olafgeibig/phi-2-OpenHermes-2.5-GGUF

phi-2-OpenHermes-2.5

I converted minghaowu/phi-2-OpenHermes-2.5 to GGUF and quantized it to my favorite quantizations. A phi-2 fine-tuned on OpenHermes-2.5.

I quickly quantized this model using a modified version of AutoGGUF from Maxime Labonne

The prompt format is a little bit guesswork but it seems to work. Here is my Ollama modelfile:

FROM ./phi-2-openhermes-2.5.Q5_K_M.gguf
PARAMETER num_ctx 2048
TEMPLATE """{{ .System }}
### USER: {{ .Prompt }}<|endoftext|>
### ASSISTANT:
"""
PARAMETER stop "<|endoftext|>"

Many Kudos to Microsoft, Teknium and Minghao Wu

---

Original Modelcard

This model is a fine-tuned version of microsoft/phi-2 on the teknium/OpenHermes-2.5 dataset.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 16
• total_train_batch_size: 128
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Framework versions

• Transformers 4.37.2
• Pytorch 2.0.1+cu117
• Datasets 2.16.1
• Tokenizers 0.15.1

Inference

from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

model_id = "minghaowu/phi-2-OpenHermes-2.5"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, device_map="auto")

your_instruction = <your_instruction>
infer_prompt = f"### USER: {your_instruction} <|endoftext|>\n### ASSISTANT:"
output = pipe(infer_prompt, do_sample=True, max_new_tokens=256)[0]["generated_text"]
print(output)