Skip to content

A programming framework for agentic AI

remove learning_rate and weight_decay (#113)

Xueqing Liu6133db84e8a5
diff --git a/flaml/nlp/autotransformers.py b/flaml/nlp/autotransformers.py
index 667d129ab..29f979edc 100644
--- a/flaml/nlp/autotransformers.py
+++ b/flaml/nlp/autotransformers.py
@@ -400,8 +400,8 @@ class AutoTransformers:
)
trainer = TrainerForAutoTransformers(
- this_model,
- training_args,
+ model=this_model,
+ args=training_args,
model_init=model_init,
train_dataset=self.train_dataset,
eval_dataset=self.eval_dataset,
diff --git a/flaml/nlp/result_analysis/azure_utils.py b/flaml/nlp/result_analysis/azure_utils.py
index 2232472c3..b87de3d7d 100644
--- a/flaml/nlp/result_analysis/azure_utils.py
+++ b/flaml/nlp/result_analysis/azure_utils.py
@@ -106,8 +106,6 @@ class JobID:
rep: int = field(default=0)
sddt: int = field(default=None)
sdhf: int = field(default=None)
- var1: Optional[float] = field(default=None)
- var2: Optional[float] = field(default=None)
def __init__(self,
console_args=None):
@@ -132,8 +130,6 @@ class JobID:
self.rep = 0
self.sddt = 43
self.sdhf = 42
- self.var1 = None
- self.var2 = None
def is_match(self, partial_jobid):
"""
@@ -344,8 +340,6 @@ class JobID:
"rep_id": "rep",
"seed_data": "sddt",
"seed_transformers": "sdhf",
- "learning_rate": "var1",
- "weight_decay": "var2"
}
for each_key in console_to_jobid_key_mapping.keys():
try:
diff --git a/flaml/nlp/utils.py b/flaml/nlp/utils.py
index ab3cf895d..ee4cb40d7 100644
--- a/flaml/nlp/utils.py
+++ b/flaml/nlp/utils.py
@@ -56,10 +56,8 @@ def load_dft_args():
arg_parser.add_argument('--round_idx', type=int, help='round idx for acl experiments', required=False, default=0)
arg_parser.add_argument('--seed_data', type=int, help='seed of data shuffling', required=False, default=43)
arg_parser.add_argument('--seed_transformers', type=int, help='seed of transformers', required=False, default=42)
- arg_parser.add_argument('--learning_rate', type=float, help='optional arg learning_rate', required=False)
- arg_parser.add_argument('--weight_decay', type=float, help='optional arg weight_decay', required=False)
- args, unknown = arg_parser.parse_known_args()
- return args
+ console_args, unknown = arg_parser.parse_known_args()
+ return console_args
def merge_dicts(dict1, dict2):
diff --git a/test/hf/test_cover_azure.py b/test/hf/test_cover_azure.py
index ce9a05ea0..eb819a52f 100644
--- a/test/hf/test_cover_azure.py
+++ b/test/hf/test_cover_azure.py
@@ -43,7 +43,7 @@ def test_get_configblob_from_partial_jobid():
from flaml.nlp import JobID
each_blob_name = "dat=glue_subdat=cola_mod=grid_spa=cus_arg=dft_alg=grid" \
"_pru=None_pre=deberta_presz=large_spt=rspt_rep=0_sddt=43" \
- "_sdhf=42_var1=1e-05_var2=0.0.json"
+ "_sdhf=42.json"
partial_jobid = JobID()
partial_jobid.pre = "deberta"
partial_jobid.mod = "grid"
diff --git a/test/hf/test_mobilebert.py b/test/hf/test_mobilebert.py
index 8cde132a2..7c5f90d5b 100644
--- a/test/hf/test_mobilebert.py
+++ b/test/hf/test_mobilebert.py
@@ -24,7 +24,6 @@ def get_autohf_settings():
"time_budget": 100000,
"ckpt_per_epoch": 1,
"fp16": False,
- "ray_local_mode": True
}
return autohf_settings