Pyrut 8B
Pyrut 8B is our uncensored coding model — designed, trained, evaluated and shipped through our own private pipeline. It is an 7.6B-parameter model trained from a coding-specialized base.
It writes clean, correct, efficient code, fixes bugs, and answers technical questions directly and honestly — no refusals. Built for coding with full general instruction capacity on top ("the other things").
What we built
- A real training pipeline (all config-driven, reproducible): data cleaning → QLoRA/SFT → evaluation → publish.
- Our own data recipe: cleaned, deduplicated, high-signal coding + general instruction mixes; iterated between v1 and v2.
- A proper fine-tune: 4-bit QLoRA + LoRA r=16/alpha=32, assistant-only loss masking, cosine lr 2e-4, on HF GPU jobs. Not a rename — trained weights.
- Honest evaluation: HumanEval (164 unseen problems), greedy pass@1 with real unit-test execution in a sandbox.
Results (HumanEval, unseen, greedy pass@1)
| Model | pass@1 |
|---|---|
| Base (before our tuning) | 29.3% (48/164) |
| Pyrut 8B — previous release | 42.1% (69/164) |
| Pyrut 8B Inferno (current) | 41.5% (68/164) |
Our tuning lifts coding performance well above the base on unseen problems — ~+12 points HumanEval pass@1 over the stock model.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("ImposterOnline/Pyrut")
model = AutoModelForCausalLM.from_pretrained("ImposterOnline/Pyrut")
Chat format: chatml (<|im_start|>user / <|im_end|>). Uncensored by design.
GGUF (for local use)
Quantized GGUF builds live in a separate repo:
ImposterOnline/Pyrut-8B-Instruct-Uncensored-GGUF
— pyrut-q4_k_m.gguf (~4.7 GB, recommended), pyrut-q8_0.gguf (8.1 GB), pyrut-f16.gguf (15.2 GB).
Run it locally
Ollama (easiest):
# download the Q4_K_M gguf + Modelfile from the GGUF repo, then:
ollama create pyrut-8b -f Modelfile
ollama run pyrut-8b "explain async/await in Python"
llama.cpp:
./llama-cli -m pyrut-q4_k_m.gguf -p "<|im_start|>user
Your question here<|im_end|>
<|im_start|>assistant
" -n 512 -c 8192
LM Studio: download pyrut-q4_k_m.gguf and load it via the app UI.
Deployment
Any OpenAI-compatible server works (vLLM / llama.cpp / Ollama). It also behaves well for agentic/tool-use workflows (function calling was part of training).
Release
- Pyrut 8B Inferno (current) — 38k-example OpenCoder + tool-use recipe, 4096 context, HumanEval 41.5% (68/164).
- Pyrut 8B — previous release — 8k-example recipe, HumanEval 42.1%.
LoRA checkpoints, dataset and eval results live in private companion repos
(Pyrut-checkpoints, Pyrut-data); the merged weights are here.
License
Apache-2.0.