Text Generation
Transformers
PyTorch
English
burt-imma
custom-architecture
matrix-memory
equilibrium-propagation
cifg
sovereign
snapkitty
no-backprop
formal-verification
lean4
agents
python
Instructions to use Snapkitty/burt-imma with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Snapkitty/burt-imma with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Snapkitty/burt-imma")# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("Snapkitty/burt-imma", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Snapkitty/burt-imma with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Snapkitty/burt-imma" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Snapkitty/burt-imma", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/Snapkitty/burt-imma
- SGLang
How to use Snapkitty/burt-imma with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "Snapkitty/burt-imma" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Snapkitty/burt-imma", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "Snapkitty/burt-imma" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Snapkitty/burt-imma", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use Snapkitty/burt-imma with Docker Model Runner:
docker model run hf.co/Snapkitty/burt-imma
Download lean4/BURT.lean from Snapkitty/burt-imma: direct link, hf CLI and curl.
- Browser
- Download file 1.95 kB
-
https://huggingface.co/Snapkitty/burt-imma/resolve/main/lean4/BURT.lean
- Command line
-
hf download hf://Snapkitty/burt-imma/lean4/BURT.lean
-
curl -L -o BURT.lean https://huggingface.co/Snapkitty/burt-imma/resolve/main/lean4/BURT.lean
1.95 kB
| /- | |
| BURT (Bi-encoder Unified Retrieval Transformer) Formalization | |
| Core retrieval and ranking properties | |
| -/ | |
| import Mathlib | |
| noncomputable section | |
| open Real | |
| -- ============================================================ | |
| -- BURT State | |
| -- ============================================================ | |
| structure BURTState (n d : Nat) where | |
| Q : Matrix (Fin n) (Fin d) Float -- query embeddings | |
| D : Matrix (Fin n) (Fin d) Float -- document embeddings | |
| I : Fin n → Float -- inverted index scores | |
| R : Fin n → Fin n -- ranking permutation | |
| C : Float -- CIFG gate value | |
| E : Float -- energy | |
| -- ============================================================ | |
| -- Core Theorems | |
| -- ============================================================ | |
| /-- Router entropy is bounded by the entropy ceiling -/ | |
| theorem router_entropy_bound {n d : Nat} (s : BURTState n d) | |
| (entropy : (Fin n → Float) → Float) | |
| (router_dist : Fin n → Float) | |
| (h_from_state : ∀ i, router_dist i = s.I i) : | |
| entropy router_dist ≤ 0.20 := sorry | |
| /-- CIFG gate conserves memory during retrieval updates -/ | |
| theorem cifg_memory_conservation {n d : Nat} | |
| (s : BURTState n d) | |
| (h_gate : 0 ≤ s.C ∧ s.C ≤ 1) | |
| (mem_prev mem_next : Float) | |
| (h_update : mem_next = s.C * mem_prev + (1 - s.C) * s.E) : | |
| mem_next ≤ max mem_prev s.E := sorry | |
| /-- The ranking function produces a valid permutation -/ | |
| theorem ranking_is_permutation {n d : Nat} (s : BURTState n d) | |
| (h_inj : Function.Injective s.R) : | |
| Function.Bijective s.R := sorry | |
| /-- Retrieval has sub-quadratic complexity via inverted index -/ | |
| theorem retrieval_complexity {n d : Nat} (s : BURTState n d) | |
| (retrieval_flops : Nat) | |
| (h_indexed : retrieval_flops ≤ n * Nat.log n + d) : | |
| retrieval_flops < n * n := sorry | |
| end | |