Model card

Qadir-v1-v2

LoRA adapters for Qwen/Qwen2.5-7B-Instruct-1M, trained for tool use. The card publishes v1 as a working baseline and says it does not pass its own 0.9 gate (71.67% raw / 88.75% with prompting, 480 tool prompts). Unpublished v2 work is not graded and is not presented here as a product.

Hugging Face
https://huggingface.co/abuz8studios/Qadir-v1-v2
Type
model
Task / pipeline
text-generation
License
apache-2.0
Size
LoRA on 7B
Local or cloud
Local open-weights (download from Hugging Face)
Catalog snapshot (18 Oct 2026)
downloads 0; updated 2026-10-14
HF API (2026-11-22T12:30:06-05:00)
downloads 0; likes 0; last modified 2026-10-14T12:17:03.000Z; repo size 357.5 MiB
Base model (from the card)
Qwen/Qwen2.5-7B-Instruct-1M

Sources: ABUZ8 catalog snapshot data/catalog.json (generated 2026-10-18T05:46Z), and where noted the public Hugging Face API plus the model card, fetched 2026-11-22T12:30:06-05:00. Unknown fields are labeled unknown. “Run in HQ” opens HQ so you can connect a local or API model; it does not auto-load these weights.