|
--- |
|
license: other |
|
language: |
|
- en |
|
pipeline_tag: text-generation |
|
datasets: |
|
- teknium/openhermes |
|
--- |
|
|
|
# Model Card for Puffin-Phi V2 |
|
|
|
Phi-1.5 fine tuned with Hermes Dataset |
|
|
|
## Model Details |
|
|
|
### Model Sources |
|
|
|
This model was trained on the OpenHermes Dataset, made by me, which is over 240,000 mostly GPT-4 generated synthetic datapoints |
|
|
|
 |
|
|
|
 |
|
|
|
## Uses |
|
|
|
Let me know! |
|
|
|
## How to Get Started with the Model |
|
|
|
Phi does not support device_map "auto", and does not seem to want to inference in fp16, so use bf16. |
|
|
|
Here is working code to inference, though it can be improved: |
|
|
|
```python |
|
import torch |
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
model = AutoModelForCausalLM.from_pretrained("teknium/Puffin-Phi-v2", trust_remote_code=True, torch_dtype=torch.bfloat16).to("cuda") |
|
tokenizer = AutoTokenizer.from_pretrained("teknium/Puffin-Phi-v2", trust_remote_code=True, torch_dtype=torch.bfloat16) |
|
inputs = tokenizer(f"### Instruction:\nWrite a negative review for the website, Twitter.\n### Response:\n", return_tensors="pt", return_attention_mask=False) |
|
outputs = model.generate(**inputs, max_length=128, do_sample=True, temperature=0.2, top_p=0.9, use_cache=True, repetition_penalty=1.2, eos_token_id=tokenizer.eos_token_id) |
|
text = tokenizer.batch_decode(outputs)[0] |
|
print(text) |
|
``` |
|
|
|
The prompt format is Alpaca, |
|
then is prompted like so: |
|
|
|
``` |
|
### Instruction: |
|
<prompt> |
|
### Response: |
|
``` |
|
|
|
## Training Details |
|
|
|
### Training Procedure |
|
|
|
Trained with Axolotl. View the wandb runs for all my puffin runs (this is puffin-phi-4 on wandb): |
|
https://wandb.ai/teknium1/hermes-phi/runs/hermes-phi-1 |
|
|
|
## Evaluation |
|
|
|
 |
|
|
