[Doc][Misc] Standardize hardware specifications, fix typos, and correct links in model tutorials (vllm-project#13100)

### What this PR does / why we need it?

This pull request performs a comprehensive cleanup and standardization
of the model tutorial documentation. Specifically, it:
- Standardizes memory unit representations (e.g., changing `64G` to
`64GB` and `32G` to `32GB`).
- Standardizes context and token length casing (e.g., changing `128K` to
`128k`, `16K` to `16k`, etc.).
- Corrects various typos (e.g., `ominidocbench` to `omnidocbench`,
`multinodes` to `multi-node clusters`, and `Rope` to `RoPE`).
- Fixes broken or outdated documentation URLs and GitHub issue links.
- Improves grammar, spacing, and formatting across multiple markdown
files.

These changes ensure consistency, accuracy, and readability across all
model deployment guides.

### Does this PR introduce _any_ user-facing change?

No, this PR only updates documentation and does not introduce any
user-facing API or behavioral changes.

### How was this patch tested?

Documentation changes only; verified by reviewing the rendered markdown
files.
- vLLM version: v0.25.1
- vLLM main:
vllm-project/vllm@d02df74

Signed-off-by: x60087760 <xiechangxiu3@h-partners.com>
Co-authored-by: x60087760 <xiechangxiu3@h-partners.com>

Author: ella1107

docs/source/tutorials/models/DeepSeek-R1.md

@@ -165,7 +165,7 @@ Key Parameter Descriptions:
 
 - Setting the environment variable `VLLM_ASCEND_BALANCE_SCHEDULING=1` enables balance scheduling. This may help increase output throughput and reduce TPOT in v1 scheduler. However, TTFT may degrade in some scenarios. Furthermore, enabling this feature is not recommended in scenarios where PD is separated.
 - For single-node deployment, we recommend using `dp4tp4` instead of `dp2tp8`.
-- `--max-model-len` specifies the maximum context length - that is, the sum of input and output tokens for a single request. For performance testing with an input length of 3.5K and output length of 1.5K, a value of `16384` is sufficient, however, for precision testing, please set it to at least `35000`.
+- `--max-model-len` specifies the maximum context length - that is, the sum of input and output tokens for a single request. For performance testing with an input length of 3.5k and output length of 1.5k, a value of `16384` is sufficient, however, for precision testing, please set it to at least `35000`.
 - `--no-enable-prefix-caching` indicates that prefix caching is disabled. To enable it, remove this option.
 - If you use the w4a8 weight, more memory will be allocated to kvcache, and you can try to increase system throughput to achieve greater throughput.
 

docs/source/tutorials/models/DeepSeek-V3.1.md

@@ -129,7 +129,7 @@ If you want to deploy multi-node environment, you need to set up environment on
 
 ### 5.1 Single-Node Online Deployment
 
-Single-node deployment completes both Prefill and Decode within the same node. The quantized model `DeepSeek-V3.1-w8a8-mtp-QuaRot` can be deployed on 1 Atlas 800 A3 (64G × 16).
+Single-node deployment completes both Prefill and Decode within the same node. The quantized model `DeepSeek-V3.1-w8a8-mtp-QuaRot` can be deployed on 1 Atlas 800 A3 (64GB × 16).
 
 Startup Command:
 
@@ -177,7 +177,7 @@ Key Parameter Descriptions:
 
 - Setting the environment variable `VLLM_ASCEND_BALANCE_SCHEDULING=1` enables balance scheduling. This may help increase output throughput and reduce TPOT in v1 scheduler. However, TTFT may degrade in some scenarios. Furthermore, enabling this feature is not recommended in scenarios where PD is separated.
 - For single-node deployment, we recommend using `dp4tp4` instead of `dp2tp8`.
-- `--max-model-len` specifies the maximum context length - that is, the sum of input and output tokens for a single request. For performance testing with an input length of 3.5K and output length of 1.5K, a value of `16384` is sufficient, however, for precision testing, please set it at least `35000`.
+- `--max-model-len` specifies the maximum context length - that is, the sum of input and output tokens for a single request. For performance testing with an input length of 3.5k and output length of 1.5k, a value of `16384` is sufficient, however, for precision testing, please set it at least `35000`.
 - `--no-enable-prefix-caching` indicates that prefix caching is disabled. To enable it, remove this option.
 - If you use the w4a8 weight, more memory will be allocated to kvcache, and you can try to increase system throughput to achieve greater throughput.
 
@@ -233,7 +233,7 @@ The service returns HTTP 200 OK with a JSON response containing the `choices` fi
 
 ### 5.2 Multi-Node Data Parallel Deployment
 
-- `DeepSeek-V3.1-w8a8-mtp-QuaRot`: require at least 2 Atlas 800 A2 (64G × 8).
+- `DeepSeek-V3.1-w8a8-mtp-QuaRot`: require at least 2 Atlas 800 A2 (64GB × 8).
 
 Run the following scripts on two nodes respectively.
 
@@ -393,9 +393,9 @@ In the standard single-node deployment mode, Prefill (prompt processing) and Dec
 
 This architecture is recommended for production deployments with concurrent multi-user workloads, where stable latency and high throughput are both required.
 
-Take Atlas 800 A3 (64G × 16) for example, we recommend to deploy 2P1D (4 nodes) rather than 1P1D (2 nodes), because there is no enough NPU memory to serve high concurrency in 1P1D case.
+Take Atlas 800 A3 (64GB × 16) for example, we recommend to deploy 2P1D (4 nodes) rather than 1P1D (2 nodes), because there is no enough NPU memory to serve high concurrency in 1P1D case.
 
-- `DeepSeek-V3.1-w8a8-mtp-QuaRot 2P1D Layerwise` require 4 Atlas 800 A3 (64G × 16).
+- `DeepSeek-V3.1-w8a8-mtp-QuaRot 2P1D Layerwise` require 4 Atlas 800 A3 (64GB × 16).
 
 To run the vllm-ascend `Prefill-Decode Disaggregation` service, you need to deploy a `launch_online_dp.py` script and a `run_dp_template.sh` script on each node and deploy a `proxy.sh` script on prefill master node to forward requests.
 
@@ -873,8 +873,8 @@ Here is one accuracy evaluation method.
 
 | dataset | version | metric | mode | vllm-api-general-chat | note |
 |----- | ----- | ----- | ----- | -----| ----- |
-| ceval | - | accuracy | gen | 90.94 | 1 Atlas 800 A3 (64G × 16) |
-| gsm8k | - | accuracy | gen | 96.28 | 1 Atlas 800 A3 (64G × 16) |
+| ceval | - | accuracy | gen | 90.94 | 1 Atlas 800 A3 (64GB × 16) |
+| gsm8k | - | accuracy | gen | 96.28 | 1 Atlas 800 A3 (64GB × 16) |
 
 ### Using Language Model Evaluation Harness
 
@@ -924,7 +924,7 @@ After about several minutes, you can get the performance evaluation result.
 
 #### Table 1: Scenario Overview
 
-> `*Total NPUs` indicates the total number of NPUs used across all nodes. 1 node = 1 Atlas 800 A3 server (64G × 16 NPUs).
+> `*Total NPUs` indicates the total number of NPUs used across all nodes. 1 node = 1 Atlas 800 A3 server (64GB × 16 NPUs).
 
 |Scenario|Deployment Mode|*Total NPUs|Weight Version|Key Considerations|
 |--------|---------------|-----------|--------------|------------------|

docs/source/tutorials/models/DeepSeek-V3.2.md

@@ -492,7 +492,7 @@ Parameter descriptions:
     export ASCEND_TRANSPORT_PRINT=1
     export ACL_OP_INIT_MODE=1
     export ASCEND_A3_ENABLE=1
-    # Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request.
+    # Timeout (in seconds) for automatically releasing the prefiller's KV cache for a particular request.
     export VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT=480
 
     export ASCEND_RT_VISIBLE_DEVICES=$1

docs/source/tutorials/models/DeepSeek-V4-Flash.md

@@ -23,7 +23,7 @@ Refer to [feature guide](../../user_guide/feature_guide/index.md) to get the fea
 
 ### 3.1 Model Weight
 
-- `DeepSeek-V4-Flash-w8a8-mtp` (Quantized version): requires 1 Atlas 800 A3 (128G × 8) node or 1 Atlas 800 A2 (64G × 8) node. [Download model weight](https://www.modelscope.cn/models/Eco-Tech/DeepSeek-V4-Flash-w8a8-mtp)
+- `DeepSeek-V4-Flash-w8a8-mtp` (Quantized version): requires 1 Atlas 800 A3 (128GB × 8) node or 1 Atlas 800 A2 (64GB × 8) node. [Download model weight](https://www.modelscope.cn/models/Eco-Tech/DeepSeek-V4-Flash-w8a8-mtp)
 
 It is recommended to download the model weight to the shared directory of multiple nodes, such as `/root/.cache/`.
 
@@ -137,7 +137,7 @@ If you want to deploy a multi-node environment, you need to set up the environme
 
 ### 5.1 Single-Node Online Deployment
 
-Single-node deployment completes both Prefill and Decode within the same node. The quantized model `DeepSeek-V4-Flash-w8a8-mtp` can be deployed on 1 Atlas 800 A3 (128G × 8) or 1 Atlas 800 A2 (64G × 8).
+Single-node deployment completes both Prefill and Decode within the same node. The quantized model `DeepSeek-V4-Flash-w8a8-mtp` can be deployed on 1 Atlas 800 A3 (128GB × 8) or 1 Atlas 800 A2 (64GB × 8).
 
 === "A2 series"
 
@@ -306,11 +306,11 @@ In the standard single-node deployment mode, Prefill (prompt processing) and Dec
 
 PD (Prefill-Decode) separation addresses these issues by running Prefill and Decode on dedicated node groups, each configured independently. This architecture is recommended for production deployments with concurrent multi-user workloads, where stable latency and high throughput are both required.
 
-The following sections describe PD separation deployment on both Atlas 800 A3 (128G × 8) and Atlas 800 A2 (64G × 8) multi-node environments.
+The following sections describe PD separation deployment on both Atlas 800 A3 (128GB × 8) and Atlas 800 A2 (64GB × 8) multi-node environments.
 
 #### 5.2.1 A3 Series PD Separation Deployment
 
-This section shows the deployment guide of DeepSeek-V4-Flash on Atlas 800 A3 (128G × 8) multi-node environment with 1P1D for better performance.
+This section shows the deployment guide of DeepSeek-V4-Flash on Atlas 800 A3 (128GB × 8) multi-node environment with 1P1D for better performance.
 
 Before you start, please:
 
@@ -601,7 +601,7 @@ Before you start, please:
 
 #### 5.2.2 A2 Series PD Separation Deployment
 
-This section shows the deployment guide of DeepSeek-V4-Flash on Atlas 800 A2 (64G × 8) multi-node environment with 4\*1P 1\*4D for better performance.
+This section shows the deployment guide of DeepSeek-V4-Flash on Atlas 800 A2 (64GB × 8) multi-node environment with 4\*1P 1\*4D for better performance.
 
 Before you start, please:
 
@@ -916,8 +916,8 @@ Here are two accuracy evaluation methods.
 
 | dataset | version | metric | mode | vllm-api-general-chat | note |
 | ----- | ----- | ----- | ----- | ----- | ----- |
-| GPQA | - | accuracy | gen | 88.17 | 1 Atlas 800 A3 (128G × 8) |
-| GSM8K | - | accuracy | gen | 96.30 | 1 Atlas 800 A3 (128G × 8) |
+| GPQA | - | accuracy | gen | 88.17 | 1 Atlas 800 A3 (128GB × 8) |
+| GSM8K | - | accuracy | gen | 96.30 | 1 Atlas 800 A3 (128GB × 8) |
 
 ## 8 Performance Evaluation
 
@@ -961,7 +961,7 @@ Refer to [vllm benchmark](https://docs.vllm.ai/en/latest/contributing/) for more
 
 `max-model-len` and `max-num-seqs` need to be set according to the actual usage scenario. For other settings, please refer to the [Deployment](#5-online-service-deployment) chapter.
 
-Currently, we support 4K prefix cache hit in an experimental manner. You only need to change the value of --block-size from 128 to 32 in the service.
+Currently, we support 4k prefix cache hit in an experimental manner. You only need to change the value of --block-size from 128 to 32 in the service.
 
 ### 9.2 Tuning Guidelines
 

docs/source/tutorials/models/DeepSeek-V4-Pro.md

@@ -23,7 +23,7 @@ Refer to [feature guide](../../user_guide/feature_guide/index.md) to get the fea
 
 ### 3.1 Model Weight
 
-- `DeepSeek-V4-Pro-w4a8-mtp` (Quantized version): requires 2 Atlas 800 A3 (128G × 8) nodes or 4 Atlas 800 A2 (64G × 8) nodes. [Download model weight](https://www.modelscope.cn/models/Eco-Tech/DeepSeek-V4-Pro-w4a8-mtp)
+- `DeepSeek-V4-Pro-w4a8-mtp` (Quantized version): requires 2 Atlas 800 A3 (128GB × 8) nodes or 4 Atlas 800 A2 (64GB × 8) nodes. [Download model weight](https://www.modelscope.cn/models/Eco-Tech/DeepSeek-V4-Pro-w4a8-mtp)
 
 It is recommended to download the model weight to the shared directory of multiple nodes, such as `/root/.cache/`.
 
@@ -133,7 +133,7 @@ If you want to deploy a multi-node environment, you need to set up the environme
 
 ### 5.1 Multi-Node Online Deployment
 
-The quantized model `DeepSeek-V4-Pro-w4a8-mtp` requires at least 2 Atlas 800 A3 (128G × 8) nodes or 4 Atlas 800 A2 (64G × 8) nodes. Run the following scripts on each node respectively.
+The quantized model `DeepSeek-V4-Pro-w4a8-mtp` requires at least 2 Atlas 800 A3 (128GB × 8) nodes or 4 Atlas 800 A2 (64GB × 8) nodes. Run the following scripts on each node respectively.
 
 === "A2 series"
 
@@ -447,11 +447,11 @@ We recommend using Mooncake for deployment: [Mooncake](../features/pd_disaggrega
 
 In the standard deployment mode, Prefill (prompt processing) and Decode (token generation) tasks run on the same set of NPUs. PD (Prefill-Decode) separation addresses this by running Prefill and Decode on dedicated node groups, each configured independently. This architecture is recommended for production deployments with concurrent multi-user workloads, where stable latency and high throughput are both required.
 
-The following sections describe PD separation deployment on both Atlas 800 A3 (128G × 8) and Atlas 800 A2 (64G × 8) multi-node environments.
+The following sections describe PD separation deployment on both Atlas 800 A3 (128GB × 8) and Atlas 800 A2 (64GB × 8) multi-node environments.
 
 #### 5.2.1 A3 Series PD Separation Deployment
 
-This section shows the deployment guide of DeepSeek-V4-Pro on Atlas 800 A3 (128G × 8) multi-node environment with 1P1D for better performance.
+This section shows the deployment guide of DeepSeek-V4-Pro on Atlas 800 A3 (128GB × 8) multi-node environment with 1P1D for better performance.
 
 Before you start, please:
 
@@ -826,7 +826,7 @@ Before you start, please:
 
 #### 5.2.2 A2 Series PD Separation Deployment
 
-This section shows the deployment guide of DeepSeek-V4-Pro on Atlas 800 A2 (64G × 8) multi-node environment with 1P1D for better performance.
+This section shows the deployment guide of DeepSeek-V4-Pro on Atlas 800 A2 (64GB × 8) multi-node environment with 1P1D for better performance.
 
 Before you start, please:
 
@@ -1202,8 +1202,8 @@ Here are two accuracy evaluation methods.
 
 | dataset | version | metric | mode | vllm-api-general-chat | note |
 | ----- | ----- | ----- | ----- | ----- | ----- |
-| GPQA | - | accuracy | gen | 89.90 | 1 Atlas 800 A3 (128G × 8) |
-| GSM8K | - | accuracy | gen | 96.21 | 1 Atlas 800 A3 (128G × 8) |
+| GPQA | - | accuracy | gen | 89.90 | 1 Atlas 800 A3 (128GB × 8) |
+| GSM8K | - | accuracy | gen | 96.21 | 1 Atlas 800 A3 (128GB × 8) |
 
 ## 8 Performance Evaluation
 

docs/source/tutorials/models/DeepSeekOCR2.md

@@ -196,7 +196,7 @@ curl http://<node0_ip>:<port>/v1/completions \
 | dataset | version | metric | mode | vllm-api-general-chat | note |
 |----- | ----- | ----- | ----- | -----| ----- |
 | textvqa | - | accuracy | gen | 50.28 | 1 Atlas 800 A2 |
-| ominidocbench | - | accuracy | gen | 66.86 | 1 Atlas 800 A2 |
+| omnidocbench | - | accuracy | gen | 66.86 | 1 Atlas 800 A2 |
 
 ## 8 Performance Evaluation
 
@@ -210,7 +210,7 @@ The performance result is:
 
 **Input/Output**: 1080P/256
 
-**Performance**: TTFT = 2s, TPOT = 200ms, Average performance of each card is 864 TPS (Token Per Second).
+**Performance**: TTFT = 2s, TPOT = 200ms, Average performance of each card is 864TPS (Token Per Second).
 
 ## 9 Performance Tuning
 
@@ -220,7 +220,7 @@ The performance result is:
 
 #### Table 1: Scenario Overview
 
-> `*Total NPUs` indicates the total number of NPUs used across all nodes. 1 node = 1 Atlas 800 A3 server (64G × 16 NPUs).
+> `*Total NPUs` indicates the total number of NPUs used across all nodes. 1 node = 1 Atlas 800 A3 server (64GB × 16 NPUs).
 
 |Scenario|Deployment Mode|*Total NPUs|Weight Version|Key Considerations|
 |--------|---------------|-----------|--------------|------------------|

docs/source/tutorials/models/GLM4.x.md

@@ -76,7 +76,7 @@ You can use our official docker image to run `GLM-4.x` directly.
 
 === "A2 series"
 
-    Start the docker image on your each node.
+    Start the docker image on each node.
 
     ```bash
 
@@ -262,7 +262,7 @@ While the previous documentation advises against multi-node deployment on the At
 
 ### 5.3 Prefill-Decode Disaggregation
 
-We'd like to show the deployment guide of `GLM-4.7` on multi-node environment with 2P1D for better performance.
+We'd like to show the deployment guide of `GLM-4.7` in a multi-node environment with 2P1D for better performance.
 
 Before you start, please
 

docs/source/tutorials/models/GLM5.2.md

@@ -16,9 +16,9 @@ Refer to [feature guide](../../user_guide/feature_guide/index.md) to get the fea
 
 ### Model Weight
 
-- `GLM-5.2`(BF16 version): requires 2 Atlas 800 A3 (128G × 8) node or 4 Atlas 800 A2 (64G × 8) node.[Download model weight](https://www.modelscope.cn/models/ZhipuAI/GLM-5.2).
-- `GLM-5.2-w8a8`: requires 1 Atlas 800 A3 (128G × 8) node or 2 Atlas 800 A2 (64G × 8) node.[Download model weight](https://www.modelscope.cn/models/Eco-Tech/GLM-5.2-w8a8).
-- `GLM-5.2-w4a8c8`: requires 1 Atlas 800 A3 (128G × 8) node or 2 Atlas 800 A2 (64G × 8) node.[Download model weight](https://modelscope.cn/models/Eco-Tech/GLM-5.2-w4a8c8).
+- `GLM-5.2`(BF16 version): requires 2 Atlas 800 A3 (128GB × 8) node or 4 Atlas 800 A2 (64GB × 8) node.[Download model weight](https://www.modelscope.cn/models/ZhipuAI/GLM-5.2).
+- `GLM-5.2-w8a8`: requires 1 Atlas 800 A3 (128GB × 8) node or 2 Atlas 800 A2 (64GB × 8) node.[Download model weight](https://www.modelscope.cn/models/Eco-Tech/GLM-5.2-w8a8).
+- `GLM-5.2-w4a8c8`: requires 1 Atlas 800 A3 (128GB × 8) node or 2 Atlas 800 A2 (64GB × 8) node.[Download model weight](https://modelscope.cn/models/Eco-Tech/GLM-5.2-w4a8c8).
 - You can use [msmodelslim](https://gitcode.com/Ascend/msmodelslim) to quantize the model directly.
 
 It is recommended to download the model weight to the shared directory of multiple nodes, such as `/root/.cache/`
@@ -110,7 +110,7 @@ If you want to deploy multi-node environment, you need to set up environment on
 
 ### Single-node Deployment
 
-- Quantized model `GLM-5.2-w4a8c8` can be deployed on 1 Atlas 800 A3 (64G × 16) .
+- Quantized model `GLM-5.2-w4a8c8` can be deployed on 1 Atlas 800 A3 (64GB × 16) .
 
 Run the following script to execute online inference.
 
@@ -157,7 +157,7 @@ If you want to deploy multi-node environment, you need to verify multi-node comm
 
 === "A3 series"
 
-    - `GLM-5.2-w4a8c8`: can be deployed on 2 Atlas 800 A3 (64G × 16).
+    - `GLM-5.2-w4a8c8`: can be deployed on 2 Atlas 800 A3 (64GB × 16).
 
     Run the following scripts on two nodes respectively.
 
@@ -263,7 +263,7 @@ If you want to deploy multi-node environment, you need to verify multi-node comm
 
 === "A2 series"
 
-    - `GLM-5.2-w4a8c8`: can be deployed on 2 Atlas 800 A2 (64G × 32).
+    - `GLM-5.2-w4a8c8`: can be deployed on 2 Atlas 800 A2 (64GB × 32).
 
     **node 0**
 
@@ -378,9 +378,9 @@ If you want to deploy multi-node environment, you need to verify multi-node comm
 
 ### Prefill-Decode Disaggregation
 
-We'd like to show the deployment guide of `GLM-5` on multi-node environment with 1P1D for better performance.
+We'd like to show the deployment guide of `GLM-5.2` on multi-node environment with 1P1D for better performance.
 
-Prefill-Decode disaggregation can be deployed on 4 Atlas 800 A3 (64G × 32).
+Prefill-Decode disaggregation can be deployed on 4 Atlas 800 A3 (64GB × 32).
 
 Before you start, please
 
@@ -1122,12 +1122,12 @@ Please refer to the following python file for further explanation and restrictio
 
 ### 1M Context Configuration
 
-Recommended configurations for serving `GLM-5.2` with a 1M context window on Atlas 800 A3 (64G x 16) and quantized GLM-5.2(W4A8C8) weights:
+Recommended configurations for serving `GLM-5.2` with a 1M context window on Atlas 800 A3 (64GB x 16) and quantized GLM-5.2(W4A8C8) weights:
 
 | Mode | Hardware | Parallelism | Context |
 | ---- | -------- | ----------- | ------- |
-| Single-node co-located | 1 Atlas 800 A3 (64G x 16) | `DP1 PP1 TP16 PCP1 DCP16` | `1024000` |
-| Dual-node co-located | 2 Atlas 800 A3 (64G x 16) | `DP4 PP1 TP8 PCP1 DCP8` | `1024000` |
+| Single-node co-located | 1 Atlas 800 A3 (64GB x 16) | `DP1 PP1 TP16 PCP1 DCP16` | `1024000` |
+| Dual-node co-located | 2 Atlas 800 A3 (64GB x 16) | `DP4 PP1 TP8 PCP1 DCP8` | `1024000` |
 | 1P1D PD disaggregation | 1 prefiller with 2 A3 nodes + 1 decoder with 2 A3 nodes | Prefill `DP4 PP1 TP8 PCP1 DCP8`, Decode `DP4 PP1 TP8 PCP1 DCP8` | `1024000` |
 
 #### Single-Node 1M Deployment