14 packages match "determinism"
- total 15
- drwxr-xr-x-../
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-logits/# 0.1.0 GPT-2's logits helpers: the greedy pick (first largest, FTZ compare) and the FNV-1a 64 digest over every step's logits
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-norm/# 0.1.0 GPT-2's layer normalization of a row: 256-thread strided sums, then the block tree, in Z
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-weights/# 0.1.0 GPT-2 124M synthetic weights from a counter hash: kcheck's checks without a checkpoint
- drwxr-xr-x-Oct 27 01:59@ajax/pathtrace-z/# 0.1.0 programs/pathtrace's renderer instantiated in the cluster float mode Z (every op flushed): pt_*_z, the CPU and wasm reference
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-kv/# 0.1.0 GPT-2's KV cache: per layer, K and V rows of the full width for each position, and its digest
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-shape/# 0.1.0 GPT-2 124M's shape: layers, heads, widths, vocabulary, context, and the flat weight buffer's layout in checkpoint order
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-linear/# 0.1.0 GPT-2's linear layer, one output row: sequential dots over k from +0 in Z, then bias, bias-GELU or a residual add
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-scalar/# 0.1.0 GPT-2's scalar functions from + - * / in the cluster float mode: exp by a degree-10 polynomial, tanh, the tanh GELU, Newton sqrt
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2-attend/# 0.1.0 GPT-2's attention for one head and query: scaled dots, block-tree max and sum, softmax, the 4-group value sum, in Z
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-matmul/# 0.1.0 programs/matmul's exact mode on the CPU: counter-hash matrices, C = A B and y = A x in the GPU's summation order, the FNV digest
- drwxr-xr-x-Oct 27 01:59@ajax/pathtrace-math/# 0.1.0 programs/pathtrace's float contract and its math, written once over a float's operations: vectors, Newton sqrt, polynomial cis, constants
- drwxr-xr-x-Oct 27 04:27@ajax/wasm-float-guard/# 0.1.0 meshd's load-time float guard for wasm CPU programs: f32 arithmetic only in exported mesh_ftz_* helpers, no threads, no relaxed SIMD, only mesh imports
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ref-gpt2/# 0.1.0 GPT-2 124M in the cluster float mode on the CPU: the forward pass, prefill and decode, greedy generation, bitwise equal to the GPU kernels
- drwxr-xr-x-Oct 27 01:59@ajax/mesh-ftz/# 0.1.0 The cluster float mode Z: f32 arithmetic with subnormal inputs and results flushed to signed zero, as the PTX .ftz ops