# Trainer() doesnt print any output or eval during training

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftrainer-doesnt-print-any-output-or-eval-during-training%2F69971%3E%5C
**Category:** Beginners\
**Created:** [January 21, 2024, 1:26pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftrainer-doesnt-print-any-output-or-eval-during-training%2F69971 "2024-01-21T13:26:22Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![orko](https://avatars.discourse-cdn.com/v4/letter/o/b5a626/32.png) [@orko](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Forko)%5C
**Post date:** [January 21, 2024, 1:26pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftrainer-doesnt-print-any-output-or-eval-during-training%2F69971%2F1 "2024-01-21T13:26:22Z")

</div>

I am following the notebook to fine-tune roberta on MLM task:

> <https://github.com/huggingface/notebooks/blob/main/examples/language_modeling.ipynb>

However, when I am running trainer.train(),  
the standard output of eval metrics during training is not showed.  
Also there is nothing in wandb.  
Any ideas?  
This is the code:

> def tokenize\_function(examples):  
> return tokenizer(examples[“text”])  
> def group\_texts(examples, block\_size = 514):

> ```
> concatenated_examples = {k: sum(examples[k], []) for k in examples.keys()}
> total_length = len(concatenated_examples[list(examples.keys())[0]])
> 
> total_length = (total_length // block_size) * block_size
> 
> result = {
> k: [t[i : i + block_size] for i in range(0, total_length, block_size)]
> for k, t in concatenated_examples.items()
> }
> result["labels"] = result["input_ids"].copy()
> return result
> 
> ```
> 
> datasets = load\_dataset(‘wikitext’, ‘wikitext-2-raw-v1’)  
> model\_checkpoint = “distilroberta-base”  
> tokenizer = AutoTokenizer.from\_pretrained(model\_checkpoint, use\_fast=True)  
> tokenized\_datasets = datasets.map(tokenize\_function, batched=True, num\_proc=4, remove\_columns=[“text”])  
> model = AutoModelForMaskedLM.from\_pretrained(model\_checkpoint)  
> lm\_datasets = tokenized\_datasets.map(  
> group\_texts,  
> batched=True,  
> batch\_size=514,  
> num\_proc=4,  
> load\_from\_cache\_file=False  
> )
> 
> numBatch = 8  
> numEpoch = 2  
> numStep = tokenized\_datasets.num\_rows[‘train’] \* numEpoch / numBatch  
> WANDB\_API\_KEY = “…”  
> out\_dir = “/mlm1/”  
> args = TrainingArguments(  
> run\_name=“robert\_mlm1”,  
> num\_train\_epochs=numEpoch,  
> per\_device\_train\_batch\_size=numBatch,  
> per\_device\_eval\_batch\_size=numBatch\*4,  
> gradient\_accumulation\_steps=2,  
> learning\_rate=2e-5,  
> weight\_decay=1e-2,  
> warmup\_ratio=1e-2,  
> warmup\_steps=int(numStep/100),  
> save\_total\_limit=3,  
> save\_steps=200,  
> logging\_steps=4,  
> eval\_steps=4,  
> do\_eval=True,  
> save\_strategy=‘steps’,  
> evaluation\_strategy=‘steps’,  
> output\_dir= out\_dir + “weights”,  
> logging\_dir= out\_dir + “logs”,  
> seed=23,  
> report\_to=‘wandb’,  
> load\_best\_model\_at\_end=True  
> )
> 
> trainer = Trainer(  
> model=model,  
> args=args,
> 
> ```
> train_dataset=lm_datasets["train"],
> eval_dataset=lm_datasets["validation"],
> 
> ```
> 
> )
> 
> trainer.train()