Quantization made by Richard Erkhov.

astrollama - GGUF

Model creator: https://huggingface.co/universeTBD/
Original model: https://huggingface.co/universeTBD/astrollama/

Name	Quant method	Size
astrollama.Q2_K.gguf	Q2_K	2.36GB
astrollama.IQ3_XS.gguf	IQ3_XS	2.6GB
astrollama.IQ3_S.gguf	IQ3_S	2.75GB
astrollama.Q3_K_S.gguf	Q3_K_S	2.75GB
astrollama.IQ3_M.gguf	IQ3_M	2.9GB
astrollama.Q3_K.gguf	Q3_K	3.07GB
astrollama.Q3_K_M.gguf	Q3_K_M	3.07GB
astrollama.Q3_K_L.gguf	Q3_K_L	3.35GB
astrollama.IQ4_XS.gguf	IQ4_XS	3.4GB
astrollama.Q4_0.gguf	Q4_0	3.56GB
astrollama.IQ4_NL.gguf	IQ4_NL	3.58GB
astrollama.Q4_K_S.gguf	Q4_K_S	3.59GB
astrollama.Q4_K.gguf	Q4_K	3.8GB
astrollama.Q4_K_M.gguf	Q4_K_M	3.8GB
astrollama.Q4_1.gguf	Q4_1	3.95GB
astrollama.Q5_0.gguf	Q5_0	4.33GB
astrollama.Q5_K_S.gguf	Q5_K_S	4.33GB
astrollama.Q5_K.gguf	Q5_K	4.45GB
astrollama.Q5_K_M.gguf	Q5_K_M	4.45GB
astrollama.Q5_1.gguf	Q5_1	4.72GB
astrollama.Q6_K.gguf	Q6_K	5.15GB
astrollama.Q8_0.gguf	Q8_0	6.67GB

Original model description:

license: mit datasets: - universeTBD/arxiv-astro-abstracts-all language: - en metrics: - perplexity pipeline_tag: text-generation tags: - llama-2 - astronomy - astrophysics - arxiv inference: false

AstroLLaMA

Play with the model in our Hugging Face space! https://huggingface.co/spaces/universeTBD/astrollama

Loading the model

from transformers import AutoModelForCausalLM
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained(
    pretrained_model_name_or_path="universeTBD/astrollama"
)
model = AutoModelForCausalLM.from_pretrained(
    pretrained_model_name_or_path="universeTBD/astrollama",
    device_map="auto",
)

Generating text from a prompt

import torch
from transformers import pipeline

generator = pipeline(
    task="text-generation",
    model=model,
    tokenizer=tokenizer,
    device_map="auto"
)

# Taken from https://arxiv.org/abs/2308.12823
prompt = "In this letter, we report the discovery of the highest redshift, " \
    "heavily obscured, radio-loud QSO candidate selected using JWST NIRCam/MIRI, " \
    "mid-IR, sub-mm, and radio imaging in the COSMOS-Web field. "

# For reproducibility
torch.manual_seed(42)

generated_text = generator(
    prompt,
    do_sample=True,
    max_length=512
)

Embedding text with AstroLLaMA

texts = [
    "Abstract 1",
    "Abstract 2"
]
inputs = tokenizer(
    texts,
    return_tensors="pt",
    return_token_type_ids=False,
    padding=True,
    truncation=True,
    max_length=4096
)
inputs.to(model.device)
outputs = model(**inputs, output_hidden_states=True)

# Last layer of the hidden states. Get average embedding of all tokens
embeddings = outputs["hidden_states"][-1][:, 1:, ...].mean(1).detach().cpu().numpy()