[Docs] Optimal Deployment #2768
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for your contribution! |
| > **gpu-memory-utilization** | ||
| - **参数：** `--gpu-memory-utilization` | ||
| - **用处：** 用于控制 FastDeploy 初始化服务的可用显存，默认0.9，即预留10%的显存备用。 | ||
| - **推荐：** A卡上推荐0.9，H卡上推荐0.8～0.9。如果服务压测时提示显存不足，可以尝试调低该值。 |
There was a problem hiding this comment.
A卡-> A100/A800
H卡 -> H100/H800?
| --quantization wint4 \ | ||
| --enable-mm \ | ||
| ``` | ||
| ### **Example**: Dual-GPU Wint8 with 128K Context Length Configuration |
There was a problem hiding this comment.
Context Length Configuration -> context length，Wint8 -> ?
| |:----------:|:----------:|:------:|:------:| | ||
| | A30 | wint4 | 432.99 | 17396.92 | | ||
| | L20 | wint4<br>wint8 | 3311.34<br>2423.36 | 46566.81<br>60790.91 | | ||
| | H20 | wint4<br>wint8<br>bfloat16 | 3827.27<br>3578.23<br>4100.83 | 89770.14<br>95434.02<br>84543.00 | |
There was a problem hiding this comment.
Why the bf16 is best?
qingqing01
reviewed
Jul 29, 2025
| | L20 [48G] | wint4<br>wint8<br>bfloat16 | 1, 2, 4<br>1, 2, 4<br>2, 4 | | ||
| | H20 [144G] | wint4<br>wint8<br>bfloat16 | 1, 2, 4<br>1, 2, 4<br>1, 2, 4 | | ||
| | A100 [80G] | wint4<br>wint8<br>bfloat16 | 1, 2, 4<br>1, 2, 4<br>1, 2, 4 | | ||
| | H800 [80G] | wint4<br>wint8<br>bfloat16 | 1, 2, 4<br>1, 2, 4<br>1, 2, 4 | |
There was a problem hiding this comment.
和其他文档保持一致，WINT4 WINT8都大写。
| ### 2.1 基础：启动服务 | ||
| **示例1：** 4090上单卡部署32K上下文的服务 | ||
| ```shell | ||
| python -m fastdeploy.entrypoints.openai.api_server --model baidu/ERNIE-4.5-VL-28B-A3B-Paddle --port 8180 --metrics-port 8181 --engine-worker-queue-port 8182 --tensor-parallel-size 1 --max-model-len 32768 --max-num-seqs 256 --limit-mm-per-prompt '{"image": 100, "video": 100}' --reasoning-parser ernie-45-vl --gpu-memory-utilization 0.9 --kv-cache-ratio 0.75 --enable-chunked-prefill --max-num-batched-tokens 384 --quantization wint4 --enable-mm |
| ``` | ||
| **示例2：** H800上双卡部署128K上下文的服务 | ||
| ```shell | ||
| python -m fastdeploy.entrypoints.openai.api_server --model baidu/ERNIE-4.5-VL-28B-A3B-Paddle --port 8180 --metrics-port 8181 --engine-worker-queue-port 8182 --tensor-parallel-size 2 --max-model-len 131072 --max-num-seqs 256 --limit-mm-per-prompt '{"image": 100, "video": 100}' --reasoning-parser ernie-45-vl --gpu-memory-utilization 0.9 --kv-cache-ratio 0.75 --enable-chunked-prefill --max-num-batched-tokens 384 --quantization wint4 --enable-mm |
There was a problem hiding this comment.
同上，建议直接按照最新的不设置 kv-cache-ratioe 的来，跟随2.1版本
| ⚠️ 注：更长的上下文会显著增加GPU显存需求，设置更长的上下文之前确保硬件资源是满足的。 | ||
| > **最大序列数量** | ||
| - **参数：** `--max-num-seqs` | ||
| - **描述：** 控制服务可以处理的最大序列数量，支持1～256。 |
There was a problem hiding this comment.
当前不支持256以上的 batch-size？如果不支持，后续需要排查
| > **上下文长度** | ||
| - **参数：** `--max-model-len` | ||
| - **描述：** 控制模型可处理的最大上下文长度。 | ||
| - **推荐：** 更长的上下文会导致吞吐降低，根据实际情况设置，最长支持**128k**（131072）长度的上下文。 |
|
|
||
| - **其他相关配置**: | ||
|
|
||
| `--max-num-batched-tokens`：限制每个chunk的最大token数量，推荐设置为384。 |
| - bfloat16 (未设置 `--quantization` 参数时，默认使用bfloat16) | ||
|
|
||
| - **推荐：** | ||
| - 除非您有极其严格的精度要求，否则我们强烈建议使用wint4量化。这将显著降低内存占用并提升吞吐量。 |
| **示例1：** H800上8卡部署128K上下文的服务 | ||
| ```shell | ||
| python -m fastdeploy.entrypoints.openai.api_server --model baidu/ERNIE-4.5-VL-424B-A47B-Paddle --port 8180 --metrics-port 8181 --engine-worker-queue-port 8182 --tensor-parallel-size 8 --max-model-len 131072 --max-num-seqs 16 --limit-mm-per-prompt '{"image": 100, "video": 100}' --reasoning-parser ernie-45-vl --gpu-memory-utilization 0.8 --kv-cache-ratio 0.75 --enable-chunked-prefill --max-num-batched-tokens 384 --quantization wint4 --enable-mm | ||
| ``` |
qingqing01
approved these changes
Aug 1, 2025
| | A100 [80G] | wint4<br>wint8<br>bfloat16 | 1, 2, 4<br>1, 2, 4<br>1, 2, 4 | | ||
| | H800 [80G] | wint4<br>wint8<br>bfloat16 | 1, 2, 4<br>1, 2, 4<br>1, 2, 4 | | ||
|
|
||
| ### 1.2 安装fastdeploy |
There was a problem hiding this comment.
fastdeploy -> FastDeploy
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add ERNIE-4.5-VL-28B-A3B-Paddle Optimal Deployment