Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support save/load model for hf generate #12499

Merged
merged 3 commits into from
Dec 4, 2024
Merged

Support save/load model for hf generate #12499

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
a814d8f
change dummy model
hkvision Dec 4, 2024
e7262b4
style
hkvision Dec 4, 2024
1270a0b
meet review
hkvision Dec 4, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 20 additions & 11 deletions python/llm/src/ipex_llm/transformers/npu_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -423,23 +423,32 @@ def load_low_bit(cls, pretrained_model_name_or_path: str, *model_args, **kwargs)

if enable_cpp_backend:
from .npu_models.npu_llm_cpp import load_model_from_file
from .npu_models.convert import generate
dummy_model = torch.nn.Module()
from .npu_models.convert import generate, general_convert, prepare_input_ids, causal_lm_forward
config = AutoConfig.from_pretrained(os.path.join(pretrained_model_name_or_path, "config.json"),
trust_remote_code=trust_remote_code)
with torch.device('meta'):
model = transformers.AutoModelForCausalLM.from_config(config, trust_remote_code=trust_remote_code)
try:
model_ptr = load_model_from_file(pretrained_model_name_or_path)
dummy_model.config = PretrainedConfig.from_dict(config_dict)
dummy_model.model_ptr = model_ptr
dummy_model.save_directory = pretrained_model_name_or_path
dummy_model.kv_len = config_dict['kv_len']
dummy_model.vocab_size = config_dict['vocab_size']
model.config = PretrainedConfig.from_dict(config_dict)
Copy link
Contributor

@plusbang plusbang Dec 4, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Other LGTM, is this still necessary?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Cool, upated.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Probably not, but I see it may be used here: https://github.com/intel-analytics/ipex-llm/blob/main/python/llm/src/ipex_llm/transformers/npu_models/convert.py#L364
Maybe keep it for now.

model.model_ptr = model_ptr
model.save_directory = pretrained_model_name_or_path
model.kv_len = config_dict['kv_len']
model.vocab_size = config_dict['vocab_size']
model.logits_buffer = torch.empty(1, 1, model.vocab_size, dtype=torch.float32)
except:
invalidInputError(False,
"False to InitLLMPipeline.")
dummy_model.eval()
"Fail to InitLLMPipeline.")
model.eval()
# patch model forward
from transformers.modeling_utils import PreTrainedModel
general_convert(model, PreTrainedModel, prepare_input_ids, "prepare_inputs_for_generation")
general_convert(model, PreTrainedModel, causal_lm_forward)
# patch generate function
import types
dummy_model.generate = types.MethodType(generate, dummy_model)
return dummy_model
model.original_generate = model.generate
model.generate = types.MethodType(generate, model)
return model

has_remote_code = hasattr(config, "auto_map") and cls.HF_Model.__name__ in config.auto_map
has_local_code = type(config) in cls.HF_Model._model_mapping.keys()
Expand Down
Loading