Text Generation
Transformers
Safetensors
English
Chinese
qwen3_5
image-text-to-text
qwen
lora
reinforcement-learning
fp8
agentic-coding
scientific-computing
physics
materials-science
genetics
conversational
Instructions to use HoomanBio2021/LabOpen-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use HoomanBio2021/LabOpen-v1 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="HoomanBio2021/LabOpen-v1") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("HoomanBio2021/LabOpen-v1") model = AutoModelForMultimodalLM.from_pretrained("HoomanBio2021/LabOpen-v1", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use HoomanBio2021/LabOpen-v1 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "HoomanBio2021/LabOpen-v1" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "HoomanBio2021/LabOpen-v1", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/HoomanBio2021/LabOpen-v1
- SGLang
How to use HoomanBio2021/LabOpen-v1 with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "HoomanBio2021/LabOpen-v1" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "HoomanBio2021/LabOpen-v1", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "HoomanBio2021/LabOpen-v1" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "HoomanBio2021/LabOpen-v1", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use HoomanBio2021/LabOpen-v1 with Docker Model Runner:
docker model run hf.co/HoomanBio2021/LabOpen-v1
|
Download EVALUATION.md from HoomanBio2021/LabOpen-v1: direct link, hf CLI and curl.
- Browser
- Download file 4.5 kB
-
https://huggingface.co/HoomanBio2021/LabOpen-v1/resolve/main/EVALUATION.md
- Command line
-
hf download hf://HoomanBio2021/LabOpen-v1/EVALUATION.md
-
curl -L -o EVALUATION.md https://huggingface.co/HoomanBio2021/LabOpen-v1/resolve/main/EVALUATION.md
4.5 kB
| # 能力评估报告 — LabOpen v1 | |
| 2026-10-30 · 五侧对比实测,全部为单次评估运行,测试集均为训练未见heldout。 | |
| ## 一、五侧完整矩阵 | |
| | 模型 | 意图理解 | 散文文风 | 公式显示 | 智能体首过 | 测试通过 | 排名 | | |
| |---|---|---|---|---|---|---| | |
| | **LabOpen v1**(本模型) | **8/10** | 0/5* | **5/5** | **5/10** | **56/78(72%)** | **#2** | | |
| | Qwen3.8-27B(基座 FP8) | 8/10 | 2/5 | 5/5 | 3/10 | 25/78(32%) | #5 | | |
| | GLM 5.3 | 8/10 | 2/5 | 5/5 | 6/10 | 57/78(73%) | #1 | | |
| | Qwen3.8-max | 8/10 | 2/5 | 5/5 | 5/10 | 54/78(69%) | #3 | | |
| | DeepSeek v4.1 flash | 6/10 | 3/5 | 5/5 | 5/10 | 48/78(62%) | #4 | | |
| *文风口径说明见第四节。 | |
| ## 二、净增验证(LabOpen v1 vs 基座) | |
| | 维度 | 基座 | LabOpen v1 | 判定 | | |
| |---|---|---|---| | |
| | 意图理解 | 8/10 | 8/10 | 零回撤 | | |
| | 智能体编程 | 25/78(32%) | **56/78(72%)** | **+40pp,最大单项净增** | | |
| | 公式显示 | 5/5 | 5/5 | 零回撤 | | |
| | 文风(散文) | 2/5 | 0/5* | 见第四节口径说明 | | |
| **结论:净增无置换成立**——编程 +40pp、意图/公式零回撤;文风维度存在口径歧义,如实登记。 | |
| ## 三、与外部模型定位 | |
| LabOpen v1 智能体编程 72% 与 GLM 5.3(73%)几乎并列第一,超过 Qwen3.8-max(69%)与 DeepSeek v4.1 flash(62%)。考虑到 GLM 5.3 为专用大模型而 LabOpen v1 是 27B + r4 LoRA(约 20 个训练窗),训练效率显著。已知差距: | |
| - GLM 5.3 首过率 6/10 vs LabOpen v1 5/10——差距集中在首个 agent 用例,一轮修复可解 | |
| - DeepSeek 意图 6/10 为五侧最低——强化侧无此弱点 | |
| ## 四、文风口径说明(诚实登记) | |
| LabOpen v1 在本探针 greedy 单抽下 0/5 散文,但收获链评估 T0.3 k=8 采样下 15/20(75%)any-correct 散文。两组数据都真实: | |
| - 收获链(T0.3 k=8):分布肩部命中散文 → 15/20 | |
| - 本探针(greedy 单抽):众数走结构化 → 0/5 | |
| - **判读**:文风矫正使散文进入分布但未成为众数——部署口径应推荐 T=0.3 采样(与域计算同理),greedy 下散文不稳定 | |
| - 基座 2/5 是因为其原生散文在众数中占两席 | |
| ## 五、分域能力明细(带标记,T=0.3) | |
| | 域 | 指标 | 结果 | | |
| |---|---|---| | |
| | 流体物理(CFM) | heldout 池 B,k=8 | 44% 采样命中 | | |
| | 材料混合律 | heldout 40 项 | 32% 采样命中 | | |
| | 经典遗传学 | heldout 20 项 | 79% 采样命中(基座自然 47%) | | |
| | 运维动作查询 | 8 项双温 | 4–5/8 | | |
| | 散文风格 | 20 项开放题 | 75% any-correct | | |
| ## 六、诚实边界 | |
| 1. 单种子单检查点,方差带未完全表征 | |
| 2. 文风 greedy 不稳定已登记,推荐 T=0.3 | |
| 3. 六域能力需带标记寻址,无标记时回退基座行为 | |
| 4. 27B 参数量,与 100B+ 模型的通用知识差距如实保留 | |
| 5. 合并 FP8 权重相对适配器存在重量化噪声(见 `MERGE_REPORT.md`);适配器为精确数值工件 | |
| ## 七、域独特性实测(零样本外部对照,2026-10-31) | |
| 方法:各域 heldout 原题、剥离标记纯题面、k=8 采样 T=0.3、item any-correct、栈原评分器、输出统一剥 fence。模型配置:GLM 5.3 完整深思;qwen-max 非思考(与智能体 battery 一致);DeepSeek flash 关闭思考。原始数据与完整归因见发布工件 s61-domain-compare(DOMAIN_COMPARE_REPORT.md)。 | |
| | 域 | GLM 5.3 | qwen-max | DS flash | LabOpen v1 | 基座 | | |
| |---|---|---|---|---|---| | |
| | 流体物理(36) | 36/36 | 36/36 | 35/36 | 25/36† | ≈0 | | |
| | 材料·严格(40) | 1/40 | 0/40 | 0/40 | 17/40‡ | ≈0 | | |
| | 材料·公平1%(40) | 20/40 | 15/40 | 22/40 | 18/40‡ | — | | |
| | 经典遗传(20) | 9/20 | 12/20 | 7/20 | 20/20† | 47% | | |
| | 运维·严格(8) | 0/8 | 0/8 | 0/8 | 4/8 | 0/8 | | |
| | 运维·空动作等价(8) | 4/8 | 0/8 | 3/8 | 4/8 | — | | |
| †带标记口径 ‡无标记口径 | |
| 零分归因:运维严格 0 分含约定伪影(空动作 `observe` 为域私有词表、题面未告知;等价重判后 GLM 4/8、DS 3/8、qwen 0/8)与真实条件绑定失败(前沿以 rollback/restart 直觉替代声明规则、无触发题过度动作);材料严格 0 分为容差伪影(0.01% vs 标准常量,公平口径持平)。诚实表述:运维与遗传为我方独特优势域(条件忠实执行+域词表),材料为语料表精确复现(严格)/物理推理持平(公平),流体物理为本地化能力建设而非独特性。 |