Xilabs/instructmix-llama-3b

Model Card for "InstructMix Llama 3B"

Model Name: InstructMix Llama 3B

Description:

InstructMix Llama 3B is a language model fine-tuned on the InstructMix dataset using parameter-efficient fine-tuning (PEFT), using the base model "openlm-research/open_llama_3b_v2," which can be found at https://huggingface.co/openlm-research/open_llama_3b_v2.

An easy way to use InstructMix Llama 3B is via the API: https://replicate.com/ritabratamaiti/instructmix-llama-3b

Usage:

import torch
from transformers import LlamaForCausalLM, LlamaTokenizer
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from transformers import LlamaTokenizer, LlamaForCausalLM, GenerationConfig
from peft import PeftModel, PeftConfig

# Hugging Face model_path
model_path = 'openlm-research/open_llama_3b_v2'
peft_model_id = 'Xilabs/instructmix-llama-3b'
tokenizer = LlamaTokenizer.from_pretrained(model_path)
model = LlamaForCausalLM.from_pretrained(
    model_path, device_map="auto"
)
model = PeftModel.from_pretrained(model, peft_model_id)
def generate_prompt(instruction, input=None):
    if input:
        return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
{instruction}

### Input:
{input}

### Response:"""
    else:
        return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{instruction}

### Response:"""
def evaluate(
    instruction,
    input=None,
    temperature=0.5,
    top_p=0.75,
    top_k=40,
    num_beams=5,
    max_new_tokens=128,
    **kwargs,
):
    prompt = generate_prompt(instruction, input)
    inputs = tokenizer(prompt, return_tensors="pt")
    input_ids = inputs["input_ids"].to("cuda")
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        num_beams=num_beams,
        early_stopping=True,
        repetition_penalty=1.1,
        **kwargs,
    )
    with torch.no_grad():
        generation_output = model.generate(
            input_ids=input_ids,
            generation_config=generation_config,
            return_dict_in_generate=True,
            output_scores=True,
            max_new_tokens=max_new_tokens,
        )
    s = generation_output.sequences[0]
    output = tokenizer.decode(s, skip_special_tokens = True)
    #print(output)
    return output.split("### Response:")[1]

instruction = "What is the meaning of life?"
print(evaluate(instruction, num_beams=3, temperature=0.1, max_new_tokens=256))

Training procedure

The following bitsandbytes quantization config was used during training:

• 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: bfloat16

Framework versions

• PEFT 0.4.0