File size: 12,067 Bytes
46a0f9c 9820fad 46a0f9c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 |
---
license: apache-2.0
inference: false
---
# Mistral-7b-300k-gguf models
Since only two formats are useful, I have converted model into those formats only.
# MegaBeam-Mistral-7B-300k Model
MegaBeam-Mistral-7B-300k is a fine-tuned [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) language model that supports input contexts up to 320k tokens. MegaBeam-Mistral-7B-300k can be deployed on a single AWS `g5.48xlarge` instance using serving frameworks such as [vLLM](https://github.com/vllm-project/vllm), Sagemaker [DJL](https://docs.aws.amazon.com/sagemaker/latest/dg/deploy-models-frameworks-djl-serving.html) endpoint, and others. Similarities and differences beween MegaBeam-Mistral-7B-300k and [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) are summarized below:
|Model|Max context length| rope_theta| prompt template|
|----------|-------------:|------------:|------------:|
| [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) | 32K | 1e6 | [instruction format](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2#instruction-format)|
| MegaBeam-Mistral-7B-300k | 320K | 25e6 | AS ABOVE|
## Evaluations
**[InfiniteBench: Extending Long Context Evaluation Beyond 100K Tokens](https://github.com/OpenBMB/InfiniteBench)**
_InfiniteBench is a cutting-edge benchmark tailored for evaluating the capabilities of language models to process, understand, and reason over super long contexts (100k+ tokens)_. We therefore evaluated MegaBeam-Mistral-7B-300k, [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2), [Llama-3-8B-Instruct-262k](https://huggingface.co/gradientai/Llama-3-8B-Instruct-262k), and [Llama3-70B-1M](https://huggingface.co/gradientai/Llama-3-70B-Instruct-Gradient-1048k) on InfiniteBench. The InfiniteBench authors also evaluated SOTA proprietary and open-source LLMs on InfiniteBench. We thus combined both results in the table below.
| Task Name | MegaBeam-Mistral-7B-300k | Mistral-7B-Instruct-v0.2 | Llama-3-8B-Instruct-262k | Llama3-70B-1M | GPT-4-1106-preview | YaRN-Mistral-7B | Kimi-Chat | Claude 2 | Yi-6B-200K | Yi-34B-200K | Chatglm3-6B-128K |
| ---------------- | ---------------- | ---------------- | ---------------- | ---------------- | ------ | --------------- | --------- | -------- | -----------| -----------| -----------|
| Retrieve.PassKey | 100% | 75.76% | 98.30% | 81.35% | 100% | 92.71% | 98.14% | 97.80% | 100.00% | 100.00% | 92.20% |
| Retrieve.Number | 96.10% | 25.25% | 97.79% | 97.62% | 100% | 56.61% | 95.42% | 98.14% | 94.92% | 100.00% | 80.68% |
| Retrieve.KV | 0% | 0% | 3.40% | 3% | 89.00% | < 5% | 53.60% | 65.40% | < 5% | < 5% | < 5% |
| En.Sum | 29.39% | 22.13% | 16.40% | 20.72% | 14.73% | 9.09% | 17.93% | 14.45% | < 5% | < 5% |< 5% |
| En.QA | 14.93% | 4.93% | 13.20% | 16.52% | 22.22% | 9.55% | 16.52% | 11.97% | 9.20% | 12.17% |< 5% |
| En.MC | 51.52% | 7.80% | 50.65% | 62% | 67.25% | 27.95% | 72.49% | 62.88% | 36.68% |38.43% |10.48% |
| En.Dia | 9.50% | 3.50% | 1% | 12.50% | 8.50% | 7.50% | 11.50% | 46.50% | < 5% |< 5% |< 5% |
| Zh.QA | 10.71% | 3.43% | 19.02% | 26% | 25.96% | 14.43% | 17.93% | 9.64% | 15.07% |13.61% |< 5% |
| Code.Debug | 27.41% | 11.60% | 22.08% | 23.85% | 39.59% | < 5% | 18.02% | < 5% | < 5% |< 5% |< 5% |
| Code.Run | 1.75% | 0.25% | 0% | 0% | 23.25% | < 5% | < 5% | < 5% | < 5% |< 5% |< 5% |
| Math.Calc | 0% | 0% | 0% | 0% | < 5% | < 5% | < 5% | < 5% | < 5% |< 5% |< 5% |
| Math.Find | 24.28% | 26.28% | 15.40% | 30% | 60.00% | 17.14% | 12.57% | 32.29% | < 5% |25.71% |7.71% |
| **Average** | 30.70% | 15.08% | 28.10% | 31.13% | 46.08% | 20.41% | 34.93% | 37.21% | 22.78% |25.41% |17.59% |
The 12 evaluation tasks are summarized below (as per [InfiniteBench]((https://github.com/OpenBMB/InfiniteBench)))
| Task Name | Context | # Examples | Avg Input Tokens | Avg Output Tokens | Description |
| -------------------- | ------------- | ---------- | ---------------- | ----------------- | ------------------------------------------------------------------------------------------- |
| En.Sum | Fake Book | 103 | 171.5k | 1.1k | Summarization of a fake book created with core entity substitution. |
| En.QA | Fake Book | 351 | 192.6k | 4.8 | Free-form question answering based on the fake book. |
| En.MC | Fake Book | 229 | 184.4k | 5.3 | Multiple choice questions derived from the fake book. |
| En.Dia | Script | 200 | 103.6k | 3.4 | Identification of talkers in partially anonymized scripts. |
| Zh.QA | New Book | 175 | 2068.6k | 6.3 | Question answering on a set of newly collected books. |
| Code.Debug | Code Document | 394 | 114.7k | 4.8 | Finding which function in a code repo contains an crashing error (in multiple choice form). |
| Code.Run | Synthetic | 400 | 75.2k | 1.3 | Simulating execution of multiple simple, synthetic functions. |
| Math.Calc | Synthetic | 50 | 43.9k | 43.9k | Calculations involving super-long arithmetic equations. |
| Math.Find | Synthetic | 350 | 87.9k | 1.3 | Finding special integers in a lengthy list. |
| Retrieve.PassKey | Synthetic | 590 | 122.4k | 2.0 | Retrieving hidden keys in a noisy long context. |
| Retrieve.Number | Synthetic | 590 | 122.4k | 4.0 | Locating repeated hidden numbers in a noisy long context. |
| Retrieve.KV | Synthetic | 500 | 89.9k | 22.7 | Finding the corresponding value from a dictionary and a key. |
## Serve MegaBeam-Mistral-7B-300k on EC2 instances ##
On an AWS `g5.48xlarge` instance, upgrade vLLM to the latest version as per [documentation on vLLM](https://vllm.readthedocs.io/en/latest/).
### Start the server
```shell
python3 -m vllm.entrypoints.openai.api_server --model amazon/MegaBeam-Mistral-7B-300k --tensor-parallel-size 8
```
**Important Note** - We have set the `max_position_embeddings` in the [`config.json`](config.json) to 288,800 in order to fit model's KV-cache on a single `g5.48xlarge` instance, which has 8 x A10 GPUs (24GB RAM per GPU).
On an instance with larger GPU RAM (e.g. `p4d.24xlarge`), feel free to increase the value of the `max_position_embeddings`(e.g. to 350K), which the model should be able to process.
### Run the client
```python
from openai import OpenAI
# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
client = OpenAI(
# defaults to os.environ.get("OPENAI_API_KEY")
api_key=openai_api_key,
base_url=openai_api_base,
)
models = client.models.list()
model = models.data[0].id
chat_completion = client.chat.completions.create(
messages = [
{"role": "user", "content": "What is your favourite condiment?"}, # insert your long context here
{"role": "assistant", "content": "Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!"},
{"role": "user", "content": "Do you have mayonnaise recipes?"} # insert your long context here
],
model=model,
)
print("Chat completion results:")
print(chat_completion)
```
### Deploy the model on a SageMaker Endpoint ###
To deploy MegaBeam-Mistral-7B-300k on a SageMaker endpoint, please follow this [SageMaker DJL deployment guide](https://docs.djl.ai/docs/demos/aws/sagemaker/large-model-inference/sample-llm/vllm_deploy_mistral_7b.html).
Run the following Python code in a SageMaker notebook (with each block running in a separate cell)
```python
import sagemaker
from sagemaker import Model, image_uris, serializers, deserializers
sagemaker_session = sagemaker.Session()
region = sagemaker_session.boto_region_name
role = sagemaker.get_execution_role()
%%writefile serving.properties
engine=Python
option.model_id=amazon/MegaBeam-Mistral-7B-300k
option.dtype=bf16
option.task=text-generation
option.rolling_batch=vllm
option.tensor_parallel_degree=8
option.device_map=auto
%%sh
mkdir mymodel
mv serving.properties mymodel/
tar czvf mymodel.tar.gz mymodel/
rm -rf mymodel
image_uri = image_uris.retrieve(
framework="djl-deepspeed",
region=region,
version="0.27.0"
)
s3_code_prefix = "megaBeam-mistral-7b-300k/code"
bucket = sagemaker_session.default_bucket() # bucket to house artifacts
code_artifact = sagemaker_session.upload_data("mymodel.tar.gz", bucket, s3_code_prefix)
print(f"S3 Code or Model tar ball uploaded to --- > {code_artifact}")
model = Model(image_uri=image_uri, model_data=code_artifact, role=role)
instance_type = "ml.g5.48xlarge"
endpoint_name = sagemaker.utils.name_from_base("megaBeam-mistral-7b-300k")
model.deploy(initial_instance_count=1,
instance_type=instance_type,
endpoint_name=endpoint_name
)
# our requests and responses will be in json format so we specify the serializer and the deserializer
predictor = sagemaker.Predictor(
endpoint_name=endpoint_name,
sagemaker_session=sagemaker_session,
serializer=serializers.JSONSerializer(),
)
# test the endpoint
input_str = """<s>[INST] What is your favourite condiment? [/INST]
Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!</s> "
[INST] Do you have mayonnaise recipes? [/INST]"""
predictor.predict(
{"inputs": input_str, "parameters": {"max_new_tokens": 75}}
)
```
## Limitations ##
Before using the MegaBeam-Mistral-7B-300k 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.
## The AWS Contributors ##
Chen Wu, Yin Song, Verdi March, Eden Duthie |