You can deploy this model and start earning money today!

nluick/
alao-qwen3-4b-step-20000

Name: alao-qwen3-4b-step-20000
Availability: InStock

Views

No views yet

Frozen Context Oracle LoRA Adapter

This is a LoRA (Low-Rank Adaptation) adapter trained for the Frozen Context Oracle architecture.

Base Model

Base Model: Qwen/Qwen3-4B
Adapter Type: LoRA
Task: Activation Interpretation via Frozen KV Cache

Architecture

The Frozen Context Oracle uses a two-phase approach:

Phase 1: Target model processes context → frozen KV cache
Phase 2: Oracle model (with LoRA) processes question attending to frozen KV → response

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model and tokenizer
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "nluick/alao-qwen3-4b-step-20000")

Training Details

This adapter was trained using the Frozen Context Oracle training script to help the model interpret activations from a target model through attention to frozen KV caches.