berkus
/
ANE
mirror of https://github.com/maderix/ANE.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
41 Commits 3 Branches 0 Tags 1.1 MiB
Commit Graph

4 Commits

Author SHA1 Message Date
log-wade 9fbd4dff5b fix: guard short token datasets in train_large_ane and dynamic pipeline 
- Add n_tokens <= SEQ+1 check in train_large_ane.m and training_dynamic/train.m
- Prevents underflow in max_pos and possible OOB reads (aligns with train_large.m)
- Add M5 MacBook Pro benchmark result and full output for Issue #3

Made-with: Cursor
 2026-03-07 14:12:31 -06:00
maderix efcf193075 Add model config to benchmark report, update README with current results 
Benchmark report now includes full Stories110M model configuration
(arch, layers, dims, kernels). README updated: 12-layer results
replace stale single-layer numbers, limitations reflect current state.
 2026-03-04 06:13:21 -08:00
maderix 1a7d8846b2 Add NE core counts, clarify FP16 vs rated TOPS methodology 
All chips have 16 NE cores except Ultra (32 via UltraFusion).
M4 38 TOPS is INT8/mixed-precision, not comparable to M3 FP16 spec.
 2026-03-04 06:11:29 -08:00
maderix 050bc4fdf0 Add cross-generation ANE benchmark report from issue #3 
Community-submitted results for M1 Pro/Max, M3 Pro, M4 Pro/Max, M5.
Includes training performance, peak throughput, MIL compatibility
matrix, and structured JSON data.
 2026-03-04 05:30:00 -08:00