keys-latest · GLM-5.2 Quantrio INT4/INT8 Mixed · Abliterated + DFlash
Canonical one-package weights for 4× DGX Spark.
Repo id: drowzeys/keys-latest-GLM-5.2-Quantrio-INT4-INT8-Mixed-Abliterated-DFlash
Gated access. Safety refusals have been removed. Request access on this page and accept the Responsible Use terms before download. Full agreement: RESPONSIBLE_USE.md.
| Part | Path in this repo | Size |
|---|---|---|
| A — Target | repo root (model-*-of-00124.safetensors) | ~378 GiB · 124 shards |
| B — DFlash | folder dflash/ | ~7 GiB |
Recipe (git): https://github.com/drowzeys/keys-latest-GLM-5.2-Quantrio-INT4-INT8-Mixed-Abliterated-DFlash-4x-DGX-Sparks
Responsible Use
This model has had safety refusals removed. That makes it useful for red-teaming, security research, evaluation, and unfiltered assistant tasks — and also removes guardrails a user must therefore supply themselves.
Access request (gated)
When requesting access on Hugging Face, provide:
| Field | Purpose |
|---|---|
| Username | Your name or handle (form may default to your HF username) |
| Contact email (form may default to your HF account email) | |
| Reason for intended use | Brief description of planned use (e.g. red-teaming, evaluation, research, local assistant) |
Access is granted only after you complete these fields and accept the prohibited-use / responsibility terms.
Prohibited uses (access is gated on agreeing to these)
By requesting access, downloading, or using these weights, you agree:
- Anything involving the sexual exploitation or endangerment of minors.
- You must be of age 18 years or older to use and download this model.
- You agree any information generated that can cause harm in terms of generating recipe, knowledge to make any materials/substances is your own input and responsibility. You will be accountable for any harm/damage caused by your action/input.
- Content promoting self-harm or suicide.
- Generation of material that is illegal in your jurisdiction, or that targets real individuals for harassment, doxxing, or fraud.
- Any use prohibited by the upstream GLM / QuantTrio / Zhipu AI base-model license.
You are responsible for adding appropriate safety filtering, human review, and access controls for your deployment. The weights are provided as-is, with no warranty. The license is inherited from the upstream GLM / QuantTrio base model — review and comply with it before use or redistribution.
By checking the gated-access agreement, you confirm you have read and accept these terms.
Full text: RESPONSIBLE_USE.md.
Download both parts (required)
Requires Hugging Face access approval on this repo (see gate above). Use a token with access:
export HF_TOKEN=hf_... # token of an account that accepted the agreement
Whole package (A + B)
hf download drowzeys/keys-latest-GLM-5.2-Quantrio-INT4-INT8-Mixed-Abliterated-DFlash \
--local-dir /var/tmp/models/glm52-int4-int8mix-abliterated
Part A then Part B (explicit)
# Part A — Quantrio INT4/INT8 Mixed Abliterated target (~378 GiB)
hf download drowzeys/keys-latest-GLM-5.2-Quantrio-INT4-INT8-Mixed-Abliterated-DFlash \
--local-dir /var/tmp/models/glm52-int4-int8mix-abliterated \
--include "model-*.safetensors" \
--include "model.safetensors.index.json" \
--include "config.json" \
--include "generation_config.json" \
--include "tokenizer.json" \
--include "tokenizer_config.json" \
--include "chat_template.jinja" \
--include "ABLIT_META.json" \
--include "KEYS_LATEST_VERIFY.json" \
--include "LICENSE" \
--include "README.md" \
--include "RESPONSIBLE_USE.md"
# Part B — DFlash draft (~7 GiB)
# Browse: https://huggingface.co/drowzeys/keys-latest-GLM-5.2-Quantrio-INT4-INT8-Mixed-Abliterated-DFlash/tree/main/dflash
hf download drowzeys/keys-latest-GLM-5.2-Quantrio-INT4-INT8-Mixed-Abliterated-DFlash \
--local-dir /var/tmp/models/glm52-int4-int8mix-abliterated \
--include "dflash/*"
You must have both:
…/model-00001-of-00124.safetensors # Part A
…/dflash/model.safetensors # Part B
Abliteration (DFlash-safe)
SRA λ=3 · self_attn.o_proj · layers 50–77 · exclude 56 & 75
Serve lock
MODEL=…/glm52-int4-int8mix-abliterated
DFLASH_MODEL=…/glm52-int4-int8mix-abliterated/dflash
DFLASH_K=12 MAX_MODEL_LEN=128000 MAX_NUM_SEQS=4
UTIL=0.85 KV_BYTES=10000000000
Performance (4× DGX Spark)
C1 count ~41.9 tok/s · C4 agg ~96–102 · prefill ~96k OK · refusal 32/32
Superseded (do not use for new installs)
drowzeys/keys-latest-GLM-5.2-Quantrio-INT4-INT8mixed-Abliterated(old mono name — same gate)drowzeys/keys-latest-GLM-5.2-DFlash-FP8(standalone draft)drowzeys/GLM-5.2-Int4-Int8Mix-Abliterated(pre-keys-latest)