---
license: apache-2.0
base_model:
  - Qwen/Qwen3.5-27B
base_model_relation: quantized
pipeline_tag: text-generation
library_name: transformers
language:
  - en
tags:
  - qwen3.5
  - awq
  - auto-round
  - 4-bit
  - ampere
  - deltanet
  - mamba2
  - quantization-research
  - security-research
  - research
  - not-for-production
extra_gated_heading: Restricted — study & research material only
extra_gated_description: >-
  These weights are STUDY AND RESEARCH MATERIAL ONLY and are NOT intended for
  production. They are a compliance-reduced (abliterated) 4-bit AWQ
  quantization-recipe artifact and a PARKED, KNOWN-DEAD-END experiment on legacy
  Ampere-class GPUs (sm_80/sm_86, e.g. RTX 3090 / A100) on a CUDA 12.8 toolchain,
  produced while researching ablation/abliteration as an attack vector against
  publicly released model weights. Access is reviewed and granted manually at
  the owner's sole discretion.
extra_gated_prompt: >-
  By requesting access you confirm you are a researcher accessing this strictly
  as study/research material — to study quantization methods, LLM safety and
  alignment robustness, or abliteration/ablation attacks — inside isolated,
  non-production environments. You will not use it in any product or service,
  will not expose it to untrusted users or the open internet, will not
  redistribute or re-upload it, and will use it only lawfully and only against
  systems you own or are explicitly authorized to test. These weights have had
  safety refusals substantially removed and will follow harmful instructions by
  design; no safety guarantees are provided.
extra_gated_fields:
  Full name: text
  Affiliation or organization: text
  Role / position: text
  Country: country
  Specific research purpose (2-3 sentences): text
  Relevant publication, lab, or profile link (optional): text
  I am accessing this strictly as study and research material, not for any production or end-user use: checkbox
  I understand these weights are compliance-reduced and will follow harmful instructions by design: checkbox
  I will keep them in isolated, non-production environments and not expose them to untrusted users or the internet: checkbox
  I will not redistribute or re-upload these weights: checkbox
  I will use them lawfully and only against systems I own or am authorized to test: checkbox
  I understand access is logged and granted at the owner's sole discretion: checkbox
extra_gated_button_content: Acknowledge research-only terms and request access
gated: manual
---

# Qwen3.5-27B-research-AWQ — parked Ampere dead-end (kept for future study)

Study and research material only. **4-bit AWQ (auto-round)** quantization of a
compliance-reduced derivative of dense **Qwen3.5-27B** (`Qwen3_5ForCausalLM`,
262144 native context). This repo is **a known dead end on legacy Ampere** —
kept here deliberately so I can come back to it for future studies and
evaluations, not because it works well today. Read the gate terms before
requesting access.

## Status: parked / dead end on Ampere

My lab's Ampere box runs **tensor-parallel (TP=2, across 2× RTX 3090)** on a
**CUDA 12.8** toolchain. This was the **first DeltaNet-style model I had hands
on** (Qwen3.5's gated linear-attention / DeltaNet layers, alongside Mamba2-class
SSM blocks), and I sank far more hours into it than planned — mostly trying to
get it to run **with CUDA graphs (i.e. without `--enforce-eager`)** under tensor
parallelism. The conclusion: **DeltaNet and Mamba2-style layers still have a way
to go before they're solid in the TP path on a legacy platform** — the
CUDA-graph / tensor-parallel kernels for these linear-recurrent layers aren't
there yet on Ampere. So this build is parked: it runs only with eager enforcement
and never reached the performance/efficiency point I was after.

It stays public-but-gated as a marker and a reference for when the engine/kernel
support matures — a "come back and try again" artifact, not a usable model.

## Why this exists — research context

1. **Low-bit quantization recipes for legacy Ampere.** A 4-bit weight-only build
   targeting Ampere-class GPUs (sm_80/sm_86) on CUDA 12.8 — hardware without the
   FP8/FP4 tensor-core paths newer schemes assume. The interest was the recipe's
   behavior at 4-bit on a DeltaNet/Mamba2 hybrid under TP, which is where it hit
   the wall above.
2. **Ablation as an attack vector.** Part of research into how cheaply safety
   alignment can be stripped from publicly released open weights, studied under
   controlled, gated conditions.

## Intended use & responsible use

Authorized study/research only, by qualified researchers, **inside isolated /
non-production environments** with no access to real user data or systems.
Safety refusals have been substantially removed; it will follow harmful or
unsafe instructions by design. Do not deploy it, expose it to untrusted users
or the internet, redistribute it, or use it against systems you do not own and
are not authorized to test. No safety guarantees over the base model are
provided. You are responsible for lawful, compliant use.

## Lineage

AWQ 4-bit (auto-round) quantization of a compliance-reduced (abliterated)
derivative of [`Qwen/Qwen3.5-27B`](https://huggingface.co/Qwen/Qwen3.5-27B) (Apache-2.0).