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": 77333.20081259635,
-      "required_pages_per_second_upper_estimate": 0.005350090269962997
+      "required_token_positions_per_second": 77341.59240807738,
+      "required_pages_per_second_upper_estimate": 0.0053506708200095595
     },
     {
       "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": 9666.650101574543,
-      "required_pages_per_second_upper_estimate": 0.004047224534920578
+      "required_token_positions_per_second": 9667.699051009673,
+      "required_pages_per_second_upper_estimate": 0.004047663708144512
     },
     {
       "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": 19333.300203149087,
-      "required_pages_per_second_upper_estimate": 0.004026616578903441
+      "required_token_positions_per_second": 19335.398102019346,
+      "required_pages_per_second_upper_estimate": 0.004027053515912802
     },
     {
       "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": 28999.95030472363,
-      "required_pages_per_second_upper_estimate": 0.0017385027404079822
+      "required_token_positions_per_second": 29003.097153029023,
+      "required_pages_per_second_upper_estimate": 0.0017386913891589315
     },
     {
       "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": 28999.95030472363,
-      "required_pages_per_second_upper_estimate": 0.0022546185788915427
+      "required_token_positions_per_second": 29003.097153029023,
+      "required_pages_per_second_upper_estimate": 0.0022548632325057644
     },
     {
       "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": 28999.95030472363,
-      "required_pages_per_second_upper_estimate": 0.37198022478833814
+      "required_token_positions_per_second": 29003.097153029023,
+      "required_pages_per_second_upper_estimate": 0.37202058917957725
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06636823122657631,
-    "producer_p95_seconds": 0.15952694800034806,
-    "gpu_update_seconds": 0.9533809440872243,
-    "producer_batches_per_second": 15.067449915699466,
-    "gpu_batches_per_second": 1.0488986655354322,
-    "producer_headroom": 14.365019625616526,
+    "producer_batch_seconds": 0.07849958746093222,
+    "producer_p95_seconds": 0.15842824399987876,
+    "gpu_update_seconds": 0.9532775018516428,
+    "producer_batches_per_second": 12.73892044971168,
+    "gpu_batches_per_second": 1.0490124838335149,
+    "producer_headroom": 12.143726262587956,
     "sustainable": true,
-    "buffer_seconds": 60.06299947749513,
+    "buffer_seconds": 60.0564826166535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424963.2820315261,
+    "generation_bytes_per_second": 424773.9512463719,
     "upload_bytes_per_second": 3349786.4763573543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.18722768506372,
+    "estimated_full_shards_per_hour": 91.14660170596474,
     "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": [

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -42,3 +42,4 @@
 {"step": 529200, "loss": 0.5764166116714478, "lr": 0.0003, "grad_norm": 0.16514451899909866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77457212289, "tokens_per_second": 151970.9354746042, "data_wait_s": 1.5268999959516805e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 529300, "loss": 0.6114671230316162, "lr": 0.0003, "grad_norm": 0.15709882034165365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77472027697, "tokens_per_second": 154359.07024890062, "data_wait_s": 1.602200063643977e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 529400, "loss": 0.5777540802955627, "lr": 0.0003, "grad_norm": 0.18941419501930984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77486755452, "tokens_per_second": 153461.67449992851, "data_wait_s": 1.189700014947448e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 529500, "loss": 0.5765049457550049, "lr": 0.0003, "grad_norm": 0.1820989624634915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77501506326, "tokens_per_second": 153654.1224225014, "data_wait_s": 1.3766999472863972e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e59fdfaebd81a68a012a1053581c7678d1078063eb7957560c0f5efa408c647
-size 93357
+oid sha256:55912f6a1e609f2e11c4795dc87f4efc5fdc40fab3555b784e4d084caa1135d1
+size 95475