Upload train_glm_qlora_v4.py with huggingface_hub
Browse files- train_glm_qlora_v4.py +3 -1
train_glm_qlora_v4.py
CHANGED
|
@@ -16,8 +16,11 @@ Agent Zero SFT: zai-org/GLM-4.7-Flash (30B MoE)
|
|
| 16 |
QLoRA (4-bit) with CPU offloading for layers that don't fit in 24GB VRAM.
|
| 17 |
"""
|
| 18 |
|
|
|
|
| 19 |
import torch
|
| 20 |
import trackio
|
|
|
|
|
|
|
| 21 |
from datasets import load_dataset
|
| 22 |
from peft import LoraConfig
|
| 23 |
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
|
@@ -36,7 +39,6 @@ bnb_config = BitsAndBytesConfig(
|
|
| 36 |
llm_int8_enable_fp32_cpu_offload=True,
|
| 37 |
)
|
| 38 |
|
| 39 |
-
import os
|
| 40 |
offload_dir = "/tmp/offload"
|
| 41 |
os.makedirs(offload_dir, exist_ok=True)
|
| 42 |
|
|
|
|
| 16 |
QLoRA (4-bit) with CPU offloading for layers that don't fit in 24GB VRAM.
|
| 17 |
"""
|
| 18 |
|
| 19 |
+
import os
|
| 20 |
import torch
|
| 21 |
import trackio
|
| 22 |
+
from huggingface_hub import login
|
| 23 |
+
login(token=os.environ["HF_TOKEN"])
|
| 24 |
from datasets import load_dataset
|
| 25 |
from peft import LoraConfig
|
| 26 |
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
|
|
|
| 39 |
llm_int8_enable_fp32_cpu_offload=True,
|
| 40 |
)
|
| 41 |
|
|
|
|
| 42 |
offload_dir = "/tmp/offload"
|
| 43 |
os.makedirs(offload_dir, exist_ok=True)
|
| 44 |
|