Harmony Structurer β€” GGUF

QLoRA fine-tune of Qwen/Qwen2.5-3B-Instruct on the ADE Corpus v2 for clinical entity extraction. Merged and exported to GGUF for local inference via LM Studio or llama.cpp.

Files

File Size Use
harmony-structurer-Q4_K_M.gguf ~2 GB Load this in LM Studio
harmony-structurer-f16.gguf ~6 GB Full precision reference

LM Studio setup

  1. Open LM Studio β†’ Search β†’ paste PranavKeshav/harmony-structurer-gguf
  2. Download harmony-structurer-Q4_K_M.gguf
  3. Load model β†’ set context length to 4096
  4. Start local server on port 1234 (default)

The Harmony backend reads LMSTUDIO_BASE_URL=http://localhost:1234/v1 from .env.

Training source

  • Adapter: PranavKeshav/harmony-structurer-qlora-v1
  • Dataset: ade-benchmark-corpus/ade_corpus_v2 (4,271 unique sentences, drug/ADE/dosage relations)
  • Method: QLoRA (4-bit NF4, rank=16, alpha=32) via Unsloth on T4 GPU
Downloads last month
25
GGUF
Model size
3B params
Architecture
qwen2
Hardware compatibility
Log In to add your hardware

4-bit

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for PranavKeshav/harmony-structurer-gguf

Base model

Qwen/Qwen2.5-3B
Quantized
(243)
this model