biondizzle 9d96c2fbbf CRITICAL FIX: FP32 RoPE cache + FP32 arithmetic for inverse RoPE round-trip ...

BF16 cos/sin cache destroys cos²+sin²=1 identity (can be 0.996 in BF16).
This causes ~3% error per RoPE→inverse RoPE round-trip, accumulating
across 61 layers into garbage output. FP32 cache + FP32 arithmetic
gives exact round-trip (diff < 1e-7).

Also fixes: MoE expert loop indentation (was only running last expert).

2026-05-31 09:14:59 +00:00

..

e2e

E3: model construction test

2026-05-30 21:22:34 +00:00

integration

Restructure: cutedsl/ -> dsv4/ with proper layering

2026-05-21 17:30:44 +00:00

unit

Cleanup C1-C7: delete dead CuTeDSL FMHA, test probes, scratch files

2026-05-30 21:08:12 +00:00

check_log.sh

Add check_log.sh convenience script

2026-05-22 17:07:23 +00:00

requirements.txt

test: add standalone layer 0 comparison test (no vLLM, no Docker)

2026-05-16 02:13:18 +00:00

run_test.sh

run_test.sh: SIGKILL all children of screen session on cleanup

2026-05-22 17:08:12 +00:00

test_minimal_e2e.py

CRITICAL FIX: FP32 RoPE cache + FP32 arithmetic for inverse RoPE round-trip

2026-05-31 09:14:59 +00:00

verify_attention.py

fix verify_attention: proper multi-head SDPA + GQA

2026-05-31 05:55:10 +00:00