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": 76474.16247511572,
-      "required_pages_per_second_upper_estimate": 0.00529066000453253
+      "required_token_positions_per_second": 77051.75240703656,
+      "required_pages_per_second_upper_estimate": 0.005330619016216098
     },
     {
       "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": 9559.270309389465,
-      "required_pages_per_second_upper_estimate": 0.004002266858277732
+      "required_token_positions_per_second": 9631.46905087957,
+      "required_pages_per_second_upper_estimate": 0.004032494963658473
     },
     {
       "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": 19118.54061877893,
-      "required_pages_per_second_upper_estimate": 0.003981887821070237
+      "required_token_positions_per_second": 19262.93810175914,
+      "required_pages_per_second_upper_estimate": 0.004011962008757311
     },
     {
       "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": 28677.810928168394,
-      "required_pages_per_second_upper_estimate": 0.0017191909766618432
+      "required_token_positions_per_second": 28894.40715263871,
+      "required_pages_per_second_upper_estimate": 0.0017321755895955534
     },
     {
       "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": 28677.810928168394,
-      "required_pages_per_second_upper_estimate": 0.0022295736593057436
+      "required_token_positions_per_second": 28894.40715263871,
+      "required_pages_per_second_upper_estimate": 0.0022464130630522046
     },
     {
       "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": 28677.810928168394,
-      "required_pages_per_second_upper_estimate": 0.36784816675220167
+      "required_token_positions_per_second": 28894.40715263871,
+      "required_pages_per_second_upper_estimate": 0.37062643055680033
     }
   ],
   "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": 0.1326903271954336,
-    "producer_p95_seconds": 0.324750548999873,
-    "gpu_update_seconds": 0.9640903229766092,
-    "producer_batches_per_second": 7.5363443676429025,
-    "gpu_batches_per_second": 1.0372472123903498,
-    "producer_headroom": 7.265716675463796,
+    "producer_batch_seconds": 0.26085108975797766,
+    "producer_p95_seconds": 0.31102060999910464,
+    "gpu_update_seconds": 0.9568633768447163,
+    "producer_batches_per_second": 3.8336048391740207,
+    "gpu_batches_per_second": 1.0450812772221756,
+    "producer_headroom": 3.6682360719002993,
     "sustainable": true,
-    "buffer_seconds": 60.73769034752638,
+    "buffer_seconds": 60.28239274121713,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390634.2383319517,
-    "upload_bytes_per_second": 4178450.9563841834,
+    "generation_bytes_per_second": 399204.61848926067,
+    "upload_bytes_per_second": 4235909.755661079,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.82101404637254,
+    "estimated_full_shards_per_hour": 85.66001811989179,
     "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": 1.5119859270016605,
-      "bytes": 1025135,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.464499204997992,
       "bytes": 17226254,
@@ -562,6 +556,12 @@
       "bytes": 6129871,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.2406462889994145,
+      "bytes": 5768030,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 875300, "loss": 0.5060501098632812, "lr": 3e-05, "grad_norm": 0.3257328238374927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128119024039, "tokens_per_second": 151256.37447600192, "data_wait_s": 1.0635001672198996e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 875400, "loss": 0.5454472303390503, "lr": 3e-05, "grad_norm": 0.30757069373546403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128133730324, "tokens_per_second": 152738.99208659367, "data_wait_s": 1.3038999895798042e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 875500, "loss": 0.5480207800865173, "lr": 3e-05, "grad_norm": 0.3173103615842183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128148376617, "tokens_per_second": 152324.82775209213, "data_wait_s": 1.1696996807586402e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 875600, "loss": 0.5303516387939453, "lr": 3e-05, "grad_norm": 0.3000887577362835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128163101951, "tokens_per_second": 152797.7142825479, "data_wait_s": 7.768998329993337e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa8601a178180a7363467844ab1f33142a55ae9e1a920c9dbe40a4bdfa2eaf6c
-size 704586
+oid sha256:15b59b8eb3e3b126f90c1f6e3fd44d1494d15fcbd364b0e64193e20162a997c8
+size 706704