metadata
license: mit
tags:
- rag
- vismem
- fine-tuning
VisMem RAG Fine-tuned: gemma-3-270m-vismem-rag-QA-1-vismem-rag-0322-1623-vismem-rag-0323-1532
Inference
import requests
from vismem_core import VisMem
from sentence_transformers import SentenceTransformer
from transformers import pipeline
# 1. Load VisMem
data = requests.get(
"https://huggingface.co/datasets/broadfield-dev/gemma-3-270m-vismem-rag-QA-1-vismem-rag-0322-1623-vismem-kb-0323-1532/resolve/main/vismem.png",
headers={"Authorization":"Bearer <HF_TOKEN>"}).content
mem = VisMem.from_png_bytes(data)
emb = SentenceTransformer('all-MiniLM-L6-v2')
# 2. RAG query
q_vec = emb.encode([your_question])[0]
results = mem.search(q_vec, k=3)
context = "\n---\n".join(results)
# 3. Prompt
system = (
"You are a helpful AI Assistant with visual memory.\n"
"### RAG MEMORY (Vector Database):\n"
"[Uploaded Doc]: None\n"
f"[Knowledge Base]: {context}\n"
"### EPISODIC MEMORY (Past Chat):\n"
"[History]: None"
)
pipe = pipeline("text-generation", model="broadfield-dev/gemma-3-270m-vismem-rag-QA-1-vismem-rag-0322-1623-vismem-rag-0323-1532")
print(pipe([
{"role":"system","content":system},
{"role":"user","content":your_question}
], max_new_tokens=200))
Config
{ "dataset_name": "AnonymousSub/MedQuAD_47441_Context_Question_Answer_Triples", "rag_columns": [ "Contexts", "Questions" ], "question_col": "Questions", "answer_col": "Answers", "split": "train", "data_config": null, "total_kb_docs": 47441, "vismem_dim": 384, "vismem_width": 13056, "vismem_height": 13056, "kb_repo": "broadfield-dev/gemma-3-270m-vismem-rag-QA-1-vismem-rag-0322-1623-vismem-kb-0323-1532" }