phoenix-oss/llama-stack
1
0
Fork 0
forked from phoenix-oss/llama-stack-mirror
Code Pull requests Releases Packages 2 Activity Actions

Add init files to post training folders (#711)

Browse source 
add init files to post training folders to make pkg build pick up those
files

## Test
WIP colab notebook
https://colab.research.google.com/drive/1K4Q2wZq232_Bpy2ud4zL9aRxvCWAwyQs?usp=sharing
to sharecase the post training APIs
This commit is contained in:
Botao Chen 2025-01-13 20:19:18 -08:00 • committed by GitHub
parent f320eede2b
commit 747683a8a2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 41 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

5
llama_stack/providers/inline/post_training/torchtune/recipes/__init__.py Normal file
View file

@ -0,0 +1,5 @@
# 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.

36
llama_stack/providers/inline/post_training/torchtune/recipes/lora_finetuning_single_device.py
View file

@ -14,6 +14,24 @@ from typing import Any, Dict, List, Optional, Tuple
import torch
from llama_models.sku_list import resolve_model
from torch import nn
from torch.optim import Optimizer
from torch.utils.data import DataLoader, DistributedSampler
from torchtune import modules, training, utils as torchtune_utils
from torchtune.data import AlpacaToMessages, padded_collate_sft
from torchtune.modules.loss import CEWithChunkedOutputLoss
from torchtune.modules.peft import (
get_adapter_params,
get_adapter_state_dict,
get_lora_module_names,
get_merged_lora_ckpt,
set_trainable_params,
validate_missing_and_unexpected_for_lora,
)
from torchtune.training.lr_schedulers import get_cosine_schedule_with_warmup
from torchtune.training.metric_logging import DiskLogger
from tqdm import tqdm
from llama_stack.apis.common.training_types import PostTrainingMetric
from llama_stack.apis.datasetio import DatasetIO
@ -38,24 +56,6 @@ from llama_stack.providers.inline.post_training.torchtune.config import (
TorchtunePostTrainingConfig,
)
from llama_stack.providers.inline.post_training.torchtune.datasets.sft import SFTDataset
from torch import nn
from torch.optim import Optimizer
from torch.utils.data import DataLoader, DistributedSampler
from torchtune import modules, training, utils as torchtune_utils
from torchtune.data import AlpacaToMessages, padded_collate_sft
from torchtune.modules.loss import CEWithChunkedOutputLoss
from torchtune.modules.peft import (
get_adapter_params,
get_adapter_state_dict,
get_lora_module_names,
get_merged_lora_ckpt,
set_trainable_params,
validate_missing_and_unexpected_for_lora,
)
from torchtune.training.lr_schedulers import get_cosine_schedule_with_warmup
from torchtune.training.metric_logging import DiskLogger
from tqdm import tqdm
log = logging.getLogger(__name__)