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": 77082.11287858774,
-      "required_pages_per_second_upper_estimate": 0.005332719424084008
+      "required_token_positions_per_second": 77094.57056189376,
+      "required_pages_per_second_upper_estimate": 0.005333581275520927
     },
     {
       "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": 9635.264109823467,
-      "required_pages_per_second_upper_estimate": 0.004034083875588439
+      "required_token_positions_per_second": 9636.82132023672,
+      "required_pages_per_second_upper_estimate": 0.004034735847070218
     },
     {
       "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": 19270.528219646934,
-      "required_pages_per_second_upper_estimate": 0.004013542830148468
+      "required_token_positions_per_second": 19273.64264047344,
+      "required_pages_per_second_upper_estimate": 0.004014191481873829
     },
     {
       "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": 28905.792329470405,
-      "required_pages_per_second_upper_estimate": 0.0017328581135624551
+      "required_token_positions_per_second": 28910.46396071016,
+      "required_pages_per_second_upper_estimate": 0.0017331381707221104
     },
     {
       "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": 28905.792329470405,
-      "required_pages_per_second_upper_estimate": 0.002247298210472768
+      "required_token_positions_per_second": 28910.46396071016,
+      "required_pages_per_second_upper_estimate": 0.002247661409253325
     },
     {
       "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": 28905.792329470405,
-      "required_pages_per_second_upper_estimate": 0.3707724673807469
+      "required_token_positions_per_second": 28910.46396071016,
+      "required_pages_per_second_upper_estimate": 0.3708323900502836
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2006315053281611,
-    "producer_p95_seconds": 0.20145634899927245,
-    "gpu_update_seconds": 0.9564864953316625,
-    "producer_batches_per_second": 4.984262059761547,
-    "gpu_batches_per_second": 1.0454930674721645,
-    "producer_headroom": 4.767379349355895,
+    "producer_batch_seconds": 0.0846117359921692,
+    "producer_p95_seconds": 0.19852954000089085,
+    "gpu_update_seconds": 0.9563319370306242,
+    "producer_batches_per_second": 11.818691441250536,
+    "gpu_batches_per_second": 1.0456620356159636,
+    "producer_headroom": 11.302592079178385,
     "sustainable": true,
-    "buffer_seconds": 60.25864920589474,
+    "buffer_seconds": 60.248912032929326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 261615.86372824135,
+    "generation_bytes_per_second": 263111.2863306049,
     "upload_bytes_per_second": 5170654.754032765,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.13667425046378,
+    "estimated_full_shards_per_hour": 56.45755712927447,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -117,3 +117,4 @@
 {"step": 656700, "loss": 0.540803074836731, "lr": 0.0001, "grad_norm": 0.18111740985091815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96123322377, "tokens_per_second": 138456.61213399825, "data_wait_s": 8.408998837694526e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 656800, "loss": 0.5317116379737854, "lr": 0.0001, "grad_norm": 0.1801931357084602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96137897915, "tokens_per_second": 152325.8180540724, "data_wait_s": 1.5690000509493984e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 656900, "loss": 0.5579832792282104, "lr": 0.0001, "grad_norm": 0.1867220756938115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96152450838, "tokens_per_second": 152308.63615351458, "data_wait_s": 8.926999726099893e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 657000, "loss": 0.5376891493797302, "lr": 0.0001, "grad_norm": 0.19578492926159133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96167131594, "tokens_per_second": 153361.8409411702, "data_wait_s": 1.2873000741819851e-05, "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 sha256:7e95c61d761f8fd0a0948e7a59406b19a9207ea7356ac9e1dba07f6ef9248e51
-size 253773
+oid sha256:cfe57521e22f0cd35d31e376fc030da9cfd55d6d7e559dddfc7066c3e02327d5
+size 255891