kacperwikiel
slayer-v31-qwen3.5-27b
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Get help setting up a custom Dedicated Endpoints.
Talk with our engineer to get a quote for reserved GPU instances with discounts.
README
License: apache-2.0Result (proxy)
On the local Open PL closed-book proxy (--limit 100, no RAG/open-book), v31 ckpt16:
| Model | Fair 18-task avg |
|---|---|
| Qwen3.5-27B base | 61.19 (broad30) |
| Slayer 9B v16 (prev best) | 64.61 (broad30) |
| Slayer v31 ckpt16 | 66.58 |
| Bielik (published target) | 65.93 |
v31 clears the Bielik line on this proxy. This is a --limit 100 proxy, not a
reviewer-proof full leaderboard run — treat as a strong indicative result pending
no-limit confirmation.
Lineage: 27B base → v30 PSC/KLEJ SFT calibration (ckpt10) → v31 +DYK anchors (ckpt16, 16 steps, LR 5e-6).
Usage
python
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizerbase = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-27B", device_map="auto")model = PeftModel.from_pretrained(base, "kacperwikiel/slayer-v31-qwen3.5-27b")tok = AutoTokenizer.from_pretrained("kacperwikiel/slayer-v31-qwen3.5-27b")
Use enable_thinking=False in the chat template for benchmark-style answers.
A GGUF Q4_K_M build is at kacperwikiel/slayer-v31-qwen3.5-27b-GGUF.
Model provider
kacperwikiel
Model tree
Base
Qwen/Qwen3.5-27B
Adapter
this model
Modalities
Input
Video, Text, Image
Output
Text
Pricing
Dedicated Endpoints
View detailsSupported Functionality
Model APIs
Dedicated Endpoints
Container
More information