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": 76856.3238130518,
-      "required_pages_per_second_upper_estimate": 0.00531709881262742
+      "required_token_positions_per_second": 76169.34117638374,
+      "required_pages_per_second_upper_estimate": 0.005269571759803394
     },
     {
       "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": 9607.040476631475,
-      "required_pages_per_second_upper_estimate": 0.004022267229747434
+      "required_token_positions_per_second": 9521.167647047967,
+      "required_pages_per_second_upper_estimate": 0.003986314069229416
     },
     {
       "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": 19214.08095326295,
-      "required_pages_per_second_upper_estimate": 0.004001786353175205
+      "required_token_positions_per_second": 19042.335294095934,
+      "required_pages_per_second_upper_estimate": 0.0039660162616083254
     },
     {
       "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": 28821.12142989443,
-      "required_pages_per_second_upper_estimate": 0.0017277822224178532
+      "required_token_positions_per_second": 28563.502941143903,
+      "required_pages_per_second_upper_estimate": 0.001712338387377915
     },
     {
       "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": 28821.12142989443,
-      "required_pages_per_second_upper_estimate": 0.002240715420458664
+      "required_token_positions_per_second": 28563.502941143903,
+      "required_pages_per_second_upper_estimate": 0.002220686716102288
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28821.12142989443,
-      "required_pages_per_second_upper_estimate": 0.36968639999351505
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28563.502941143903,
+      "required_pages_per_second_upper_estimate": 0.3663819466290056
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07195431516400674,
-    "producer_p95_seconds": 0.20641076499850897,
-    "gpu_update_seconds": 0.9592964682950325,
-    "producer_batches_per_second": 13.89770714543919,
-    "gpu_batches_per_second": 1.0424306072733807,
-    "producer_headroom": 13.332021382018452,
+    "producer_batch_seconds": 0.057508882671982064,
+    "producer_p95_seconds": 0.17832049400021788,
+    "gpu_update_seconds": 0.9679485060697798,
+    "producer_batches_per_second": 17.388618132328855,
+    "gpu_batches_per_second": 1.0331128089244757,
+    "producer_headroom": 16.831286943805598,
     "sustainable": true,
-    "buffer_seconds": 60.43567750258705,
+    "buffer_seconds": 60.01280737632634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320072.301174809,
-    "upload_bytes_per_second": 5429452.783823526,
+    "generation_bytes_per_second": 319678.7584812908,
+    "upload_bytes_per_second": 5341106.381350947,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.6800649302788,
+    "estimated_full_shards_per_hour": 68.59561982945483,
     "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": [
@@ -412,6 +412,12 @@
       "bytes": 6116466,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.993601779000528,
+      "bytes": 1093689,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 768200, "loss": 0.5287339091300964, "lr": 3e-05, "grad_norm": 0.2479985593271508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112441856491, "tokens_per_second": 151739.89184902445, "data_wait_s": 1.1863001418532804e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768300, "loss": 0.5232650637626648, "lr": 3e-05, "grad_norm": 0.24733559985406833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112456519879, "tokens_per_second": 152929.25510498046, "data_wait_s": 1.6621001122985035e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768400, "loss": 0.5567591190338135, "lr": 3e-05, "grad_norm": 0.25034149847708914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112471228766, "tokens_per_second": 152735.4847341141, "data_wait_s": 1.2127000445616432e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 768500, "loss": 0.5103076100349426, "lr": 3e-05, "grad_norm": 0.25224451025330713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112485802586, "tokens_per_second": 151647.1536539731, "data_wait_s": 1.4532000932376832e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41b5052a9c71a91c637078ac3bd1e679af1aba1de7f210f9cee8e3974f3f1add
-size 178860
+oid sha256:af832f3aa51f4f7b563cc8a837e7a97b9cdf89481e92cb98a1405fd563583fd2
+size 180978