library_name: peft
base_model: google/gemma-2b
license: apache-2.0
datasets:
- b-mc2/sql-create-context
language:
- en
pipeline_tag: text-generation
Model Card for Model ID
This is an SFT-based (Supervised Fine-Tuned) Gemma-2B model for SQL-based tasks without applying flash-attention or using other methods libraries to reduce inference. We used LoRa(Low-Ranking Adaptors) method for Fine-Tuning.
Model Details
Model Description
This is SFT based Fine-Tuned Gemma-2B model for SQL-based tasks by providing prompts to the model in the format given below(an Example): """ Question: What is the average number of cows per farm with more than 100 acres of land? Context: CREATE TABLE farm (Cows INTEGER, Acres INTEGER) """.
Formatting (Prompting) was applied to dataset to improve training loss over time during training as well reducing basic inference speed.
- Finetuned from model : "google/gemma-2b"
Inference Code:
do the necessary imports then
device_map = {"": 0} model_id = "google/gemma-2b" new_model = "Akil15/Gemma_SQL_v.0.1"
Reload model in FP16 and merge it with LoRA weights
base_model = AutoModelForCausalLM.from_pretrained( model_id, torch_dtype=torch.float16, device_map=device_map, )
model = PeftModel.from_pretrained(base_model, new_model) model = model.merge_and_unload()
Reload tokenizer to save it
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True) tokenizer.pad_token = tokenizer.eos_token tokenizer.padding_side = "right"
text = input() inputs = tokenizer(text, return_tensors="pt").to(device) outputs = model.generate(**inputs, max_new_tokens=20) print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Framework versions
- PEFT 0.9.0