Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77444.33452203314,
-      "required_pages_per_second_upper_estimate": 0.005357778757847567
+      "required_token_positions_per_second": 76744.43891099581,
+      "required_pages_per_second_upper_estimate": 0.005309358355494447
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9680.541815254142,
-      "required_pages_per_second_upper_estimate": 0.004053040705346257
+      "required_token_positions_per_second": 9593.054863874477,
+      "required_pages_per_second_upper_estimate": 0.004016411745738884
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19361.083630508285,
-      "required_pages_per_second_upper_estimate": 0.004032403134124111
+      "required_token_positions_per_second": 19186.109727748953,
+      "required_pages_per_second_upper_estimate": 0.0039959606845514604
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29041.625445762427,
-      "required_pages_per_second_upper_estimate": 0.0017410011014790023
+      "required_token_positions_per_second": 28779.164591623427,
+      "required_pages_per_second_upper_estimate": 0.0017252669740278892
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29041.625445762427,
-      "required_pages_per_second_upper_estimate": 0.0022578586377977367
+      "required_token_positions_per_second": 28779.164591623427,
+      "required_pages_per_second_upper_estimate": 0.0022374534608317212
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29041.625445762427,
-      "required_pages_per_second_upper_estimate": 0.3725147887503037
+      "required_token_positions_per_second": 28779.164591623427,
+      "required_pages_per_second_upper_estimate": 0.3691482227219177
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10950133801583206,
-    "producer_p95_seconds": 0.3251234160052263,
-    "gpu_update_seconds": 0.9520128290213943,
-    "producer_batches_per_second": 9.132308500699935,
-    "gpu_batches_per_second": 1.0504060129399027,
-    "producer_headroom": 8.694074851247471,
+    "producer_batch_seconds": 0.10509886827304626,
+    "producer_p95_seconds": 0.30829189799987944,
+    "gpu_update_seconds": 0.9606950164233513,
+    "producer_batches_per_second": 9.514850316009166,
+    "gpu_batches_per_second": 1.0409130711669354,
+    "producer_headroom": 9.140869280604155,
     "sustainable": true,
-    "buffer_seconds": 60.928821057369234,
+    "buffer_seconds": 60.52378603467113,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 518300.1898336327,
-    "upload_bytes_per_second": 5718693.409986577,
+    "generation_bytes_per_second": 467432.2061769441,
+    "upload_bytes_per_second": 5718201.65809837,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.21515532738434,
+    "estimated_full_shards_per_hour": 100.30007018071404,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.317680020998523,
-      "bytes": 10858764,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.2862951650022296,
       "bytes": 2384330,
@@ -562,6 +556,12 @@
       "bytes": 5829196,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.570511810001335,
+      "bytes": 12268120,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -362,3 +362,4 @@
 {"step": 721200, "loss": 0.5369404554367065, "lr": 8e-05, "grad_norm": 0.21757630034294323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105563713753, "tokens_per_second": 154697.9656602676, "data_wait_s": 1.6921003407333046e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721300, "loss": 0.5238915681838989, "lr": 8e-05, "grad_norm": 0.21782143119792022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105578467144, "tokens_per_second": 154161.2509054728, "data_wait_s": 1.4673001714982092e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721400, "loss": 0.5490963459014893, "lr": 8e-05, "grad_norm": 0.20821286446828496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105592944425, "tokens_per_second": 150433.45233805044, "data_wait_s": 1.0651994671206921e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 721500, "loss": 0.5709020495414734, "lr": 8e-05, "grad_norm": 0.21111324364175713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105607461004, "tokens_per_second": 151306.48656521892, "data_wait_s": 1.5398996765725315e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcbc37512124561fabbe5d4d35569cbaeadc9abca619050214177cf5bc9662c8
-size 776598
+oid sha256:aa685769f96200d8bbc10063129648a26042bd6601570eaa1412547b8483314f
+size 778716