` | No |  |  |
+
+## Sample Configuration
+
+```yaml
+checkpoint_format: huggingface
+distributed_backend: null
+device: cpu
+dpo_output_dir: ~/.llama/dummy/dpo_output
+
+```
+

docs/source/providers/post_training/inline_torchtune-cpu.md

@@ -0,0 +1,20 @@
+# inline::torchtune-cpu
+
+## Description
+
+TorchTune-based post-training provider for fine-tuning and optimizing models using Meta's TorchTune framework.
+
+## Configuration
+
+| Field | Type | Required | Default | Description |
+|-------|------|----------|---------|-------------|
+| `torch_seed` | `int \| None` | No |  |  |
+| `checkpoint_format` | `Literal['meta', 'huggingface'` | No | meta |  |
+
+## Sample Configuration
+
+```yaml
+checkpoint_format: meta
+
+```
+

docs/source/providers/post_training/inline_torchtune-gpu.md

@@ -0,0 +1,20 @@
+# inline::torchtune-gpu
+
+## Description
+
+TorchTune-based post-training provider for fine-tuning and optimizing models using Meta's TorchTune framework.
+
+## Configuration
+
+| Field | Type | Required | Default | Description |
+|-------|------|----------|---------|-------------|
+| `torch_seed` | `int \| None` | No |  |  |
+| `checkpoint_format` | `Literal['meta', 'huggingface'` | No | meta |  |
+
+## Sample Configuration
+
+```yaml
+checkpoint_format: meta
+
+```
+

llama_stack/distributions/ci-tests/build.yaml

@@ -34,7 +34,7 @@ distribution_spec:
     telemetry:
     - provider_type: inline::meta-reference
     post_training:
-    - provider_type: inline::huggingface
+    - provider_type: inline::huggingface-cpu
     eval:
     - provider_type: inline::meta-reference
     datasetio:

llama_stack/distributions/ci-tests/run.yaml

@@ -156,8 +156,8 @@ providers:
       sqlite_db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/trace_store.db
       otel_exporter_otlp_endpoint: ${env.OTEL_EXPORTER_OTLP_ENDPOINT:=}
   post_training:
-  - provider_id: huggingface
-    provider_type: inline::huggingface
+  - provider_id: huggingface-cpu
+    provider_type: inline::huggingface-cpu
     config:
       checkpoint_format: huggingface
       distributed_backend: null

llama_stack/distributions/starter-gpu/__init__.py

@@ -0,0 +1,7 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from .starter_gpu import get_distribution_template  # noqa: F401

llama_stack/distributions/starter-gpu/build.yaml

@@ -0,0 +1,59 @@
+version: 2
+distribution_spec:
+  description: Quick start template for running Llama Stack with several popular providers.
+    This distribution is intended for GPU-enabled environments.
+  providers:
+    inference:
+    - provider_type: remote::cerebras
+    - provider_type: remote::ollama
+    - provider_type: remote::vllm
+    - provider_type: remote::tgi
+    - provider_type: remote::fireworks
+    - provider_type: remote::together
+    - provider_type: remote::bedrock
+    - provider_type: remote::nvidia
+    - provider_type: remote::openai
+    - provider_type: remote::anthropic
+    - provider_type: remote::gemini
+    - provider_type: remote::vertexai
+    - provider_type: remote::groq
+    - provider_type: remote::sambanova
+    - provider_type: inline::sentence-transformers
+    vector_io:
+    - provider_type: inline::faiss
+    - provider_type: inline::sqlite-vec
+    - provider_type: inline::milvus
+    - provider_type: remote::chromadb
+    - provider_type: remote::pgvector
+    files:
+    - provider_type: inline::localfs
+    safety:
+    - provider_type: inline::llama-guard
+    - provider_type: inline::code-scanner
+    agents:
+    - provider_type: inline::meta-reference
+    telemetry:
+    - provider_type: inline::meta-reference
+    post_training:
+    - provider_type: inline::torchtune-gpu
+    eval:
+    - provider_type: inline::meta-reference
+    datasetio:
+    - provider_type: remote::huggingface
+    - provider_type: inline::localfs
+    scoring:
+    - provider_type: inline::basic
+    - provider_type: inline::llm-as-judge
+    - provider_type: inline::braintrust
+    tool_runtime:
+    - provider_type: remote::brave-search
+    - provider_type: remote::tavily-search
+    - provider_type: inline::rag-runtime
+    - provider_type: remote::model-context-protocol
+    batches:
+    - provider_type: inline::reference
+image_type: venv
+additional_pip_packages:
+- aiosqlite
+- asyncpg
+- sqlalchemy[asyncio]

llama_stack/distributions/starter-gpu/run.yaml

@@ -0,0 +1,238 @@
+version: 2
+image_name: starter-gpu
+apis:
+- agents
+- batches
+- datasetio
+- eval
+- files
+- inference
+- post_training
+- safety
+- scoring
+- telemetry
+- tool_runtime
+- vector_io
+providers:
+  inference:
+  - provider_id: ${env.CEREBRAS_API_KEY:+cerebras}
+    provider_type: remote::cerebras
+    config:
+      base_url: https://api.cerebras.ai
+      api_key: ${env.CEREBRAS_API_KEY:=}
+  - provider_id: ${env.OLLAMA_URL:+ollama}
+    provider_type: remote::ollama
+    config:
+      url: ${env.OLLAMA_URL:=http://localhost:11434}
+  - provider_id: ${env.VLLM_URL:+vllm}
+    provider_type: remote::vllm
+    config:
+      url: ${env.VLLM_URL:=}
+      max_tokens: ${env.VLLM_MAX_TOKENS:=4096}
+      api_token: ${env.VLLM_API_TOKEN:=fake}
+      tls_verify: ${env.VLLM_TLS_VERIFY:=true}
+  - provider_id: ${env.TGI_URL:+tgi}
+    provider_type: remote::tgi
+    config:
+      url: ${env.TGI_URL:=}
+  - provider_id: fireworks
+    provider_type: remote::fireworks
+    config:
+      url: https://api.fireworks.ai/inference/v1
+      api_key: ${env.FIREWORKS_API_KEY:=}
+  - provider_id: together
+    provider_type: remote::together
+    config:
+      url: https://api.together.xyz/v1
+      api_key: ${env.TOGETHER_API_KEY:=}
+  - provider_id: bedrock
+    provider_type: remote::bedrock
+  - provider_id: ${env.NVIDIA_API_KEY:+nvidia}
+    provider_type: remote::nvidia
+    config:
+      url: ${env.NVIDIA_BASE_URL:=https://integrate.api.nvidia.com}
+      api_key: ${env.NVIDIA_API_KEY:=}
+      append_api_version: ${env.NVIDIA_APPEND_API_VERSION:=True}
+  - provider_id: openai
+    provider_type: remote::openai
+    config:
+      api_key: ${env.OPENAI_API_KEY:=}
+      base_url: ${env.OPENAI_BASE_URL:=https://api.openai.com/v1}
+  - provider_id: anthropic
+    provider_type: remote::anthropic
+    config:
+      api_key: ${env.ANTHROPIC_API_KEY:=}
+  - provider_id: gemini
+    provider_type: remote::gemini
+    config:
+      api_key: ${env.GEMINI_API_KEY:=}
+  - provider_id: ${env.VERTEX_AI_PROJECT:+vertexai}
+    provider_type: remote::vertexai
+    config:
+      project: ${env.VERTEX_AI_PROJECT:=}
+      location: ${env.VERTEX_AI_LOCATION:=us-central1}
+  - provider_id: groq
+    provider_type: remote::groq
+    config:
+      url: https://api.groq.com
+      api_key: ${env.GROQ_API_KEY:=}
+  - provider_id: sambanova
+    provider_type: remote::sambanova
+    config:
+      url: https://api.sambanova.ai/v1
+      api_key: ${env.SAMBANOVA_API_KEY:=}
+  - provider_id: sentence-transformers
+    provider_type: inline::sentence-transformers
+  vector_io:
+  - provider_id: faiss
+    provider_type: inline::faiss
+    config:
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/faiss_store.db
+  - provider_id: sqlite-vec
+    provider_type: inline::sqlite-vec
+    config:
+      db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec.db
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/sqlite_vec_registry.db
+  - provider_id: ${env.MILVUS_URL:+milvus}
+    provider_type: inline::milvus
+    config:
+      db_path: ${env.MILVUS_DB_PATH:=~/.llama/distributions/starter}/milvus.db
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/milvus_registry.db
+  - provider_id: ${env.CHROMADB_URL:+chromadb}
+    provider_type: remote::chromadb
+    config:
+      url: ${env.CHROMADB_URL:=}
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter/}/chroma_remote_registry.db
+  - provider_id: ${env.PGVECTOR_DB:+pgvector}
+    provider_type: remote::pgvector
+    config:
+      host: ${env.PGVECTOR_HOST:=localhost}
+      port: ${env.PGVECTOR_PORT:=5432}
+      db: ${env.PGVECTOR_DB:=}
+      user: ${env.PGVECTOR_USER:=}
+      password: ${env.PGVECTOR_PASSWORD:=}
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/pgvector_registry.db
+  files:
+  - provider_id: meta-reference-files
+    provider_type: inline::localfs
+    config:
+      storage_dir: ${env.FILES_STORAGE_DIR:=~/.llama/distributions/starter/files}
+      metadata_store:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/files_metadata.db
+  safety:
+  - provider_id: llama-guard
+    provider_type: inline::llama-guard
+    config:
+      excluded_categories: []
+  - provider_id: code-scanner
+    provider_type: inline::code-scanner
+  agents:
+  - provider_id: meta-reference
+    provider_type: inline::meta-reference
+    config:
+      persistence_store:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/agents_store.db
+      responses_store:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/responses_store.db
+  telemetry:
+  - provider_id: meta-reference
+    provider_type: inline::meta-reference
+    config:
+      service_name: "${env.OTEL_SERVICE_NAME:=\u200B}"
+      sinks: ${env.TELEMETRY_SINKS:=console,sqlite}
+      sqlite_db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/trace_store.db
+      otel_exporter_otlp_endpoint: ${env.OTEL_EXPORTER_OTLP_ENDPOINT:=}
+  post_training:
+  - provider_id: torchtune-gpu
+    provider_type: inline::torchtune-gpu
+    config:
+      checkpoint_format: meta
+  eval:
+  - provider_id: meta-reference
+    provider_type: inline::meta-reference
+    config:
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/meta_reference_eval.db
+  datasetio:
+  - provider_id: huggingface
+    provider_type: remote::huggingface
+    config:
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/huggingface_datasetio.db
+  - provider_id: localfs
+    provider_type: inline::localfs
+    config:
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/localfs_datasetio.db
+  scoring:
+  - provider_id: basic
+    provider_type: inline::basic
+  - provider_id: llm-as-judge
+    provider_type: inline::llm-as-judge
+  - provider_id: braintrust
+    provider_type: inline::braintrust
+    config:
+      openai_api_key: ${env.OPENAI_API_KEY:=}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:=}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:=}
+      max_results: 3
+  - provider_id: rag-runtime
+    provider_type: inline::rag-runtime
+  - provider_id: model-context-protocol
+    provider_type: remote::model-context-protocol
+  batches:
+  - provider_id: reference
+    provider_type: inline::reference
+    config:
+      kvstore:
+        type: sqlite
+        db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/batches.db
+metadata_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/registry.db
+inference_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/inference_store.db
+models: []
+shields:
+- shield_id: llama-guard
+  provider_id: ${env.SAFETY_MODEL:+llama-guard}
+  provider_shield_id: ${env.SAFETY_MODEL:=}
+- shield_id: code-scanner
+  provider_id: ${env.CODE_SCANNER_MODEL:+code-scanner}
+  provider_shield_id: ${env.CODE_SCANNER_MODEL:=}
+vector_dbs: []
+datasets: []
+scoring_fns: []
+benchmarks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::rag
+  provider_id: rag-runtime
+server:
+  port: 8321

llama_stack/distributions/starter-gpu/starter_gpu.py

@@ -0,0 +1,22 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+
+from llama_stack.distributions.template import BuildProvider, DistributionTemplate
+
+from ..starter.starter import get_distribution_template as get_starter_distribution_template
+
+
+def get_distribution_template() -> DistributionTemplate:
+    template = get_starter_distribution_template()
+    name = "starter-gpu"
+    template.name = name
+    template.description = "Quick start template for running Llama Stack with several popular providers. This distribution is intended for GPU-enabled environments."
+
+    template.providers["post_training"] = [
+        BuildProvider(provider_type="inline::torchtune-gpu"),
+    ]
+    return template

llama_stack/distributions/starter/build.yaml

@@ -1,6 +1,7 @@
 version: 2
 distribution_spec:
-  description: Quick start template for running Llama Stack with several popular providers
+  description: Quick start template for running Llama Stack with several popular providers.
+    This distribution is intended for CPU-only environments.
   providers:
     inference:
     - provider_type: remote::cerebras
@@ -34,7 +35,7 @@ distribution_spec:
     telemetry:
     - provider_type: inline::meta-reference
     post_training:
-    - provider_type: inline::huggingface
+    - provider_type: inline::huggingface-cpu
     eval:
     - provider_type: inline::meta-reference
     datasetio:

llama_stack/distributions/starter/run.yaml

@@ -156,8 +156,8 @@ providers:
       sqlite_db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/trace_store.db
       otel_exporter_otlp_endpoint: ${env.OTEL_EXPORTER_OTLP_ENDPOINT:=}
   post_training:
-  - provider_id: huggingface
-    provider_type: inline::huggingface
+  - provider_id: huggingface-cpu
+    provider_type: inline::huggingface-cpu
     config:
       checkpoint_format: huggingface
       distributed_backend: null

llama_stack/distributions/starter/starter.py

@@ -120,7 +120,7 @@ def get_distribution_template() -> DistributionTemplate:
         ],
         "agents": [BuildProvider(provider_type="inline::meta-reference")],
         "telemetry": [BuildProvider(provider_type="inline::meta-reference")],
-        "post_training": [BuildProvider(provider_type="inline::huggingface")],
+        "post_training": [BuildProvider(provider_type="inline::huggingface-cpu")],
         "eval": [BuildProvider(provider_type="inline::meta-reference")],
         "datasetio": [
             BuildProvider(provider_type="remote::huggingface"),
@@ -178,7 +178,7 @@ def get_distribution_template() -> DistributionTemplate:
     return DistributionTemplate(
         name=name,
         distro_type="self_hosted",
-        description="Quick start template for running Llama Stack with several popular providers",
+        description="Quick start template for running Llama Stack with several popular providers. This distribution is intended for CPU-only environments.",
         container_image=None,
         template_path=None,
         providers=providers,

llama_stack/providers/registry/post_training.py

@@ -5,15 +5,15 @@
 # the root directory of this source tree.
 
 
+from typing import cast
+
 from llama_stack.providers.datatypes import AdapterSpec, Api, InlineProviderSpec, ProviderSpec, remote_provider_spec
 
-
-def available_providers() -> list[ProviderSpec]:
-    return [
-        InlineProviderSpec(
+# We provide two versions of these providers so that distributions can package the appropriate version of torch.
+# The CPU version is used for distributions that don't have GPU support -- they result in smaller container images.
+torchtune_def = dict(
     api=Api.post_training,
-            provider_type="inline::torchtune",
-            pip_packages=["torch", "torchtune==0.5.0", "torchao==0.8.0", "numpy"],
+    pip_packages=["torchtune==0.5.0", "torchao==0.8.0", "numpy"],
     module="llama_stack.providers.inline.post_training.torchtune",
     config_class="llama_stack.providers.inline.post_training.torchtune.TorchtunePostTrainingConfig",
     api_dependencies=[
@@ -21,11 +21,11 @@ def available_providers() -> list[ProviderSpec]:
         Api.datasets,
     ],
     description="TorchTune-based post-training provider for fine-tuning and optimizing models using Meta's TorchTune framework.",
-        ),
-        InlineProviderSpec(
+)
+
+huggingface_def = dict(
     api=Api.post_training,
-            provider_type="inline::huggingface",
-            pip_packages=["torch", "trl", "transformers", "peft", "datasets"],
+    pip_packages=["trl", "transformers", "peft", "datasets"],
     module="llama_stack.providers.inline.post_training.huggingface",
     config_class="llama_stack.providers.inline.post_training.huggingface.HuggingFacePostTrainingConfig",
     api_dependencies=[
@@ -33,6 +33,46 @@ def available_providers() -> list[ProviderSpec]:
         Api.datasets,
     ],
     description="HuggingFace-based post-training provider for fine-tuning models using the HuggingFace ecosystem.",
+)
+
+
+def available_providers() -> list[ProviderSpec]:
+    return [
+        InlineProviderSpec(
+            **{
+                **torchtune_def,
+                "provider_type": "inline::torchtune-cpu",
+                "pip_packages": (
+                    cast(list[str], torchtune_def["pip_packages"])
+                    + ["torch torchtune==0.5.0 torchao==0.8.0 --index-url https://download.pytorch.org/whl/cpu"]
+                ),
+            },
+        ),
+        InlineProviderSpec(
+            **{
+                **huggingface_def,
+                "provider_type": "inline::huggingface-cpu",
+                "pip_packages": (
+                    cast(list[str], huggingface_def["pip_packages"])
+                    + ["torch --index-url https://download.pytorch.org/whl/cpu"]
+                ),
+            },
+        ),
+        InlineProviderSpec(
+            **{
+                **torchtune_def,
+                "provider_type": "inline::torchtune-gpu",
+                "pip_packages": (
+                    cast(list[str], torchtune_def["pip_packages"]) + ["torch torchtune==0.5.0 torchao==0.8.0"]
+                ),
+            },
+        ),
+        InlineProviderSpec(
+            **{
+                **huggingface_def,
+                "provider_type": "inline::huggingface-gpu",
+                "pip_packages": (cast(list[str], huggingface_def["pip_packages"]) + ["torch"]),
+            },
         ),
         remote_provider_spec(
             api=Api.post_training,