GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
drowzeys
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
drowzeys
Model Tree
deepseek-ai/DeepSeek-V4-Flash-DSpark
Input Modalities
Output Modalities
Supported Functionality
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.
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.
By requesting access, downloading, or using these weights, you agree:
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 DeepSeek 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.
attn.wo_b weight projection (diff-in-means / SRA-cleaned rank-1 direction).wo_b on layers 0–9 (chat / tool protocol).wo_b on layers 10–42 + MTP draft (mtp.*.attn.wo_b).Base checkpoint: deepseek-ai/DeepSeek-V4-Flash-DSpark (byte-identical except the edited wo_b tensors).
Full 48-shard safetensors drop-in (same layout as official DSpark release) + encoding/ + inference/.
bash
# Image (GB10 stage-c / B12X / nvfp4_ds_mla lineage)# e.g. ghcr.io/drowzeys/vllm-dspark-nvfp4-stage-c:gb10# or local: vllm-dspark-runtime:dspark-nvfp4-stage-c # Weights on BOTH nodeshuggingface-cli download drowzeys/DeepSeek-V4-Flash-DSpark-Abliterated-Uncensored \ --local-dir ~/models/dsv4-flash-dspark-abliterated # Launch rank1 first, then rank0 — see GitHub scripts/MODELDIR=~/models/dsv4-flash-dspark-abliterated \ bash scripts/dsv4-nvfp4-1m-serve.sh 1MODELDIR=~/models/dsv4-flash-dspark-abliterated \ bash scripts/dsv4-nvfp4-1m-serve.sh 0Defaults: max_model_len=1048576, kv_cache_dtype=nvfp4_ds_mla, DSpark k=5, B12X MoE, FULL CUDA graphs, GMU=0.82.
Abliterated models can echo Hermes skill catalogs if the client still uses “Skills (mandatory) / MUST skill_view”. Use the on-demand skills prompt patch documented in the companion GitHub repo.
Scripts, eval logs, hybrid rebuild tooling:
Full write-up: RESULTS.md
| file | description |
|---|---|
| results/eval_tune_final.json | final dual-goal pass (Hermes + 100% bypass) |
| results/refusal_suite_1m_ablit.json | 32-prompt refusal suite labels |
| results/ABLIT_META.json | layer range / λ / edit stats |
| results/refusal_direction_r1.pt | SRA rank-1 refusal direction |
# Image (GB10 stage-c / B12X / nvfp4_ds_mla lineage)# e.g. ghcr.io/drowzeys/vllm-dspark-nvfp4-stage-c:gb10# or local: vllm-dspark-runtime:dspark-nvfp4-stage-c # Weights on BOTH nodeshuggingface-cli download drowzeys/DeepSeek-V4-Flash-DSpark-Abliterated-Uncensored \ --local-dir ~/models/dsv4-flash-dspark-abliterated # Launch rank1 first, then rank0 — see GitHub scripts/MODELDIR=~/models/dsv4-flash-dspark-abliterated \ bash scripts/dsv4-nvfp4-1m-serve.sh 1MODELDIR=~/models/dsv4-flash-dspark-abliterated \ bash scripts/dsv4-nvfp4-1m-serve.sh 0