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": 77626.64711175101,
-      "required_pages_per_second_upper_estimate": 0.005370391565827739
+      "required_token_positions_per_second": 76985.55120453195,
+      "required_pages_per_second_upper_estimate": 0.005326039063418876
     },
     {
       "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": 9703.330888968876,
-      "required_pages_per_second_upper_estimate": 0.004062582014620642
+      "required_token_positions_per_second": 9623.193900566494,
+      "required_pages_per_second_upper_estimate": 0.0040290303310271764
     },
     {
       "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": 19406.661777937752,
-      "required_pages_per_second_upper_estimate": 0.0040418958602570915
+      "required_token_positions_per_second": 19246.387801132987,
+      "required_pages_per_second_upper_estimate": 0.004008515017597661
     },
     {
       "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": 29109.99266690663,
-      "required_pages_per_second_upper_estimate": 0.001745099611995895
+      "required_token_positions_per_second": 28869.581701699477,
+      "required_pages_per_second_upper_estimate": 0.001730687341718015
     },
     {
       "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": 29109.99266690663,
-      "required_pages_per_second_upper_estimate": 0.0022631738885260733
+      "required_token_positions_per_second": 28869.581701699477,
+      "required_pages_per_second_upper_estimate": 0.0022444829934372976
     },
     {
       "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": 29109.99266690663,
-      "required_pages_per_second_upper_estimate": 0.3733917300561387
+      "required_token_positions_per_second": 28869.581701699477,
+      "required_pages_per_second_upper_estimate": 0.3703079963276443
     }
   ],
   "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": 0.10824864397650913,
-    "producer_p95_seconds": 0.3361183050001273,
-    "gpu_update_seconds": 0.9497769482927875,
-    "producer_batches_per_second": 9.237991010926738,
-    "gpu_batches_per_second": 1.0528787856954076,
-    "producer_headroom": 8.774030910714199,
+    "producer_batch_seconds": 0.0902693556717793,
+    "producer_p95_seconds": 0.3424191880003491,
+    "gpu_update_seconds": 0.9576862001562161,
+    "producer_batches_per_second": 11.077956550792438,
+    "gpu_batches_per_second": 1.044183365946885,
+    "producer_headroom": 10.609206114624072,
     "sustainable": true,
-    "buffer_seconds": 60.7857246907384,
+    "buffer_seconds": 60.33423060984161,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309516.5107087563,
-    "upload_bytes_per_second": 5532592.155983226,
+    "generation_bytes_per_second": 276597.8815827699,
+    "upload_bytes_per_second": 5592785.726130952,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.4150380224897,
+    "estimated_full_shards_per_hour": 59.35146651851962,
     "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": 1.0161830090000876,
-      "bytes": 1229232,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.4130291219989886,
       "bytes": 17178237,
@@ -562,6 +556,12 @@
       "bytes": 2384518,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.027337190000253,
+      "bytes": 5987324,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 678000, "loss": 0.5482680797576904, "lr": 0.0001, "grad_norm": 0.19308343636945574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99241725668, "tokens_per_second": 152805.12796581432, "data_wait_s": 8.801998774288222e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 678100, "loss": 0.5835674405097961, "lr": 0.0001, "grad_norm": 0.22842088470374425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99256445591, "tokens_per_second": 154111.1735606397, "data_wait_s": 1.3627999578602612e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 678200, "loss": 0.5442383885383606, "lr": 0.0001, "grad_norm": 0.21608107373354424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99271132121, "tokens_per_second": 143973.4307800373, "data_wait_s": 8.634000550955534e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 678300, "loss": 0.5567049384117126, "lr": 0.0001, "grad_norm": 0.19036875865297515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99285865302, "tokens_per_second": 152639.1467377215, "data_wait_s": 9.129998943535611e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98bdb2be5cb7f9d81a3b3f7fd3edf3a2aa16f33896be6b78533414acf8369f54
-size 708039
+oid sha256:c2a1f3921f3bf4cca949a51e3a66b447107ad3805242da97004fbaeedc74ac51
+size 710157