This utility extracts the KV cache scaling factors from a quantized HF (Hugging Face) model. The extracted scaling factors are saved to a JSON file, which can later be used by vLLM (variable-length language model) during runtime. This tool is particularly useful when the KV cache data type is FP8 and is intended for use on ROCm (hcu) platforms.
## Prerequisites
- Python 3.x
- PyTorch
- NumPy
- Hugging Face Transformers
- Hugging Face Hub
- AMMO
Before incorporating the FP8 datatype for inference workloads, you must adhere to the following steps:
1. Install all necessary prerequisites and dependencies.
2. Convert HF model into a quantized HF model.
3. Extract KV Cache Scaling Factors from quantized HF model.
4. Load KV Cache Scaling Factors into VLLM.
### 2. Convert HF model into a quantized HF model.
Note: The following steps are adapted from the [TensorRT-LLM repository](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/quantization/README.md).
`quantize.py` (examples/fp8/quantizer/quantize.py) uses the quantization toolkit (AMMO) to calibrate the PyTorch models and export TensorRT-LLM checkpoints. Each TensorRT-LLM checkpoint contains a config file (in .json format) and one or several rank weight files (in .safetensors format).
The detailed quantization toolkit (AMMO) conversion guide for FP8 can be found at `examples/fp8/quantizer/README.md`.
`extract_scales.py` (examples/fp8/extract_scales.py) can be utilized to extract the KV cache scaling factors from your quantized HF model, however at the moment, this tool exclusively supports Llama 2 models. It is also important to note the following:
1.**File Structure**: The utility operates under the assumption that all parameters, including KV cache scaling factors, corresponding to a particular Tensor Parallelism (TP) rank are stored in a single file. These files must adhere to a specific naming convention where the TP rank is immediately identified after a specific keyword (e.g., "rank") in the filename.
2.**TP Decomposition**: The utility assumes consistency between the TP decomposition employed by the quantizer tool and that used by vLLM.
3.**AMMO Compatibility**: Currently, the generated KV cache scaling factors for AMMO remain uniform across all TP ranks.
This script evaluates the inference throughput of language models using various backends such as vLLM. It measures the time taken to process a given number of prompts and generate sequences for each prompt. The recently generated KV cache scaling factors are now integrated into the benchmarking process and allow for KV cache scaling factors to be utilized for FP8.
# The model has an audio-specific lora directly in its model dir;
# it should be enabled whenever you pass audio inputs to the model.
speech_lora_path=model_name
audio_placeholder="<|audio|>"*audio_count
prompts=f"<|start_of_role|>system<|end_of_role|>Knowledge Cutoff Date: April 2024.\nToday's Date: December 19, 2024.\nYou are Granite, developed by IBM. You are a helpful AI assistant<|end_of_text|>\n<|start_of_role|>user<|end_of_role|>{audio_placeholder}{question}<|end_of_text|>\n<|start_of_role|>assistant<|end_of_role|>"# noqa: E501
The `LLM` class provides the primary Python interface for doing offline inference, which is interacting with a model without using a separate model inference server.
## Usage
The first script in this example shows the most basic usage of vLLM. If you are new to Python and vLLM, you should start here.
```bash
python examples/offline_inference/basic/basic.py
```
The rest of the scripts include an [argument parser](https://docs.python.org/3/library/argparse.html), which you can use to pass any arguments that are compatible with [`LLM`](https://docs.vllm.ai/en/latest/api/offline_inference/llm.html). Try running the script with `--help` for a list of all available arguments.
The chat and generate scripts also accept the [sampling parameters](https://docs.vllm.ai/en/latest/api/inference_params.html#sampling-parameters): `max_tokens`, `temperature`, `top_p` and `top_k`.
In the scripts that support passing arguments, you can experiment with the following features.
### Default generation config
The `--generation-config` argument specifies where the generation config will be loaded from when calling `LLM.get_default_sampling_params()`. If set to ‘auto’, the generation config will be loaded from model path. If set to a folder path, the generation config will be loaded from the specified folder path. If it is not provided, vLLM defaults will be used.
> If max_new_tokens is specified in generation config, then it sets a server-wide limit on the number of output tokens for all requests.
Try it yourself with the following argument:
```bash
--generation-config auto
```
### Quantization
#### AQLM
vLLM supports models that are quantized using AQLM.
Try one yourself by passing one of the following models to the `--model` argument:
> Some of these models are likely to be too large for a single GPU. You can split them across multiple GPUs by setting `--tensor-parallel-size` to the number of required GPUs.
#### GGUF
vLLM supports models that are quantized using GGUF.
Try one yourself by downloading a quantized GGUF model and using the following arguments:
The `--cpu-offload-gb` argument can be seen as a virtual way to increase the GPU memory size. For example, if you have one 24 GB GPU and set this to 10, virtually you can think of it as a 34 GB GPU. Then you can load a 13B model with BF16 weight, which requires at least 26GB GPU memory. Note that this requires fast CPU-GPU interconnect, as part of the model is loaded from CPU memory to GPU memory on the fly in each model forward pass.
