phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-12-26 19:18:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge remote-tracking branch 'origin/main' into support_more_data_format

Browse source
This commit is contained in:
Botao Chen 2025-01-06 14:19:10 -08:00
commit 2a992d4f05
10 changed files with 76 additions and 55 deletions
Download patch file Download diff file Expand all files Collapse all files

30
llama_stack/providers/inline/post_training/torchtune/common/checkpointer.py
View file

@ -90,18 +90,24 @@ class TorchtuneCheckpointer:
model_file_path.mkdir(parents=True, exist_ok=True)
# copy the related files for inference
shutil.copy(
Path.joinpath(self._checkpoint_dir, "params.json"),
Path.joinpath(model_file_path, "params.json"),
)
shutil.copy(
Path.joinpath(self._checkpoint_dir, "tokenizer.model"),
Path.joinpath(model_file_path, "tokenizer.model"),
)
shutil.copy(
Path.joinpath(self._checkpoint_dir, "orig_params.json"),
Path.joinpath(model_file_path, "orig_params.json"),
)
source_path = Path.joinpath(self._checkpoint_dir, "params.json")
if source_path.exists():
shutil.copy(
source_path,
Path.joinpath(model_file_path, "params.json"),
)
source_path = Path.joinpath(self._checkpoint_dir, "tokenizer.model")
if source_path.exists():
shutil.copy(
source_path,
Path.joinpath(model_file_path, "tokenizer.model"),
)
source_path = Path.joinpath(self._checkpoint_dir, "orig_params.json")
if source_path.exists():
shutil.copy(
source_path,
Path.joinpath(model_file_path, "orig_params.json"),
)
if not adapter_only:
model_state_dict = state_dict[training.MODEL_KEY]

7
llama_stack/providers/inline/post_training/torchtune/common/utils.py
View file

@ -29,8 +29,9 @@ from torchtune.data._messages import (
ShareGPTToMessages,
)
from torchtune.models.llama3 import llama3_tokenizer, lora_llama3_8b
from torchtune.models.llama3 import llama3_tokenizer
from torchtune.models.llama3._tokenizer import Llama3Tokenizer
from torchtune.models.llama3_1 import lora_llama3_1_8b
from torchtune.models.llama3_2 import lora_llama3_2_3b
from torchtune.modules.transforms import Transform
@ -63,8 +64,8 @@ MODEL_CONFIGS: Dict[str, ModelConfig] = {
tokenizer_type=llama3_tokenizer,
checkpoint_type="LLAMA3_2",
),
"Llama-3-8B-Instruct": ModelConfig(
model_definition=lora_llama3_8b,
"Llama3.1-8B-Instruct": ModelConfig(
model_definition=lora_llama3_1_8b,
tokenizer_type=llama3_tokenizer,
checkpoint_type="LLAMA3",
),