LFM2.5-8B-A1B-Coder-LoRA
LoRA adapter for LiquidAI/LFM2.5-8B-A1B.
This is the LFM coding LoRA. It was trained to improve practical coding behavior across real-world and multilingual coding tasks.
Merged model:
josephmayo/LFM2.5-8B-A1B-Coder
GGUF target:
josephmayo/LFM2.5-8B-A1B-Coder-GGUF
Training
- Base model:
LiquidAI/LFM2.5-8B-A1B - Method: supervised LoRA fine-tuning
- Training rows:
123 - LoRA rank:
16 - LoRA alpha:
32 - LoRA dropout:
0.03 - Target modules:
q_proj,k_proj,v_proj,o_proj
Validation
Validation metric:
- Base heldout mean NLL:
3.3903426826000214 - Adapter heldout mean NLL:
1.8745103478431702 - Absolute NLL decrease:
1.5158323347568512 - Relative NLL reduction:
44.71%
Public benchmark pass-rate scores are not claimed for this adapter.
Usage
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base_id = "LiquidAI/LFM2.5-8B-A1B"
adapter_id = "josephmayo/LFM2.5-8B-A1B-Coder-LoRA"
tokenizer = AutoTokenizer.from_pretrained(adapter_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(base_id, trust_remote_code=True)
model = PeftModel.from_pretrained(model, adapter_id)
model.eval()
Notes
- This is an adapter, not a standalone merged model.
- The values above are validation NLL metrics, not benchmark pass-rate claims.
Evidence files
Run evidence for this release is stored in the repository under evidence/:
These files are compact local/Kaggle run artifacts used to document training, evaluation, merge, or quantization evidence for this model family.
- Downloads last month
- 80