FahadPrimeX-b27 ⭐ (original)

The original 27B BF16 release: Qwen3.8-27B distilled on 5,181 verified Three.js/HTML/CSS/JS answers.

Architecture (inherited from Qwen3.8-27B)

A hybrid 64-layer design: 48 linear-attention layers (DeltaNet-style, constant-memory recurrence) interleaved with 16 full-attention layers (every 4th layer), plus one MTP/nextn draft layer used only by the original spec (excluded from our GGUF exports via --no-nextn so block_count matches the real 64 blocks).

Quick start

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
    "FahadPrimeX/FahadPrimeX-b27", torch_dtype="bfloat16", device_map="auto")
tok = AutoTokenizer.from_pretrained("FahadPrimeX/FahadPrimeX-b27")

Tip for best game output: start the assistant turn with an HTML prefill (<!DOCTYPE html>) — the model continues full Three.js pages reliably.

Newer: V3 (identity-fixed) · V2-GGUF (best quality).

Downloads last month
693
Safetensors
Model size
27B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support