Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
base_model: meta-llama/Llama-2-7b-hf
|
3 |
+
inference: true
|
4 |
+
model_type: llama
|
5 |
+
datasets:
|
6 |
+
- cerebras/SlimPajama-627B
|
7 |
+
tags:
|
8 |
+
- sparse
|
9 |
+
---
|
10 |
+
|
11 |
+
# Llama-2-7b-pruned50-retrained
|
12 |
+
|
13 |
+
This repo contains model files for a [Llama 2 7B](https://huggingface.co/meta-llama/Llama-2-7b-hf) model that has had 50% of the parameters pruned in one-shot with [SparseGPT](https://arxiv.org/abs/2301.00774), then retrained by [Cerebras](https://huggingface.co/cerebras) with 45B tokens from SlimPajama while maintaining sparsity.
|
14 |
+
|
15 |
+
**Authors**: Neural Magic, Cerebras
|
16 |
+
|
17 |
+
## Usage
|
18 |
+
|
19 |
+
Below we share some code snippets on how to get quickly started with running the model.
|
20 |
+
|
21 |
+
### Fine-tuning examples
|
22 |
+
|
23 |
+
Coming soon.
|
24 |
+
|
25 |
+
### Running the model
|
26 |
+
|
27 |
+
```python
|
28 |
+
# pip install transformers accelerate
|
29 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
30 |
+
|
31 |
+
tokenizer = AutoTokenizer.from_pretrained("neuralmagic/Llama-2-7b-pruned50-retrained")
|
32 |
+
model = AutoModelForCausalLM.from_pretrained("neuralmagic/Llama-2-7b-pruned50-retrained", device_map="auto")
|
33 |
+
|
34 |
+
input_text = "Write me a poem about Machine Learning."
|
35 |
+
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
|
36 |
+
|
37 |
+
outputs = model.generate(**input_ids)
|
38 |
+
print(tokenizer.decode(outputs[0]))
|
39 |
+
```
|
40 |
+
|
41 |
+
## Evaluation Benchmark Results
|
42 |
+
|
43 |
+
Model evaluation metrics and results.
|
44 |
+
|
45 |
+
| Benchmark | Metric | Llama-2-7b | Llama-2-7b-pruned50-retrained |
|
46 |
+
|------------------------------------------------|---------------|-------------|-------------------------------|
|
47 |
+
| [MMLU](https://arxiv.org/abs/2009.03300) | 5-shot, top-1 | xxxx | xxxx |
|
48 |
+
| [HellaSwag](https://arxiv.org/abs/1905.07830) | 0-shot | xxxx | xxxx |
|
49 |
+
| [WinoGrande](https://arxiv.org/abs/1907.10641) | partial score | xxxx | xxxx |
|
50 |
+
| [ARC-c](https://arxiv.org/abs/1911.01547) | | xxxx | xxxx |
|
51 |
+
| [TruthfulQA](https://arxiv.org/abs/2109.07958) | 5-shot | xxxx | xxxx |
|
52 |
+
| [HumanEval](https://arxiv.org/abs/2107.03374) | pass@1 | xxxx | xxxx |
|
53 |
+
| [GSM8K](https://arxiv.org/abs/2110.14168) | maj@1 | xxxx | xxxx |
|
54 |
+
| ------------------------------ | ------------- | ----------- | --------- |
|
55 |
+
| **Average** | | xxxx | xxxx |
|
56 |
+
|
57 |
+
## Model Training Data
|
58 |
+
|
59 |
+
Coming soon.
|
60 |
+
|
61 |
+
## Sparsification
|
62 |
+
|
63 |
+
This model was pruned with [SparseGPT](https://arxiv.org/abs/2301.00774), using [SparseML](https://github.com/neuralmagic/sparseml).
|