GPT-2 Medium

Model Details

Model Description: GPT-2 Medium is the 355M parameter version of GPT-2, a transformer-based language model created and released by OpenAI. The model is a pretrained model on English language using a causal language modeling (CLM) objective.

Parameter-Efficient Fine-tuning (PEFT)

Parameter-Efficient Fine-tuning (PEFT) is a technique used to improve the performance of pre-trained language models (LLMs) on specific downstream tasks without fine-tuning all the model's parameters. This is done by freezing most of the model's parameters and only fine-tuning a small number of parameters that are specific to the downstream task.

Training Data

the model is trained on 'b-mc2/sql-create-context' dataset upto 5000rows

Usage:

please install transformers, and peft:

!pip install transformers peft

To use the model, you can run the following:

import torch
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM,  AutoTokenizer

config = PeftConfig.from_pretrained("Naveengo/gpt2-medium-on-sql-create-context")
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, return_dict=True, load_in_8bit=False)
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)


# Load the Lora model
model = PeftModel.from_pretrained(model,"Naveengo/gpt2-medium-on-sql-create-context")

from IPython.display import display, Markdown

def make_inference(question, context):
  batch = tokenizer(f"Below is an SQL instruction that describes a task, paired with an input that provides further context. Write an SQL query that appropriately completes the request using your expertise in SQL.  ### Instruction: {question}### Input: {context}### Response:", return_tensors='pt')
  
  with torch.cuda.amp.autocast():
    output_tokens = model.generate(**batch, max_new_tokens=100)

  display(Markdown((tokenizer.decode(output_tokens[0], skip_special_tokens=True))))  
#give question and context to function
make_inference(your_question_here, your_context_here)

Training procedure

The following bitsandbytes quantization config was used during training:

quant_method: bitsandbytes
load_in_8bit: False
load_in_4bit: True
llm_int8_threshold: 6.0
llm_int8_skip_modules: None
llm_int8_enable_fp32_cpu_offload: False
llm_int8_has_fp16_weight: False
bnb_4bit_quant_type: nf4
bnb_4bit_use_double_quant: True
bnb_4bit_compute_dtype: float16