Merge pull request #2 from soumik12345/wandb-dev

Adding wandb integration to text classification, image classification & lm finetuning

src/autotrain/cli/run_image_classification.py

@@ -203,6 +203,12 @@ def register_subcommand(parser: ArgumentParser):
                 "required": False,
                 "type": str,
             },
+            {
+                "arg": "--log-to-wandb",
+                "help": "Use Weights & Biases tracking",
+                "required": False,
+                "action": "store_true",
+            },
         ]
         run_text_classification_parser = parser.add_parser(
             "image-classification", description="✨ Run AutoTrain Image Classification"
@@ -236,6 +242,7 @@ def __init__(self, args):
             "auto_find_batch_size",
             "fp16",
             "push_to_hub",
+            "log_to_wandb"
         ]
         for arg_name in store_true_arg_names:
             if getattr(self.args, arg_name) is None:
@@ -291,6 +298,7 @@ def run(self):
                 fp16=self.args.fp16,
                 push_to_hub=self.args.push_to_hub,
                 repo_id=self.args.repo_id,
+                log_to_wandb=self.args.log_to_wandb,
             )
             params.save(output_dir=self.args.project_name)
             if self.num_gpus == 1:

src/autotrain/cli/run_llm.py

@@ -332,6 +332,13 @@ def register_subcommand(parser: ArgumentParser):
                 "action": "store_true",
                 "alias": ["--use-flash-attention-2", "--use-fa2"],
             },
+            {
+                "arg": "--log_to_wandb",
+                "help": "Use Weights & Biases tracking",
+                "required": False,
+                "action": "store_true",
+                "alias": ["--log-to-wandb"],
+            },
         ]
         run_llm_parser = parser.add_parser("llm", description="✨ Run AutoTrain LLM")
         for arg in arg_list:
@@ -372,6 +379,7 @@ def __init__(self, args):
             "use_int4",
             "merge_adapter",
             "use_flash_attention_2",
+            "log_to_wandb",
         ]
         for arg_name in store_true_arg_names:
             if getattr(self.args, arg_name) is None:
@@ -466,6 +474,7 @@ def run(self):
                 merge_adapter=self.args.merge_adapter,
                 username=self.args.username,
                 use_flash_attention_2=self.args.use_flash_attention_2,
+                log_to_wandb=self.args.log_to_wandb,
             )
 
             # space training

src/autotrain/cli/run_text_classification.py

@@ -232,6 +232,12 @@ def register_subcommand(parser: ArgumentParser):
                 "required": False,
                 "type": str,
             },
+            {
+                "arg": "--log-to-wandb",
+                "help": "Use Weights & Biases tracking",
+                "required": False,
+                "action": "store_true",
+            },
         ]
         run_text_classification_parser = parser.add_parser(
             "text-classification", description="✨ Run AutoTrain Text Classification"
@@ -265,6 +271,7 @@ def __init__(self, args):
             "auto_find_batch_size",
             "fp16",
             "push_to_hub",
+            "log_to_wandb",
         ]
         for arg_name in store_true_arg_names:
             if getattr(self.args, arg_name) is None:
@@ -326,6 +333,7 @@ def run(self):
                 repo_id=self.args.repo_id,
                 token=self.args.token,
                 username=self.args.username,
+                log_to_wandb=self.args.log_to_wandb,
             )
 
             if self.args.backend.startswith("spaces"):

src/autotrain/project.py

@@ -81,6 +81,7 @@ def _munge_common_params(self, job_idx):
         _params["repo_id"] = f"{self.username}/{self.project_name}-{job_idx}"
         _params["data_path"] = self.data_path
         _params["username"] = self.username
+        _params["log_to_wandb"] = True
         return _params
 
     def _munge_params_llm(self, job_idx):

src/autotrain/trainers/clm/__main__.py

@@ -225,6 +225,10 @@ def train(config):
 
     else:
         logging_steps = config.logging_steps
+
+    report_to = "tensorboard"
+    if config.log_to_wandb:
+        report_to = "wandb"
 
     training_args = dict(
         output_dir=config.project_name,
@@ -237,7 +241,7 @@ def train(config):
         save_total_limit=config.save_total_limit,
         save_strategy=config.save_strategy,
         gradient_accumulation_steps=config.gradient_accumulation,
-        report_to="tensorboard",
+        report_to=report_to,
         auto_find_batch_size=config.auto_find_batch_size,
         lr_scheduler_type=config.scheduler,
         optim=config.optimizer,

src/autotrain/trainers/clm/params.py

@@ -45,6 +45,7 @@ class LLMTrainingParams(BaseModel):
     merge_adapter: bool = Field(False, title="Merge adapter")
     username: str = Field(None, title="Hugging Face Username")
     use_flash_attention_2: bool = Field(False, title="Use flash attention 2")
+    log_to_wandb: bool = Field(False, title="Logging using Weights & Biases")
 
     def save(self, output_dir):
         os.makedirs(output_dir, exist_ok=True)

src/autotrain/trainers/image_classification.py

@@ -233,6 +233,10 @@ def train(co2_tracker, payload, huggingface_token, model_path):
     fp16 = True
     if device == "cpu":
         fp16 = False
+
+    report_to = "none"
+    if job_config.log_to_wandb:
+        report_to = "wandb"
 
     training_args = dict(
         output_dir=model_path,
@@ -248,7 +252,7 @@ def train(co2_tracker, payload, huggingface_token, model_path):
         save_strategy="epoch",
         disable_tqdm=not bool(os.environ.get("ENABLE_TQDM", 0)),
         gradient_accumulation_steps=job_config.gradient_accumulation_steps,
-        report_to="none",
+        report_to=report_to,
         auto_find_batch_size=True,
         lr_scheduler_type=job_config.scheduler,
         optim=job_config.optimizer,

src/autotrain/trainers/image_classification/__main__.py

@@ -98,6 +98,10 @@ def train(config):
 
     else:
         logging_steps = config.logging_steps
+
+    report_to = "tensorboard"
+    if config.log_to_wandb:
+        report_to = "wandb"
 
     training_args = dict(
         output_dir=config.project_name,
@@ -111,7 +115,7 @@ def train(config):
         save_total_limit=config.save_total_limit,
         save_strategy=config.save_strategy,
         gradient_accumulation_steps=config.gradient_accumulation,
-        report_to="tensorboard",
+        report_to=report_to,
         auto_find_batch_size=config.auto_find_batch_size,
         lr_scheduler_type=config.scheduler,
         optim=config.optimizer,

src/autotrain/trainers/image_classification/params.py

@@ -30,6 +30,7 @@ class ImageClassificationParams(BaseModel):
     evaluation_strategy: str = Field("epoch", title="Evaluation strategy")
     image_column: str = Field("image", title="Image column")
     target_column: str = Field("target", title="Target column")
+    log_to_wandb: bool = Field(False, title="Logging using Weights & Biases")
 
     def __str__(self):
         data = self.dict()

src/autotrain/trainers/lm_trainer.py

@@ -387,6 +387,10 @@ def group_texts(examples):
     logging_steps = int(0.2 * len(valid_data) / job_config.train_batch_size)
     if logging_steps == 0:
         logging_steps = 1
+
+    report_to = "none"
+    if job_config.log_to_wandb:
+        report_to = "wandb"
 
     training_args = dict(
         output_dir=model_path,
@@ -400,7 +404,7 @@ def group_texts(examples):
         save_strategy="epoch",
         disable_tqdm=not bool(os.environ.get("ENABLE_TQDM", 0)),
         gradient_accumulation_steps=job_config.gradient_accumulation_steps,
-        report_to="none",
+        report_to=report_to,
         auto_find_batch_size=True,
         lr_scheduler_type=job_config.scheduler,
         optim=job_config.optimizer,

src/autotrain/trainers/text_classification.py

@@ -204,6 +204,10 @@ def train(co2_tracker, payload, huggingface_token, model_path):
     fp16 = True
     if model_config.model_type in FP32_MODELS or device == "cpu":
         fp16 = False
+
+    report_to = "none"
+    if job_config.log_to_wandb:
+        report_to = "wandb"
 
     training_args = dict(
         output_dir="/tmp/autotrain",
@@ -219,7 +223,7 @@ def train(co2_tracker, payload, huggingface_token, model_path):
         save_strategy="epoch",
         disable_tqdm=not bool(os.environ.get("ENABLE_TQDM", 0)),
         gradient_accumulation_steps=job_config.gradient_accumulation_steps,
-        report_to="none",
+        report_to=report_to,
         auto_find_batch_size=True,
         lr_scheduler_type=job_config.scheduler,
         optim=job_config.optimizer,

src/autotrain/trainers/text_classification/__main__.py

@@ -113,6 +113,10 @@ def train(config):
 
     else:
         logging_steps = config.logging_steps
+
+    report_to = "tensorboard"
+    if config.log_to_wandb:
+        report_to = "wandb"
 
     training_args = dict(
         output_dir=config.project_name,
@@ -126,7 +130,7 @@ def train(config):
         save_total_limit=config.save_total_limit,
         save_strategy=config.save_strategy,
         gradient_accumulation_steps=config.gradient_accumulation,
-        report_to="tensorboard",
+        report_to=report_to,
         auto_find_batch_size=config.auto_find_batch_size,
         lr_scheduler_type=config.scheduler,
         optim=config.optimizer,

src/autotrain/trainers/text_classification/params.py

@@ -34,6 +34,7 @@ class TextClassificationParams(BaseModel):
     repo_id: str = Field(None, title="Repo id")
     evaluation_strategy: str = Field("epoch", title="Evaluation strategy")
     username: str = Field(None, title="Hugging Face Username")
+    log_to_wandb: bool = Field(False, title="Logging using Weights & Biases")
 
     def __str__(self):
         data = self.dict()