Text Generation
Transformers
PyTorch
English
burt-imma
custom-architecture
matrix-memory
equilibrium-propagation
cifg
sovereign
snapkitty
no-backprop
formal-verification
lean4
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")# 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
File size: 3,441 Bytes
b88c26d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 | /-
SparkDeterministicExecutor
Deterministic execution with contracts and sparse transitions
-/
import Mathlib
import Mathlib.Data.Matrix.Basic
import Mathlib.LinearAlgebra.Matrix.Basic
noncomputable section
open Real
-- ============================================================
-- State
-- ============================================================
structure State (n : Nat) where
values : Fin n β β
invariant_holds : Bool
-- ============================================================
-- Contract
-- ============================================================
structure Contract (n : Nat) where
precondition : State n β Prop
postcondition : State n β State n β Prop
invariant : State n β Prop
-- ============================================================
-- Perceptron Dispatch
-- ============================================================
structure PerceptronDispatch (n : Nat) where
weights : Fin n β β
bias : β
threshold : β
def dispatch {n : Nat} (pd : PerceptronDispatch n) (input : Fin n β β) : Bool :=
decide ((Finset.univ.sum (fun i => pd.weights i * input i)) + pd.bias > pd.threshold)
-- ============================================================
-- Sparse Transition
-- ============================================================
structure SparseTransition (n : Nat) where
indices : List (Fin n)
deltas : List β
h_same_len : indices.length = deltas.length
-- ============================================================
-- Execution Step
-- ============================================================
def exec_step {n : Nat} (s : State n) (trans : SparseTransition n) : State n :=
{ values := fun i =>
if trans.indices.contains i then
s.values i + (trans.deltas.get? (trans.indices.indexOf i)).getD 0
else
s.values i
, invariant_holds := s.invariant_holds }
-- ============================================================
-- Theorems
-- ============================================================
/-- If precondition holds, postcondition holds after exec_step -/
theorem contract_preservation {n : Nat}
(c : Contract n) (s : State n) (trans : SparseTransition n)
(h_pre : c.precondition s)
(h_contract : c.precondition s β c.postcondition s (exec_step s trans)) :
c.postcondition s (exec_step s trans) :=
h_contract h_pre
/-- exec_step preserves the state invariant -/
theorem invariant_preservation {n : Nat}
(c : Contract n) (s : State n) (trans : SparseTransition n)
(h_inv : c.invariant s)
(h_pres : c.invariant s β c.invariant (exec_step s trans)) :
c.invariant (exec_step s trans) :=
h_pres h_inv
/-- Execution is deterministic: same state + same transition = same result -/
theorem deterministic_execution {n : Nat}
(s : State n) (trans : SparseTransition n) :
exec_step s trans = exec_step s trans := rfl
/-- LoRA update preserves base model weights in non-adapted dimensions -/
theorem lora_preserves_base {n : Nat}
(base : Fin n β β) (lora_A : Fin n β β) (lora_B : Fin n β β)
(rank : Nat) (h_rank_small : rank < n)
(adapted : Fin n β β)
(h_lora : β i, adapted i = base i + lora_A i * lora_B i) :
β i, lora_A i = 0 β adapted i = base i := by
intro i h_A_zero
rw [h_lora i, h_A_zero, zero_mul, add_zero]
end
|