RichardErkhov/aloobun_-_TinyAiroboros-2.2.1-awq

Quantization made by Richard Erkhov.

TinyAiroboros-2.2.1 - AWQ

Model creator: https://huggingface.co/aloobun/
Original model: https://huggingface.co/aloobun/TinyAiroboros-2.2.1/

Original model description:

datasets: - jondurbin/airoboros-2.2.1 tags: - airoboros - tinyllama

This model is a fine-tuned version of PY007/TinyLlama-1.1B-Chat-v0.3 (finetuned on 15k rows of airoboros-2.2.1 dataset)

lm-eval

|    Task     |Version| Metric |Value |   |Stderr|
|-------------|------:|--------|-----:|---|-----:|
|arc_challenge|      0|acc     |0.2671|±  |0.0129|
|             |       |acc_norm|0.2850|±  |0.0132|
|arc_easy     |      0|acc     |0.5673|±  |0.0102|
|             |       |acc_norm|0.5109|±  |0.0103|
|boolq        |      1|acc     |0.6040|±  |0.0086|
|hellaswag    |      0|acc     |0.4155|±  |0.0049|
|             |       |acc_norm|0.5420|±  |0.0050|
|openbookqa   |      0|acc     |0.2200|±  |0.0185|
|             |       |acc_norm|0.3420|±  |0.0212|
|piqa         |      0|acc     |0.7057|±  |0.0106|
|             |       |acc_norm|0.6970|±  |0.0107|
|winogrande   |      0|acc     |0.5714|±  |0.0139|

Usage:

from transformers import AutoTokenizer
import transformers
import torch

model = "aloobun/TinyAiroboros-2.2.1"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

prompt = "Write a short story about a dystopian society."

sequences = pipeline(
    f'[INST] {prompt} [/INST]',
    do_sample=True,
    top_k=10,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id,
    max_length=1024,
)
for seq in sequences:
    print(f"Result: {seq['generated_text']}")