# Dataset map method - how to pass argument to the function
**URL:**
**Author:**  [@sssingh](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fsssingh)%5C
**Post date:** [March 30, 2022, 6:36pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fdataset-map-method-how-to-pass-argument-to-the-function%2F16274%2F1 "2022-03-30T18:36:37Z")
Hi, just started using the Huggingface library. I am wondering how can I pass `model` and `tokenizer` to my processing function along with the batch when using the `map` method.
def my\_processing\_func(batch, model, tokenizer):
–code–
I am using map like this…
new\_dataset = my\_dataset.map(my\_processing\_func, model, tokenizer, batched=True)
when I do this it does not fail but instead of passing the dictionary with input\_ids and attention\_mask, it passes a `list` of just input\_ids as the batch to my\_processing\_func. When I remove the model and tokenizer argument then it sends the dictionary as expected.
Where am I going wrong?
Thanks in advance.
---
**Author:**  [@mariosasko](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmariosasko)%5C
**Post date:** [March 31, 2022, 11:24am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fdataset-map-method-how-to-pass-argument-to-the-function%2F16274%2F2 "2022-03-31T11:24:25Z")
Hi! You can use `fn_kwargs` to pass the arguments to the map function:
```python
new_dataset = my_dataset.map(my_processing_func, batched=True, fn_kwargs={"model": model, "tokenizer": tokenizer})
```
Or you can use `partial`:
```python
from functools import partial
new_dataset = my_dataset.map(partial(my_processing_func, model=model, tokenizer=tokenizer), batched=True)
```
---
**Author:**  [@BramVanroy](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FBramVanroy)%5C
**Post date:** [March 31, 2022, 11:51am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fdataset-map-method-how-to-pass-argument-to-the-function%2F16274%2F3 "2022-03-31T11:51:59Z")
Is there any downside to using either options? If I remember correctly (?) lambdas are not picklable. So my assumption would be that if you do something like
```auto
new_dataset = my_dataset.map(lambda batch: my_processing_func(batch, model, tokenizer), batched=True)
```
it won’t be cached. Is that correct?
---
**Author:**  [@sssingh](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fsssingh)%5C
**Post date:** [March 31, 2022, 1:05pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fdataset-map-method-how-to-pass-argument-to-the-function%2F16274%2F4 "2022-03-31T13:05:33Z")
Super!! this works for me … thanks 🙏
---
**Author:**  [@mariosasko](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmariosasko)%5C
**Post date:** [April 5, 2022, 12:16pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fdataset-map-method-how-to-pass-argument-to-the-function%2F16274%2F5 "2022-04-05T12:16:51Z")
1. There shouldn’t be a significant difference in speed between these two approaches.
2. We use `dill`, which knows how to pickle lamdas in most situations.