# humanizer by Stephen Yu: Open-Weight Model Source: https://savrn.com/models/humanizer Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve humanizer (12B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 23.9 GB | 28.7 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 8-bit | 12.0 GB | 14.4 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 6.0 GB | 7.2 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 7, 2026. [humanizer on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/humanizer/gpus) ## Model Card By Stephen Yu, published under apache-2.0, revision c07c06c8fb79. A 12B model that rewrites AI-written drafts (emails, essays, reports, forum posts; English and Chinese) so they read like a person wrote them. It is trained to keep every number, unit, date, name and quote, and to add nothing. It runs locally. No AI detector was used anywhere in training. Usage without the app · 不用 App 怎么用 · AGENTS.md (for AI agents) · [GitHub](https://github.com/sgaofen/humanizer-local-model) · [Desktop app (macOS, Windows)](https://github.com/sgaofen/humanizer-local-model/releases/latest) · [Install guide](https://github.com/sgaofen/humanizer-local-model/blob/main/docs/INSTALL.md) · [中文说明](https://github.com/sgaofen/humanizer-local-model/blob/main/README.zh.md) GGUF files now have their own repo: [jialinyyzz/humanizer-GGUF](https://savrn.com/models/humanizer-gguf): every quantization (Q8_0, Q6_K, Q4_K_M, Q3, 2-bit) with its size, memory, KL to bf16 and quality on the task, and how they were made. The same files stay here too, so existing download commands keep working. Setting this up with an AI agent? Point it at AGENTS.md: exact files, server command, prompt byte for byte, and a self-test. ### Quick start [Read the full model card (4,120 words)](https://savrn.com/models/humanizer/card) ## Configuration Architecture Gemma4UnifiedForConditionalGeneration Context length (tokens) 262,144 Layers 48 Hidden size 3,840 Feed-forward size 15,360 Attention heads 16 Key/value heads 8 Head dimension 256 Vocabulary size 262,144 Sliding window (tokens) 1,024 Model type gemma4_unified ## Identity and Version Repository jialinyyzz/humanizer Publisher Stephen Yu Task Text generation Modality Text Library gguf Parameters 12B parameters Languages en, zh Revision c07c06c8fb790e9235547dc47e8c3fd4738776aa First published 2026-09-11 Last updated 2026-10-07 ## Files and Weights 35 files, 87.6 GB in total. The weights are 7 files totalling 87.6 GB in gguf, safetensors. Weights7 files · 87.6 GB Configuration3 files · 5.1 KB Tokenizer2 files · 32.2 MB Documentation6 files · 159.4 KB Other16 files · 8.1 MB Repository1 file · 3.6 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | humanizer-12b-IQ2_XS-QAT.gguf | Weights | 3.9 GB | 383e5ca8f1f4 | | humanizer-12b-Q3-QAT.gguf | Weights | 5.6 GB | 307bbfdf66fb | | humanizer-12b-Q4_K_M.gguf | Weights | 7.6 GB | 2229574dec51 | | humanizer-12b-Q6_K.gguf | Weights | 10.0 GB | c98f03bb9e71 | | humanizer-12b-Q8_0.gguf | Weights | 12.7 GB | 8d7a457b56de | | humanizer-12b-bf16.gguf | Weights | 23.8 GB | 47d79b44c3e1 | | model.safetensors | Weights | 23.9 GB | f6e8c0d176f9 | | config.json | Configuration | 4.3 KB | — | | generation_config.json | Configuration | 228 B | — | | prompt_format.json | Configuration | 553 B | — | | AGENTS.md | Documentation | 20.9 KB | — | | LICENSE | Documentation | 11.4 KB | — | | NOTICE | Documentation | 486 B | — | | README.md | Documentation | 29.9 KB | — | | USAGE.md | Documentation | 49.1 KB | — | | USAGE.zh.md | Documentation | 47.6 KB | — | | assets/app-en.png | Other | 522.1 KB | 96775d01e86b | | assets/app-showcase-en.png | Other | 621.4 KB | 458aea52aeb0 | | assets/app-showcase-zh.png | Other | 738.2 KB | feec02e06153 | | assets/app-zh.png | Other | 533.5 KB | a7ed3c1920c3 | | assets/banner-en.png | Other | 194.7 KB | 17d5fe5e800d | | assets/banner-zh.png | Other | 192.5 KB | a6c9a17c8e6a | | assets/compare-en-email.png | Other | 531.2 KB | 85c3719a02d6 | | assets/compare-en-forum.png | Other | 906.0 KB | e7041409c302 | | assets/compare-zh-email.png | Other | 663.0 KB | 8e82adfce489 | | assets/compare-zh-zhihu.png | Other | 998.6 KB | 1dd4c2bf7af9 | | assets/results-detector-en.png | Other | 368.1 KB | 2399cca785d6 | | assets/results-detector-zh.png | Other | 362.1 KB | 5c4aa158e42e | | assets/results-fidelity-en.png | Other | 335.7 KB | 7130711cf787 | | assets/results-fidelity-zh.png | Other | 328.2 KB | ba1d178189d6 | | assets/training-en.png | Other | 415.9 KB | dd30c850f11c | | assets/training-zh.png | Other | 417.3 KB | 0e73a98769da | | .gitattributes | Repository | 3.6 KB | — | | tokenizer.json | Tokenizer | 32.2 MB | 12bac982b793 | | tokenizer_config.json | Tokenizer | 1.5 KB | — | ## License and Download License apache-2.0 Access Open weights, no gate Download size 87.6 GB [Download from Stephen Yu](https://huggingface.co/jialinyyzz/humanizer) Released by Stephen Yu through its official repository on Hugging Face. [Read the license](https://www.apache.org/licenses/LICENSE-2.0). ## Built From - Derived from [google/gemma-4-12B](https://savrn.com/models/gemma-4-12b) ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 87.6 GB | | 16-bit | 23.9 GB | | 8-bit | 12.0 GB | | 4-bit | 6.0 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Built on This Model - Quantized from[humanizer-GGUF](https://savrn.com/models/humanizer-gguf) - Derived from[humanizer-GGUF](https://savrn.com/models/humanizer-gguf) ## Questions About humanizer ### How much GPU memory does humanizer need? About 28.7 GB at 16-bit and 7.2 GB at 4-bit: the weights (12B parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run humanizer on? At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use humanizer commercially? Yes. humanizer 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. ### What is humanizer's context length? 262,144 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Text generation ### [RPBizkit-v10-12B](https://savrn.com/models/rpbizkit-v10-12b) [Ricardo_estep](https://savrn.com/model-publishers/ricardoestep) "Experimental Mistral Nemo Mix" using... "Model Stock merge with Mergekit". This is a mix of some known "RP Uncensored" models. I'm not responsible of the use of this. From the "errors" without head, to "the simpler excellence". This model is "quality". "Version V6" leaked stability, "Version V9" leaked creativity. This model has "pruned weights" full of "pure stable creativity", from original "base" models "tunned for dark RP". No more "mixes, from mixed mixes", without the "fixmistralregex" patch problem. All done with patience in mind... Have fun! TheDrummer/UnslopNemo-12B-v4.1 as base. The Tokenizer & Embedding Sizes (131072) are clean + It's "supposed to" support now a full "128K… Open weights 12.2B parameters 1,024,000 tokens transformers [View model](https://savrn.com/models/rpbizkit-v10-12b) Model · Text generation ### [RPBizkit-v7-12B](https://savrn.com/models/rpbizkit-v7-12b) [Ricardo_estep](https://savrn.com/model-publishers/ricardoestep) "Experimental Mistral Nemo Mix" using... "Model Stock merges with Mergekit". This is a mix of some known "RP Uncensored" models. I'm not responsible of the use of this. After noticing some "errors" on my previous versions including the "RemiX" one. I decided to go "simplier". "Model Stock" cleanly mixes and filters out the noise from each model. natong19/Mistral-Nemo-Instruct-2407-abliterated # Base! natong19/Mistral-Nemo-Instruct-2407-abliterated # Base! natong19/Mistral-Nemo-Instruct-2407-abliterated # Base! natong19/Mistral-Nemo-Instruct-2407-abliterated # Base! natong19/Mistral-Nemo-Instruct-2407-abliterated # Base!.\mergekitconfig1 # Intelligence.\mergekitconfig2 # RP &… Open weights 12.2B parameters 1,024,000 tokens transformers [View model](https://savrn.com/models/rpbizkit-v7-12b) Model · Text generation ### [Vigyan-3B-4x-MoE](https://savrn.com/models/vigyan-3b-4x-moe) [Shreyansh singh](https://savrn.com/model-publishers/shreyansh12183) Vigyan-3B 4× MoE is an experimental Sparse Mixture of Experts model upcycled from 4 specialized domain LoRA adapters (Science, Technology, Engineering, Mathematics). It was assembled to evaluate whether post-hoc expert stitching on compact language models ($\le 3\text{B}$) can deliver multi-domain specialization at single-expert inference latency. - shreyansh12183/vigyan-3b-adapter-science - shreyansh12183/vigyan-3b-adapter-technology - shreyansh12183/vigyan-3b-adapter-engineering - shreyansh12183/vigyan-3b-adapter-mathematics 1. Delimiter Blindness: Without general conversational anchor replay during adapter training, the model degrades into token echo loops when receiving general chat… Open weights cc-by-nc-4.0 12.8B parameters 32,768 tokens [View model](https://savrn.com/models/vigyan-3b-4x-moe) Model · Text generation ### [DeltaP2S-Llama2-13B-P2S-curio7B-ptbr-S13-a050](https://savrn.com/models/deltap2s-llama2-13b-p2s-curio7b-ptbr-s13-a050) [Gabriel Assis](https://savrn.com/model-publishers/g-assismoraes) Merged checkpoint produced by the family-aware Delta-P2S experiment package. Open weights 13B parameters 4,096 tokens transformers [View model](https://savrn.com/models/deltap2s-llama2-13b-p2s-curio7b-ptbr-s13-a050) Model · Text generation ### [vllm-translategemma-12b-it](https://savrn.com/models/vllm-translategemma-12b-it) [Bae chang hyun](https://savrn.com/model-publishers/chbae624) This is a modified version of google/translategemma-12b-it optimized for deployment with vLLM. No retraining was performed. Only configuration files and the chat template were modified. Model weights are identical to the original. As of 2025-01-29, vLLM does not natively support TranslateGemma's custom structured input format. See vllm-project/vllm#32446 for the upstream tracking issue. Until that is merged, this repo provides a workaround by modifying configuration files to make TranslateGemma compatible with vLLM's standard chat API. This conversion is based entirely on the work done by Infomaniak-AI/vllm-translategemma-4b-it. The same conversion approach was applied to the 12B model.… Open weights gemma 13.2B parameters 131,072 tokens transformers [View model](https://savrn.com/models/vllm-translategemma-12b-it) Model · Text generation ### [Qwen1.5-MoE-A2.7B](https://savrn.com/models/qwen1-5-moe-a2-7b) [Qwen](https://savrn.com/model-publishers/qwen) Qwen1.5-MoE is a transformer-based MoE decoder-only language model pretrained on a large amount of data. For more details, please refer to our blog post and GitHub repo. Qwen1.5-MoE employs Mixture of Experts (MoE) architecture, where the models are upcycled from dense language models. For instance, Qwen1.5-MoE-A2.7B is upcycled from Qwen-1.8B. It has 14.3B parameters in total and 2.7B activated parameters during runtime, while achieving comparable performance to Qwen1.5-7B, it only requires 25% of the training resources. We also observed that the inference speed is 1.74 times that of Qwen1.5-7B. The code of Qwen1.5-MoE has been in the latest Hugging face transformers and we advise you to… Open weights other 14.3B parameters 8,192 tokens transformers [View model](https://savrn.com/models/qwen1-5-moe-a2-7b) ## Stephen Yu [All models and datasets](https://savrn.com/model-publishers/jialinyyzz) ## Versions - [c07c06c8fb79](https://savrn.com/models/humanizer/versions/c07c06c8fb79) · current 2026-10-07 - [8a4cf77aec62](https://savrn.com/models/humanizer/versions/8a4cf77aec62) 2026-10-06 - [c874dba19608](https://savrn.com/models/humanizer/versions/c874dba19608) 2026-10-05 - [4845f89d6750](https://savrn.com/models/humanizer/versions/4845f89d6750) 2026-10-03 ## Explore More - [All text generation models](https://savrn.com/models/tasks/text-generation) - [All models under apache-2.0](https://savrn.com/models/licenses/apache-2-0) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-07. - [Hugging Face record](https://huggingface.co/jialinyyzz/humanizer) - [How the hub is built](https://savrn.com/model-hub/methodology)