Phiter
Phiter is a merge of the following models using LazyMergekit:
Thanks to the great Maxime Labonne we have evaluation results on YALL.
The model tops all other phi-2 finetunes on the leaderboard, even most MoE implementations like Phixtral(Date: 27th February 2024)
License: MIT
This model wouldn't have been possible without the support of:
Maxime Labonne - he helped me troubleshoot the merge process
brittlewis12 - helped me troubleshooting the creation of GGUF files
Prompt template: ChatML
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
GGUF: Phiter-GGUF
🧩 Configuration
models:
- model: mixedbread-ai/phi-2
# no parameters necessary for base model
- model: rhysjones/phi-2-orange
parameters:
density: 0.5
weight: 0.5
- model: cognitivecomputations/dolphin-2_6-phi-2
parameters:
density: 0.5
weight: 0.3
merge_method: ties
base_model: mixedbread-ai/phi-2
parameters:
normalize: true
dtype: float16
💻 Usage
!pip install -qU transformers accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "Venkman42/Phiter"
messages = [{"role": "user", "content": "What is a large language model?"}]
tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
"text-generation",
model=model,
torch_dtype=torch.float16,
device_map="auto",
)
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])