Skip to content

Updated results, removed max_seq_len_to_capture #727

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Oct 8, 2025
Merged

Updated results, removed max_seq_len_to_capture #727

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
da06753
Updated results, removed max_seq_len_to_capture
Mcirino Oct 6, 2025
5baa407
Updated Python version and Docker Manifest section
Mcirino Oct 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
114 changes: 63 additions & 51 deletions docs/dev-docker/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,19 +10,20 @@ This documentation includes information for running the popular Llama 3.1 series

The pre-built image includes:

- ROCm™ 6.4.1
- ROCm™ 7.0.0
- HipblasLT 0.15
- vLLM 0.10.1
- PyTorch 2.7
- vLLM 0.10.2
- PyTorch 2.9

## Pull latest Docker Image

Pull the most recent validated docker image with `docker pull rocm/vllm-dev:main`

## What is New

- vLLM version 0.10.1
- Flag enabled by default in the docker -VLLM_V1_USE_PREFILL_DECODE_ATTENTION
- Support for FP4 models
- GPT-OSS support
- Support for MI35x

## Known Issues and Workarounds

Expand All @@ -39,14 +40,14 @@ The table below shows performance data where a local inference client is fed req

| Model | Precision | TP Size | Input | Output | Num Prompts | Max Num Seqs | Throughput (tokens/s) |
|-------|-----------|---------|-------|--------|-------------|--------------|-----------------------|
| Llama 3.1 70B (amd/Llama-3.1-70B-Instruct-FP8-KV) | FP8 | 8 | 128 | 2048 | 3200 | 3200 | 13818.7 |
| | | | 128 | 4096 | 1500 | 1500 | 11612.0 |
| | | | 500 | 2000 | 2000 | 2000 | 11408.7 |
| | | | 2048 | 2048 | 1500 | 1500 | 7800.5 |
| Llama 3.1 405B (amd/Llama-3.1-405B-Instruct-FP8-KV) | FP8 | 8 | 128 | 2048 | 1500 | 1500 | 4134.0 |
| | | | 128 | 4096 | 1500 | 1500 | 3177.6 |
| | | | 500 | 2000 | 2000 | 2000 | 3034.1 |
| | | | 2048 | 2048 | 500 | 500 | 2214.2 |
| Llama 3.1 70B (amd/Llama-3.1-70B-Instruct-FP8-KV) | FP8 | 8 | 128 | 2048 | 3200 | 3200 | 13212.5 |
| | | | 128 | 4096 | 1500 | 1500 | 11312.8 |
| | | | 500 | 2000 | 2000 | 2000 | 11376.7 |
| | | | 2048 | 2048 | 1500 | 1500 | 7252.1 |
| Llama 3.1 405B (amd/Llama-3.1-405B-Instruct-FP8-KV) | FP8 | 8 | 128 | 2048 | 1500 | 1500 | 4201.7 |
| | | | 128 | 4096 | 1500 | 1500 | 3176.3 |
| | | | 500 | 2000 | 2000 | 2000 | 2992.0 |
| | | | 2048 | 2048 | 500 | 500 | 2153.7 |

*TP stands for Tensor Parallelism.*

Expand All @@ -58,38 +59,38 @@ The table below shows latency measurement, which typically involves assessing th

| Model | Precision | TP Size | Batch Size | Input | Output | MI300X Latency (sec) |
|-------|-----------|----------|------------|--------|---------|-------------------|
| Llama 3.1 70B (amd/Llama-3.1-70B-Instruct-FP8-KV) | FP8 | 8 | 1 | 128 | 2048 | 15.254 |
| | | | 2 | 128 | 2048 | 18.157 |
| | | | 4 | 128 | 2048 | 18.549 |
| | | | 8 | 128 | 2048 | 20.547 |
| | | | 16 | 128 | 2048 | 22.164 |
| | | | 32 | 128 | 2048 | 25.426 |
| | | | 64 | 128 | 2048 | 33.297 |
| | | | 128 | 128 | 2048 | 45.792 |
| | | | 1 | 2048 | 2048 | 15.299 |
| | | | 2 | 2048 | 2048 | 18.194 |
| | | | 4 | 2048 | 2048 | 18.942 |
| | | | 8 | 2048 | 2048 | 20.526 |
| | | | 16 | 2048 | 2048 | 23.211 |
| | | | 32 | 2048 | 2048 | 26.516 |
| | | | 64 | 2048 | 2048 | 34.824 |
| | | | 128 | 2048 | 2048 | 52.211 |
| Llama 3.1 405B (amd/Llama-3.1-405B-Instruct-FP8-KV) | FP8 | 8 | 1 | 128 | 2048 | 47.150 |
| | | | 2 | 128 | 2048 | 50.933 |
| | | | 4 | 128 | 2048 | 52.521 |
| | | | 8 | 128 | 2048 | 55.233 |
| | | | 16 | 128 | 2048 | 59.065 |
| | | | 32 | 128 | 2048 | 68.786 |
| | | | 64 | 128 | 2048 | 88.094 |
| | | | 128 | 128 | 2048 | 118.512 |
| | | | 1 | 2048 | 2048 | 47.675 |
| | | | 2 | 2048 | 2048 | 50.788 |
| | | | 4 | 2048 | 2048 | 52.405 |
| | | | 8 | 2048 | 2048 | 55.459 |
| | | | 16 | 2048 | 2048 | 59.923 |
| | | | 32 | 2048 | 2048 | 70.388 |
| | | | 64 | 2048 | 2048 | 91.218 |
| | | | 128 | 2048 | 2048 | 127.004 |
| Llama 3.1 70B (amd/Llama-3.1-70B-Instruct-FP8-KV) | FP8 | 8 | 1 | 128 | 2048 | 15.882 |
| | | | 2 | 128 | 2048 | 17.934 |
| | | | 4 | 128 | 2048 | 18.487 |
| | | | 8 | 128 | 2048 | 20.251 |
| | | | 16 | 128 | 2048 | 22.307 |
| | | | 32 | 128 | 2048 | 29.933 |
| | | | 64 | 128 | 2048 | 32.359 |
| | | | 128 | 128 | 2048 | 45.419 |
| | | | 1 | 2048 | 2048 | 15.959 |
| | | | 2 | 2048 | 2048 | 18.177 |
| | | | 4 | 2048 | 2048 | 18.684 |
| | | | 8 | 2048 | 2048 | 20.716 |
| | | | 16 | 2048 | 2048 | 23.136 |
| | | | 32 | 2048 | 2048 | 26.969 |
| | | | 64 | 2048 | 2048 | 34.359 |
| | | | 128 | 2048 | 2048 | 52.351 |
| Llama 3.1 405B (amd/Llama-3.1-405B-Instruct-FP8-KV) | FP8 | 8 | 1 | 128 | 2048 | 49.098 |
| | | | 2 | 128 | 2048 | 51.009 |
| | | | 4 | 128 | 2048 | 52.979 |
| | | | 8 | 128 | 2048 | 55.675 |
| | | | 16 | 128 | 2048 | 58.982 |
| | | | 32 | 128 | 2048 | 67.889 |
| | | | 64 | 128 | 2048 | 86.844 |
| | | | 128 | 128 | 2048 | 117.440 |
| | | | 1 | 2048 | 2048 | 49.033 |
| | | | 2 | 2048 | 2048 | 51.316 |
| | | | 4 | 2048 | 2048 | 52.947 |
| | | | 8 | 2048 | 2048 | 55.863 |
| | | | 16 | 2048 | 2048 | 60.103 |
| | | | 32 | 2048 | 2048 | 69.632 |
| | | | 64 | 2048 | 2048 | 89.826 |
| | | | 128 | 2048 | 2048 | 126.433 |

*TP stands for Tensor Parallelism.*

Expand Down Expand Up @@ -206,7 +207,6 @@ Below is a list of a few of the key vLLM engine arguments for performance; these
- **--max-model-len** : Maximum context length supported by the model instance. Can be set to a lower value than model configuration value to improve performance and gpu memory utilization.
- **--max-num-batched-tokens** : The maximum prefill size, i.e., how many prompt tokens can be packed together in a single prefill. Set to a higher value to improve prefill performance at the cost of higher gpu memory utilization. 65536 works well for LLama models.
- **--max-num-seqs** : The maximum decode batch size (default 256). Using larger values will allow more prompts to be processed concurrently, resulting in increased throughput (possibly at the expense of higher latency). If the value is too large, there may not be enough GPU memory for the KV cache, resulting in requests getting preempted. The optimal value will depend on the GPU memory, model size, and maximum context length.
- **--max-seq-len-to-capture** : Maximum sequence length for which Hip-graphs are captured and utilized. It's recommended to use Hip-graphs for the best decode performance. The default value of this parameter is 8K, which is lower than the large context lengths supported by recent models such as LLama. Set this parameter to max-model-len or maximum context length supported by the model for best performance.
- **--gpu-memory-utilization** : The ratio of GPU memory reserved by a vLLM instance. Default value is 0.9. Increasing the value (potentially as high as 0.99) will increase the amount of memory available for KV cache. When running in graph mode (i.e. not using `--enforce-eager`), it may be necessary to use a slightly smaller value of 0.92 - 0.95 to ensure adequate memory is available for the HIP graph.

### Latency Benchmark
Expand Down Expand Up @@ -271,7 +271,6 @@ python3 /app/vllm/benchmarks/benchmark_throughput.py \
--model $MODEL \
--max-model-len 8192 \
--max-num-batched-tokens 131072 \
--max-seq-len-to-capture 131072 \
--input-len $IN \
--output-len $OUT \
--tensor-parallel-size $TP \
Expand Down Expand Up @@ -471,13 +470,21 @@ Please refer to the [Benchmarking Machine Learning using ROCm and AMD GPUs: Repr

## Docker Manifest

To reproduce the release docker:
Clone the vLLM repository:

```bash
git clone https://github.com/ROCm/vllm.git
git clone https://github.com/vllm-project/vllm.git
cd vllm
git checkout 6663000a391911eba96d7864a26ac42b07f6ef29
docker build -f docker/Dockerfile.rocm -t <your_tag> --build-arg USE_CYTHON=1 .
```

Use the following command to build the image directly from the specified commit.

```bash
docker build -f docker/Dockerfile.rocm \
--build-arg REMOTE_VLLM=1 \
--build-arg VLLM_REPO=https://github.com/ROCm/vllm \
--build-arg VLLM_BRANCH="790d22168820507f3105fef29596549378cfe399" \
-t vllm-rocm .
```

### Building AITER Image
Expand All @@ -493,6 +500,11 @@ Use AITER release candidate branch instead:

## Changelog

rocm7.0.0_vllm_0.10.2_20251002:
- Support for FP4 models
- GPT-OSS support
- Support for MI35x

rocm6.4.1_vllm_0.10.1_20250909:
- vLLM version 0.10.1
- Flag enabled by default in the docker -VLLM_V1_USE_PREFILL_DECODE_ATTENTION
Expand Down