Modelpublic

gyung/lfm25-local-harness-rlvr-http-vllm-lora-browsecomp-20260619-base

sourceHugging Faceotherupdated 3mo agoView on Hugging Face

0likes4downloads

Model Card

LFM2.5-8B-A1B_localharnessrlvrhttpvllmlorabrowsecomp20260619_base

This public artifact is part of the ko-law-retriever work. It is intended for Korean legal retrieval, evidence selection, or related tool-use experiments.

Scope

—Artifact type: PEFT LoRA adapter
—This is not a standalone legal-advice model.
—Use it to retrieve, rank, or prepare evidence for a separate answer model.

Basic Use

Install:

bash

pip install -U "huggingface_hub[cli]" peft transformers

Download:

bash

hf download gyung/lfm25-local-harness-rlvr-http-vllm-lora-browsecomp-20260619-base --local-dir lfm25-local-harness-rlvr-http-vllm-lora-browsecomp-20260619-base

For LoRA adapters, load with the matching base model:

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_id = "LiquidAI/LFM2.5-8B-A1B"
adapter_id = "gyung/lfm25-local-harness-rlvr-http-vllm-lora-browsecomp-20260619-base"

tokenizer = AutoTokenizer.from_pretrained(base_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    base_id,
    device_map="auto",
    torch_dtype="auto",
    trust_remote_code=True,
)
model = PeftModel.from_pretrained(model, adapter_id)

For vLLM LoRA serving:

bash

python -m vllm.entrypoints.openai.api_server \
  --model LiquidAI/LFM2.5-8B-A1B \
  --enable-lora \
  --lora-modules lfm25-local-harness-rlvr-http-vllm-lora-browsecomp-20260619-base=./lfm25-local-harness-rlvr-http-vllm-lora-browsecomp-20260619-base \
  --served-model-name lfm25-local-harness-rlvr-http-vllm-lora-browsecomp-20260619-base \
  --max-model-len 12288

Related Repositories

—Main code: gyung/ko-law-retriever
—Large data/eval artifacts: gyung/ko-law-retriever-artifacts-20260622