metadata

license: apache-2.0
inference: false

MistralLite Model

MistralLite is a fine-tuned Mistral-7B-v0.1 language model, with enhanced capblities of processing long context (up to 32K tokens). By utilizing an adapted Rotary Embedding and sliding window during fine-tuning, MistralLight is able to perform signficantly better on several long context retrieve and answering tasks, while keeping the simple model structure of the original model. MistralLite is useful for applications such as long context line and topic retrieval, summarization, question-answering, and etc. MistralLite can be deployed on a single AWS g5.2x instance with Sagemaker Huggingface Text Generation Inference (TGI) endpoint, making it suitable for applications that require high performance in resource-constrained environments. You can also serve the MistralLite model directly using TGI docker containers. Also, MistralLite supports other ways of serving like vLLM, and you can use MistralLite in Python by using the HuggingFace transformers and FlashAttention-2 library.

MistralLight evolves from Mistral-7B-v0.1, and their similarities and differences are summarized below:

Model	Fine-tuned on long contexts	Max context length	RotaryEmbedding adaptation	Sliding Window Size
Mistral-7B-v0.1	No	32K	rope_theta = 10000	4096
MistralLite	Yes	32K	rope_theta = 1000000	16384

Motivation of Developing MistralLite

Since the release of Mistral-7B-Instruct-v0.1, the model became increasingly popular because its strong performance on a wide range of benchmarks. But most of the benchmarks are evaluated on short context, and not much has been investigated on its performance on long context tasks. Then We evaluated Mistral-7B-Instruct-v0.1 against benchmarks that are specifically designed to assess the capabilities of LLMs in handling longer context. Although the performance of the models on long context was fairly competitive on long context less than 4096 tokens, there were some limitations on its performance on longer context. Motivated by improving its performance on longer context, we finetuned the Mistral 7B model, and got Mistrallite. The model managed to signifantly boost the performance of long context handling over Mistral-7B-Instruct-v0.1. The detailed long context evalutaion results are as below:

Topic Retrieval

Model Name	Input length	Input length	Input length	Input length	Input length
	2851	5568	8313	11044	13780
Mistral-7B-Instruct-v0.1	90%	0%	0%	0%	0%
MistralLite	100%	100%	100%	100%	98%

Line Retrieval

Model Name	Input length	Input length	Input length	Input length	Input length	Input length
	3818	5661	7505	9354	11188	12657
Mistral-7B-Instruct-v0.1	98%	62%	42%	42%	32%	30%
MistralLite	98%	92%	88%	76%	70%	60%

Pass key Retrieval

Model Name	Input length	Input length	Input length	Input length
	3264	5396	8329	10197
Mistral-7B-Instruct-v0.1	100%	50%	20%	30%
MistralLite	100%	100%	100%	100%

Question Answering with Long Input Texts

Model Name	Test set Accuracy	Hard subset Accuracy
Mistral-7B-Instruct-v0.1	44.3%	39.7%
MistralLite	64.4%	56.2%

Model Details

Developed by: AWS Contributors
Model type: Mistral-7B-v0.1
Language: English
Finetuned from weights: Mistral-7B-v0.1
Finetuned on data:
Supported Serving Framework:
Model License: Apache 2.0
Contact: GitHub issues

How to Use MistralFlite from Python Code

Install the necessary packages

Requires: transformers 4.34.0 or later, and flash-attn 2.3.1.post1 or later.

pip install transformers==4.34.0
pip install flash-attn==2.3.1.post1 --no-build-isolation

You can then try the following example code

from transformers import AutoModelForCausalLM, AutoTokenizer
import transformers
import torch

model_id = "amazon/MistralLite"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id,
                                             torch_dtype=torch.bfloat16,
                                             use_flash_attention_2=True,
                                             device_map="auto",)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
)
prompt = "<|prompter|>What are the main challenges to support a long context for LLM?</s><|assistant|>"

sequences = pipeline(
    prompt,
    max_new_tokens=200,
    do_sample=False,
    return_full_text=False,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id,
)
for seq in sequences:
    print(f"{seq['generated_text']}")

Important - Use the prompt template below for MistralLite:

<|prompter|>What are the main challenges to support a long context for LLM?</s><|assistant|>

How to Deploy MistralFlite on Amazon SageMaker

Install the necessary packages

Requires: sagemaker 2.192.1 or later.

pip install sagemaker==2.192.1

Deploy the Model as A SageMaker Endpoint

To deploy MistralLite on a SageMaker endpoint, please follow the example code as below.

import sagemaker
from sagemaker.huggingface import HuggingFaceModel, get_huggingface_llm_image_uri
import time

sagemaker_session = sagemaker.Session()
region = sagemaker_session.boto_region_name
role = sagemaker.get_execution_role()

image_uri = get_huggingface_llm_image_uri(
  backend="huggingface", # or lmi
  region=region,
 version="1.1.0"
)

model_name = "MistralLite-" + time.strftime("%Y-%m-%d-%H-%M-%S", time.gmtime())

hub = {
    'HF_MODEL_ID':'amazon/MistralLite',
    'HF_TASK':'text-generation',
    'SM_NUM_GPUS':'1',
    'HF_MODEL_QUANTIZE':'true'
}

model = HuggingFaceModel(
    name=model_name,
    env=hub,
    role=role,
    image_uri=image_uri
)
predictor = model.deploy(
  initial_instance_count=1,
  instance_type="ml.g5.2xlarge",
  endpoint_name=model_name
)

Perform Inference

To call the endpoint, please follow the example code as below:

input_data = {
  "inputs": "<|prompter|>What are the main challenges to support a long context for LLM?</s><|assistant|>",
  "parameters": {
    "do_sample": False,
    "max_new_tokens": 100,
  }
}
predictor.predict(input_data)

or via boto3, and the example code is shown as below:

import boto3
import json
def call_endpoint(client, prompt, endpoint_name, paramters):
    client = boto3.client("sagemaker-runtime")
    payload = {"inputs": prompt,
               "parameters": parameters}
    response = client.invoke_endpoint(EndpointName=endpoint_name,
                                      Body=json.dumps(payload), 
                                      ContentType="application/json")
    output = json.loads(response["Body"].read().decode())
    result = output[0]["generated_text"]
    return result

client = boto3.client("sagemaker-runtime")
parameters = {
        "max_new_tokens": 250,
        "do_sample": True,
        "temperature": None,
        "use_cache": True,
        "seed": 1,
}
endpoint_name = "your-endpoint-name-here""
prompt = "<|prompter|>What are the main challenges to support a long context for LLM?</s><|assistant|>"
result = call_endpoint(client, prompt, endpoint_name, paramters)
print(result)

How to Serve MistralFlite on TGI

Start TGI server

Use TGI version 1.1.0 or later. The official Docker container is: ghcr.io/huggingface/text-generation-inference:1.1.0

Example Docker parameters:

--model-id amazon/MistralLite --port 3000 --max-input-length 8192 --max-total-tokens 16384 --max-batch-prefill-tokens 16384

Perform Inference

Example Python code for inference with TGI (requires huggingface-hub 0.17.0 or later):

pip3 install huggingface-hub==0.17.0

from huggingface_hub import InferenceClient

endpoint_url = "https://your-endpoint-url-here"

prompt = "<|prompter|>What are the main challenges to support a long context for LLM?</s><|assistant|>"

client = InferenceClient(endpoint_url)
response = client.text_generation(prompt,
                                  max_new_tokens=100,
                                  do_sample=False,
                                  temperature=None,
                                  )

print(f"Model output: {response}")

Important - When using MistralLite for inference for the first time, it may require a brief 'warm-up' period that can take 10s of seconds. However, subsequent inferences should be faster and return results in a more timely manner. This warm-up period is normal and should not affect the overall performance of the system once the initialisation period has been completed.

How to Serve MistralFlite on vLLM

Documentation on installing and using vLLM can be found here.

Using vLLM as a server

When using vLLM as a server, pass the --model amazon/MistralLite parameter, for example:

python3 python -m vllm.entrypoints.api_server --model amazon/MistralLite

Using vLLM in Python Code

When using vLLM from Python code, Please see the example code as below:

from vllm import LLM, SamplingParams

prompts = [
   "<|prompter|>What are the main challenges to support a long context for LLM?</s><|assistant|>",
]
sampling_params = SamplingParams(temperature=0, max_tokens=100)

llm = LLM(model="amazon/MistralLite",)

outputs = llm.generate(prompts, sampling_params)

# Print the outputs.
for output in outputs:
    prompt = output.prompt
    generated_text = output.outputs[0].text
    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")

Limitations

Before using the MistralLite model, it is important to perform your own independent assessment, and take measures to ensure that your use would comply with your own specific quality control practices and standards, and that your use would comply with the local rules, laws, regulations, licenses and terms that apply to you, and your content.