eptwts-x-posts-gguf

eptwts X posts (GGUF Q4_K_M)

Fine-tuned Qwen2.5-3B-Instruct (QLoRA adapter qwen25-3b-ep-sft-v3 → 16-bit merge → Q4_K_M GGUF) for local Ollama.

Attribution: fine-tuned style inspired by EP (@eptwts) public signal / knowledge-base writing. Not an official EP model. Attribute the inspiration; do not impersonate deceptively on X.

Prompt UX (natural language)

please give 3 posts for niche dental AI automation

Replies are plain-text posts (numbered or blank-line separated). No JSON. No “Sure, here are…”.

Install with Ollama (primary)

Download the GGUF + Modelfile, then create a local Ollama model. This works on Mac/Linux when the native hf.co pull fails:

huggingface-cli download sairahul1/eptwts-x-posts-gguf \
  eptwts-x-posts-q4_k_m.gguf Modelfile \
  --local-dir ./eptwts-gguf
cd eptwts-gguf
# ensure Modelfile FROM matches local gguf filename
# (this repo's Modelfile uses: FROM ./eptwts-x-posts-q4_k_m.gguf)
ollama create eptwts-x-posts-gguf -f Modelfile
ollama run eptwts-x-posts-gguf

Requires huggingface_hub (pip install huggingface_hub) and Ollama.

Optional: native Hub pull

ollama run hf.co/sairahul1/eptwts-x-posts-gguf

May fail with blocked redirect to a different host when Ollama follows the HF AWS CDN / Xet host. Prefer the download → ollama create path above.

Files

Training

• Base: Qwen/Qwen2.5-3B-Instruct
• Adapter run: qwen25-3b-ep-sft-v3 (niche-faithful NL prompts)
• Quant: llama.cpp Q4_K_M