Skip to content

Commit de0c8ec

Browse files
armstrongttwalker-altarmstrongttwalker-alt
authored
Merge pull request #429 from flagos-ai/update/modelscope-docs-20260723-142026
ModelScope Documentation Update - 2026-07-23 14:20
2 parents 7e92d33 + 4b5825d commit de0c8ec

26 files changed

Lines changed: 3465 additions & 35 deletions

‎docs/flagrelease_en/model_list.txt‎

Lines changed: 10 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,12 @@
1+
FlagRelease/AI21-Jamba-1.5-Mini-hygon-FlagOS
2+
FlagRelease/AI21-Jamba-1.5-Mini-nvidia-FlagOS
13
FlagRelease/BAAI-Cardiac-Agent-hygon-FlagOS
24
FlagRelease/Baguettotron-metax-FlagOS
35
FlagRelease/C2S-Scale-Gemma-2-27B-hygon-FlagOS
46
FlagRelease/C2S-Scale-Gemma-2-27B-nvidia-FlagOS
7+
FlagRelease/DeepSeek-R1-0528-Qwen3-8B-metax-FlagOS
8+
FlagRelease/DeepSeek-R1-Distill-Qwen-1.5B-hygon-FlagOS
9+
FlagRelease/DeepSeek-R1-Distill-Qwen-1.5B-nvidia-FlagOS
510
FlagRelease/DeepSeek-R1-Distill-Qwen-32B-FlagOS-Cambricon
611
FlagRelease/DeepSeek-R1-Distill-Qwen-32B-FlagOS-NVIDIA
712
FlagRelease/DeepSeek-R1-FlagOS-Cambricon-BF16
@@ -29,6 +34,8 @@ FlagRelease/DeepSeek-V4-Pro-nvidia-FlagOS
2934
FlagRelease/ERNIE-4.5-0.3B-PT
3035
FlagRelease/ERNIE-4.5-300B-A47B-PT-FlagOS
3136
FlagRelease/Emu3.5-FlagOS
37+
FlagRelease/GLM-4-32B-Base-0414-hygon-FlagOS
38+
FlagRelease/GLM-4-32B-Base-0414-nvidia-FlagOS
3239
FlagRelease/GLM-4.5-FlagOS
3340
FlagRelease/GLM-5-FP8-FlagOS
3441
FlagRelease/GLM-5-ascend-FlagOS
@@ -107,6 +114,8 @@ FlagRelease/QwQ-32B-FlagOS-Nvidia
107114
FlagRelease/Qwen2-7B-FlagOS-Arm
108115
FlagRelease/Qwen2-7B-Instruct-FlagOS
109116
FlagRelease/Qwen2.5-32B-Instruct-FlagOS-Nvidia
117+
FlagRelease/Qwen2.5-7B-Instruct-metax-FlagOS
118+
FlagRelease/Qwen2.5-Coder-7B-Instruct-metax-FlagOS
110119
FlagRelease/Qwen2.5-VL-32B-Instruct-FlagOS-Metax-BF16
111120
FlagRelease/Qwen2.5-VL-32B-Instruct-FlagOS-Nvidia
112121
FlagRelease/Qwen3-235B-A22B-FlagOS-nvidia
@@ -137,7 +146,7 @@ FlagRelease/Qwen3.5-397B-A17B-metax-FlagOS
137146
FlagRelease/Qwen3.5-397B-A17B-nvidia-FlagOS
138147
FlagRelease/Qwen3.5-397B-A17B-zhenwu-FlagOS
139148
FlagRelease/Qwen3.6-27B-hygon-FlagOS
140-
FlagRelease/Qwen3.6-27B-metax-FlagOS
149+
FlagRelease/Qwen3.6-27B-metax-FlagOS-Express
141150
FlagRelease/Qwen3.6-35B-A3B-nomtp-ascend-FlagOS
142151
FlagRelease/Qwen3.6-35B-A3B-nomtp-hygon-FlagOS
143152
FlagRelease/Qwen3.6-35B-A3B-nomtp-iluvatar-FlagOS

‎docs/flagrelease_en/model_readmes/FlagRelease_AI21-Jamba-1.5-Mini-hygon-FlagOS.md‎

Lines changed: 127 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,127 @@
1+
---
2+
frameworks:
3+
- ""
4+
tasks: []
5+
---
6+
# Introduction
7+
AI21-Jamba-1.5-Mini is an open-source large language model released by AI21. This release completes adaptation and validation on the Nvidia platform and is published based on the FlagOS software stack.
8+
9+
### Integrated Deployment
10+
- Out-of-the-box inference scripts with pre-configured hardware and software parameters
11+
- Released **FlagOS-Hygon** container image supporting deployment within minutes
12+
### Consistency Validation
13+
- Rigorously evaluated through benchmark testing: Performance and results from the FlagOS software stack are compared against native stacks on multiple public.
14+
15+
16+
# Evaluation Results
17+
## Benchmark Result
18+
| Metrics | AI21-Jamba-1.5-Mini-Nvidia-Origin | AI21-Jamba-1.5-Mini-Hygon-FlagOS |
19+
|---------------------------------|-----------------------------------|----------------------------------|
20+
| GPQA_Diamond | 0.177 | 0.222 |
21+
| GPQA_Generative_CoT | 0.218 | 0.223 |
22+
| LiveBench_New | 0.275 | 0.267 |
23+
| MUSR_Generative | 0.290 | 0.300 |
24+
| MMLU_Pro | 0.401 | 0.406 |
25+
26+
27+
# User Guide
28+
Environment Setup
29+
30+
| Item | Version |
31+
|------------------|----------------------|
32+
| Docker Version | Docker version 20.10.24, build 297e128 |
33+
| Operating System | Sugon OS 8.9 |
34+
35+
## Operation Steps
36+
37+
### Download FlagOS Image
38+
```bash
39+
docker pull harbor.baai.ac.cn/external-cooperation/ai21-jamba-1.5-mini-hygon-tree_0.5.0-gems_5.0.2-vllm_0.13.0-plugin_0.1.1-cx_none-python_3.10.12-torch_2.9.0-pcp_hygon-dpu_hygon-x86_64-driver_1.11.0:2607101038
40+
```
41+
42+
### Download Open-source Model Weights
43+
```bash
44+
pip install modelscope
45+
modelscope download \
46+
--model FlagRelease/AI21-Jamba-1.5-Mini-hygon-FlagOS \
47+
--local_dir /data/models/vllm-plugin-fl/AI21-Jamba-1.5-Mini-hygon-FlagOS
48+
```
49+
50+
### Start the Container
51+
```bash
52+
docker run \
53+
--name ai21-jamba-hygon \
54+
--network=host \
55+
--privileged=true \
56+
--shm-size=16g \
57+
-v /data/models/vllm-plugin-fl:/data/vllm-plugin-fl \
58+
-v /opt/hyhal:/opt/hyhal:ro \
59+
-itd \
60+
harbor.baai.ac.cn/external-cooperation/ai21-jamba-1.5-mini-hygon-tree_0.5.0-gems_5.0.2-vllm_0.13.0-plugin_0.1.1-cx_none-python_3.10.12-torch_2.9.0-pcp_hygon-dpu_hygon-x86_64-driver_1.11.0:2607101038 \
61+
sleep infinity
62+
63+
docker exec -it ai21-jamba-hygon bash
64+
65+
```
66+
### Start the Server
67+
```bash
68+
nohup env \
69+
HIP_VISIBLE_DEVICES=0,1 \
70+
VLLM_PLUGINS=fl \
71+
TRITON_ALL_BLOCKS_PARALLEL=1 \
72+
vllm serve \
73+
--model /data/vllm-plugin-fl/AI21-Jamba-1.5-Mini-hygon-FlagOS \
74+
--tensor-parallel-size 2 \
75+
--enforce-eager \
76+
--max-cudagraph-capture-size 0 \
77+
--served-model-name ai21_flagos \
78+
--port 8131 \
79+
--gpu-memory-utilization 0.85 \
80+
> /workspace/ai21-flagos.log 2>&1 &
81+
```
82+
83+
## Service Invocation
84+
### Invocation Script
85+
```bash
86+
curl http://localhost:8131/v1/chat/completions \
87+
-H "Content-Type: application/json" \
88+
-d '{
89+
"model": "ai21_flagos",
90+
"messages": [
91+
{
92+
"role": "user",
93+
"content": "你好"
94+
}
95+
]
96+
}'
97+
```
98+
99+
100+
# Technical Overview
101+
**FlagOS** is a fully open-source system software stack designed to unify the "model–system–chip" layers and foster an open, collaborative ecosystem. It enables a “develop once, run anywhere” workflow across diverse AI accelerators, unlocking hardware performance, eliminating fragmentation among vendor-specific software stacks, and substantially lowering the cost of porting and maintaining AI workloads. With core technologies such as the **FlagScale**, together with vllm-plugin-fl, distributed training/inference framework, **FlagGems** universal operator library, **FlagCX** communication library, and **FlagTree** unified compiler, the **FlagRelease** platform leverages the **FlagOS** stack to automatically produce and release various combinations of \<chip + open-source model\>. This enables efficient and automated model migration across diverse chips, opening a new chapter for large model deployment and application.
102+
## FlagGems
103+
FlagGems is a high-performance, generic operator libraryimplemented in [Triton](https://github.com/openai/triton) language. It is built on a collection of backend-neutralkernels that aims to accelerate LLM (Large-Language Models) training and inference across diverse hardware platforms.
104+
## FlagTree
105+
FlagTree is an open source, unified compiler for multipleAI chips project dedicated to developing a diverse ecosystem of AI chip compilers and related tooling platforms, thereby fostering and strengthening the upstream and downstream Triton ecosystem. Currently in its initial phase, the project aims to maintain compatibility with existing adaptation solutions while unifying the codebase to rapidly implement single-repository multi-backend support. Forupstream model users, it provides unified compilation capabilities across multiple backends; for downstream chip manufacturers, it offers examples of Triton ecosystem integration.
106+
## FlagScale and vllm-plugin-fl
107+
Flagscale is a comprehensive toolkit designed to supportthe entire lifecycle of large models. It builds on the strengths of several prominent open-source projects, including [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) and [vLLM](https://github.com/vllm-project/vllm), to provide a robust, end-to-end solution for managing and scaling large models.
108+
vllm-plugin-fl is a vLLM plugin built on the FlagOS unified multi-chip backend, to help flagscale support multi-chip on vllm framework.
109+
## **FlagCX**
110+
FlagCX is a scalable and adaptive cross-chip communication library. It serves as a platform where developers, researchers, and AI engineers can collaborate on various projects, contribute to the development of cutting-edge AI solutions, and share their work with the global community.
111+
112+
## **FlagEval Evaluation Framework**
113+
FlagEval is a comprehensive evaluation system and open platform for large models launched in 2023. It aims to establish scientific, fair, and open benchmarks, methodologies, and tools to help researchers assess model and training algorithm performance. It features:
114+
- **Multi-dimensional Evaluation**: Supports 800+ modelevaluations across NLP, CV, Audio, and Multimodal fields,covering 20+ downstream tasks including language understanding and image-text generation.
115+
- **Industry-Grade Use Cases**: Has completed horizonta1 evaluations of mainstream large models, providing authoritative benchmarks for chip-model performance validation.
116+
117+
# Contributing
118+
119+
We warmly welcome global developers to join us:
120+
121+
1. Submit Issues to report problems
122+
2. Create Pull Requests to contribute code
123+
3. Improve technical documentation
124+
4. Expand hardware adaptation support
125+
# License
126+
The model weights are derived from AI-ModelScope/AI21-Jamba-1.5-Mini and are open‑sourced under the Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0.txt
127+

‎docs/flagrelease_en/model_readmes/FlagRelease_AI21-Jamba-1.5-Mini-nvidia-FlagOS.md‎

Lines changed: 125 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,125 @@
1+
---
2+
frameworks:
3+
- ""
4+
language:
5+
- zh
6+
- en
7+
license: apache-2.0
8+
tasks: []
9+
---
10+
# Introduction
11+
AI21-Jamba-1.5-Mini is an open-source large language model released by AI21. This release completes adaptation and validation on the Nvidia platform and is published based on the FlagOS software stack.
12+
13+
### Integrated Deployment
14+
- Out-of-the-box inference scripts with pre-configured hardware and software parameters
15+
- Released **FlagOS-Nvidia** container image supporting deployment within minutes
16+
### Consistency Validation
17+
- Rigorously evaluated through benchmark testing: Performance and results from the FlagOS software stack are compared against native stacks on multiple public.
18+
19+
20+
# Evaluation Results
21+
## Benchmark Result
22+
| Metrics | AI21-Jamba-1.5-Mini-Nvidia-Origin | AI21-Jamba-1.5-Mini-Nvidia-FlagOS |
23+
|--------------|-----------------------------------|-----------------------------------|
24+
| GPQA_Diamond | 0.1869 | 0.1717 |
25+
| GPQA | 0.2324 | 0.2106 |
26+
| LiveBench | 0.2760 | 0.2642 |
27+
| MMLU_Pro | 0.4033 | 0.4044 |
28+
| MUSR | 0.3082 | 0.2712 |
29+
30+
# User Guide
31+
Environment Setup
32+
33+
| Item | Version |
34+
|------------------|----------------------|
35+
| Docker Version | Docker version 24.0.0, build 98fdcd7 |
36+
| Operating System | 22.04.4 LTS (Jammy Jellyfish) |
37+
38+
## Operation Steps
39+
40+
### Download FlagOS Image
41+
```bash
42+
docker pull harbor.baai.ac.cn/external-cooperation/ai21-jamba-1.5-mini-nvidia-gems_5.0.2-vllm_0.13.0-plugin_0.0.0-python_3.12.3-torch_2.9.0_cu128-pcp_cuda13.2-gpu_h20_3e-arc_amd64-driver_570.133.20:20260513152138
43+
```
44+
45+
### Download Open-source Model Weights
46+
```bash
47+
pip install modelscope
48+
modelscope download --model FlagRelease/AI21-Jamba-1.5-Mini-nvidia-FlagOS --local_dir /data/AI21-Jamba-1.5-Mini-nvidia-FlagOS
49+
```
50+
51+
### Start the Container
52+
```bash
53+
docker run --init --detach \
54+
--name ai21_jamba_1_5_mini_flagos \
55+
--gpus all \
56+
--shm-size=16g \
57+
-p 8000:8000 \
58+
-w /workspace \
59+
-v /data/AI21-Jamba-1.5-Mini-nvidia-FlagOS:/data/AI21-Jamba-1.5-Mini-nvidia-FlagOS \
60+
harbor.baai.ac.cn/external-cooperation/ai21-jamba-1.5-mini-nvidia-gems_5.0.2-vllm_0.13.0-plugin_0.0.0-python_3.12.3-torch_2.9.0_cu128-pcp_cuda13.2-gpu_h20_3e-arc_amd64-driver_570.133.20:20260513152138 \
61+
sleep infinity
62+
63+
docker exec -it ai21_jamba_1_5_mini_flagos bash
64+
```
65+
### Start the Server
66+
```bash
67+
nohup env CUDA_VISIBLE_DEVICES=0 \
68+
VLLM_PLUGINS=fl \
69+
TRITON_ALL_BLOCKS_PARALLEL=1 \
70+
VLLM_FL_FLAGOS_WHITELIST=exponential_,sort,rand_like,masked_fill_,argmax,lt_scalar,neg,where_self,exp,fill_tensor_,fill_scalar_,gt_scalar,gather,softmax,zeros,zeros_like,where_self,where_self_out,ones,sub,cumsum_out,sort_stable,index,to_copy,copy_,rms_norm \
71+
vllm serve \
72+
--model /data/AI21-Jamba-1.5-Mini-nvidia-FlagOS \
73+
--served-model-name ai21-jamba-1.5-mini-flagos \
74+
--port 8000 \
75+
--max-num-batched-tokens 8192 \
76+
> ai21-jamba-1.5-mini-flagos.log 2>&1 &
77+
```
78+
79+
## Service Invocation
80+
### Invocation Script
81+
```bash
82+
curl http://localhost:8000/v1/chat/completions \
83+
-H "Content-Type: application/json" \
84+
-d '{
85+
"model": "ai21-jamba-1.5-mini-flagos",
86+
"messages": [{"role": "user", "content": "你好"}]
87+
}'
88+
```
89+
90+
#### 3. Model Interaction
91+
92+
- After model loading is complete:
93+
- Click **"New Conversation"**
94+
- Enter your question (e.g., “Explain the basics of quantum computing”)
95+
- Click the send button to get a response
96+
# Technical Overview
97+
**FlagOS** is a fully open-source system software stack designed to unify the "model–system–chip" layers and foster an open, collaborative ecosystem. It enables a “develop once, run anywhere” workflow across diverse AI accelerators, unlocking hardware performance, eliminating fragmentation among vendor-specific software stacks, and substantially lowering the cost of porting and maintaining AI workloads. With core technologies such as the **FlagScale**, together with vllm-plugin-fl, distributed training/inference framework, **FlagGems** universal operator library, **FlagCX** communication library, and **FlagTree** unified compiler, the **FlagRelease** platform leverages the **FlagOS** stack to automatically produce and release various combinations of \<chip + open-source model\>. This enables efficient and automated model migration across diverse chips, opening a new chapter for large model deployment and application.
98+
## FlagGems
99+
FlagGems is a high-performance, generic operator libraryimplemented in [Triton](https://github.com/openai/triton) language. It is built on a collection of backend-neutralkernels that aims to accelerate LLM (Large-Language Models) training and inference across diverse hardware platforms.
100+
## FlagTree
101+
FlagTree is an open source, unified compiler for multipleAI chips project dedicated to developing a diverse ecosystem of AI chip compilers and related tooling platforms, thereby fostering and strengthening the upstream and downstream Triton ecosystem. Currently in its initial phase, the project aims to maintain compatibility with existing adaptation solutions while unifying the codebase to rapidly implement single-repository multi-backend support. Forupstream model users, it provides unified compilation capabilities across multiple backends; for downstream chip manufacturers, it offers examples of Triton ecosystem integration.
102+
## FlagScale and vllm-plugin-fl
103+
Flagscale is a comprehensive toolkit designed to supportthe entire lifecycle of large models. It builds on the strengths of several prominent open-source projects, including [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) and [vLLM](https://github.com/vllm-project/vllm), to provide a robust, end-to-end solution for managing and scaling large models.
104+
vllm-plugin-fl is a vLLM plugin built on the FlagOS unified multi-chip backend, to help flagscale support multi-chip on vllm framework.
105+
## **FlagCX**
106+
FlagCX is a scalable and adaptive cross-chip communication library. It serves as a platform where developers, researchers, and AI engineers can collaborate on various projects, contribute to the development of cutting-edge AI solutions, and share their work with the global community.
107+
108+
## **FlagEval Evaluation Framework**
109+
FlagEval is a comprehensive evaluation system and open platform for large models launched in 2023. It aims to establish scientific, fair, and open benchmarks, methodologies, and tools to help researchers assess model and training algorithm performance. It features:
110+
- **Multi-dimensional Evaluation**: Supports 800+ modelevaluations across NLP, CV, Audio, and Multimodal fields,covering 20+ downstream tasks including language understanding and image-text generation.
111+
- **Industry-Grade Use Cases**: Has completed horizonta1 evaluations of mainstream large models, providing authoritative benchmarks for chip-model performance validation.
112+
113+
# Contributing
114+
115+
We warmly welcome global developers to join us:
116+
117+
1. Submit Issues to report problems
118+
2. Create Pull Requests to contribute code
119+
3. Improve technical documentation
120+
4. Expand hardware adaptation support
121+
# License
122+
The model weights are derived from AI-ModelScope/AI21-Jamba-1.5-Mini and are open‑sourced under the Apache License 2.0: https://www.apache.org/licenses/LICENSE-2.0.txt
123+
124+
125+

0 commit comments

Comments
 (0)