Search public pages, research tools, and SAVRN solutions.
Open-weight model · Text classification
by Han Wang HanWangTJU/haidass1.5-143m-jev
haidass1.5-143m-jev is an open-weight model for text classification from Han Wang, released under Apache License 2.0. Its published files total 28.3 MB.
Haidass1.5-143M-Jev 是一个研究性质的 Jev 类决策模型。它以 为冻结基座，通过 LoRA 和一个可训练选项指针头输出 choice、noul 和 score 模型不会生成答案文本。加载时使用 Haidass 的 Transformer backbone，不加载语言模型词表投影头； 多个问题共享状态前缀，并通过 block-causal attention mask 隔离问题分支。指针头读取每个选项边界和 位置的隐藏状态，为请求中的现有选项直接打分。 本仓库发布的是 LoRA…
By Han Wang, published under apache-2.0, revision 0063a7e9da51.
Haidass1.5-143M-Jev 是一个研究性质的 Jev 类决策模型。它以 为冻结基座，通过 LoRA 和一个可训练选项指针头输出 choice、noul 和 score 模型不会生成答案文本。加载时使用 Haidass 的 Transformer backbone，不加载语言模型词表投影头； 多个问题共享状态前缀，并通过 block-causal attention mask 隔离问题分支。指针头读取每个选项边界和 位置的隐藏状态，为请求中的现有选项直接打分。 本仓库发布的是 LoRA adapter、指针头、tokenizer、训练配置、开发集评测结果和复现实验代码。 运行时仍需下载原始 Haidass 基座。 发布 checkpoint 来自两阶段继续训练： 1. 从初始 Kev-style Haidass-Jev checkpoint 开始，使用 Kev decision-v7 数据继续训练； 2. 选择学习率 2e-5 的 step 150，再以 90% Kev + 10% 规则合成数据训练 150 step。 这些是 development 指标。test split 没有用于选模，本次发布也不声明 test 指标。 最终 checkpoint 的监督目标是问题内平均交叉熵，再按源请求平均。它没有采用未完成退火实验中的 Brier/RPS/permutation 辅助损失。 主要训练数据来自 Kev 的 frozen decision-v7 suite。经过精确去重和 tokenizer 长度检查后， 规则合成集包含 5,000 个训练请求和 800…
Haidass1.5-143M-Jev 是一个研究性质的 Jev 类决策模型。它以
DALabCommunity/Haidass1.5-143M
为冻结基座，通过 LoRA 和一个可训练选项指针头输出 choice、noul 和 score
问题的概率分布。
模型不会生成答案文本。加载时使用 Haidass 的 Transformer backbone，不加载语言模型词表投影头；
多个问题共享状态前缀，并通过 block-causal attention mask 隔离问题分支。指针头读取每个选项边界和
<decide> 位置的隐藏状态，为请求中的现有选项直接打分。
本仓库发布的是 LoRA adapter、指针头、tokenizer、训练配置、开发集评测结果和复现实验代码。 运行时仍需下载原始 Haidass 基座。
发布 checkpoint 来自两阶段继续训练：
2e-5 的 step 150，再以 90% Kev + 10% 规则合成数据训练 150 step。最终选择依据是开发集综合分数：
0.6 × Decision-v7 accuracy
+ 0.4 × Transfer-v4 accuracy
- 0.1 × (Decision-v7 Brier + Transfer-v4 Brier)
| Development suite | Accuracy | Brier ↓ | NLL ↓ | ECE ↓ |
|---|---|---|---|---|
| Decision-v7 | 0.5866 | 0.5057 | 0.9325 | 0.0378 |
| Transfer-v4 | 0.4771 | 0.5838 | 0.9692 | 0.0700 |
| Haidass synthetic v1 | 0.4125 | 0.6255 | 1.0376 | 0.0685 |
这些是 development 指标。test split 没有用于选模，本次发布也不声明 test 指标。
| 配置 | 值 |
|---|---|
| Base | DALabCommunity/Haidass1.5-143M |
| LoRA | rank 16, alpha 32, dropout 0.05 |
| LoRA targets | q/k/v/o 和 gate/up/down projections |
| Pointer head | hidden 576 → projection 256 |
| Global batch | 128 |
| Peak learning rate | 2e-5 |
| Scheduler | 5% warmup + cosine decay to 10% peak |
| Optimizer | AdamW, weight decay 0.01 |
| Precision | FP32 master weights, BF16 autocast |
| Checkpoint interval | 50 optimizer steps |
| Seed | 0 |
| Context admission | state ≤ 384 tokens; each branch must fit with state in 1024 tokens; packed ≤ 2048 |
最终 checkpoint 的监督目标是问题内平均交叉熵，再按源请求平均。它没有采用未完成退火实验中的 Brier/RPS/permutation 辅助损失。
主要训练数据来自 Kev 的 frozen decision-v7 suite。经过精确去重和 tokenizer 长度检查后，
12,049 个请求被保留，420 个重复请求和 107 个超长请求被移除。
规则合成集包含 5,000 个训练请求和 800 个开发请求，中英文各半。每个请求包含 Choice、Noul、
Score 三类问题；行动标签在 proceed、review、block 之间严格均衡。每两个样本组成只修改一个
决策关键字段的反事实对，标签由确定性规则重新计算。合成数据不包含模型自标注或人为伪造的软概率。
生成器见 code/generate_haidass_synthetic.py。训练数据本身放在独立
dataset 仓库 HanWangTJU/haidass-jev-synthetic-v1。
本模型依赖 Kev 的自定义 block mask、输入编码和指针头，不能直接用
AutoModelForCausalLM.generate()。
git clone https://github.com/jaredpalmer/kev.git
cd kev
uv sync --extra serve
uv run --extra serve python -m kev.serve \
--run HanWangTJU/haidass1.5-143m-jev \
--port 8009
也可以从 Python 加载：
from kev.evaluate import load
tokenizer, model = load(
"HanWangTJU/haidass1.5-143m-jev",
"cuda", # Ascend 环境中使用 "npu"
merge=False,
)
输入必须走 Kev 的 SystemOneRequest / materialize() 路径，确保服务和训练使用相同格式，并防止用户文本
伪造内部边界 token。
option_isolation=False：问题分支彼此隔离，但同一问题内的选项不是结构性隔离。完整实验历史、失败分析和后续建议见
TRAINING_REPORT_ZH.md。
架构和研究工具基于 Apache-2.0 许可的
jaredpalmer/kev，模型基座
DALabCommunity/Haidass1.5-143M
同样采用 Apache-2.0。本仓库新增代码和权重按 Apache-2.0 发布。
24 files, 28.3 MB in total. The weights are 2 files totalling 20.8 MB in pt, safetensors.
| File | Type | Size | SHA-256 |
|---|---|---|---|
| adapter_model.safetensors | Weights | 19.6 MB | a7b8fb92b1ed |
| head.pt | Weights | 1.2 MB | 57f3448542fb |
| adapter_config.json | Configuration | 1.2 KB | — |
| code/final_test_eval.py | Configuration | 4.7 KB | — |
| code/generate_haidass_synthetic.py | Configuration | 7.8 KB | — |
| code/inference.py | Configuration | 1.2 KB | — |
| code/train_v5.py | Configuration | 15.4 KB | — |
| code/v5_data.py | Configuration | 6.7 KB | — |
| data/training_manifest.json | Configuration | 14.8 KB | — |
| metrics/development-decision-v7.json | Configuration | 19.7 KB | — |
| metrics/development-synthetic-v1.json | Configuration | 5.6 KB | — |
| metrics/development-transfer-v4.json | Configuration | 13.3 KB | — |
| special_tokens_map.json | Configuration | 610 B | — |
| training_config.json | Configuration | 1.3 KB | — |
| LICENSE | Documentation | 11.3 KB | — |
| NOTICE.md | Documentation | 497 B | — |
| README.md | Documentation | 5.2 KB | — |
| TRAINING_REPORT_ZH.md | Documentation | 12.9 KB | — |
| chat_template.jinja | Other | 4.2 KB | — |
| requirements.txt | Other | 82 B | — |
| .gitattributes | Repository | 1.5 KB | — |
| tokenizer.json | Tokenizer | 6.1 MB | — |
| tokenizer.model | Tokenizer | 1.3 MB | dddb2e3e527f |
| tokenizer_config.json | Tokenizer | 5.8 KB | — |
Released by Han Wang through its official repository on Hugging Face. Read the license.
| Precision | Weights in memory |
|---|---|
| As published | 20.8 MB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Yes. haidass1.5-143m-jev is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
This is a RoBERTa-base model trained on ~124M tweets from January 2018 to December 2021, and finetuned for sentiment analysis with the TweetEval benchmark. The original Twitter-based RoBERTa model can be found here and the original reference paper is TweetEval. This model is suitable for English. 0 -> Negative; 1 -> Neutral; 2 -> Positive This sentiment analysis model has been integrated into TweetNLP. You can access the demo here.
FinBERT is a BERT model pre-trained on financial communication text. The purpose is to enhance financial NLP research and practice. It is trained on the following three financial communication corpus. The total corpora size is 4.9B tokens. More technical details on FinBERT: Click Link This released finbert-tone model is the FinBERT model fine-tuned on 10,000 manually annotated (positive, negative, neutral) sentences from analyst reports. This model achieves superior performance on financial tone analysis task. If you are simply interested in using FinBERT for financial tone analysis, give it a try. If you use the model in your academic work, please cite the following paper: Huang, Allen H.…
https://huggingface.co/cross-encoder/ms-marco-MiniLM-L-6-v2 with ONNX weights to be compatible with Transformers.js. If you haven't already, you can install the Transformers.js JavaScript library from NPM using: Note: Having a separate repo for ONNX weights is intended to be a temporary solution until WebML gains more traction. If you would like to make your models web-ready, we recommend converting to ONNX using Optimum and structuring your repo like this one (with ONNX weights located in a subfolder named onnx).
This is a multilingual XLM-roBERTa-base model trained on ~198M tweets and finetuned for sentiment analysis. The sentiment fine-tuning was done on 8 languages (Ar, En, Fr, De, Hi, It, Sp, Pt) but it can be used for more languages (see paper for details). This model has been integrated into the TweetNLP library.
The huggingface models currently give different results to the detoxify library (see issue here). For the most up to date models we recommend using the models from https://github.com/unitaryai/detoxify Trained models & code to predict toxic comments on 3 Jigsaw challenges: Toxic comment classification, Unintended Bias in Toxic comments, Multilingual toxic comment classification. Built by Laura Hanu at Unitary, where we are working to stop harmful content online by interpreting visual content in context. Score not directly comparable since it is obtained on the validation set provided and not on the test set. To update when the test labels are made available. It is also noteworthy to mention…
With this model, you can classify emotions in English text data. The model was trained on 6 diverse datasets (see Appendix below) and predicts Ekman's 6 basic emotions, plus a neutral class: 1) anger 2) disgust 3) fear 4) joy 5) neutral 6) sadness 7) surprise The model is a fine-tuned checkpoint of DistilRoBERTa-base. For a 'non-distilled' emotion model, please refer to the model card of the RoBERTa-large version. a) Run emotion model with 3 lines of code on single text example using Hugging Face's pipeline command on Google Colab: b) Run emotion model on multiple examples and full datasets (e.g.,.csv files) on Google Colab: Please reach out to [email protected] if you have any…