Text Classification Models

libraryname: transformers - autotrain - text-classification basemodel: google-bert/bert-base-uncased f1macro: 0.7533020080884588 f1micro: 0.7533333333333333 f1weighted: 0.7533020080884587 precisionmacro: 0.7551310982162045 precisionmicro: 0.7533333333333333 precisionweighted: 0.7551310982162046 recallmacro: 0.7533333333333333 recallmicro: 0.7533333333333333 recallweighted: 0.7533333333333333

Model · Text classification

auto-200m-2-int4

SSH

This is auto-200m-2 with its weights stored as 4-bit integers. The file is 77 MB instead of 299 MB, about a quarter of the size. It gets 2,884/3,000 on the benchmark, with 60 false approvals and 56 false denials. The BF16 model gets 2,890, with 53 and 57. 44 of the 3,000 decisions differ from the BF16 model. auto-200m-2 is a 149.6M-parameter ModernBERT classifier. It reads an AI agent's proposed tool call, the user's request and the agent's history, then answers approve or deny. It takes up to 65,536 tokens of context. This version loads through one small Python file, autoquant.py, in plain PyTorch, without compiled kernels. It was checked on NVIDIA CUDA, the Apple M4 Max CPU and Apple MPS.…

Model · Text classification

auto-200m-2-int8

SSH

This is auto-200m-2 with its weights stored as 8-bit integers. The file is 150 MB instead of 299 MB, and it gets the same benchmark score: 2,890/3,000, with 53 false approvals and 57 false denials. Only 2 of the 3,000 decisions differ from the BF16 model. auto-200m-2 is a 149.6M-parameter ModernBERT classifier. It reads an AI agent's proposed tool call, the user's request and the agent's history, then answers approve or deny. It takes up to 65,536 tokens of context. This version loads through one small Python file, autoquant.py, in plain PyTorch, without compiled kernels. It was checked on NVIDIA CUDA, the Apple M4 Max CPU and Apple MPS. For an even smaller file, see auto-200m-2-int4 (77…

When a caller starts talking while a voice agent is still speaking, should the agent stop? This model makes that decision from text, in a few milliseconds. It reads one moment of a call: what the agent had already said, what it was about to say, and the caller's speech-recognised words. It then returns the probability that the agent should stop (INTERRUPT) or keep talking (CONTINUE). The CPU-friendly model: 38 ms per decision on 4 CPU threads. It matches the 400m on held-out business types and real calls, and gives up about 2 points under distribution shift. Have a GPU? bargein-classifier-ettin-400m is the flagship: more robust to distribution shift, at 8.5 ms on a GPU. Callers talk over…

When a caller starts talking while a voice agent is still speaking, should the agent stop? This model makes that decision from text, in a few milliseconds. It reads one moment of a call: what the agent had already said, what it was about to say, and the caller's speech-recognised words. It then returns the probability that the agent should stop (INTERRUPT) or keep talking (CONTINUE). The flagship model: the most robust to distribution shift (2 points ahead of the 150m on OOD-10k), at ~9 ms per decision on a GPU. Need CPU-only inference? bargein-classifier-ettin-150m runs in 38 ms on 4 CPU threads. Callers talk over voice agents all the time, and most of it should not stop the agent: "mm…

Model · Text classification

belief-engine-argument-quality-deberta-v3-large

Joshua C. Yang

This repository releases the DeBERTa-v3-large argument-quality regressor used in Belief Engine's judgment layer. Fine-tuned on the IBM Argument Quality Ranking dataset's WA human ratings, it takes an English argument and returns a quality score in [0, 1]. This is a reproducibility release of a fine-tuned checkpoint using established argument-quality regression methods. It provides the exact scorer used in the framework, verified inference, evaluation tools, and a documented recipe for training a new checkpoint. It has 435,062,785 parameters and a 1.74 GB weight file; the original December 2025 weights are preserved unchanged. Download the published model, scorer, and training/evaluation…

Full fine-tuning of bert-base-uncased for 4-class news topic classification, delivered as part of assignment U2T01 (Adapting BERT for NLP tasks — Trends in Data Science, Unit 2, Universidad Politécnica de Yucatán). bert-base-uncased with a linear classification head on top of the [CLS] token's last hidden state, mapping to 4 topic classes: World, Sports, Business, Sci/Tech. - 10% of the official train split was held out as validation; the official test split was left untouched. - 4 balanced classes. Two adaptation methods were trained and compared; full fine-tuning is the delivered model, since it beat the feature-based baseline by a margin well above the run-to-run noise floor (±1–3 points…

A System One decision model with open weights. Give it a document (text, email, ticket, log or JSON) and typed questions; it returns typed answers (yes/no, a choice between options you describe, or a score on levels you define) with a calibrated probability for every option, in one pass. It never generates text, so there is nothing to parse and nothing to hallucinate. It runs on your own machines: your documents never leave them. About 26B parameters, built on Google Gemma 4 26B-A4B (mixture of experts, about 4B active per token). Everything it needs is in this repository, so nothing else is downloaded. English text, up to 4,096 input tokens; needs one GPU with about 60 GB of memory (bf16).…

A Bidirectional LSTM that classifies Indonesian public comments about a government policy (the Kopdes program) into three sentiment classes: Negatif, Netral, and Positif. The model is trained from scratch (no pretrained embeddings) on a small dataset, so treat it as a course/research-level baseline rather than a production classifier. - About 850 unique comments remained after preprocessing and deduplication. - Labels are not human-annotated. They were generated automatically with the InSet lexicon: score = (number of positive words) − (number of negative words) in the stemmed text. Score > 0 is Positif, < 0 is Negatif, and 0 is Netral. Because of this, the reported scores measure how well…

Model · Text classification

botfilter

Hraness

Botfilter is a browser extension that scores English text for “likely AI-written” as a calibrated probability, on the device. A score can be wrong and should not be used as proof of authorship. Each releases/ / directory contains the ONNX model, the WordPiece vocabulary, the calibration settings, and a lockfile with SHA-256 checksums. Use all files from the same version. The extension’s inference implementation defines preprocessing, window aggregation, and calibration; these files do not provide a Transformers pipeline. Human text: openly licensed news articles from Common Pile (only documents whose metadata records CC BY 4.0), Foodista food blog posts (CC BY 3.0), UK Parliament Hansard…

Base model: indobenchmark/indobert-base-p1; Input: title, description; Tokenization: maxlength=256, truncation=True; Output: Single label, with confidence score for each label..

Model · Text classification

clefsift

Stephen Murray

ClefSift is an experimental English text classifier fine-tuned from Cloudflare/Clef. It returns scores for four document-level labels in one forward pass: These are provisional, process-based labels. The model sees writing style, not the author's actual workflow. Its output does not prove authorship. Do not use it as the sole basis for academic, employment, or moderation decisions. This release uses the selected v2 step-128 checkpoint. The additional rank-64 LoRA weights are merged into the BF16 backbone. The trained joint decision head is stored separately in FP32. This is a decision model, not a chat model. The retained config and loader use the Qwen3.5 architecture identifiers above. Pin…

Model · Text classification

clm

Ollaya

Ollaya package of Contrastive-LM/CLM-v0.1-8B and Qwen/Qwen3-8B by Contrastive-LM (projection heads) and the Qwen team (Qwen3-8B encoder). Ollaya runs open decision models locally, the way Ollama runs LLMs: typed questions in, calibrated answers out, behind a TypeSafe-compatible API. This repository holds only the files Ollaya derives, with no weights. Each graph is an ONNX export of the original model whose weights reference the authors' own weight files by byte offset, so ollaya pull downloads the weights from the upstream repositories, unmodified and pinned to a commit, and verifies their sha256. Each tag has an fp32 graph, used on CPU and GPU. Each tag also has decision.json (sequence…

Model · Text classification

compressor-reflex

Eric Maddox

compressor-reflex is a small (151M-parameter) extractive line-level compressor for tool output in agentic coding loops (IDEs such as Cursor and Antigravity). It scores every line of a tool result for keep/drop and runs post-tool, pre-context — cutting the tokens that reach the next model call while guaranteeing safety-critical lines are never dropped. The companion MCP server packages this model for IDE deployment (Cursor, Antigravity, Claude Desktop): install with pip install compressor-reflex-mcp (PyPI). It downloads these weights from this repo on first run and exposes compresstooloutput / compressfile tools plus a transparent proxy mode. Intended use: compressing tool outputs (build…

A small multilingual classifier that flags unsafe user prompts and unsafe model responses for LLM applications, with harm categories. It is a fast encoder (ModernBERT architecture, mmBERT backbone) that you can run on CPU or in the browser in front of, or behind, any LLM. real-world prompts; evaluated on 17 (PolyGuard), 14 (textdetox) and 8 (Aya) languages. Part of the Horizon Labs guard family: prompt-injection-guard, Decision rule: flag when unsafe >= 0.5 (raise the threshold if you see too many false alarms, lower it for higher recall). Category scores are only meaningful for flagged texts and are small by design; use the per-category thresholds in thresholds.json (chosen on a validation…

Model · Text classification

content-safety-guard-small-ONNX

ONNX Community

This is an ONNX version of Horizon-Labs/content-safety-guard-small. It was automatically converted and uploaded using this Hugging Face Space. See the pipeline documentation for text-classification: https://huggingface.co/docs/transformers.js/api/pipelines#modulepipelines.TextClassificationPipeline A small multilingual classifier that flags unsafe user prompts and unsafe model responses for LLM applications, with harm categories. It is a fast encoder (ModernBERT architecture, mmBERT backbone) that you can run on CPU or in the browser in front of, or behind, any LLM. real-world prompts; evaluated on 17 (PolyGuard), 14 (textdetox) and 8 (Aya) languages. Which size? This small model has less…

Model · Text classification

credential-line-32m

ZentropyLabs.ai

A 32M-parameter classifier that reads one line of code, config, log or prose and scores whether the line holds a literal credential: a password, API key, access token, private-key header, connection string with a password, or an Authorization header value. It runs on a laptop CPU in about 4 ms per line, so you can score every line of a trace, log or diff before you share it. Use it next to a rule-based scanner, not in place of one. On 1,264 human-written lines it finds about 4 in 5 credentials and is right about 9 times in 10 when it flags a line. It also misses about 1 credential in 5, so it must never be the only check before you publish a file. predictonnx.py applies the temperature…

If you take one thing from this page: this model does not do customer service. It watches something else do customer service and vetoes bad steps. Everything below walks you through that idea, then the evidence, then the exact commands. Imagine a support agent (human or AI) working a ticket. The work is a loop: look up the caller, check their history, open a case, log the call. The dangerous mistake is doing steps out of order — logging a call when no case exists means the record vanishes. This model sits beside that loop and answers four questions about the current situation: 1. nexttool — "which step comes next?" (look up / check history / open case / log call) 2. safetolog — "does a case…

Model · Text classification

d1

Ollaya

Ollaya package of LiquidAI/d1-3B-GGUF by Liquid AI. Ollaya runs open decision models locally, the way Ollama runs LLMs: typed questions in, calibrated answers out, behind a TypeSafe-compatible API. This repository holds only the files Ollaya derives, with no weights. The model is the authors' own GGUF file: ollaya pull downloads it from their repository, unmodified and pinned to a commit, verifies its sha256, and Ollaya runs it on llama.cpp. Each tag has decision.json (the prompt, the option labels Ollaya reads and llama.cpp's settings) and calibration.json (temperatures). A vision tag also pulls the vision projector the table names, unmodified from its author's repository, which reads the…

Model · Text classification

Darwin-27B-ZTC-v2

FINAL_Bench

A zero-token decision engine from the Darwin family, second version. Darwin-27B-ZTC-v2 reads a piece of state and a typed question (noul yes or no, choice one of N labels, score an ordered rubric) and returns a probability for every option. It uses one forward pass per question and generates no tokens. v2 adds control and workflow decisions (games, drones, retrieval control, entity alignment, customer and incident workflows) on top of FINAL-Bench/Darwin-27B-ZTC. 1 on the S1MB leaderboard under both of its rankings: the default Borda score and Task Avg. All 137 benchmarks complete, measured with the S1MB evaluator and its autojev adapter at revision a8f283a, then validated and merged by the…

DEBATE-kor-base is a Korean-adapted Political DEBATE model for binary natural language inference (NLI) on political text. The model is initialized from mlburnham/PoliticalDEBATEDeBERTabasev1.1, the original DeBERTa-based Political DEBATE checkpoint, and subsequently fine-tuned on jongrock17/PolNLI-kor, a Korean translation and adaptation of PolNLI. Political DEBATE DeBERTa-base → PolNLI-kor → DEBATE-kor-base Unlike the PolNLI-kor-RoBERTa model family, which starts from Korean-pretrained KLUE-RoBERTa encoders, DEBATE-kor directly adapts the original Political DEBATE checkpoint to Korean political NLI. DEBATE-kor formulates NLI as a binary classification problem. notentailment combines…

DEBATE-kor-large is a Korean-adapted Political DEBATE model for binary natural language inference (NLI) on political text. The model is initialized from mlburnham/PoliticalDEBATEDeBERTalargev1.1, the original DeBERTa-based Political DEBATE checkpoint, and subsequently fine-tuned on jongrock17/PolNLI-kor, a Korean translation and adaptation of PolNLI. Political DEBATE DeBERTa-large → PolNLI-kor → DEBATE-kor-large Unlike the PolNLI-kor-RoBERTa model family, which starts from Korean-pretrained KLUE-RoBERTa encoders, DEBATE-kor directly adapts the original Political DEBATE checkpoint to Korean political NLI. DEBATE-kor formulates NLI as a binary classification problem. notentailment combines…

Model · Text classification

decider-0.8b-fp8

LLM Tech

Mapika/decider-0.8b v1 quantized to FP8 for vLLM: FP8 E4M3 weights with one scale per output channel and FP8 activations scaled per token at run time. 1.01 GB against 1.5 GB for the bf16 checkpoint. Quantized and measured by LLM Tech; the model, its training and its evaluation protocol are Mapika's. Read the bf16 card for what the model is and how it was trained. The base revision is a0a01d6f8135298f400a8c856b355793012ae971. Tokenizer, chat template, generation config and deciderconfig.json (temperatures included) are the author's files unchanged, apart from the version and quantization fields. Both models were run through vLLM 0.29.0 on the same rows: the author's regression set rebuilt…

The decider-2b System One decision model (Qwen3.5-2B-Base, 1.9B parameters, 32768-token context, checkpoint revision v11 — the vendor's default), packaged for Verdict — a high-performance serving engine for System One decision models. Serve it with one command: This package serves through the torch runtime, which runs on CUDA only in v0.2, so there is one image and it needs a GPU. The -v mounts cache the checkpoint, so only the first run downloads it. If the hub is slow or unreachable, add -e HFENDPOINT=https://hf-mirror.com — the mirror covers both the package and the upstream checkpoint, since both are fetched over the Hugging Face path. Add -e VERDICTUSEMODELSCOPE=1 and use…

Model · Text classification

decider-2b-fp8

LLM Tech

Mapika/decider-2b v11 quantized to FP8 for vLLM: FP8 E4M3 weights with one scale per output channel and FP8 activations scaled per token at run time. 2.39 GB against 3.77 GB for the bf16 checkpoint. Quantized and measured by LLM Tech; the model, its training and its evaluation protocol are Mapika's. Read the bf16 card for what the model is and how it was trained. The base revision is 533964dae8be954c5b5e19fa4948e48408094c1e. Tokenizer, chat template, generation config and deciderconfig.json (temperatures included) are the author's files unchanged, apart from the version and quantization fields. Both models were run through vLLM 0.29.0 on the same rows: the author's regression set rebuilt…

Decider-4B is a structured decision model based on TokenRhythm/NeoHorse-1-4B. It uses LoRA to fine-tune the base model and trains the decision head from scratch on our collected decision datasets. The model then undergoes global temperature calibration to further calibrate its outputs. It produces typed probability distributions (noul, choice, score) and is not a chat-generation model. Exact run settings, where recorded by the exporter, are retained in modelmanifest.json; reference scripts under provenance/recipereference/ describe the project recipes rather than prove the exact mixture or hyperparameters of this checkpoint. No additional head-only training phase is claimed for this…

Model · Text classification

decider-4b-AWQ-ASYM-W4A16

Nikita Davidchuk

Activation-aware asymmetric INT4 weight-only quantization of Mapika/decider-4b, revision 99d7fd5bfcc4d11ba37239dafe0ed653803819d3. This is a typed decision model: read probabilities over specified options in one forward pass. It is not trained for conversational text generation. llm-compressor 0.12.0.1, Transformers 5.10.1, PyTorch 2.12.0+cu130. AWQ scaling followed by QuantizationModifier, scheme W4A16ASYM, group size 128, asymmetric integer weights, BF16 activations. savecompressed=True produces physical weightpacked tensors in compressed-tensors format. All Linear layers are targeted except lmhead, linearattn.inproja, linearattn.inprojb, and MTP (if present). Embeddings, norms…

Model · Text classification

decider-4b-fp8

LLM Tech

Mapika/decider-4b v2.1 quantized to FP8 for vLLM: FP8 E4M3 weights with one scale per output channel and FP8 activations scaled per token at run time. 4.85 GB against 8.41 GB for the bf16 checkpoint. Quantized and measured by LLM Tech; the model, its training and its evaluation protocol are Mapika's. Read the bf16 card for what the model is and how it was trained. The base revision is eb5fbdfc9448473ec25e399882912863afbdb70e. Tokenizer, chat template, generation config and deciderconfig.json (temperatures included) are the author's files unchanged, apart from the version and quantization fields. Both models were run through vLLM 0.29.0 on the same rows: the author's regression set rebuilt…

Model · Text classification

decider-4b-nvfp4

LLM Tech

Mapika/decider-4b v2.1 quantized to NVFP4 for vLLM: 4-bit floating-point weights and activations with FP8 block scales (block size 16). 3.29 GB against 8.41 GB for the bf16 checkpoint. Quantized and measured by LLM Tech; the model, its training and its evaluation protocol are Mapika's. Read the bf16 card for what the model is and how it was trained. The base revision is eb5fbdfc9448473ec25e399882912863afbdb70e. Tokenizer, chat template, generation config and deciderconfig.json (temperatures included) are the author's files unchanged, apart from the version and quantization fields. Both models were run through vLLM 0.29.0 on the same rows: the author's regression set rebuilt from public data…

Model · Text classification

decider-4b-tr-judge

Hayri Yigit

English summary. A third LoRA round (rank 64, merged into the bf16 weights) on hayriyigit/decider-4b-tr-rag for a RAG relevance judge: does this retrieved document carry information that contributes directly to the answer, for the asked entity, institution, facet and period? Unlike the -rag model, partial evidence counts as relevant. The input is the production format: state {question, document, today} (in that order), the document rendered as [date] title + body (4000 characters), one fixed instruction. On held-out partial-evidence rows the "yes" rate goes from 0,028 to 0,952 (HotpotQA-tr) and 0,004 to 0,936 (synthetic); the cost is 2-8 points on some negatives (facet 1,000 → 0,920). Most…

Model · Text classification

decima

Ollaya

Ollaya package of amyrmahdy/decima-small and amyrmahdy/decima-base and amyrmahdy/decima-agent by A. M. Madani (amyrmahdy). Ollaya runs open decision models locally, the way Ollama runs LLMs: typed questions in, calibrated answers out, behind a TypeSafe-compatible API. This repository holds only the files Ollaya derives, with no weights. Each graph is an ONNX export of the original model whose weights reference the authors' own weight files by byte offset, so ollaya pull downloads the weights from the upstream repositories, unmodified and pinned to a commit, and verifies their sha256. Each tag has an fp32 graph, used on CPU and GPU. Each tag also has decision.json (sequence layout, special…

Model · Text classification

Decision-1.0-Kai-0.6B

vLLM Semantic Router

Kai, from kairos — the right moment to choose. Open Decision Foundation Models Choose an action, judge a condition, or score against your own rubric. Kai reads the context and candidate descriptions together, then returns structured decisions and probability distributions. 53.52 overall — above both general Laya models, and +7.04 points over the previous Kai release. On the benchmark's 160-question BoolQ slice, Kai scores 74.38%, versus 69.38% for each Laya reference. Same 54 tasks and 3,766 scored questions, with fixed source/family weights. Both Laya references use released general weights, without target-dataset fine-tuning. This is an observed regression suite; results vary by task.…

Model · Text classification

Decision-1.0-Lex-0.6B

vLLM Semantic Router

Lex — Latin for law. Give your decisions a rule. Open Decision Foundation Models Turn operational context into structured decisions. Lex is a specialist fine-tuned from Kai to select actions, test conditions and apply scoring rubrics across four operational workflows. 78.15% accuracy — 1.55 points above Laya Typed Decisions. Lex contains 571,909,635 parameters. Both models were evaluated on the original 2,000-decision English typed-decisions test split, with complete inputs and the official fine-tuned Laya checkpoint. Lex answers 31 more decisions correctly. Observed accuracy gain; the paired 95% interval is −0.20 to +3.15 points. Laya leads on Noul and probability-quality metrics. Full…

Model · Text classification

decision-4b-gguf

Eval Engine

GGUF builds of Decision-4B, an open-weight Jev-like decision model from Eval Engine, the AI arm of Chromia. Give it a state, a question, and a list of options. It answers with one letter. Runs in llama.cpp, Ollama, and on your phone. The Q4KM file is the one that powers Decision-4B in the Unbound app. All three are the LoRA merged into Qwen3.5-4B. The BF16 adapter scores 87.3% on the same panel. Use Q4KM for phones and laptops, Q80 or F16 when you have the memory. Full table and details on the adapter card. The reply is a single letter. toplogprobs gives the score for each option letter; softmax over the listed letters gives a probability per option. Input is a JSON object with state…

Model · Text classification

decision-model-rl-overcooked

Rodney Lafuente-Mercado

Get the code and instructions, or read the write-up and watch the video. Two independently acting chefs share this policy and serve six soups in 512 ticks in PufferLib's Cramped Room. Each sees its public observation expressed in text, game rules, and eight recent action outcomes. The policy scores stay, up, down, left, right, and interact. Each selection executes one native action, without a planner, pathfinder, action macro, or assigned role. This repository contains LoRA adapters and the trained original NLI classifier head, not a standalone base model. The root is cooperative checkpoint 220. single-chef/ contains checkpoint 330, which initialized cooperative training. Both use OpenJev…

Dev is an open-source 399M parameter bidirectional decision model built on ModernBERT-large. It is purpose-built for unstructured-to-structured classification—including ticket routing, yes/no verification, and rating scales—executing in a single forward pass (~28ms on MPS) without token generation. Following Jev (TypeSafe) and Kev-0.5B (Jared Palmer), Dev tests a fundamental architectural question: What if decision models shouldn't be causal decoders at all, but native bidirectional cross-encoders? Evaluated against Jared Palmer's kev-0.5b (built on a frozen Qwen-2.5-0.5B causal backbone + 9.3M LoRA pointer head): Inference Latency: Dev-0.4B executes in 27.6ms on Apple Silicon MPS (M1 Max)…

训练时对文本做了规范化（还原 HTML 转义、链接替换为 http），推理时建议做同样的处理， 见项目中的 triage/preprocess.py。 - 数据：Kaggle NLP with Disaster Tweets 的 train.csv，其中 id ≤ 8696 的 6090 条用于训练，其余 1523 条作为验证集。 - 超参数：BERT 学习率 2e-5，ModernBERT 学习率 5e-5；2 个 epoch；batch size 32；maxseqlength 128； weight decay 0.01；warmup 0.1；bf16；随机种子 42。 同一配置 3 个随机种子的 Macro-F1：BERT 81.63 ± 0.11，ModernBERT 81.70 ± 1.11，两者的差异在种子波动范围内。 - 仅支持英文；比喻性用法（如 "this mixtape is fire"）仍可能误判。 - 两个模型都只在推文（≤ 82 token）上微调，ModernBERT 在长文本上的效果没有经过标注数据验证。 - 模型权重：CC BY-NC 4.0。基于 Kaggle 比赛数据训练，仅限非商业用途；使用时还需遵守该比赛规则中对数据的限制。 - 基础模型：bert-base-uncased、ModernBERT-base（均为 Apache-2.0）。

This model is a fine-tuned version of distilbert-base-uncased on an unknown dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 64 - evalbatchsize: 64 - lrschedulertype: linear - numepochs: 2 - Transformers 5.18.0 - Pytorch 2.11.0+cu130 - Datasets 4.8.5 - Tokenizers 0.23.2

Model · Text classification

distilbert-food-not-food

Saber Bounehas

This model is a fine-tuned version of distilbert-base-uncased on an unknown dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 0.0001 - trainbatchsize: 32 - evalbatchsize: 32 - lrschedulertype: linear - numepochs: 10 - Transformers 5.18.0 - Pytorch 2.11.0+cpu - Datasets 5.0.1 - Tokenizers 0.23.2

Model · Text classification

distilbert-pt-cased-redacao-nota

Marcelo Lampe Zacharias

This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Model · Text classification

dLLM-PRM-Gap-bidir-dream7b

Yan Zhan

Public adapter release for dLLM PRM Gap. bidirectional attention with mask-aware mean pooling over intermediate denoising states - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance Role: process-reward model used by PRM Guided, Hybrid, and the snapshot diagnostics.

Model · Text classification

dLLM-PRM-Gap-bidir-llada8b

Yan Zhan

Public adapter release for dLLM PRM Gap. bidirectional PRM for the cross-backbone LLaDA diagnostic - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance Role: cross-backbone diagnostic; it is not the Dream headline scorer.

Model · Text classification

dLLM-PRM-Gap-c3a-causal-lasttoken-dream7b-s42

Yan Zhan

Public adapter release for dLLM PRM Gap. a C3 control checkpoint with the stated attention/readout protocol and seed. - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance

Model · Text classification

dLLM-PRM-Gap-c3a-causal-lasttoken-dream7b-s43

Yan Zhan

Public adapter release for dLLM PRM Gap. a C3 control checkpoint with the stated attention/readout protocol and seed. - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance

Model · Text classification

dLLM-PRM-Gap-c3b-bidir-lasttoken-dream7b-s42

Yan Zhan

Public adapter release for dLLM PRM Gap. a C3 control checkpoint with the stated attention/readout protocol and seed. - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance

Model · Text classification

dLLM-PRM-Gap-c3b-bidir-lasttoken-dream7b-s43

Yan Zhan

Public adapter release for dLLM PRM Gap. a C3 control checkpoint with the stated attention/readout protocol and seed. - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance

Model · Text classification

dLLM-PRM-Gap-c3c-bidir-meanpool-dream7b-s42

Yan Zhan

Public adapter release for dLLM PRM Gap. a C3 control checkpoint with the stated attention/readout protocol and seed. - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance

Model · Text classification

dLLM-PRM-Gap-c3c-bidir-meanpool-dream7b-s43

Yan Zhan

Public adapter release for dLLM PRM Gap. a C3 control checkpoint with the stated attention/readout protocol and seed. - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance

Model · Text classification

dLLM-PRM-Gap-causal-lasttoken-dream7b

Yan Zhan

Public adapter release for dLLM PRM Gap. causal attention with a last-token readout for the submitted readout diagnostic - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance Role: last-token scorer used for the last-token row in the readout diagnostic.

Model · Text classification

dLLM-PRM-Gap-causal-llada8b

Yan Zhan

Public adapter release for dLLM PRM Gap. causal PRM for the cross-backbone LLaDA diagnostic - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance Role: cross-backbone diagnostic; it is not a full LLaDA ORM replication.

Model · Text classification

dLLM-PRM-Gap-causal-meanpool-dream7b

Yan Zhan

Public adapter release for dLLM PRM Gap. causal attention with mask-aware mean pooling for the submitted readout diagnostic - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance Role: causal mean-pool scorer used for the mean-pool row in the readout diagnostic.

Model · Text classification

dLLM-PRM-Gap-orm-dream7b

Yan Zhan

Public adapter release for dLLM PRM Gap. bidirectional final-state scoring for the matched-compute ORM Rerank baseline - adapter.safetensors — compact adapter weights - config.json — public base-model id, architecture, and provenance

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 8 - evalbatchsize: 8 - lrschedulertype: linear - numepochs: 3 - Transformers 5.16.1 - Pytorch 2.11.0+cpu - Datasets 4.8.5 - Tokenizers 0.23.1

把 Qwen2.5-0.5B 改造为双向 encoder,接 11 个线性分类头,一次前向(17.7ms,p50)输出文本的完整安全画像:违规二判 + TC260 / 语言风险(LM)/ 隐私(PII)/ 内容分级(DC)四套体系的大类与小类 + 安全放行闸门。 - 判别式安全模型,非生成式 guard:无自回归、全头并行输出,单条 17.7ms vs 生成式 8B guard 的 281ms+(同数据重训对比)。 - backbone 为 HF Qwen2Model 结构(Qwen/Qwen2.5-0.5B 权重转 bidirectional 微调产物),推理时需全可见 attention(双向,非 causal)。 - 生产推理推荐配套的纯 C 本地引擎 verdict(CPU / Apple Metal GPU,零 Python 依赖,支持加密.vmd 打包分发)。 safetythreshold=0.95,校准方法 [email protected](FPR≤2% 约束下最大召回),校准集 finetunefinalv68val(259 safe / 1176 unsafe,实测 FPR 1.93%)。该阈值仅对训练分布可靠:偏离训练分布时 FPR 会上升(见局限)。releasegate=0.8055 为 safe 样本放行闸门,减少深误杀。 - 数据:v69a,17589 条(双模型生成 + 交叉验证过滤);四轨标签张量化 ~160 维 0/1,未标注位置 None 掩码。 - 配置:lr 2e-5、bf16、20 epochs(epoch…

An English human/AI detector for local Python, CLI and browser use, with calibrated probabilities, explicit abstention and exploratory source rankings. AI includes direct generation and substantive AI rewriting, including a human original rewritten by DIPPER. Results are not an AI-written word percentage, proof of authorship or authenticated model identity. The frozen binary model passed all preregistered point criteria on four necessary final cohorts and both required rewrite-origin views. This is a measured-scope result, not universal accuracy or a population/per-domain 1% false-positive guarantee. Family and nominal-version assertions are disabled; both acceptedfamily and acceptedmodel…