File size: 3,039 Bytes
213596c
 
 
 
 
 
 
 
9602f6a
213596c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
datasets:
- gsm8k
tags:
- deepsparse
---
# mpt-7b-gsm8k-pruned75-quant

**Paper**: [Sparse Finetuning for Inference Acceleration of Large Language Models](https://arxiv.org/abs/2310.06927)  
**Code**: https://github.com/neuralmagic/deepsparse/tree/main/research/mpt 

This model was produced from a [MPT-7B base model](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-pt) finetuned on the GSM8k dataset with pruning applied using [SparseGPT](https://arxiv.org/abs/2301.00774) and retrain for 4 epochs with L2 distillation. Then it was exported for optimized inference with [DeepSparse](https://github.com/neuralmagic/deepsparse/tree/main/research/mpt).

GSM8k zero-shot accuracy with [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness) : 26.61% (FP32 baseline is 28.2%)

### Usage

```python
from deepsparse import TextGeneration
model_path = "hf:neuralmagic/mpt-7b-gsm8k-pruned75-quant" # or use a sparsezoo stub (zoo:mpt-7b-gsm8k_mpt_pretrain-pruned75_quantized)
model = TextGeneration(model=model_path)
model("There are twice as many boys as girls at Dr. Wertz's school. If there are 60 girls and 5 students to every teacher, how many teachers are there?", max_new_tokens=50)
```

All MPT model weights are available on [SparseZoo](https://sparsezoo.neuralmagic.com/?datasets=gsm8k&ungrouped=true) and CPU speedup for generative inference can be reproduced by following the instructions at [DeepSparse](https://github.com/neuralmagic/deepsparse/tree/main/research/mpt)


| Model Links                                                                                               | Compression                       |
| --------------------------------------------------------------------------------------------------------- | --------------------------------- |
| [neuralmagic/mpt-7b-gsm8k-quant](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-quant)                   | Quantization (W8A8)               |
| [neuralmagic/mpt-7b-gsm8k-pruned40-quant](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-pruned40-quant) | Quantization (W8A8) & 40% Pruning |
| [neuralmagic/mpt-7b-gsm8k-pruned50-quant](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-pruned50-quant) | Quantization (W8A8) & 50% Pruning |
| [neuralmagic/mpt-7b-gsm8k-pruned60-quant](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-pruned60-quant) | Quantization (W8A8) & 60% Pruning |
| [neuralmagic/mpt-7b-gsm8k-pruned70-quant](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-pruned70-quant) | Quantization (W8A8) & 70% Pruning |
| [neuralmagic/mpt-7b-gsm8k-pruned70-quant](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-pruned75-quant) | Quantization (W8A8) & 75% Pruning |
| [neuralmagic/mpt-7b-gsm8k-pruned80-quant](https://huggingface.co/neuralmagic/mpt-7b-gsm8k-pruned80-quant) | Quantization (W8A8) & 80% Pruning |


For general questions on these models and sparsification methods, reach out to the engineering team on our [community Slack](https://join.slack.com/t/discuss-neuralmagic/shared_invite/zt-q1a1cnvo-YBoICSIw3L1dmQpjBeDurQ).