LLMs
Large-language models are AI models that can understand and generate text, primarily using transformer architectures. This page is about running them on a HPC cluster. This requires programming experience and knowledge of using the cluster (Tutorials), but allows maximum computational power for the least cost. Aalto RSE maintains these models and can provide help with using these, even to users who aren’t computational experts.
Because the size of model weights are typically very large and the interest in the models are high, so we provide our users with pre-downloaded model weights in various format, along with instructions on how to load these weights for inference purposes, retraining, and fine-tuning tasks.
HuggingFace Models
The simplest way to use an open-source LLM(Large Language Model) is through the tools and pre-trained models hub from huggingface.
Huggingface is a popular platform for NLP(Natural Language Processing) tasks. It provides a user-friendly interface through the transformers library to load and run various pre-trained models.
Most open-source models from Huggingface are widely supported and integrated with the transformers library.
We are keeping our eyes on the latest models and have downloaded some of them for you.
The full list of all the available models are located at /scratch/shareddata/dldata/huggingface-hub-cache/models.txt
. Please contact us if you need any other models.
The following table lists only a few example from the hosted models:
Model type |
Huggingface model identifier |
---|---|
Text Generation |
meta-llama/Meta-Llama-3-8B |
Text Generation |
meta-llama/Meta-Llama-3-8B-Instruct |
Text Generation |
mistralai/Mixtral-8x22B-v0.1 |
Text Generation |
mistralai/Mixtral-8x22B-Instruct-v0.1 |
Text Generation |
tiiuae/falcon-40b |
Text Generation |
tiiuae/falcon-40b-instruct |
Text Generation |
google/gemma-2b-it |
Text Generation |
google/gemma-7b |
Text Generation |
google/gemma-7b-it |
Text Generation |
google/gemma-7b |
Text Generation |
LumiOpen/Poro-34B |
Text Generation |
meta-llama/Llama-2-7b-hf |
Text Generation |
meta-llama/Llama-2-13b-hf |
Text Generation |
meta-llama/Llama-2-70b-hf |
Text Generation |
codellama/CodeLlama-7b-hf |
Text Generation |
codellama/CodeLlama-13b-hf |
Text Generation |
codellama/CodeLlama-34b-hf |
Translation |
Helsinki-NLP/opus-mt-en-fi |
Translation |
Helsinki-NLP/opus-mt-fi-en |
Translation |
t5-base |
Fill Mask |
bert-base-uncased |
Fill Mask |
bert-base-cased |
Fill Mask |
distilbert-base-uncased |
Text to Speech |
microsoft/speecht5_hifigan |
Text to Speech |
facebook/hf-seamless-m4t-large |
Automatic Speech Recognition |
openai/whisper-large-v3 |
Token Classification |
dslim/bert-base-NER-uncased |
To access Huggingface models:
Load the module to setup the environment variable HF_HOME:
module load model-huggingface/all
# this will set HF_HOME to /scratch/shareddata/dldata/huggingface-hub-cache
In jupyter notebook, one can set up HF_HOME directly:
import os
os.environ['TRANSFORMERS_OFFLINE'] = '1'
os.environ['HF_HOME']='/scratch/shareddata/dldata/huggingface-hub-cache'
Here is a Python script using huggingface model.
## Force transformer to load model(s) from local hub instead of download and load model(s) from remote hub. NOTE: this must be run before importing transformers.
import os
os.environ['TRANSFORMERS_OFFLINE'] = '1'
from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1")
prompt = "How many stars in the space?"
model_inputs = tokenizer([prompt], return_tensors="pt")
input_length = model_inputs.input_ids.shape[1]
generated_ids = model.generate(**model_inputs, max_new_tokens=20)
print(tokenizer.batch_decode(generated_ids[:, input_length:], skip_special_tokens=True)[0])
Raw model weights
We also downloaded the following raw llama model weights (PyTorch model checkpoints), and they are managed by the following modules.
Model type |
Model version |
Module command to load |
Description |
---|---|---|---|
Llama 2 |
Raw Data |
|
Raw weights of Llama 2. |
Llama 2 |
7b |
|
Raw weights of 7B parameter version of Llama 2. |
Llama 2 |
7b-chat |
|
Raw weights of 7B parameter chat optimized version of Llama 2. |
Llama 2 |
13b |
|
Raw weights of 13B parameter version of Llama 2. |
Llama 2 |
13b-chat |
|
Raw weights of 13B parameter chat optimized version of Llama 2. |
Llama 2 |
70b |
|
Raw weights of 70B parameter version of Llama 2. |
Llama 2 |
70b-chat |
|
Raw weights of 70B parameter chat optimized version of Llama 2. |
CodeLlama |
Raw Data |
|
Raw weights of CodeLlama. |
CodeLlama |
7b |
|
Raw weights of 7B parameter version of CodeLlama. |
CodeLlama |
7b-Python |
|
Raw weights of 7B parameter version CodeLlama, specifically designed for Python. |
CodeLlama |
7b-Instruct |
|
Raw weights of 7B parameter version CodeLlama, designed for instruction following. |
CodeLlama |
13b |
|
Raw weights of 13B parameter version of CodeLlama. |
CodeLlama |
13b-Python |
|
Raw weights of 13B parameter version CodeLlama, specifically designed for Python. |
CodeLlama |
13b-Instruct |
|
Raw weights of 13B parameter version CodeLlama, designed for instruction following. |
CodeLlama |
34b |
|
Raw weights of 34B parameter version of CodeLlama. |
CodeLlama |
34b-Python |
|
Raw weights of 34B parameter version CodeLlama, specifically designed for Python. |
CodeLlama |
34b-Instruct |
|
Raw weights of 34B parameter version CodeLlama, designed for instruction following. |
Each module will set the following environment variables:
MODEL_ROOT
- Folder where model weights are stored, i.e., PyTorch model checkpoint directory.TOKENIZER_PATH
- File path to the tokenizer.model.
Here is an example slurm script, using the raw weights for batch inference. For detailed environment setting up, example prompts and Python code, please check out this repo.
#!/bin/bash
#SBATCH --time=00:25:00
#SBATCH --cpus_per_task=4
#SBATCH --mem=20GB
#SBATCH --gpus=1
#SBATCH --output=llama2inference-gpu.%J.out
#SBATCH --error=llama2inference-gpu.%J.err
# get access to the model weights
module load model-llama2/7b
echo $MODEL_ROOT
# Expect output: /scratch/shareddata/dldata/llama-2/llama-2-7b
echo $TOKENIZER_PATH
# Expect output: /scratch/shareddata/dldata/llama-2/tokenizer.model
# activate your conda environment
module load mamba
source activate llama2env
# run batch inference
torchrun --nproc_per_node 1 batch_inference.py \
--prompts prompts.json \
--ckpt_dir $MODEL_ROOT \
--tokenizer_path $TOKENIZER_PATH \
--max_seq_len 512 --max_batch_size 16
llama.cpp and GGUF model weights
llama.cpp is another popular framework
for running inference on LLM models with CPUs or GPUs. It provides C++ implementations of many large language models. llama.cpp uses a format called GGUF as its storage format.
We have GGUF conversions of all Llama 2 and CodeLlama models with multiple quantization levels.
Please contact us if you need any other GGUF models.
NOTE: Before loading the following modules, one must first load a module for the raw model weights. For example, run module load model-codellama/34b
first, and then run module load codellama.cpp/q8_0-2023-12-04
to get the 8-bit integer version of CodeLlama weights in a .gguf file.
Model type |
Model version |
Module command to load |
Description |
---|---|---|---|
Llama 2 |
f16-2023-08-28 |
|
Half precision version of Llama 2 weights done with llama.cpp on 4th of Dec 2023. |
Llama 2 |
q4_0-2023-08-28 |
|
4-bit integer version of Llama 2 weights done with llama.cpp on 4th of Dec 2023. |
Llama 2 |
q4_1-2023-08-28 |
|
4-bit integer version of Llama 2 weights done with llama.cpp on 4th of Dec 2023. |
Llama 2 |
q8_0-2023-08-28 |
|
8-bit integer version of Llama 2 weights done with llama.cpp on 4th of Dec 2023. |
CodeLlama |
f16-2023-08-28 |
|
Half precision version of CodeLlama weights done with llama.cpp on 4th of Dec 2023. |
CodeLlama |
q4_0-2023-08-28 |
|
4-bit integer version of CodeLlama weights done with llama.cpp on 4th of Dec 2023. |
CodeLlama |
q8_0-2023-08-28 |
|
8-bit integer version of CodeLlama weights done with llama.cpp on 4th of Dec 2023. |
Each module will set the following environment variables:
MODEL_ROOT
- Folder where model weights are stored.MODEL_WEIGHTS
- Path to the model weights in GGUF file format.
This Python code snippet is part of a ‘Chat with Your PDF Documents’ example, utilizing LangChain and leveraging model weights stored in a .gguf file. For detailed environment setting up and Python code, please check out this repo.
import os
from langchain.llms import LlamaCpp
model_path = os.environ.get('MODEL_WEIGHTS')
llm = LlamaCpp(model_path=model_path, verbose=False)
More examples
Starting a local API
With the pre-downloaded model weights, you are also able create an API endpoint locally. For detailed examples, you can checkout this repo.