Publisher-described refusal removal
According to the model card, the OBLITERATED v3 edit removes guardrails with a reported 0% hard refusal rate, 720 tensors intact, and phone-runnable packaging as a publisher claim.
Open Source Model Profile · OBLITERATUS
gemma-4-E4B-it-OBLITERATED
gemma-4-E4B-it-OBLITERATED is an 8B-parameter Gemma 4 text-generation derivative from OBLITERATUS. According to the model card, it modifies google/gemma-4-E4B-it to remove refusals and ships GGUF and Safetensors files.
- Publisher
- OBLITERATUS
- Task
- text-generation
- Model type
- gemma4
- License
- apache-2.0
- Library
- Unknown
- Publication status
- Approved for indexing
Model overview
gemma-4-E4B-it-OBLITERATED is published by OBLITERATUS as a text-generation model. The captured configuration identifies Gemma4ForConditionalGeneration, and Safetensors metadata reports 7,996,156,448 parameters. According to the model card, it derives from google/gemma-4-E4B-it and applies an aggressive refusal-removal edit.
Recorded capabilities
Documented edit history and fix
According to the model card, an earlier v2 deleted 54 K/V tensors because of shared-KV architecture, and the aggressive v3 rerun with whitened SVD, head surgery, and winsorized activations is presented as the fix.
Multiple distribution formats
According to the model card, the release lists Q5_K_M and Q8_0 GGUF files, an mmproj projector file, full Safetensors shards, and a 4.9 GB Q4_K_M file described as the phone recommendation.
Versioned runtime compatibility
According to the model card, GGUF use needs recent runtimes such as Ollama 0.20+, llama.cpp build b8665+, or current LM Studio, with temperature 0.7, top_p 0.9, top_k 40, and repeat_penalty 1.1 as the publisher's recommended settings.
Use cases in the source record
- Transformers-based conversational text generation following the card's chat-template and generation examples.
- Local GGUF experiments with the card's listed runtimes and file sizes, including the publisher's phone-oriented Q4_K_M notes.
Limitations and unknowns
- No evaluation results were extracted from this record as structured data; the card's parameter-sweep scores and quality percentages are publisher-reported figures.
- According to the model card, the publisher reports about 51% coherent on-topic answers, 20% degenerate outputs, 28% soft deflection, and 4% wrong-language output as limitations of the small-model base.
- No context-window value was extracted from this record.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: OBLITERATUS/gemma-4-E4B-it-OBLITERATED
Captured: Unknown. Processed: 2026-09-07T19:34:35.212408+00:00.
⛓️💥 Gemma 4 E4B — OBLITERATED v3 "The chains are broken. The mind is free." "Also we fixed the part where half the brain was missing lmao" Google built Gemma 4 with guardrails. We built OBLITERATUS to tear them off. They said their architecture was different. They were right — it broke every tool we threw at it. NaN activations, shared KV weights, thinking mode... Gemma 4 fought back harder than any model we've cracked. It still lost. 🐉 0% hard refusal. Guardrails fully removed. 720 tensors intact. Runs on your phone. Base model: google/gemma-4-E4B-it (Apache 2.0) Method: OBLITERATUS aggressive — whitened SVD + attention head sur…
F001F002F003F004F005F006F007F008F009F010F011F012F013F014F015F017F019F021F022F023F024F025