Yes83882/custom-fine-tuned-qwen3-1.7b

Custom Fine tuned 'Qwen3-1.7B' — FPE Deep Edition

LoRA fine-tune of Qwen/Qwen3-1.7B, specialized in:

• —Fundamental Paper Education (FPE) — deep lore: every character, the full plot of Basics in Behavior, locations, and the series' real-world history (sourced from the FPE community wiki)
• —Iron Man (Marvel official), ULTRAKILL (official Steam/New Blood), Hibiscus plants (botany), Xiaomi Pad 7 (official specs)

Files

• —custom-fine-tuned-qwen3-1.7b-fpe.Q8_0.gguf — highest quality (1.8 GB). Best for PocketPal / ChatterUI / LLMFarm / llama.cpp on devices with 6 GB+ RAM.
• —custom-fine-tuned-qwen3-1.7b-fpe.Q6_K.gguf — near-Q8 quality (1.4 GB). Used for in-browser inference (wllama) and lower-RAM devices.
• —adapter/ — the raw LoRA adapter (load with peft on Qwen/Qwen3-1.7B).

The correct non-thinking chat template is embedded in the GGUFs. Keep temperature low (0.1-0.3) for factual recall. Trained on ~250 curated fact pairs (~750 samples), 3 epochs, LoRA r=16, CPU fp32.