| --- |
| license: other |
| license_name: sovereign-source-license-v2 |
| library_name: custom |
| tags: |
| - code |
| - sovereign-compute |
| --- |
| |
| # Assembly Bite β ML Macromodel Corpus |
|
|
| [](LICENSE) |
| [](LICENSE) |
| []() |
| []() |
| []() |
| []() |
|
|
| **Author:** Ahmad Ali Parr |
| **Trust:** Bel Esprit D'Accord Irrevocable Trust Β· EIN 42-697643 |
|
|
| > Low-level pseudo-assembly language for representing ML models at the instruction level. Token matchers, tree matchers, full transformer macromodels with multiplicity β plus production SASS/PTX kernels for sm_89 (RTX 4090). |
| |
| --- |
| |
| ## What Is Assembly Bite |
| |
| Assembly Bite is Ahmad's custom pseudo-assembly language for expressing ML model computations at the byte-code level. Four instruction types: |
| |
| - `.DATA` β memory layout declarations (`.word`, `.repl`, `.float`) |
| - `.CODE` β instruction stream (`LOAD`, `STORE`, `CALL`, `MATMUL`, `ADD`, `SUB`, `CMP`, `JEQ`, etc.) |
| - Subroutine calls for primitive operations (`MATMUL`, `SOFTMAX`, `RELU`, `LAYER_NORM`, `ADD_BIAS`) |
| - No invented syntax β grounded in standard CS algorithms |
| |
| --- |
| |
| ## Contents |
| |
| ``` |
| examples/ |
| token-matcher/token_matcher.asm β Literal token ID sequence matcher (sliding window) |
| tree-matcher/tree_matcher.asm β DFS path matcher on [id, child, sibling] trees |
| transformer/transformer_macromodel.asm β Full transformer: 24L Γ 12H Γ 768D Γ M=4 |
|
|
| python/ |
| deberta_encoder.py β DeBERTa-v3 encoder wrapper + instruction token |
| gguf_dag_pipeline.py β GGUF load + networkx DAG pipeline |
| |
| sass/ |
| flash_attention.ptx β Flash attention paged (sm_89 Hopper) |
| dequant_q4k.ptx β GGUF Q4_K dequant PTX (sm_89) |
| dequant_q4k.sass β GGUF Q4_K dequant SASS |
| cuda_kernels.c β Host launch wrappers |
| mamba_bind.h β Mamba SSM + GGUF binding header |
| qemu_arm64_holyc.HC β HolyC QEMU ARM64 integration |
| ``` |
| |
| --- |
| |
| ## Transformer Macromodel β Multiplicity Architecture |
| |
| The key insight in `transformer_macromodel.asm`: each neuron block has **M copies** (default M=4). The outer loop is `LΓAΓM` β layer Γ head Γ multiplicity. Each copy computes Q/K/V independently. Results are summed across copies before the next layer. |
| |
| ``` |
| Parameters: L=24, A=12, D=768, H=3072, M=4 |
| Weight layout: W_Q[L][A][M][D][D] = 24Γ12Γ4Γ768Γ768 |
| Per copy: full attention + FFN + residual + layer norm |
| Aggregate: LAYER_ACC += LAYER_OUT for each M copy |
| ``` |
| |
| This is distinct from standard multi-head attention β it's multiplicity within each head, not across heads. |
| |
| --- |
| |
| ## SASS/PTX Kernels β sm_89 (RTX 4090) |
|
|
| ### flash_attention.ptx |
| Full paged flash attention with TMA async copy and WMMA tensor core tiles. |
| |
| ### dequant_q4k.ptx / dequant_q4k.sass |
| GGUF Q4_K block dequantization. 32-value blocks β FP16. Each thread processes 8 values (4 packed bytes). 36 bytes per block layout: `[32 bytes packed][2 bytes scale][2 bytes min]`. |
|
|
| ``` |
| Grid: ceil(num_blocks / 256) blocks |
| Block: 256 threads |
| Dequant: q_val * scale + min β FP16 |
| ``` |
|
|
| --- |
|
|
| ## Build |
|
|
| ```bash |
| # PTX β Cubin (requires CUDA 12.x + SM89 GPU) |
| ptxas -arch=sm_89 sass/dequant_q4k.ptx -o dequant_q4k.cubin |
| ptxas -arch=sm_89 sass/flash_attention.ptx -o flash_attention.cubin |
| |
| # Host wrappers |
| nvcc -arch=sm_89 sass/cuda_kernels.c -o kernels.so |
| |
| # Python |
| pip install torch transformers llama_cpp_python networkx |
| python python/deberta_encoder.py |
| ``` |
|
|
| --- |
|
|
| Β© 2026 Bel Esprit D'Accord Irrevocable Trust Β· Patent Pending Β· ΞΈ = 89/2462 |
|
|
