Update README, deployment guide, and figures

.gitattributes

@@ -38,3 +38,5 @@ figs/efficiency.jpg filter=lfs diff=lfs merge=lfs -text
 figs/title.png filter=lfs diff=lfs merge=lfs -text
 figs/performance.png filter=lfs diff=lfs merge=lfs -text
 figs/case-study.png filter=lfs diff=lfs merge=lfs -text
+figs/chain-of-thought.png filter=lfs diff=lfs merge=lfs -text
+figs/efficiency.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -12,7 +12,7 @@ pipeline_tag: image-text-to-text
 
   <div>&nbsp;</div>
 
-[💻Github Repo](https://github.com/InternLM/Intern-S2-Mobius) • [🤗Model Collections](https://huggingface.co/collections/internlm/intern-s2) • [💬Online Chat](https://chat.intern-ai.org.cn/)
+[💻Github Repo](https://github.com/InternLM/Intern-S2-Mobius) • [🤗Model Collections](https://huggingface.co/collections/internlm/intern-s2) • [🌳Arch Space](https://github.com/InternLM/archspace)
 
 </div>
 
@@ -40,16 +40,27 @@ This knowledge-reasoning separation gives Intern-S2-Mobius two native capabiliti
   <figcaption>Fig1: Inference efficiency on reasoning benchmarks. Intern-S2-Mobius improves request throughput over the Transformer baseline while maintaining strong reasoning performance, with gains largely coming from shorter, more compact reasoning traces.</figcaption>
 </figure>
 
-### Performance
+<figure>
+  <img src="./figs/chain-of-thought.png" alt="chain of thought">
+  <figcaption>Fig2: The average output length of Mobius continual pre-trained from Qwen3.5.</figcaption>
+</figure>
 
-We evaluate the Intern-S2-Mobius on various benchmarks, including general datasets and scientific datasets. We report the performance comparison with Qwen3.5-35B below. 
+### Performance
 
-![performance](./figs/performance.png)
+We evaluate the Intern-S2-Mobius on various benchmarks, including general datasets and scientific datasets. We report the performance comparison with Qwen3.5-35B below. We use the [OpenCompass](https://github.com/open-compass/OpenCompass/) to evaluate all models. For text benchmarks, Intern-S2-Mobius is evaluated with a maximum inference length of 64K tokens on MMLU Pro, SimpleQA, and HLE, and 128K tokens on the remaining text benchmarks. 
 
+<figure>
+  <img src="./figs/performance.png" alt="performance">
+  <figcaption>Fig3: Performance comparison across general, scientific, and multimodal benchmarks. The higher score in each row is highlighted in <strong>bold</strong>.</figcaption>
+</figure>
 
-> **Note**: **Bold** indicates the best performance among all models.
+<figure>
+  <img src="./figs/case-study.png" alt="case study">
+  <figcaption>Fig4: Step-aligned comparison between Intern-S2-Mobius-35B and Qwen3.5-35B
+    on a linear-algebra multiple-choice question. Both models select the correct
+    answer (Option C). Token counts are computed using the Qwen3.5-35B tokenizer. Mobius completes the same reasoning steps with fewer tokens, which mainly benefits from the model's elimination of repeated derivation and checks.</figcaption>
+</figure>
 
-We use the [OpenCompass](https://github.com/open-compass/OpenCompass/) and [VLMEvalKit](https://github.com/open-compass/vlmevalkit) to evaluate all models. For text benchmarks, Intern-S2-Mobius is evaluated with a maximum inference length of 64K tokens on MMLU Pro, SimpleQA, and HLE, and 128K tokens on the remaining text benchmarks. For multimodal benchmarks, Intern-S2-Mobius is evaluated with a maximum inference length of 64K tokens.
 
 
 ## Quick Start
@@ -72,7 +83,6 @@ Intern-S2-Mobius can be deployed using any of the following LLM inference framew
 - LMDeploy
 - Transformer
 - vLLM
-- SGLang
 
 Detailed deployment examples for these frameworks are available in the [Model Deployment Guide](./deployment_guide.md).
 

deployment_guide.md

@@ -97,56 +97,26 @@ For example, you can use the vLLM nightly Docker image `docker pull vllm/vllm-op
 - Serving With MTP (Recommended)
 
 ```bash
-vllm serve internlm/Intern-S2-Preview \
+vllm serve \
+    internlm/Intern-S2-Mobius \
     --trust-remote-code \
     --tensor-parallel-size 2 \
     --reasoning-parser qwen3 \
     --enable-auto-tool-choice \
     --tool-call-parser qwen3_coder \
-    --speculative-config '{"method":"mtp","num_speculative_tokens":4}'
+    --spec-method mtp \
+    --spec-tokens 4
 ```
 
 - Basic Serving Without MTP
 
 ```bash
-vllm serve internlm/Intern-S2-Preview \
+vllm serve \
+    internlm/Intern-S2-Mobius \
     --trust-remote-code \
     --tensor-parallel-size 2 \
     --reasoning-parser qwen3 \
     --enable-auto-tool-choice \
-    --tool-call-parser qwen3_coder
-```
-
-## SGLang
-
-Use the latest SGLang Docker image or source build with Intern-S2-Preview support.
-
-For example, you can use the SGLang nightly Docker image `docker pull lmsysorg/sglang:nightly-dev-cu12-20260520-425dffbd`
-
-- Serving With MTP (Recommended)
-
-```bash
-SGLANG_ENABLE_SPEC_V2=1 \
-python3 -m sglang.launch_server \
-  --model-path internLM/Intern-S2-Preview \
-  --trust-remote-code \
-  --tp-size 2 \
-  --reasoning-parser qwen3 \
-  --tool-call-parser qwen3_coder \
-  --mamba-scheduler-strategy extra_buffer \
-  --speculative-algo 'NEXTN' \
-  --speculative-eagle-topk 1 \
-  --speculative-num-steps 3 \
-  --speculative-num-draft-tokens 4
+    --tool-call-parser qwen3_coder 
 ```
 
-- Basic Serving Without MTP
-
-```bash
-python3 -m sglang.launch_server \
-    --model-path internlm/Intern-S2-Preview \
-    --trust-remote-code \
-    --tp-size 2 \
-    --reasoning-parser qwen3 \
-    --tool-call-parser qwen3_coder
-```