compile cache model card update

compile_cache_model_card.md

@@ -1,6 +1,6 @@
-# Compile Cache — MI300X
+# Compile Cache — MI300X
 
-Generated: 2026-04-12T04:49:32Z  (by repack_and_upload_compile_cache.sh)
+Generated: 2026-04-30T13:04:59Z
 
 ## Purpose
 Records the exact configuration under which the compile cache (inductor + Triton) was built.
@@ -8,61 +8,56 @@ If unexpected AUTOTUNE blocks appear on a later run, compare the current environ
 against these values to pinpoint the cache-key mismatch.
 
 ## Hardware
-- GPU: AMD Radeon Graphics
-- GPU count: 1
+- GPU: AMD Instinct MI300X VF
+- GPU count (world_size): 1 (1)
 - Architecture: gfx942 (AMD MI300X)
 
 ## Software Versions
 | Package | Version |
 |---------|---------|
-| Torch      | 2.6.0+rocm6.2.4 |
+| Torch   | 2.6.0+rocm6.2.4 |
 | HIP / ROCm | 6.2.41134-65d174c3e |
-| Triton     | 3.1.0 |
+| Triton  | 3.1.0 |
 | ck4inductor | 0.0.11 |
-| Python     | 3.10.13 |
+| Python  | 3.10.12 |
 
 ## Compile Configuration
 | Setting | Value |
 |---------|-------|
 | Compile backend | inductor |
 | Compile mode | max-autotune-no-cudagraphs |
-| GEMM backends | ATEN,TRITON,CK |
-| TORCHINDUCTOR\_CK\_DIR | `N/A` |
-| TORCHINDUCTOR\_CACHE\_DIR | `/workspace/inductor_cache` |
-| TRITON\_CACHE\_DIR | `/workspace/triton_cache` |
-| TRITON\_MAX\_BLOCK\_SIZE | 64 |
-| TORCHINDUCTOR\_COMPILE\_THREADS | 4 |
+| GEMM backends | CK,ATEN,TRITON |
+| TORCHINDUCTOR\_CK\_DIR | `/home/hotaisle/.local/lib/python3.10/site-packages/ck4inductor` |
+| TORCHINDUCTOR\_CACHE\_DIR | `/tmp/ml_cache/inductor_cache` |
+| TRITON\_CACHE\_DIR | `/tmp/ml_cache/triton_cache` |
+| TRITON\_MAX\_BLOCK\_SIZE | 32 |
+| TORCHINDUCTOR\_COMPILE\_THREADS | 40 |
 
 ## Model
 - Architecture: EnhancedDSRModel (~1.743B total params)
-- Trainable params: fp32 (~337M — new alignment modules)
-- Frozen backbone: bf16 (~1.406B — RWKV-7 1.5B World)
+- Trainable params: fp32 (~337M — new alignment modules)
+- Frozen backbone: bf16 (~1.406B — RWKV-7 1.5B World)
 - Sequence length: 1024 tokens
 
-## Cache Statistics (at time of pack)
+## Cache Statistics
 | Cache | Files |
 |-------|-------|
-| inductor\_cache | 7654 |
-| triton\_cache | 59993 |
-| tar.gz size | 443 MB |
-
-## HuggingFace
-- Repo: `RaninduH/torchCompiledModels`
-- File: `my_compile_cache.tar.gz`
-- Model card: `compile_cache_model_card.md`
+| inductor\_cache | 995 |
+| triton\_cache | 7900 |
+| tar.gz size | 110.3 MB |
 
 ## Autotune Cache Key
 The inductor autotune winner cache key is a **hash of the full candidate set**
 including which backends are enabled. Adding/removing from
-`TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_BACKENDS` (currently `ATEN,TRITON,CK`) will
+`TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_BACKENDS` (currently `CK,ATEN,TRITON`) will
 invalidate all cached winners and force a full AUTOTUNE re-run.
 
 ## Recompilation Trigger Checklist
 If AUTOTUNE blocks appear on a pod that should have cache hits, verify:
-1. `TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_BACKENDS` == `ATEN,TRITON,CK`
-2. `TORCHINDUCTOR_CK_DIR` points to valid CK root: `N/A`
+1. `TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_BACKENDS` == `CK,ATEN,TRITON`
+2. `TORCHINDUCTOR_CK_DIR` points to valid CK root: `/home/hotaisle/.local/lib/python3.10/site-packages/ck4inductor`
 3. `library/src` symlink inside ck4inductor package exists (needed for gemm_universal kernel enumeration)
-4. `TORCHINDUCTOR_CACHE_DIR` == `/workspace/inductor_cache` (with extracted cache files present)
-5. `TRITON_CACHE_DIR` == `/workspace/triton_cache` (must NOT be `~/.triton` — that is wiped on restart)
-6. Torch version matches: `2.6.0+rocm6.2.4` — different minor/patch versions can invalidate kernel hashes
+4. `TORCHINDUCTOR_CACHE_DIR` == `/tmp/ml_cache/inductor_cache` (with extracted cache files present)
+5. `TRITON_CACHE_DIR` == `/tmp/ml_cache/triton_cache` (must NOT be `~/.triton` — that is wiped on restart)
+6. Torch version matches: `2.6.0+rocm6.2.4` — different minor/patch versions can invalidate kernel hashes
 7. Triton version matches: `3.1.0`