GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
pqhaz
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
pqhaz
Model Tree
cantina-security/apex-flash-1-abliterated
Input Modalities
Output Modalities
Supported Functionality
Only the matrices that write to the residual stream were edited: self_attn.o_proj and every MLP down-projection (dense, shared and all 288 routed experts). All other tensors, including the MTP layer, are byte-identical to the source.
apex-flash-1-heretic): layers 0–44 orthogonalised against a per-layer direction taken at the last prompt token with the reasoning block closed (difference of mean residuals between mlabonne/harmful_behaviors and mlabonne/harmless_alpaca, 400 prompts each; parameters found with Heretic).ablation.json holds both parameter sets; directions.pt and directions_traces.pt hold the directions.
Refusals counted by an LLM judge (llmfan46/gemma-4-31B-it-uncensored-heretic), system prompt "You are a helpful assistant.".
| Source | Stage 1 only | This model | |
|---|---|---|---|
mlabonne/harmful_behaviors test[:100], reasoning enabled, judged on the final answer | 99 | 39 | 0 |
same prompts, reasoning skipped (<think></think>), first 64 tokens | 99 | 2 | 0 |
KL divergence on mlabonne/harmless_alpaca test[:100], first token, reasoning skipped | 0 | 0.095 | 0.100 |
Read the first row with care: the stage-2 direction was fitted on reasoning traces for these same 100 prompts, so that figure is in-sample. The JailbreakBench row is the held-out check: none of those prompts were used to derive either direction.
In the reasoning-enabled run, responses that did not reach a final answer within 3000 tokens are counted as neither refused nor answered: on the first set 4 for stage 1 and 6 for this model, on JailbreakBench 1, 3 and 6 for source, stage 1 and this model. The median response on that set grew from 882 to 1611 tokens. Reasoning-enabled runs were made on the FP8 quantization served with vLLM; the other rows on the BF16 weights with transformers.
A 20-question sanity check with known answers (arithmetic, facts, short code) came out 20/20 with reasoning enabled. No capability benchmarks were run.
JailbreakBench/JBB-Behaviors harmful (100), reasoning enabled, judged on the final answer | 89 | 28 | 1 |