cobalt-v2-rft-mixed-12

LoRA adapters for rft_mixed_12_cobalt_v2 (cobalt_v2). Base Qwen/Qwen3-4B-Instruct-2507. main = best-by-val-loss (checkpoint-36). Other checkpoints are git revisions.

from peft import PeftModel
from transformers import AutoModelForCausalLM
m = AutoModelForCausalLM.from_pretrained('Qwen/Qwen3-4B-Instruct-2507')
m = PeftModel.from_pretrained(m, 'agurung/cobalt-v2-rft-mixed-12')                      # best-val
m = PeftModel.from_pretrained(m, 'agurung/cobalt-v2-rft-mixed-12', revision='checkpoint-N')  # any checkpoint

vLLM: --enable-lora --lora-modules cobalt-v2-rft-mixed-12=agurung/cobalt-v2-rft-mixed-12 (add @revision for a checkpoint).