ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'main' into main

Browse Source 
Former-commit-id: 7be442f37d53a0c6324728fa1fa8e2c84d7f0fa5
This commit is contained in:
hoshi-hiyouga
2024-07-01 21:01:09 +08:00
committed by GitHub
parent 6c185a2c57 973cf8e980
commit a715490c2a
176 changed files with 4760 additions and 1322 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
src/llamafactory/model/patcher.py
View File

@@ -1,3 +1,17 @@
# Copyright 2024 the LlamaFactory team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
from types import MethodType
from typing import TYPE_CHECKING, Any, Dict
@@ -46,13 +60,16 @@ def patch_config(
is_trainable: bool,
) -> None:
if model_args.compute_dtype is None: # priority: bf16 > fp16 > fp32
model_args.compute_dtype = infer_optim_dtype(model_dtype=getattr(config, "torch_dtype", None))
if model_args.infer_dtype != "auto" and not is_trainable:
model_args.compute_dtype = getattr(torch, model_args.infer_dtype)
else:
model_args.compute_dtype = infer_optim_dtype(model_dtype=getattr(config, "torch_dtype", None))
if is_torch_npu_available():
use_jit_compile = os.environ.get("JIT_COMPILE", "0").lower() in ["true", "1"]
torch.npu.set_compile_mode(jit_compile=use_jit_compile)
configure_attn_implementation(config, model_args)
configure_attn_implementation(config, model_args, is_trainable)
configure_rope(config, model_args, is_trainable)
configure_longlora(config, model_args, is_trainable)
configure_quantization(config, tokenizer, model_args, init_kwargs)
@@ -74,14 +91,17 @@ def patch_config(
# deepspeed zero3 is not compatible with low_cpu_mem_usage
init_kwargs["low_cpu_mem_usage"] = model_args.low_cpu_mem_usage and (not is_deepspeed_zero3_enabled())
if not is_deepspeed_zero3_enabled() and not is_fsdp_enabled(): # cast dtype and device if not use zero3 or fsdp
# cast data type of the model if:
# 1. not deepspeed zero3 and not fsdp (keep zero3 or fsdp in float32)
# 2. quantization_bit is not None (qlora)
if (not is_deepspeed_zero3_enabled() and not is_fsdp_enabled()) or model_args.quantization_bit is not None:
init_kwargs["torch_dtype"] = model_args.compute_dtype
if init_kwargs["low_cpu_mem_usage"]: # device map requires low_cpu_mem_usage=True
if "device_map" not in init_kwargs and model_args.device_map:
init_kwargs["device_map"] = model_args.device_map
if init_kwargs["device_map"] == "auto":
if init_kwargs.get("device_map", None) == "auto":
init_kwargs["offload_folder"] = model_args.offload_folder
if finetune_args.stage == "sft" and data_args.efficient_packing:
@@ -137,6 +157,10 @@ def patch_valuehead_model(model: "AutoModelForCausalLMWithValueHead") -> None:
if isinstance(self.pretrained_model, PreTrainedModel):
return self.pretrained_model.get_input_embeddings()
def get_output_embeddings(self: "AutoModelForCausalLMWithValueHead") -> torch.nn.Module:
if isinstance(self.pretrained_model, PreTrainedModel):
return self.pretrained_model.get_output_embeddings()
def create_or_update_model_card(self: "AutoModelForCausalLMWithValueHead", output_dir: str) -> None:
if isinstance(self.pretrained_model, PeftModel):
self.pretrained_model.create_or_update_model_card(output_dir)
@@ -145,4 +169,5 @@ def patch_valuehead_model(model: "AutoModelForCausalLMWithValueHead") -> None:
setattr(model, "_keys_to_ignore_on_save", ignore_modules)
setattr(model, "tie_weights", MethodType(tie_weights, model))
setattr(model, "get_input_embeddings", MethodType(get_input_embeddings, model))
setattr(model, "get_output_embeddings", MethodType(get_output_embeddings, model))
setattr(model, "create_or_update_model_card", MethodType(create_or_update_model_card, model))