File size: 5,328 Bytes
3e63e40 4b4e754 3e63e40 4b4e754 3e63e40 b64a154 3e63e40 b335a0b 4b4e754 b335a0b 4b4e754 b335a0b |
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 |
# Model Overview
## Description:
The NVIDIA Llama 3.1 8B Instruct FP8 model is the quantized version of the Meta's Llama 3.1 8B Instruct model, which is an auto-regressive language model that uses an optimized transformer architecture. For more information, please check [here](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct). The NVIDIA Llama 3.1 8B Instruct FP8 model is quantized with [TensorRT Model Optimizer](https://github.com/NVIDIA/TensorRT-Model-Optimizer).
This model is ready for commercial and non-commercial use. <br>
## Third-Party Community Consideration
This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case; see link to Non-NVIDIA [(Meta-Llama-3.1-8B-Instruct) Model Card](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct).
### License/Terms of Use:
[llama3.1](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct/blob/main/LICENSE)
## Model Architecture:
**Architecture Type:** Transformers <br>
**Network Architecture:** Llama3.1 <br>
## Input:
**Input Type(s):** Text <br>
**Input Format(s):** String <br>
**Input Parameters:** Sequences <br>
**Other Properties Related to Input:** Context length up to 128K <br>
## Output:
**Output Type(s):** Text <br>
**Output Format:** String <br>
**Output Parameters:** Sequences <br>
**Other Properties Related to Output:** N/A <br>
## Software Integration:
**Supported Runtime Engine(s):** <br>
* Tensor(RT)-LLM <br>
* vLLM <br>
**Supported Hardware Microarchitecture Compatibility:** <br>
* NVIDIA Blackwell <br>
* NVIDIA Hopper <br>
* NVIDIA Lovelace <br>
**Preferred Operating System(s):** <br>
* Linux <br>
## Model Version(s):
The model is quantized with nvidia-modelopt **v0.15.1** <br>
## Datasets:
* Calibration Dataset: [cnn_dailymail](https://huggingface.co/datasets/abisee/cnn_dailymail) <br>
* Evaluation Dataset: [MMLU](https://github.com/hendrycks/test) <br>
## Inference:
**Engine:** Tensor(RT)-LLM or vLLM <br>
**Test Hardware:** H100 <br>
## Post Training Quantization
This model was obtained by quantizing the weights and activations of Meta-Llama-3.1-8B-Instruct to FP8 data type, ready for inference with TensorRT-LLM and vLLM. Only the weights and activations of the linear operators within transformers blocks are quantized. This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%. On H100, we achieved 1.3x speedup.
## Usage
To deploy the quantized checkpoint with [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM), follow the sample commands below with the TensorRT-LLM GitHub repo:
* Checkpoint convertion:
```sh
python examples/llama/convert_checkpoint.py --model_dir Llama-3.1-8B-Instruct-FP8 --output_dir /ckpt --use_fp8
```
* Build engines:
```sh
trtllm-build --checkpoint_dir /ckpt --output_dir /engine
```
* Accuracy evaluation:
1) Prepare the MMLU dataset:
```sh
mkdir data; wget https://people.eecs.berkeley.edu/~hendrycks/data.tar -O data/mmlu.tar
tar -xf data/mmlu.tar -C data && mv data/data data/mmlu
```
2) Measure MMLU:
```sh
python examples/mmlu.py --engine_dir ./engine --tokenizer_dir Llama-3.1-8B-Instruct-FP8/ --test_trt_llm --data_dir data/mmlu
```
* Throughputs evaluation:
Please refer to the [TensorRT-LLM benchmarking documentation](https://github.com/NVIDIA/TensorRT-LLM/blob/main/benchmarks/Suite.md) for details.
## Evaluation
The accuracy (MMLU, 5-shot) and throughputs (tokens per second, TPS) benchmark results are presented in the table below:
<table>
<tr>
<td><strong>Precision</strong>
</td>
<td><strong>MMLU</strong>
</td>
<td><strong>TPS</strong>
</td>
</tr>
<tr>
<td>FP16
</td>
<td>68.6
</td>
<td>8,579.93
</td>
</tr>
<tr>
<td>FP8
</td>
<td>68.3
</td>
<td>11,062.90
</td>
</tr>
<tr>
</table>
We benchmarked with tensorrt-llm v0.13 on 8 H100 GPUs, using batch size 1024 for the throughputs with in-flight batching enabled. We achieved **~1.3x** speedup with FP8.
### Deploy with vLLM
To deploy the quantized checkpoint with [vLLM](https://github.com/vllm-project/vllm.git), follow the instructions below:
1. Install vLLM from directions [here](https://github.com/vllm-project/vllm?tab=readme-ov-file#getting-started).
2. To use a Model Optimizer PTQ checkpoint with vLLM, `quantization=modelopt` flag must be passed into the config while initializing the `LLM` Engine.
Example deployment on H100:
```
from vllm import LLM, SamplingParams
model_id = "nvidia/Llama-3.1-8B-Instruct-FP8"
sampling_params = SamplingParams(temperature=0.8, top_p=0.9)
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
llm = LLM(model=model_id, quantization="modelopt")
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}")
```
This model can be deployed with an OpenAI Compatible Server via the vLLM backend. Instructions [here](https://docs.vllm.ai/en/latest/getting_started/quickstart.html#openai-compatible-server). |