Phi-3-Goru / README.md
Kquant03's picture
Update README.md
48a6666 verified
|
raw
history blame
5.08 kB
metadata
license: mit
language:
  - en
thumbnail: >-
  https://cdn-uploads.huggingface.co/production/uploads/6589d7e6586088fd2784a12c/Wv1Nqd_2MKbBlaiORb479.png

Be one with nature.

image/png

Named after the method used to create it, interleaving the layers of its predecessor to become far larger, giving it much more potential.

Goru was an ancient treeant, and I couldn't think of a better naming convention for a model that was created using the passthrough method.

By concatenating layers from different LLMs, it can produce models with an exotic number of parameters (e.g., 9B with two 7B parameter models). These models are often referred to as "frankenmerges" or "Frankenstein models" by the community.

Many thanks to Microsoft for providing the fine tuned weights that were used in the creation of this base model. out this script.

This idea was brought to me by KatyTheCutie. I have her to thank if fine-tuning this model turns out to be a success.

How to run inference:

import transformers
import torch

if __name__ == "__main__":
  model_name = "Replete-AI/Phi-3-Goru"
  tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
  
  model = (
      transformers.AutoModelForCausalLM.from_pretrained(
          model_name,
      )
      .to("cuda:0")
      .eval()
  )
  
  messages = [
      {"role": "user", "content": "Hello, who are you?"}
  ]
  inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)
  input_ids_cutoff = inputs.size(dim=1)
  
  with torch.no_grad():
      generated_ids = model.generate(
          input_ids=inputs,
          use_cache=True,
          max_new_tokens=512,
          temperature=0.2,
          top_p=0.95,
          do_sample=True,
          eos_token_id=tokenizer.eos_token_id,
          pad_token_id=tokenizer.pad_token_id,
      )
  
  completion = tokenizer.decode(
      generated_ids[0][input_ids_cutoff:],
      skip_special_tokens=True,
  )
  
  print(completion)

Join the Replete AI Discord here!

The Sauce:

dtype: bfloat16
merge_method: passthrough
slices:
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [0,2]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [1,3]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [2,4]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [3,5]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [4,6]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [5,7]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [6,8]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [7,9]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [8,10]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [9,11]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [10,12]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [11,13]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [12,14]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [13,15]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [14,16]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [15,17]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [16,18]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [17,19]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [18,20]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [19,21]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [20,22]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [21,23]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [22,24]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [23,25]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [24,26]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [25,27]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [26,28]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [27,29]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [28,30]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [29,31]
  - sources:
    - model: microsoft/Phi-3-mini-4k-instruct
      layer_range: [30,32]