Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75500.42899566473,
-      "required_pages_per_second_upper_estimate": 0.006544185831256373
+      "required_token_positions_per_second": 75199.96620807982,
+      "required_pages_per_second_upper_estimate": 0.006518142478344486
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9437.55362445809,
-      "required_pages_per_second_upper_estimate": 0.0037647812447973873
+      "required_token_positions_per_second": 9399.995776009977,
+      "required_pages_per_second_upper_estimate": 0.0037497988575115595
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18875.10724891618,
-      "required_pages_per_second_upper_estimate": 0.004366423871494893
+      "required_token_positions_per_second": 18799.991552019954,
+      "required_pages_per_second_upper_estimate": 0.004349047177009065
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28312.660873374272,
-      "required_pages_per_second_upper_estimate": 0.001724092502137057
+      "required_token_positions_per_second": 28199.98732802993,
+      "required_pages_per_second_upper_estimate": 0.0017172312743779924
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28312.660873374272,
-      "required_pages_per_second_upper_estimate": 0.0021795870084938378
+      "required_token_positions_per_second": 28199.98732802993,
+      "required_pages_per_second_upper_estimate": 0.002170913087072894
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28312.660873374272,
-      "required_pages_per_second_upper_estimate": 0.3515484916668646
+      "required_token_positions_per_second": 28199.98732802993,
+      "required_pages_per_second_upper_estimate": 0.35014946332778635
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05600939007810268,
-    "producer_p95_seconds": 0.19215561099917977,
-    "gpu_update_seconds": 0.9765242526533657,
-    "producer_batches_per_second": 17.854149074031035,
-    "gpu_batches_per_second": 1.0240401068205394,
-    "producer_headroom": 17.43500958127994,
+    "producer_batch_seconds": 0.3343946196951606,
+    "producer_p95_seconds": 0.206454899000164,
+    "gpu_update_seconds": 0.9804259724797368,
+    "producer_batches_per_second": 2.9904787371029347,
+    "gpu_batches_per_second": 1.019964819445527,
+    "producer_headroom": 2.93194302400412,
     "sustainable": true,
-    "buffer_seconds": 60.544503664508674,
+    "buffer_seconds": 60.786410293743685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469940.4817917097,
-    "upload_bytes_per_second": 5964602.953666395,
+    "generation_bytes_per_second": 522015.1031728014,
+    "upload_bytes_per_second": 6077640.394827299,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.83828773797482,
+    "estimated_full_shards_per_hour": 112.01228925121337,
     "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": 6037950,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.402298138000333,
+      "bytes": 13311379,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -94,3 +94,4 @@
 {"step": 124400, "loss": 0.619434978812933, "lr": 0.0003, "grad_norm": 0.14032894751096783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18204768422, "tokens_per_second": 150585.28200775647, "data_wait_s": 1.2665001122513786e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 124500, "loss": 0.5670794546604156, "lr": 0.0003, "grad_norm": 0.13992898670251436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18219316192, "tokens_per_second": 149352.98389062865, "data_wait_s": 8.661998435854912e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 124600, "loss": 0.5929880812764168, "lr": 0.0003, "grad_norm": 0.13171493037253057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18233878200, "tokens_per_second": 148533.25672023543, "data_wait_s": 1.3689001207239926e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 124700, "loss": 0.6310654245316982, "lr": 0.0003, "grad_norm": 0.14385225200419247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18248395216, "tokens_per_second": 148602.9556673551, "data_wait_s": 1.4315999578684568e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1b69b383b5c57871fe8a7f9a9ae056094425d6e221bd89f09d6f2d16c71bc01
-size 204276
+oid sha256:b106bcc5e5c1e2c87842c4e6c1a712e74219eaee6e255ebb6f18a93d3d74bcef
+size 206394