xuebi commited on
Commit
7d718af
·
1 Parent(s): 927ea2b

update: clearify supported context length in docs

Browse files
docs/sglang_deploy_guide.md CHANGED
@@ -27,9 +27,9 @@ The deployment process is illustrated below using MiniMax-M2.1 as an example.
27
 
28
  The following are recommended configurations; actual requirements should be adjusted based on your use case:
29
 
30
- - 4x 96GB GPUs: Supported context length of up to 400K tokens.
31
 
32
- - 8x 144GB GPUs: Supported context length of up to 3M tokens.
33
 
34
  ## Deployment with Python
35
 
 
27
 
28
  The following are recommended configurations; actual requirements should be adjusted based on your use case:
29
 
30
+ - 4x 96GB GPUs: Supports 400K aggregate KV cache tokens. (Max 196K per sequence)
31
 
32
+ - 8x 144GB GPUs: Supports 3M aggregate KV cache tokens. (Max 196K per sequence)
33
 
34
  ## Deployment with Python
35
 
docs/sglang_deploy_guide_cn.md CHANGED
@@ -27,9 +27,9 @@
27
 
28
  以下为推荐配置,实际需求请根据业务场景调整:
29
 
30
- - 96G x4 GPU:支持 40 万 token 的总上下文。
31
 
32
- - 144G x8 GPU:支持长达 300 万 token 的总上下文。
33
 
34
  ## 使用 Python 部署
35
 
 
27
 
28
  以下为推荐配置,实际需求请根据业务场景调整:
29
 
30
+ - 96G x4 GPU:支持 40 万 token 的总上下文(每个序列最多 19.6 万个)。
31
 
32
+ - 144G x8 GPU:支持长达 300 万 token 的总上下文(每个序列最多 19.6 万个)。
33
 
34
  ## 使用 Python 部署
35
 
docs/vllm_deploy_guide.md CHANGED
@@ -26,9 +26,9 @@ The deployment process is illustrated below using MiniMax-M2.1 as an example.
26
 
27
  The following are recommended configurations; actual requirements should be adjusted based on your use case:
28
 
29
- - 4x 96GB GPUs: Supported context length of up to 400K tokens.
30
 
31
- - 8x 144GB GPUs: Supported context length of up to 3M tokens.
32
 
33
  ## Deployment with Python
34
 
 
26
 
27
  The following are recommended configurations; actual requirements should be adjusted based on your use case:
28
 
29
+ - 4x 96GB GPUs: Supports 400K aggregate KV cache tokens. (Max 196K per sequence)
30
 
31
+ - 8x 144GB GPUs: Supports 3M aggregate KV cache tokens. (Max 196K per sequence)
32
 
33
  ## Deployment with Python
34
 
docs/vllm_deploy_guide_cn.md CHANGED
@@ -26,9 +26,9 @@
26
 
27
  以下为推荐配置,实际需求请根据业务场景调整:
28
 
29
- - 96G x4 GPU:支持 40 万 token 的总上下文。
30
 
31
- - 144G x8 GPU:支持长达 300 万 token 的总上下文。
32
 
33
  ## 使用 Python 部署
34
 
 
26
 
27
  以下为推荐配置,实际需求请根据业务场景调整:
28
 
29
+ - 96G x4 GPU:支持 40 万 token 的总上下文(每个序列最多 19.6 万个)。
30
 
31
+ - 144G x8 GPU:支持长达 300 万 token 的总上下文(每个序列最多 19.6 万个)。
32
 
33
  ## 使用 Python 部署
34