File size: 1,738 Bytes
9d0ea1b 147404e 9d0ea1b cc65ad8 9d0ea1b f179132 9d0ea1b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
---
license: apache-2.0
language:
- en
tags:
- mamba-hf
---
# Mamba-130M
<img src="https://cdn-uploads.huggingface.co/production/uploads/63da3d7ae697e5898cb86854/A3BYIH-q7G5vz4NlsPlGJ.jpeg" width="300" height="300" alt="mamba-hf">
Mamba Models with hf_integration.
For modeling codes: [**mamba-hf**](https://github.com/LegallyCoder/mamba-hf)
# Usage:
```python
from transformers import AutoModelForCausalLM , AutoTokenizer
model = AutoModelForCausalLM.from_pretrained('Q-bert/Mamba-130M', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained('Q-bert/Mamba-130M')
text = "Hi"
input_ids = tokenizer.encode(text, return_tensors="pt")
output = model.generate(input_ids, max_length=20, num_beams=5, no_repeat_ngram_size=2)
generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
print(generated_text)
```
> Hi, I'm looking for a new job. I've been working at a company for about a year now.
# For Training:
```python
from transformers import Trainer ,TrainingArguments
import torch
import os
class MambaTrainer(Trainer):
def compute_loss(self, model, inputs, return_outputs=False):
input_ids = inputs.pop("input_ids")
lm_logits = model(input_ids)[0]
labels = input_ids.to(lm_logits.device)
shift_logits = lm_logits[:, :-1, :].contiguous()
labels = labels[:, 1:].contiguous()
loss_fct = torch.nn.CrossEntropyLoss()
lm_loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), labels.view(-1))
return lm_loss
```
You must use this class for training. And fp16 must be **False**.
# Credits:
https://huggingface.co/state-spaces
Special thanks to Albert Gu and Tri Dao for their articles. (https://arxiv.org/abs/2312.00752)
|