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": 76611.62499571896,
-      "required_pages_per_second_upper_estimate": 0.005300169980664853
+      "required_token_positions_per_second": 77339.97337202121,
+      "required_pages_per_second_upper_estimate": 0.005350558811338513
     },
     {
       "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": 9576.45312446487,
-      "required_pages_per_second_upper_estimate": 0.0040094609441318225
+      "required_token_positions_per_second": 9667.496671502651,
+      "required_pages_per_second_upper_estimate": 0.004047578976071086
     },
     {
       "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": 19152.90624892974,
-      "required_pages_per_second_upper_estimate": 0.0039890452755480385
+      "required_token_positions_per_second": 19334.993343005302,
+      "required_pages_per_second_upper_estimate": 0.00402696921528439
     },
     {
       "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": 28729.35937339461,
-      "required_pages_per_second_upper_estimate": 0.0017222812272432366
+      "required_token_positions_per_second": 29002.490014507952,
+      "required_pages_per_second_upper_estimate": 0.0017386549921316401
     },
     {
       "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": 28729.35937339461,
-      "required_pages_per_second_upper_estimate": 0.0022335813241844337
+      "required_token_positions_per_second": 29002.490014507952,
+      "required_pages_per_second_upper_estimate": 0.0022548160301562713
     },
     {
       "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": 28729.35937339461,
-      "required_pages_per_second_upper_estimate": 0.36850937485915536
+      "required_token_positions_per_second": 29002.490014507952,
+      "required_pages_per_second_upper_estimate": 0.3720128014585235
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5949651994139913,
-    "producer_p95_seconds": 0.26525801600291743,
-    "gpu_update_seconds": 0.9623604773312131,
-    "producer_batches_per_second": 1.6807705744553567,
-    "gpu_batches_per_second": 1.0391116671511362,
-    "producer_headroom": 1.6175071723171144,
+    "producer_batch_seconds": 0.07659373584360196,
+    "producer_p95_seconds": 0.26923336899926653,
+    "gpu_update_seconds": 0.9532974577758533,
+    "producer_batches_per_second": 13.055897965884792,
+    "gpu_batches_per_second": 1.0489905242515896,
+    "producer_headroom": 12.446154339858907,
     "sustainable": true,
-    "buffer_seconds": 60.62871007186642,
+    "buffer_seconds": 60.05773983987876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 223370.9084710889,
-    "upload_bytes_per_second": 4588906.080014869,
+    "generation_bytes_per_second": 223182.65970923155,
+    "upload_bytes_per_second": 4690420.650504971,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.93019714927196,
+    "estimated_full_shards_per_hour": 47.88980334718428,
     "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": 0.9132081499992637,
-      "bytes": 2381673,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0147103390008851,
       "bytes": 6222923,
@@ -562,6 +556,12 @@
       "bytes": 6202154,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2203903480003646,
+      "bytes": 12408575,
+      "pages": 47,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -280,3 +280,4 @@
 {"step": 713000, "loss": 0.5104162096977234, "lr": 8e-05, "grad_norm": 0.21975198869218812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104363301640, "tokens_per_second": 150502.55223819133, "data_wait_s": 1.6524998500244692e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713100, "loss": 0.5497522950172424, "lr": 8e-05, "grad_norm": 0.21310864575031135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104377844279, "tokens_per_second": 144989.59008227877, "data_wait_s": 1.6006997611839324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713200, "loss": 0.5147819519042969, "lr": 8e-05, "grad_norm": 0.23424810404783758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104392597617, "tokens_per_second": 153461.35665833578, "data_wait_s": 1.5017998521216214e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 713300, "loss": 0.5463555455207825, "lr": 8e-05, "grad_norm": 0.21938065296435114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104407196969, "tokens_per_second": 151590.80574856538, "data_wait_s": 1.593500201124698e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1448787d117d2f42ba5886965ac5361c5a335361e1a5ac2e88db0a5fa9106c3f
-size 601356
+oid sha256:9cd7cfad4fb19b122d888f89ee030e4b747e720a33fb2b97e48d8e24d87d9693
+size 603474