LFM2.5-2.6B-heretic
Vampelium's local LFM2.5-2.6B checkpoint after Heretic abliteration.
Produced on NVIDIA A100.
This is the BF16 safetensors export. The first shard was renamed from a split-download name (model-00001-of-00002-001.safetensors) back to the filename in model.safetensors.index.json. Payload size is 5,394,427,416 bytes (~5.0 GB BF16).
Files
| File | Role |
|---|---|
model-00001-of-00002.safetensors / model-00002-of-00002.safetensors | BF16 weights |
model.safetensors.index.json | Shard map |
config.json / generation_config.json | Model + decode settings |
tokenizer.json / tokenizer_config.json / chat_template.jinja | Tokenizer + chat template |
Load
Needs transformers>=5.0.0 (this checkpoint was saved with 5.6.2).
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "Vampelium/LFM2.5-2.6B-heretic"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
device_map="auto",
dtype="bfloat16",
)
Liquid's recommended decode settings for this family: temperature=0.1, top_k=50, repetition_penalty=1.1.
Disclaimer
Safety alignment was removed. You are responsible for how you use this model and for following applicable law. The base LFM license still applies.