Kiesel v2 — LoRA adapter
LoRA adapter for ibm-granite/granite-4.2-8b (r=64, α=128, dropout 0.05, all linear projections, task type CAUSAL_LM).
This is the standalone adapter behind gmrrh/kiesel-v2 — the merged, ready-to-use bf16 model. See the model card there for training details, usage, and the chat template's thinking controls.
Load the adapter
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained(
"ibm-granite/granite-4.2-8b", torch_dtype="bfloat16", device_map="auto"
)
model = PeftModel.from_pretrained(base, "gmrrh/kiesel-v2-lora-peft")
tok = AutoTokenizer.from_pretrained("gmrrh/kiesel-v2-lora-peft")
Merge into full weights
merged = model.merge_and_unload()
merged.save_pretrained("kiesel-v2-merged")
tok.save_pretrained("kiesel-v2-merged")
License
Apache-2.0, consistent with the base model. Kiesel is an independent fine-tune and is not an IBM product.
Kiesel · gmrrh · D. Gutz
- Downloads last month
- 17
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for gmrrh/kiesel-v2-lora-peft
Base model
ibm-granite/granite-4.1-8b-base Finetuned
ibm-granite/granite-4.2-8b