MLX
Joblib
Safetensors
English
reasoning
chain-of-thought
context-compression
soft-prompt
apple-silicon
Instructions to use baya1116/hypernet-sp-distill with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use baya1116/hypernet-sp-distill with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir hypernet-sp-distill baya1116/hypernet-sp-distill
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
| import sys, os, time | |
| sys.path.insert(0, os.getcwd()) | |
| import tiered_rag_mlx as T | |
| P="/tmp/app_session_mem.jsonl" | |
| if os.path.exists(P): os.remove(P) | |
| chat = T.ChatSession(T.TieredMemory(P)) # rw=1024, temp 0.6 | |
| def on_message(text): | |
| intent = T.classify(text) | |
| if intent == "fact": | |
| ans,_,_ = chat.turn(text, store="session", ack_only=True) # instant ack, logged | |
| return intent, "stored", ans | |
| mathish = intent in ("math","command") | |
| msg = text + (" Please reason step by step and give the final number." if mathish else "") | |
| ans, src, _ = chat.turn(msg, store="none") | |
| return intent, (src or "—"), ans | |
| turns = [ | |
| ("hey, I'm planning a weekend trip", None), | |
| ("I'm going to Kyoto for 3 days", None), | |
| ("my budget is $800", None), | |
| ("if I spend $120 a day for 3 days, what's the total food cost?", ("math","360")), | |
| ("actually make it 4 days instead", None), | |
| ("recompute the food total", ("command","480")), | |
| ("who is the current emperor of Japan?", ("question","Naruhito")), | |
| ("explain what a ryokan is", ("chitchat",None)), | |
| ("remind me how long I'm staying and my budget", ("question","4 days / 800")), | |
| ("I'm vegetarian", None), | |
| ("what dietary preference should I mention at restaurants?", ("question","vegetarian")), | |
| ("thanks, that's all!", None), | |
| ] | |
| print("# Realistic app session — full pipeline\n", flush=True) | |
| for u, chk in turns: | |
| t0=time.time(); intent, path, ans = on_message(u); dt=time.time()-t0 | |
| print("="*70, flush=True) | |
| print(f"USER: {u}", flush=True) | |
| print(f" [intent={intent} · {path} · {dt:.0f}s]", flush=True) | |
| print(f"ASSISTANT: {ans[:240]}", flush=True) | |
| if chk: print(f" >>> expect {chk[0]}" + (f" ~ '{chk[1]}'" if chk[1] else ""), flush=True) | |
| print("\nAPP_SESSION_DONE", flush=True) | |